The long non-coding RNA Meg3 mediates imprinted gene expression during stem cell differentiation

Abstract

The imprinted Dlk1-Dio3 domain comprises the developmental genes Dlk1 and Rtl1, which are silenced on the maternal chromosome in different cell types. On this parental chromosome, the domain's imprinting control region activates a polycistron that produces the lncRNA Meg3 and many miRNAs (Mirg) and C/D-box snoRNAs (Rian). Although Meg3 lncRNA is nuclear and associates with the maternal chromosome, it is unknown whether it controls gene repression in cis. We created mouse embryonic stem cells (mESCs) that carry an ectopic poly(A) signal, reducing RNA levels along the polycistron, and generated Rian^−/− mESCs as well. Upon ESC differentiation, we found that Meg3 lncRNA (but not Rian) is required for Dlk1 repression on the maternal chromosome. Biallelic Meg3 expression acquired through CRISPR-mediated demethylation of the paternal Meg3 promoter led to biallelic Dlk1 repression, and to loss of Rtl1 expression. lncRNA expression also correlated with DNA hypomethylation and CTCF binding at the 5′-side of Meg3. Using Capture Hi-C, we found that this creates a Topologically Associating Domain (TAD) organization that brings Meg3 close to Dlk1 on the maternal chromosome. The requirement of Meg3 for gene repression and TAD structure may explain how aberrant MEG3 expression at the human DLK1-DIO3 locus associates with imprinting disorders.

Graphical Abstract

Graphical Abstract.

Introduction

Genomic imprinting is an epigenetic phenomenon that plays diverse roles in development, metabolism and behavior. It mediates the mono-allelic expression of ∼150 protein-coding genes—strictly dependent on their parental origin—to critically control their expression dosage (1). In addition, hundreds of regulatory non-coding RNAs (ncRNAs) are also imprinted, whose functions often remain poorly characterized. Most imprinted domains, for instance, express at least one long non-coding RNA (lncRNA) (2–5).

The imprinted Dlk1-Dio3 locus on mouse chromosome 12qF1 (Figure 1A) is controlled by an intergenic Imprinting Control Region (ICR)—called the intergenic Differentially Methylated Region (IG-DMR)—which is methylated on the paternal chromosome (6). The unmethylated maternal copy of this ICR acts as an enhancer that activates the nearby Meg3 promoter, thereby driving the expression of a 220-kb polycistron (7–10). This complex transcription unit generates the lncRNA Meg3 (‘Maternally expressed gene 3′, also called Gtl2) and many small RNAs, including miRNAs of the Mirg (‘microRNA containing gene’) cluster and C/D-box snoRNAs of the Rian (‘RNA imprinted and accumulated in the nucleus’) locus (referred to as MEG8 in humans) (8,11–13).

Figure 1.

pAS insertion into Meg3-intron-1 abrogates Dlk1 imprinted expression in NPCs. (A) Schematic presentation of the imprinted Dlk1-Dio3 domain, at which the Meg3 promoter drives the expression of the Meg3-Rian-Mirg ncRNA polycistron. Rectangles represent genes with their maternal- (red) or paternal (blue)-specific expression. The yellow rectangle depicts the ICR, called IG-DMR (6), which is methylated (lollipops) on the paternally inherited chromosome. Primary transcripts from Meg3-Rian-Mirg polycistron are represented by a dashed line; wave-shaped lines indicate processed ncRNAs. (B) CRISPR insertion of a 49-bp synthetic p(A) signal (‘pAS’) into Meg3 intron-1, using a single-stranded DNA oligo with homology arms (200 bp in total). Blue rectangles depict the exons of Meg3. (C) Immunofluorescence (IF) staining of Nestin (green) and Tubulin-β3 (red) with DAPI counter-staining (blue) in hybrid mESC-derived NPCs at d12 of neural differentiation. (D) DNA methylation status in hybrid mESC-derived NPCs (control, PAS-A1 and PAS-A2) and in E9.5 embryos, determined by methylation-sensitive qPCR. Controls: KvDMR1 (a maternally methylated ICR on chromosome 7), ActB promoter (low methylation) and IAP elements (high methylation). Bars represent means ± SD from three experiments. (E) Levels of Meg3, Rian and Mirg spliced RNAs assessed by RT-qPCR on total-RNAs relative to housekeeping genes (β-actin and Gapdh) in control, PAS-A1 and PAS-A2 NPCs. Bars represent means ± SD from three experiments (**** P< 0.0001). (F) Dlk1 mRNA amounts relative to housekeeping genes (β-actin and Gapdh) in control, PAS-A1 and PAS-A2 NPCs. Bars represent means ± SD from three independent experiments (**P< 0.01, *** P< 0.001). (G) Sanger sequencing-based assessment of Dlk1 allelism in control, PAS-A1 and PAS-A2 NPCs. The arrow indicates the SNP used to distinguish the maternal (M) and paternal (P) alleles. (H) H3K4me3 ChIP on control, PAS-A1 and PAS-A2 NPCs. Percentile precipitation levels were determined by qPCR at the Dlk1 promoter. Hoxa11 and ActB are negative and positive control regions, respectively. The KvDMR1, a control ICR, is transcriptionally active on the paternal chromosome only. Bottom: Sanger sequencing profiles indicate the allele-specificity of H3K4me3 at Dlk1 and the KvDMR1.

On the paternal chromosome, the Dlk1-Dio3 domain predominantly expresses two protein-coding genes with a distinct developmental dynamics (8,14,15). These are the Dlk1 gene (the atypical Notch-ligand-encoding ‘Delta-like homologue-1’) and the retrotransposon-derived gene Rtl1 (‘Retrotransposon gag-like 1’), which expresses a gag-like protein (14,16). Rtl1 overlaps the Meg3–Rian–Mirg polycistron, which expresses on the maternal chromosome sequences antisense to Rtl1 RNA (Figure 1A). A third protein-coding gene located downstream of the Meg3 ncRNA polycistron, thyroxine deiodinase type III (Dio3), shows a weak paternal expression bias (6,17–19).

While there is no evidence that the IG-DMR acts as an enhancer for the Dlk1 and Rtl1 genes on the paternal chromosome, the allelic repression of these genes on the maternal chromosome does require the maternal ICR (6,20). Specifically, the 3′ part of the ICR is critically required for its repressive effects on Dlk1 (7). Since this part of the IG-DMR displays enhancer features, and activates in cis the Meg3 promoter that drives the expression of the polycistron (8,9,21), it has been hypothesized that the expression of the Meg3-Rian-Mirg polycistron could be responsible for the in-cis repression of Dlk1 during development (7,8,10,14). Indeed, gene targeting studies in the mouse have shown that loss of Meg3–Rian–Mirg expression correlates with a loss of Dlk1 imprinting (i.e. biallelic expression) (6,7,10,20,22,23).

In humans, epimutations and microdeletions that affect the IG-DMR or the MEG3 promoter are observed in two congenital imprinting disorders (IDs), Kagami-Ogata Syndrome (KOS14, OMIM 608149) and Temple Syndrome (TS14, OMIM 616222) (24–27). KOS14 and TS14 have in common that the activity of the polycistron is either fully ablated, or becomes biallelic, respectively. Similar as in mice, this evokes a putative cis role of the MEG3-RIAN (MEG8)-MIRG polycistron in DLK1 gene expression dosage. In patients with these IDs, however, methylation changes are often mosaic and can involve multiple loci, which complicates interpretations about individual genes (28,29).

In support of a putative repressive role of Meg3 RNA, fluorescence in situ hybridization (FISH) studies on mESCs have shown that this lncRNA is retained in part on the imprinted domain and ‘overlaps’ Dlk1 on the maternal chromosome (22). Concordantly, Chromosome Conformation Capture (3C) based studies indicate that the Meg3 promoter is positioned in close proximity to the Dlk1 gene on the maternal chromosome, through the formation of an allele-specific sub-TAD (Topologically Associating Domain) (7,30). The maternal-specific boundary of this sub-TAD, formed by a non-methylated binding site for the CTCF insulator protein within intron-1 of Meg3, is critical for Dlk1 repression (7,30). This allele-specific 3D-organization might thus provide a framework for (or result from) the focal accumulation of Meg3 lncRNA. In both cases, it remains unclear whether Meg3 proximity to the Dlk1 locus is functionally relevant for its imprinting, and whether Meg3 cis-retention is required. As such, the precise role of Meg3 lncRNA, and that of the other ncRNAs of the polycistron, in imprinted gene expression remains unclear.

To address these questions, we engineered genetic modifications to the Dlk1-Dio3 domain in F1-hybrid mouse embryonic stem cells (mESC, C57BL/6J x JF1) that faithfully recapitulate imprinted gene expression upon in vitro differentiation (8,22,31). Multiple cell lines with decreased expression along the Meg3–Rian–Mirg polycistron were generated, and the effects of Rian deletion were explored as well. In addition, we induced biallelic Meg3 expression by CRISPR-mediated demethylation of the (normally silenced) paternal promoter, to assess the precise effects on imprinted gene expression and chromatin structure at the Dlk1-Dio3 domain. Our findings highlight the importance of Meg3 lncRNA—and not of Rian—in differential chromatin 3D structure and the in-cis repression of Dlk1. This essential role helps to understand complex imprinting-related disorders in humans, such as TS14 and KOS14, in which patients show aberrant MEG3 expression.

Materials and methods

mESC derivation, culture and differentiation

mESCs were derived under serum-free conditions in ESGRO 2i medium (Sigma-Aldrich, #SF016-200) from blastocysts that were hybrid between M. m. domesticus strain C57BL/6 J and M. m. molossinus strain JF1 (22,32). mESCs were maintained on gelatin-coated dishes in synthetic ESGRO 1i medium (Sigma-Aldrich, #SF001-500P) without serum, systematically supplemented with l-ascorbic acid (50 ug/ml) to prevent aberrant DNA methylation (33,34). For the detachment of cells, Accutase solution (Millipore, #SCR005) was used. mESCs were differentiated into neural progenitor (NPCs) cells for 12 days in the presence of the kinase inhibitor DMH1 in substitution of cyclopamine, but otherwise following a previously published protocol (35,36). mESC differentiation into cardiomyocytes (CMCs) was similar as reported by others before (37). Briefly, control or gene-edited mESCs were cultured for 2 days and were dissociated, to then form aggregates in differentiation medium [DMEM-Glutamax (4.5 g/l d-glucose + pyruvate; Gibco, #31966-021) supplemented with 20% fetal bovine serum (Gibco, #10500-064), 0.1 mM non-essential amino acids (Gibco, #11140-035), 100 μM β-ME (Gibco, #31350-010)] using the hanging drop method. Hanging drops comprised between 700 and 1000 cells, and were cultured for 3 days in 20 μl differentiation medium. The obtained embryoid bodies (EBs) were plated individually onto 0.1% gelatin-coated 24- or 96-well plates, or on microscope cover slips placed within culture dishes, in differentiation medium that was changed every 2 days. Cultures were analysed following 12 days of differentiation when most of the outgrowing EBs comprise beating cardiomyocytes. Cardiomyocyte differentiation was monitored by assessing the developmental marker genes Tbx5, Gata4, Mesp1 and Nkx2.5 (37) and by immuno-fluorescence staining of cardiac Troponin-T.

CRISPR-Cas9 mediated pAS insertion and Rian deletion

gRNAs were designed using CRISPR Design tool (http://crispr.mit.edu/), and presented in Supplementary Table S1. gRNAs were synthesized to be flanked with BbsI sticky ends and cloned into pSpCas9(BB)-2A-GFP plasmid (Addgene, #48138). Constructs were transfected into mESCs using Amaxa nucleofector (Lonza, #VPH-1001). GFP-positive cells were sorted by flow cytometry (FACS Aria, Becton Dickinson) after 48 h of transfection, and plated as single cells in 96 well plates. As a control experiment, a non-specific, scrambled gRNA was used to generate control mESC lines that had undergone the same manipulations. Cell lines were established from single colonies. For pAS insertion, a plasmid carrying one gRNA was transfected into mESCs in addition to a homology repair sequence that ensured a proper pAS insertion at the desired location. Colony PCR testing identified cell lines harboring the pAS insertion, and two lines were selected for further study: PAS-A1 (collection name: BJ-PAS-A3) and PAS-A2 (collection name BJ-PAS-A19). A CRISPR-Cas9 NHEJ-mediated deletion genome-editing approach was used to delete the Rian snoRNA cluster, using two gRNAs. RT qPCR and RNA-seq confirmed that the entire Rian cluster was deleted in the selected ESC clones: Rian^−/−[1] (collection name BJ-Rian 59) and Rian^−/−[2] (collection name BJ-Rian 75).

CRISPR-dCas9-TET1-mediated DNA demethylation

gRNAs are described in Supplementary Table S1. The gRNA for the H19 ICR was used by others before (38). gRNAs for the Meg3 DMR were designed about 1 kb upstream of exon-1 in the annotated promoter regions. sgRNAs were annealed and cloned into the pPlatTet-gRNA2 plasmid (Addgene, #82559; a kind gift from Dr Izuho Hatada) after digestion by AflII and Gibson assembly (38). Plasmids expressing gRNA were electroporated into mESCs using Amaxa nucleofector (Lonza, #VPH-1001). GFP-positive cells were selected 48 h post-electroporation by flow cytometry (FACS Aria, Becton Dickinson), and single cells were seeded onto 96-well plates. After 10–12 days of culture, individually picked colonies were grown in 6-well plates to derive ESC clones. Their methylation levels were determined by methylation-sensitive enzymatic digestion.

Immunofluorescence staining

Immunofluorescence staining of cells was performed as reported before (8). The used primary antisera were directed against Nestin (Biolegend, #839801), Tubulin-B3 (Biolegend, #801201) and Troponin-T (Invitrogen, #13-11 MAS-12960). Secondary antibodies used: goat anti-mouse Alexa fluor 488 (Thermo-Fisher, #A-11011) or goat anti-rabbit Alexa Fluor 594 (Thermo-Fisher, #A-11012).

RNA and DNA FISH

RNA-FISH was performed as in Chaumeil et al. 2008 (39), on cells grown on gelatin-coated coverslips and fixed with 4% paraformaldehyde for 10 min at room temperature. Cells were permeabilized in 0.5% v/v Triton X-100 in 1× PBS (10 min, on ice), rinsed in 2× PBS (5 min, RT) and store at 4°C in 70% ethanol. RNA FISH against total Meg3 RNA was with a cDNA probe (Openbiosystems, #6831921), and with a probe comprising intron1 + intron8 sequences (Chr.12, mm10: 109541150–109542018, 109554267–109555402, 109556110–109557125 and 109555037–109557125). For simultaneous RNA- and DNA-FISH, cells were re-permeabilised in 0.5% v/v Triton X-100 in 1× PBS (20 min, RT), rinsed in 2× PBS (5 min, RT) and 70% ethanol (5 min, RT). Cells were dehydrated for 3 min subsequently, with respectively 80%, 95% and 100% ethanol. Before hybridization, cells were denatured in 50% formamide, 2× SSC (pH 7.2) for 30 min in water bath at 80°C; then, washed twice in 2× SSC on ice. We used a Dlk1-comprising fosmid (WIBR1-1116K16; Chr12: 110674199–110709802, BACPAC Resources Center) and Meg3 cDNA as probes, labelled with fluorescent nucleotides by nick translation using 1 μg of DNA per 50 μl of reaction following manufacturer's instructions (Abbott, #07J00-001). Per coverslip, approximately 0.1 μg of probe was ethanol-precipitated together with 10 μg of salmon sperm DNA and 1ug of cot-1 DNA, air-dried and resuspended in 12.5 μl of formamide. Dlk1 fosmid and Meg3 cDNA probes were then mixed in 12.5 μl hybridization cocktail consisting of 2× hybridization mixture (4× SSCT, 20% w/v dextran sulfate, 2 mg/ml BSA, 40 mM Vanadyl Ribonucleoside Complex). Cells were incubated in hybridization cocktail, overnight at 40°C in a dark humidified chamber. The next day, cells were washed three times with 50% formamide, 2× SSC at 42°C for 5 min, followed by three times 5 min with 2× SSC at 42°C, and stained with DAPI and mounted using Vectashield antifade 18 mounting medium (VectorLabs, #H-1000). Images were acquired on a laser scanning confocal microscope (LSM980 Airyscan 8Y, Zeiss) with 63× NA1.4 Plan-Apochromat objective (Zeiss). z stacks of 0.2-μm slices were visualized and analysed using OMERO Open Microscopy Environment (OME) and ImageJ tools. For foci co-localization measurements, we used JACoP plugin in ImageJ.

RNA expression analyses

Total RNA was extracted using RNeasy-plus mini kit (Qiagen) and transcribed into cDNA using random hexamers and Superscript III reverse transcriptase (Thermo Fisher). In subsequent RT-qPCR, measured expression levels were normalized to the geometric mean of two housekeeping genes (Actb and Gapdh) as reported before (22). For the analysis of Rtl1 expression levels, strand-specific oligonucleotides were used for cDNA synthesis in multiple experiments, followed by Rtl1-specific amplification; a one-way Anova test was used for statistical comparison (40). Primer sequences are presented in Supplementary Table S2.

For sequencing of nuclear RNAs, nuclei were purified by incubating 15.10⁶ resuspended cells in 4 ml of Buffer I (10 mM Tris–HCl pH 7.5, 10 mM NaCl, 2.5 mM MgCl₂, 0.5% IGEPAL CA-630) on ice for 5 min. Next, we carefully underlaid 1 ml of Buffer II (10 mM Tris–HCl pH 7.5, 10 mM NaCl, 2.5 mM MgCl₂, 0.5% IGEPAL CA-630, 10% sucrose) and harvested the nuclear fraction at 1400 rcf for 5 min at 4°C. Nuclear RNAs were isolated using RNeasy Plus Mini Kit (Qiagen, #74136); they were quantified by Qubit and quality was assessed using the RNA Assay kit (Agilent RNA 6000 Pico reagents, #1567-1513) with Bioanalyzer 2100 (Agilent Technologies, USA). Putative residual genomic DNA was digested using Amplification-Grade DNase I (AMPD1, Sigma) and spike-in ERCC (Thermofisher, # 4456740) was added following manufacturer instructions before library preparation of RNA samples. For library preparation we used the True-seq stranded total RNA library prep gold kit (Illumina, #220599) following the manufacturer's instructions (including ribo-depletion). Sequencing datasets were aligned to the mouse genome (mm39) using TopHat2. Aligned reads were treated using an in-house developed PASHA (version 0.99.21) R (version 3.3.1) pipeline to generate wiggle files. Read counts within the ERCC spike-in genome were determined by HTSeq-count (version 0.6.1p1) and used to normalize the wiggle files for visualization as reads per million reads of the spiked-in genome.

ChIP and CUT&RUN

Chromatin immuno-precipitation (ChIP) was performed as described before (22). Briefly, cells were fixed in 1% (v/v) formaldehyde for 10 min at RT, followed by 1 × 10 cycles (30 s on/30 s off) of sonication in a BioRuptor Pico sonicator. Around 15 million cells were used for each IP. The antisera used were: 5 μg of anti-H3K4me3 (Active Motif, #39155), 10 μg of anti-CTCF (Merck Millipore, #07-729). Precipitated DNA was purified using the ‘ChIP DNA clean and concentrator kit’ (Zymo Research, #D5205) and quantified by qPCR. PCR products were run on a 1% agarose gel, excised, and column-purified (Macherey-Nagel, #740609.10). The allelism was determined by Sanger sequencing of qPCR product across SNPs. Primer sequences are given in Supplementary Table S3. CUT&RUN (Cleavage Under Targets and Release Using Nuclease) experiments were as described before (Roidor, Syx et al. 2023 BiorXiv; doi: https://doi.org/10.1101/2023.04.25.532252). Briefly, for each sample 200 000 cells were pelleted. Nuclear Extraction Buffer (20 mM HEPES–KOH, 10 mM KCl, 0.5 mM spermidine, 0.1% Triton X-100, 20% glycerol, complete EDTA-free protease inhibitor cocktail) was gently added to the pellet, with incubation on ice for 5 min. Nuclei were then re-suspended in Nuclear Extraction Buffer and stored at –80°C for up to several weeks. The antisera used for CUT&RUN were anti-H3K27me3 (Cell signalling 36B11#9733) and anti-IgG (Sigma, #I5006), followed by addition of Protein-A-MNase fusion protein (a kind gift of Dr D. Helmlinger), to 200 ng/ul. The MNAse-digested DNA (fragments) diffused out of the cells was precipitated after 30 min, and loci of interest were quantified by qPCR. In addition, PCR products were run on a 1.5% agarose gel, excised, and column-purified (Macherey-Nagel, #740609.50), followed by Sanger sequencing to distinguish the parental alleles. PCR primer sequences are given in Supplementary Table S3.

DNA methylation analysis

Methylation levels were analysed through digestion with methylation-sensitive restriction endonucleases (AciI, HpaII or HhaI), depending on the genomic locus studied, followed by qPCR. Briefly, 1 μg of genomic DNA was digested with EcoRI in a 100 μl reaction volume. After 3 h, the reaction volume was divided into two tubes, each containing 40 μl of the initial reaction volume, with one Eppendorf tube without, and a second tube with, addition of methylation-sensitive restriction enzyme. 1 ng of DNA from each tube was used for qPCR. The percentage of methylation for each region was calculated, and the standard curve method was used to quantify values. Values were normalized to the amplification levels of two non-methylated control regions (Col1a2 and Col9a2) (30). Primer sequences are provided in Supplementary Table S4. For Reduced Representation Bisulfite Sequencing (RRBS), 200 ng of genomic DNA was digested with MspI for 5 h, followed by end-repair, A-tailing with Klenow fragment (ThermoScientific, #EPO-421), and ligation to methylated indexed Illumina adapters using T4 DNA ligase (ThermoScientific, # 15224017). Fragments were purified using AMPure XP magnetic beads (Beckman Coulter) (41). Two rounds of bisulfite conversion were then performed using the EpiTect kit (Qiagen, #59104). Final RRBS libraries were PCR-amplified with PfuTurbo Cx hot-start DNA polymerase (Agilent, #600410) as follows: 95°C for 2 min, 14 cycles (95°C for 30 s, 65°C for 30 s, and 72°C for 45 s), 72°C for 7 min. Next, the libraries were purified with AMPure XP magnetic beads, quantified with a Qubit fluorometer (Life Technologies), and verified by Fragment analyzer (Advanced Analytical) and qPCR. Directional libraries were sequenced (100 nt single-end reads) on an Illumina HiSeq2000 at the MGX facility. Reads were processed and analyzed with tools developed by the Babraham Institute, Cambridge, UK (Trim_galore, Bismark and Seqmonk) (42). For pyrosequencing, 1 μg of genomic DNA was treated with sodium bisulfite using the EpiTect kit (Qiagen, #59104). PCR products were amplified with the PyroMark PCR kit (Qiagen, #978703) and purified with Streptavidin Sepharose HP™ (GE Healthcare, #17–5113-01) using a PyroMark Q24 Workstation. Pyrosequencing was done with Gold Q24 reagents (Qiagen, #970802) using a PyroMark Q24. PCR- and sequencing primers are provided in Supplementary Table S5.

4C-seq, Capture-C and Capture Hi-C experiments

4C-seq experiments on WT and Meg3-TET-46 mESCs were as described before (30,43). For the IG-DMR viewpoint, only the JF1 (paternal allele) was analyzed. For other viewpoints, both alleles were targeted and reads were assigned to their corresponding allele using a single nucleotide polymorphism (SNP) located four nucleotides 3′ of the forward 4C-seq primers. Downstream analyses were done using the c4ctus pipeline (43), with customized Perl and R scripts to distinguish the parental alleles based on SNPs (scripts available upon request from BM). Primer sequences are given in Supplementary Table S6. The Capture-C strategy used for WT and Meg3-TET-46 mESCs was adapted from the C-TALE protocol (44). Briefly, 3C material was prepared from cross-linked cell pellets as described before (43). DNA was fragmented using a Covaris S220 apparatus (10% Duty Factor, 140W incident Power, 200 cycles per burst for 55 s) and converted into Illumina compatible libraries using separate NEBNext Ultra II modules (NEB, #E6050, #E6053 and #E6056) following the manufacturer's instructions. Three or four aliquots of 200 ng of adaptor ligated libraries were amplified using KAPA HiFi HotStart for 5 cycles. To generate biotinylated probes for enrichment, an equimolar mix of BACs (RP23-75I2, RP23-132J1, RP23-409I23, ordered from Source BioScience) was fragmented by enzymatic digestion (with MboI) followed by sonication (Covaris S220; 10% duty factor, 150 W incident power, 200 cycles per burst for 180 sec) and biotinylated as described (44). Two rounds of target enrichment were performed, using reagents and instructions from the Twist Hybridization and Wash kit (Twist, #101279) and Twist Universal Blockers (Twist, #100856). For both rounds, the enrichment was performed on a total of 1.5 μg Illumina compatible material, with the post-capture PCR reactions done using KAPA HiFi polymerase (12 PCR cycles). After the second enrichment step, size-selection was by sequential SPRI bead binding (Beckman-Coulter, 0.6× and 0.9× volumes).

For Capture Hi-C on differentiated control CMCs and NPCs, epimutated CMCs and mutant NPCs, a commercially available kit was used. Hi-C material was prepared from 1 million cross-linked cells using the Arima Hi-C+/High Coverage Hi-C kit (Arima Genomics), following the manufacturer's instructions. Sequencing libraries were prepared using the SureSelect XT HS2 DNA System kit (Agilent). Target enrichment was performed using a custom panel targeting the coordinates chr12:108930000–110030000 (Agilent; mm10), following the manufacturer's instructions. For both Capture-C and the Capture Hi-C, the material was sequenced on the Illumina NextSeq 500 (paired-end, 2 × 43) at the High-throughput sequencing facility of I2BC. Reads were processed using the Hi-C pro tool (45) using the JF1-specific SNPs from the Mouse Genome Project REL-1807. Allele-specific matrices, generated at 10-kb resolution, were displayed and analyzed using R. Indicated coordinates are from GRCm38/mm10.

Results

Insertion of a poly(A) signal into Meg3 reduces Meg3, Rian and Mirg expression and attenuates Dlk1 imprinting

The Meg3-promoter-driven expression of the Meg3–Rian–Mirg polycistron (Figure 1A) is required to prevent Dlk1 activation on the maternal chromosome (22). Whether this repressive role is mediated by the Meg3 promoter, its transcriptional activity, or by one or more of the many ncRNAs produced by this large polycistron (46), is unclear. To address this important question, we used CRISPR-Cas9-mediated recombination to insert a poly(A) signal (pAS) within Meg3, with the aim of generating mESC lines with strongly reduced Meg3–Rian–Mirg polycistron expression. We used a synthetic pAS of 49-bp in length adapted from the rabbit β-globin pAS (47) that had yielded truncated lncRNA transcripts in earlier studies by others (48,49). Experiments were performed on naïve hybrid mESCs and we systematically supplemented the serum-free ESC medium with ascorbic acid, to prevent acquisition of aberrant de novo DNA methylation (33,34).

To reduce RNA levels along the entire polycistron, while maintaining the Meg3 promoter in an active state, we inserted one copy of the synthetic pAS into intron-1 of Meg3, 240-bp downstream of the transcription start site (TSS) (Figure 1B). Following electroporation of WT mESCs [of (C57BL/6J x JF1)F1 genotype]—hereafter referred to as ‘BJ’—gene-edited cells were purified by cytometric cell sorting based on Cas9-GFP expression. Following colony formation, pAS insertion was ascertained by PCR amplification followed by DNA sequencing of the obtained PCR products. Two independent clones were selected for further studies—with biallelic and maternal PAS insertion respectively—which we named ‘PAS-A1’ and ‘PAS-A2’ (Supplementary Figure S1A). As a negative control, a non-specific scrambled gRNA was used, to generate suitable control mESCs.

Methylation levels at the IG-DMR and Meg3 promoter, as assayed using methylation-sensitive enzymatic digestion, remained at about 50%, as expected from the allelic nature of the methylation imprint (Supplementary Figure S1B). Because of the pAS insertion, the expression of Meg3, Rian and Mirg was reduced to about half compared to control mESCs (Supplementary Figure S1C). RNA FISH showed that the percentage of cells with focal lncRNA retention was similar (∼70%) as in the control cells (Supplementary Figure S1D).

To induce neural differentiation, we applied a previously published procedure to generate neural progenitor cells (NPCs) from mESCs (31,35). PAS-A1 and PAS-A2 mESCs readily differentiated into NPCs, with similar efficiency as control BJ mESCs. At day 12 (d12) of differentiation, NPCs displayed axonal outgrowth and showed similar protein expression of Nestin and Tubulin-β3 as NPCs generated from control mESCs (Figure 1C). Similarly, RT-PCR analysis on total-RNA samples confirmed that the neural marker genes Nestin, Fabp7 and Emx1 were expressed at similar levels in the PAS-A1 and PAS-A2 versus the control NPCs (Supplementary Figure S1E). DNA methylation at the Meg3 promoter and the IG-DMR remained unaltered upon neural differentiation of the PAS-A1 and PAS-A2 cells (Figure 1D).

In the obtained NPCs, the transcriptional reduction was more pronounced than in mESCs, with the Meg3 (5′ part of the spliced transcript), Rian and Mirg transcript levels reduced by 70–80% (Figure 1E). To determine whether the reduced Meg3 levels included the primary transcript, and the 3′ portion of the gene, we performed additional RT-qPCR amplifications at exon 8 and intron 8. At both these regions, there was a similar reduction (70–80%) in the PAS-A1 and PAS-A2 NPCs (Supplementary Figure S1F).

In control NPCs, substantial parts of the Meg3 RNA were present at high levels in the nucleus, presumably through post-transcriptional accumulation. RNA FISH with a Meg3 cDNA probe revealed multiple accumulation foci in the nucleoplasm of NPCs. This observation showed that besides cis-accumulation, there is extensive trans-accumulation in neuronal cells as well. The same was observed in PAS-A NPCs as well, albeit less strongly (Supplementary Figure S2A). To assess nuclear RNA levels across the entire 220-kb polycistron, we performed RNA-seq on purified nuclei. The RNA-seq confirmed that specific portions of Meg3 RNA are present at high levels in the nucleus. It also revealed that processed forms of Rian snoRNA were present at high concentration in the nucleus as well, in agreement with an earlier report on the cis-accumulation of snoRNAs in neural cells (50). Although in PAS-A1 and PAS-A2 NPCs there was still considerable presence of specific Meg3 and Rian sequences in the nucleus, overall RNA levels across the polycistron were reduced compared to WT NPCs, in accordance with the RT-qPCR analyses of total-RNA samples (Supplementary Figure S2B).

We next assessed Dlk1, which becomes activated primarily on the paternal allele during stem cell differentiation (15,22). Dlk1 mRNA levels were about twice as high in the PAS-A1 and PAS-A2 NPCs compared to control NPCs (Figure 1F). Concordantly, whereas in the control NPCs Dlk1 expression was mostly paternal, the PAS-A1 and PAS-A2 NPCs showed expression from both the parental chromosomes (Figure 1G). To confirm the biallelic Dlk1 activation (‘loss of imprinting’), we performed chromatin immunoprecipitation (ChIP) against histone H3 lysine-4 tri-methylation (H3K4me3), which marks active promoters. Whereas, as expected, H3K4me3 was strictly paternal at Dlk1 in the control NPCs, it was detected on both the parental alleles in the PAS-A1 and PAS-A2 NPCs (Figure 1H). These findings demonstrate that the reduced transcription across the Meg3-Rian-Mirg polycistron in NPCs, either directly or indirectly through its RNA products, gave strongly reduced repression of Dlk1 on the maternal chromosome.

To assess if mono-allelic activation of Dlk1 in mesodermal lineages (15) was similarly perturbed, we applied a published procedure to generate cardiomyocytes (CMCs) from mESCs [(37), see Materials & Methods]. In WT control cells, at day 12 of differentiation, a high proportion of cells had differentiated into CMCs, with expression of cardiac Troponin-T (Supplementary Figure S3A). Concordantly, the cardiac-lineage markers Tbx5, Gata4, Mesp1 and Nkx2.5 were strongly upregulated as well, and a comparable marker-gene activation occurred in the PAS-A1 and PAS-A2-derived CMCs (Supplementary Figure S3B). In the PAS-A1 and PAS-A2 cardiomyocytes, Meg3, Rian and Mirg RNA levels were reduced by about 20% only (Supplementary Figure S3C). A concordant, minor effect on Dlk1 imprinting was observed, with partial activation of the normally-silent maternal Dlk1 allele (Supplementary Figures S3D, E). These observations in CMCs complement our finding in NPCs that transcription across the polycistron is important to prevent Dlk1 activation in-cis on the maternal chromosome.

The pAS insertion study indicates that the expression level of the Meg3 polycistron is critical for the repression in cis of Dlk1 during differentiation.

Rian C/D-box snoRNAs are not required for Dlk1 imprinted expression

Earlier studies in NPCs and newborn mice have showed that deletion and overexpression of the Mirg miRNAs does not influence the level or allelism of expression of Dlk1 (51–54). However, allele-specific functional studies on the Rian snoRNAs have not been reported. Since processed forms of Rian are locally retained in the nucleus (50), and we still detected substantial expression in the PAS-A NPCs, our pAS insertion study did not exclude a possible involvement of Rian in Dlk1 imprinting.

To address this question, we deleted a region of 56 kb comprising the entire snoRNA locus in the hybrid mESCs, using a CRISPR-Cas9 NHEJ-mediated deletion approach (Figure 2A). Two independent mESC lines with biallelic Rian deletion—called Rian^−/−[1] and Rian^−/−[2]—were selected. Both these knock-out mutants had unaltered DNA methylation at the Meg3 DMR and the IG-DMR (Supplementary Figure S4A). As expected, Rian RNA was no longer detected by RT-qPCR analysis in the Rian^−/−[1] and Rian^−/−[2] mESCs, whereas the amounts of the other ncRNAs of the polycistron remained at levels similar to control cells (Supplementary Figure S4B). RNA-seq on nuclear RNA confirmed that Rian was deleted in the Rian^−/−[1] mESCs, with a complete lack of RNA signal across the entire Rian gene, and that the remainder of the polycistron remained expressed (Supplementary Figure S5C).

Figure 2.

Deletion of the Rian snoRNA cluster does not affect Dlk1 imprinting. (A) Schematic presentation of the CRISPR-Cas9-mediated Rian deletion in mESCs. (B) DNA methylation status in hybrid mESC-derived NPCs at d12 of neural differentiation (control and Rian^−/−) and in E9.5 embryo as determined by methylation-sensitive qPCR. Bars represent means ± SD from three experiments. (C) IF staining of Nestin (green) and Tubulin-β3 (red) with DAPI counterstaining (blue) in NPCs at d12 of neural differentiation. (D) RNA accumulation of the Meg3, Rian and Mirg ncRNAs (RT-qPCR on total RNA samples) relative to housekeeping genes (β-actin and Gapdh) in control, Rian^−/−[1] and Rian^−/−[2] NPCs. Bars represent means ± SD from three independent experiments (ns, non-significant; **** p< 0.0001). (E) Dlk1 mRNA amounts in control, Rian^−/−[1] and Rian^−/−[2] NPCs (ns, non-significant). (F) Sanger sequencing-based assessment of the allele-specificity of Dlk1 expression in control, Rian^−/−[1] and Rian^−/−[2] NPCs. The arrow indicates the SNP used to distinguish maternal (M) and paternal (P) alleles.

Rian^−/−[1] and Rian^−/−[2] mESCs differentiated normally into NPCs, and showed unaltered DNA methylation at the IG-DMR and the Meg3 promoter (Figure 2B), and a pattern of Nestin and Tubb3 expression similar to control NPCs (Figure 2C). RT-PCR analysis also confirmed normal expression of the neural marker genes Nestin, Fabp7 and Emx1 in the Rian-/-[1] and Rian^−/−[2] NPCs as compared to control NPCs (Supplementary Figure S4D). The Rian^−/− NPCs did not express Rian, as expected, and had unaltered levels of the Meg3 and Mirg RNAs (Figure 2D). Dlk1 expression levels and the paternal allele-specificity of Dlk1 expression were unaltered in the Rian^−/− NPCs as well (Figure 3E, F). Together, these findings demonstrate that the Rian C/D-box snoRNAs do not control the imprinted expression of Dlk1.

Figure 3.

dCas9-suntag-TET1 mediated demethylation of the Meg3-DMR induces biallelic expression of Meg3 in mESCs. (A) Schematic presentation of Meg3, with the gRNAs used for promoter demethylation in mESCs. (B) DNA methylation status in hybrid mESCs (control, Meg3-TET-26 and Meg3-TET-46) and in E9.5 embryos, determined by methylation-sensitive qPCR. Bars represent means ± SD from three experiments. (C) DNA methylation status in hybrid mESCs (control, Meg3-TET-26 and Meg3-TET-46) as determined by pyrosequencing. Data represent the percentile methylation levels of 3 CpG dinucleotides at Meg3 promoter, 5 CpG at the 5′ part of the IG-DMR and the 11 CpGs at the 3′ part of the IG-DMR. (D) Sanger sequencing-based assessment of the allelism of Meg3 expression in control, Meg3-TET-26 and Meg3-TET-46 mESCs. The arrow indicates the SNP used to distinguish the maternal (M) and paternal (P) alleles. (E) RNA accumulation of the Meg3, Rian and Mirg ncRNAs (RT-qPCR) relative to housekeeping genes (β-actin and Gapdh) in control, Meg3-TET-26 and Meg3-TET-46 mESCs. Bars represent means ± SD from three experiments (** P< 0.01, * P< 0.05). (F) RNA-FISH analysis of the Meg3 lncRNA in control mESCs (n = 140) and Meg3-TET-26 mESCs (n = 123). DNA was counter-stained with DAPI (blue); nuclei are delineated by a dashed line; scale bar, 10 μm. On the right: the proportion of nuclei with zero, one or two Meg3 focus. (∗∗ P < 0.005, two-sided Fisher's exact test). (G) Combined RNA + DNA FISH detection of Meg3 RNA (red) and the Dlk1 gene (green, with a fosmid probe) in control and Meg3-TET-26 ESCs. DNA was counter-stained with DAPI (bleu). Arrows show overlap/proximity between Meg3 and Dlk1. Dashed lines demarcate projection nuclei with foci; scale bar, 10 μm. To the right: Pearson coefficients for Meg3-Dlk1 overlap calculated for control ESCs (n = 62, maternal chromosome), and for Meg3-TET-26 ESCs with a single RNA accumulation spot (n = 33; ‘1 focus’) or two RNA spots (n = 77; ‘2 foci’, both parental chromosomes).

Combined, our analyses of the pAS insertion and the Rian^−/− mESC lines indicate that transcription across the polycistron is important for the acquisition of Dlk1 imprinted expression, but that this function does not require the expression of Rian, located downstream of Meg3. Given that Mirg miRNAs also does not influence the level or allelism of expression of Dlk1 (51–54), we conclude that the lncRNA Meg3 itself is required for Dlk1 imprinted expression, with its expression levels influencing the degree of the in-cis repression.

CRISPR-mediated demethylation of the Meg3 promoter induces biallelic ncRNA expression

Temple Syndrome (TS14) is a human imprinting disorder (ID) characterized by increased expression of the ncRNA polycistron, affecting MEG3, MEG8 (called Rian in the mouse) and MIRG RNA levels (55–57). This aberrant ncRNA expression could have a causative involvement in this developmental and aberrant-growth disorder; yet, the underlying mechanisms remain largely unknown (57). TS14 often arises through maternal uniparental disomy of chromosome 14 (mUPD14)—where the human DLK1-DIO3 locus resides—leading to biallelic MEG3 expression (55). Less frequently, the syndrome is caused by losses of methylation at the MEG3 promoter region (24,58–60). To model the increased biallelic MEG3 expression in TS14 expression and to unravel its molecular effects, we performed CRISPR-dCas9-based epigenetic editing in our hybrid mESCs. Particularly, we asked whether demethylation of the (paternal) Meg3 promoter (DMR) is sufficient to induce biallelic Meg3 lncRNA expression, and, if so, whether this gives rise to biallelic repression of Dlk1 during ESC differentiation.

We used a dCas9-SunTag system that recruits up to ten copies of the catalytic domain of TET1 to the targeted DNA sequence and neighboring DNA (38). Recently, this transient approach was applied to demethylate the intergenic ICR of the Igf2-H19 imprinted domain in mESCs (38,61). To benchmark the technology, we therefore first aimed to demethylate the H19 ICR in our hybrid mESC system. For this purpose, we designed a single guide-RNA directed against the second CTCF binding site in the H19 ICR, which is methylated on the paternal chromosome only (Supplementary Figure S5A). To screen for maintained DNA hypomethylation in the obtained clones, we used a restriction-endonuclease digestion approach to quantify DNA methylation at CTCF binding sites 2 and 4, and at the H19 promoter. Two of the clones—H19-TET-2.1 and H19-TET-2.7—showed strongly reduced methylation levels at all three regions analysed, with only about 20% of remaining methylation (Supplementary Figure S5B). For CTCF site-2 and the H19 promoter, we confirmed hypomethylation by pyrosequencing and by reduced representation bisulfite sequencing (RRBS) (Supplementary Figure S5C,D). Following differentiation of H19-TET-2.1 and H19-TET-2.7 cells into CMCs, the H19 ICR and the H19 promoter faithfully retained their hypo-methylated state, with the same low methylation levels detected as in the mESCs from which they derived (Supplementary Figure S5E). In contrast, in H19-TET-2.1 and H19-TET-2.7 derived NPCs we noted a regain of normal methylation levels (Supplementary Figure S5E).

With experimental conditions in place, we next set out to demethylate the Meg3 promoter. For this, we used three guide RNAs that target the Meg3 promoter within an interval of 380 bp (Figure 3A). Two mESC lines that showed Meg3 demethylation—Meg3-TET-26 and MEG3-TET-46—were selected for further studies (Figure 3B). As confirmed by pyrosequencing, these lines showed an about 50% reduction of methylation at the Meg3 promoter (Figure 3C). We used RRBS as another approximate of methylation levels. This confirmed that the Meg3 promoter/CpG island had become hypomethylated in both the Meg3-TET-26 and Meg3-TET-46 mESC lines, in a genomic interval comprising the Meg3 promoter, exon 1 and part of intron-1 (Supplementary Figure S6A). At the IG-DMR, the 5′-side retained its high methylation level, whereas a partial reduction in methylation was observed at the 3′ side of the IG-DMR (Figure 3C). This sensitivity of the 3′ side of the IG-DMR to the targeted demethylation, despite the genomic distance from the Meg3 promoter (∼13 kb), may be due to its reported physical proximity with the Meg3 promoter within the nucleus (7).

In agreement with the induced Meg3 promoter hypo-methylation, the Meg3-TET-26 and Meg3-TET-46 mESCs showed biallelic Meg3 expression (Figure 3D). Meg3, Rian and Mirg RNA levels were about 2-fold higher than in control ESCs (Figure 3E). To confirm the allelism of Meg3 expression, we performed RNA FISH. In WT BJ ESCs, a single accumulation focus was apparent in about 70% of the cells—with no cells showing two foci—similarly as reported before (22). On the contrary, about 30% of the Meg3-TET-26 cells showed two nuclear accumulation foci, with a similar retention on the locus on both the parental chromosomes, in agreement with the TET1-induced partial demethylation of the paternal Meg3 promoter (Figure 3F). Next, we performed RNA-FISH against Meg3 RNA combined with DNA FISH against Dlk1. In the Meg3-TET-26 ESCs that showed two Meg3 RNA foci (expression from both parental chromosomes), there was a similar overlap with Dlk1 as in the WT cells with a single focus (maternal expression only) (Figure 3G). Although this assay does not tell the parental chromosomes apart, this observation suggests that in Meg3-TET cells with two RNA foci, the Meg3 cis-accumulation was comparable on the maternal and the paternal chromosomes.

CRISPR-induced biallelic Meg3 expression leads to biallelic Dlk1 repression and loss of Rtl1 expression in differentiated cells

Having generated ECSs with Meg3 biallelic expression, we next explored its effects on the Dlk1 expression in differentiated cells. The Meg3-TET-26 and Meg3-TET-46 mESCs could be readily differentiated into CMCs with stable maintenance of the Meg3 hypomethylation (Figure 4A), similarly as observed for the CRISPR-demethylated H19 ICR in CMCs (Supplementary Figure S5E). Consequently, Meg3 expression remained biallelic in the differentiated Meg3-TET-26 and Meg3-TET-46 cells, and this correlated with a 2-fold increase in Meg3, Rian and Mirg RNA amounts (Figure 4B, C).

Figure 4.

Demethylation of the Meg3 DMR affects Dlk1 and Rtl1 imprinted expression. (A) DNA methylation status in hybrid mESC-derived CMCs (control, Meg3-TET-26 and Meg3-TET-46) and in E9.5 embryos as determined by methylation-sensitive qPCR. Bars represent means ± SD from three experiments. (B) RNA accumulation of the Meg3, Rian, and Mirg ncRNAs (RT-qPCR) relative to housekeeping genes (β-actin and Gapdh) in control, Meg3-TET-26 and Meg3-TET-46 CMCs. Bars represent means ± SD from three experiments (** P< 0.01, * P< 0.05). (C) Sanger sequencing-based assessment of Meg3 expression allelism in control, Meg3-TET-26 and Meg3-TET-46 CMCs. (D) Dlk1 mRNA amounts in control, Meg3-TET-26 and Meg3-TET-46 CMCs (*** P< 0.001). (E) Sanger sequencing-based assessment of Dlk1 allelism in control, Meg3-TET-26 and Meg3-TET-46 CMCs. (F) Rtl1 mRNA amounts in control, Meg3-TET-26 and Meg3-TET-46 CMCs (****P< 0.0001).

In the Meg3-hypomethylated CMCs, the biallelic Meg3 expression and lncRNA cis-accumulation correlated with strongly reduced Dlk1 expression levels (Figure 4D) with, as expected, the maternal allele being less strongly expressed than the paternal (Figure 4E). Combined, the above findings show that the induced loss of methylation at the Meg3 DMR had induced biallelic Meg3 expression, which gave Dlk1 repression on both the parental chromosomes.

Meg3-TET-26 and Meg3-TET-46 derived NPCs were studied as well, and showed expression of the neural markers Nestin and Tubulin-β3 as in control NPCs (Supplementary Figure S6B). However, in these neural cells, similarly as was observed at the H19 ICR in the H19-TET NPCs (Supplementary Figure S5E), we noted that the Meg3 DMR was methylated at levels similar as in the control NPCs (Supplementary Figure S6C). Concordantly, Meg3 expression levels were as in the control WT NPCs, with normal maternal allele-specific expression (Supplementary Figure S6D, E). Consequently, in the Meg3-TET-26 and Meg3-TET-46 derived NPCs, we noted unaltered levels of Dlk1 expression, from the paternal chromosome only, precisely as in the NPCs derived from the WT control mESCs (Supplementary Figure S6F, G). Given the aberrant regain of methylation at the Meg3 DMR, the Meg3-TET-26 and Meg3-TET-46 derived NPCs were not used for further imprinting studies.

Next, we explored Rtl1, an imprinted gene that overlaps the Meg3–Rian–Mirg polycistron, but that is transcribed in the opposite direction (Figure 1A). The Rtl1 gene is not expressed on the maternal chromosome, where the Meg3 promoter drives the expression of the polycistron (16). Besides its main site of expression, the placenta, Rtl1 is expressed in muscle cells where its deficiency causes distinct muscle abnormalities in mice (40).

Using a published strand-specific RT-PCR approach (62), Rtl1 mRNA was readily detected in the mESC-derived cardiomyocytes. In the Meg3-TET-26 and Meg3-TET-46 CMCs, in contrast, hardly any Rtl1 RNA was detected (Figure 4F). Thus, the biallelic high transcription across the ncRNA polycistron—including across Rtl1—had given an almost complete loss of Rtl1 expression. Though we could not distinguish the parental alleles, the loss of Rtl1 expression implies that the CRISPR-induced Meg3 promoter activity and polycistron transcription lead to a loss of Rtl1 expression on the paternal chromosome, precisely as it does on the maternal chromosome (16).

Differential Meg3 expression guides parental chromosome-specific sub-TAD organization

The Dlk1-Dio3 domain is organized into parent-of-origin specific 3D chromatin organizations, resulting in the formation of allele-specific sub-Topologically Associating Domains (sub-TADs) that contribute to the imprinted gene expression (30). Particularly, using an allelic 4C-seq approach, we recently reported that the maternal chromosome is organised into a CTCF-structured sub-TAD that coalesces the Dlk1 and the Meg3 promoters and that this contributes to the regulation of Dlk1 imprinting (30). Whether the allelic Meg3 expression and/or its methylation status instruct the functionally relevant differential sub-TAD organization is unknown.

To resolve better the sub-TAD organization along the entire imprinted domain, we used allelic Capture-C and Capture Hi-C approaches (Figure 5). The overarching region of interest was ‘captured’ using probes that covered a 650-kb region comprising Dlk1 and the entire Meg3–Rian–Mirg polycistron (see Materials and methods). Similarly as in our recent 4C-seq study (30), in the hybrid WT mESCs we detected a strong sub-TAD on the maternal chromosome that comprised both the Dlk1 gene and the Meg3 promoter (‘Dlk1-Meg3 sub-TAD’) (Figure 5A). The large majority of the polycistron itself is contained within another sub-TAD, which is most prominent on the maternal chromosome as well, and appears delineated by CTCF binding sites in Meg3 promoter region [maternal-specific; (30)] and extends to downstream of the Mirg mRNA cluster (Figure 5A). Importantly, we find that the differential sub-TAD organization between the parental chromosomes persists during ESC differentiation into CMCs (Figure 5B). Concordantly, maternal-specific binding of CTCF was observed at Meg3 promoter (intron-1) in the WT CMCs, as in WT mESCs (30) (Figure 5C).

Figure 5.

Meg3 DMR demethylation causes reorganization of sub-TADs within the Dlk1-Dio3 domain. (A) 3D chromatin organization in hybrid mESCs determined by allelic Capture-C for the maternal (top) and the paternal chromosome (middle). A comparison matrix is shown below (log₂ ratio). Bins are 10-kb. The matrixes are aligned with CTCF ChIP-seq in WT control mESCs. The interrupted lines outline maternal-specific sub-TADs, similarly as described before (30). (B) 3D chromatin organization in hybrid mESC-derived CMCs, as determined by Capture Hi-C. From top to bottom: the maternal allele, the paternal allele and the comparison matrix (log₂ ratio). (C) Left: CTCF ChIP on control and Meg3-TET-26 CMCs. Percentile precipitation was determined by qPCR at ‘binding site 2′ in the Meg3 DMR. Right: Sanger sequencing profiles assess the allele-specificity of CTCF binding at the Meg3 DMR and the KvDMR1. (D) Comparison of the 3D chromatin organization between the paternal allele of hybrid mESC-derived control and Meg3-TET-26 CMCs (Capture Hi-C). In the comparison matrix (log₂ ratio), stronger signal in the TET-26 CMC is shown in red, while stronger signal in control is shown in blue. Dashed black lines outline the maternal-specific sub-TADs. (E) Distribution of log₂ ratios between control and Meg3-TET26 CMCs for all bins comprised in the four zones highlighted in yellow. The percentage of bins with positive/negative log2 ratio is indicated. For the three zones along the x-axis, the histogram values are skewed towards positive values, indicative of increased short-range interactions in Meg3-TET-26 CMCs. On the contrary, the histogram is centered close to zero in the rightmost panel, representing the more distal interactions between the Meg3–Rian sub-TAD and the downstream Mirg sub-TADs.

CTCF binding does not occur when critical CpG dinucleotides within its binding motif are methylated (63,64). We therefore verified if biallelic CTCF binding occurred after inducing hypomethylation at the paternal Meg3 DMR. Indeed, a more pronounced CTCF precipitation was observed in the Meg3-TET-26 CMCs, with CTCF binding now detected at both the maternal and the paternal Meg3 DMR (Figure 5C). This finding shows that the CRISPR induced hypomethylation, which persisted upon differentiation (Figure 4A), led to biallelic CTCF binding to the Meg3 promoter region in the CMCs.

Next, we assessed the impact of the induced loss of methylation, the gain of CTCF binding and the gain of Meg3 expression on the 3D-organization of the paternal chromosome. For this, we compared WT and Meg3-hypomethylated cells. Initial 4C-seq observations in mESCs, using three different viewpoints, showed that the paternal allele of hypomethylated cells exhibited stronger interactions within the Dlk1-Meg3 sub-TAD (Supplementary Figure S7A, yellow highlighted area). Quantification confirmed this visual observation, with the 3D-interaction patterns of the hypomethylated paternal allele being intermediate between those of the maternal allele and those of the WT paternal allele (Supplementary Figure S7B, violet bars versus blue/red bars). To corroborate this finding in CMCs, we used Capture Hi-C to compare the 3D-organization between WT and the Meg3-TET-26 CMCs. This revealed altered interaction frequencies on the paternal chromosome upon loss of Meg3 methylation. Notably, there was an increase in short-range interactions (1.5-fold) in the Meg3-TET-26 CMCs (Figure 5D, positive log2 ratio illustrated by red pixels). This overall trend was observed across the entire domain. This was notably the case in the two sub-TADs that hinge the Meg3 promoter, but also in a sub-TAD downstream of Mirg. Quantifications confirmed this pattern, with the three sub-TADs displaying stronger interaction in Meg3-TET-26 CMCs compared to WT CMCs (Figure 5E, log₂ score skewed for positive values). These data show that the paternal chromosome of the Meg3-TET-26 CMCs adopted a more maternal-like 3D-organization. This finding agrees with the observed increase in repression of Dlk1 on the paternal chromosome, more resembling the maternal chromosome as well (Figure 4D, E).

We conclude that the allelic methylation status of the Meg3-DMR dictates the allelic binding of CTCF, and thereby instructs differential sub-TAD organization, and this correlates with Dlk1 imprinted gene expression.

Similar to CMCs, the organization of the maternal Dlk1-Dio3 domain into sub-TADs was maintained in normal mESC-derived NPCs as well (Supplementary Figure S7C), albeit with a weaker insulation between maternal sub-TADs (Supplementary Figure S7C versus Figure 5B). The premature termination of Meg3 transcription by the insertion of a poly-adenylation site in intron 1 (Figure 1, PAS-A1 cells) did not perturb CTCF binding at the Meg3 DMR, which remained exclusively maternal (Supplementary Figure S7D).

A moderate re-organization of the 3D-organization was observed by Capture Hi-C in the PAS-A1 NPCs on the maternal chromosome. In these neural cells with strongly reduced Meg3 lncRNA expression (Figure 1E), increased short-range interactions were observed within the Dlk1-Meg3 sub-TAD (Supplementary Figure S7E, red pixels illustrative of positive log2 ratio, confirmed by quantifications) while no clear trends were observed in the two other sub-TADs of the domain (Supplementary Figure S7F, equal mix of positive and negative log₂ ratios). The strongly reduced Meg3 RNA levels in PAS-A1 NPCs (Figure 1E) was thus associated with increased short-range interactions only in the sub-TAD upstream of Meg3 TSS. The abundance of Meg3 transcript therefore seemed less critical than that of the Meg3 DMR methylation level for the establishment and maintenance of the parental-specific sub-TADs covering the imprinted domain.

Our study does not address the function of Meg3 on its own, and how this lncRNA could repress the Dlk1 gene in cis. Previously, we reported that the PRC2 component EZH2 -which mediates H3K27me3- is essential for the imprinted expression of Dlk1 (22). Meg3 lncRNA seems to interact with the PRC2-associated proteins EZH2 and JARID2 (65,66), similarly as was reported for many other long nascent RNAs (67). These observations suggested that Meg3 could control Dlk1 by influencing PRC2-mediated H3K27me3. To explore this possibility, we performed ‘CUT&RUN’ experiments to assess H3K27me3 levels at Dlk1 in normal and PAS-A1 NPCs (Supplementary Figure S8). H3K27me3 enrichment was detected at the Hoxa11 locus as expected (positive control), compared to IgG and a negative control region (Actin-B) (P= 0.0014 and P= 0.0003 by Kruskal–Wallis test, in control and PAS-A1 NPCs respectively) (Supplementary Figure S8A, C). At the Dlk1 promoter, H3K27me3 enrichment was detected in control and PAS-A1 NPCs, with signal on both the parental chromosomes. However, despite a tendency for higher H3K27me3 signal for the Dlk1 promoter region in PAS-A1, no significant difference was observed. These observations extend an earlier ChIP-based study on ESCs and NPCs, in which depletion of Meg3 did not alter EZH2 levels at the Dlk1 promoter and did not give rise to loss of H3K27me3 (22).

In the Meg3-TET-26 CMCs -in which the Meg3 RNA was expressed and retained on both the parental chromosomes- there was no apparent change in biallelic low H3K27me3 levels at the Dlk1 promoter compared to control CMCs, despite significant enrichment at Hoxa11 (P= 0.0022 and P= 0.0013, by Kruskal–Wallis test) in control and Meg3-TET-26 CMCs respectively (Supplementary Figure S8B, C).

It remains to be determined whether Polycomb group proteins are important to retain the Meg3 lncRNA on the maternal chromosome -possibly in conjunction with the observed 3D chromatin structural interactions- and whether Meg3 lncRNA acts on regulatory protein complexes at the Dlk1-Dio3 domain, including the PRC1 complex.

Discussion

The main finding of this study is that Dlk1 imprinting requires the lncRNA Meg3, and not the Rian snoRNAs, nor the act of transcription by itself alone. In the NPCs that had a poly(A) signal inserted into intron-1 of Meg3, the remaining lncRNA expression was insufficient to repress Dlk1 on the maternal chromosome. In cardiomyocytes, a lesser reduction in lncRNA was achieved, and this had a concordant, smaller, effect on Dlk1 imprinting. Our studies thus indicate that a certain level of lncRNA is required to prevent the transcriptional activation of Dlk1 in-cis during differentiation. This provides a mechanistic model (Figure 6) as to why epigenetic and genetic changes affecting the Meg3–Rian–Mirg polycistron correlated with aberrant Dlk1 expression in earlier mouse studies (6–8,10,20,22), and has implications for understanding human imprinting disorders that are linked to aberrant MEG3 expression.

Figure 6.

Model of how Meg3 lncRNA expression controls imprinted gene expression. Our study provides two main insights: (A) Meg3 lncRNA expression—and not Rian—controls the maternal allele-specific repression of the Dlk1 gene during differentiation, and (B), Hypomethylation of the Meg3-DMR (on the active maternal Meg3 allele) enhances the binding of CTCF to this DMR, thereby inducing a sub-TAD that brings Dlk1 in close proximity to Meg3 (and its lncRNA). (C) A model combining this study's and earlier insights: on the maternal chromosome, the ICR (called IG-DMR) is unmethylated and is an enhancer that activates the close-by Meg3-Rian-Mirg polycistron. Meg3 lncRNA and its partial retention on the locus control the allelic repression of Dlk1, and this may involve functionally important interactions with components of Polycomb Repressive Complexes (PRCs) (22,54,65,66) and/or other chromatin regulatory proteins (orange diamonds). Polycistron transcription downstream of Meg3 prevents transcription of the developmental Rtl1 gene, possibly through transcriptional interference. On the maternal chromosome, Meg3 promoter activity protects against de novo acquisition of DNA methylation. It remains unmethylated and CTCF can bind to multiple recognition motifs at this DMR. As shown recently (30), the latter mediates about a sub-TAD structure that brings Dlk1 in close proximity to Meg3 (and Meg3 lncRNA) and these structural interactions contribute to the lncRNA-mediated Dlk1 repression. On the paternal chromosome, the domain's ICR is fully methylated and lacks enhancer activity. The polycistron is thus not activated, leading to acquisition of Meg3 de novo methylation in the early embryo (80). The Meg3-DMR methylation, in turn, prevents CTCF binding and sub-TAD structuration. The lack of Meg3 lncRNA and absence of polycistron transcription allow the developmentally controlled Dlk1 and Rtl1 activation on this parental chromosome. In panel C, only CTCF binding sites at confirmed boundaries are depicted, including only one of the two CTCF sites upstream of Dlk1 shown in panel B.

In a recent study on two other nuclear imprinted lncRNAs—Kcnq1ot1 and Airn—the level of lncRNA expression correlated with the degree of in-cis repression, and such a dosage effect was suggested to be linked to the formation of lncRNA-protein aggregates, possible involving liquid-liquid phase separation (68). Meg3, Airn and Kcnq1ot1 interact with components of Polycomb Repressive Complexes (PRCs), and these could thus be part of such aggregates (22,66,68–70). Notably, the PRC1 complex, via its subunit CBX2, can undergo phase separation and contributes to the formation of condensates (71,72). Further studies are required to determine if such aggregates form at the Dlk1-Dio3 locus and to dissect the relative contribution of PRC1 and PRC2 and Meg3 lncRNA in this process.

Another key finding is that the Meg3-DMR methylation level instructs the structural organization of the Dlk1-Dio3 domain, and mediates the formation of maternal chromosome-specific sub-TADs (Figure 6). We explored this aspect using a parental allele-specific Capture Hi-C approach. The activity of the Meg3-promoter keeps the CTCF binding sites of the Meg3-DMR unmethylated on the maternal chromosome, leading to CTCF binding on the maternal allele only (30,73). We, and others, demonstrated that this creates a maternal boundary that hinges the Dlk1-Meg3 sub-TAD and a neighbouring sub-TAD comprising the ncRNA polycistron (7,30). We find here that the distinct sub-TAD organisation on the maternal chromosome is maintained during differentiation into CMCs or NPCs.

Importantly, in the Meg3-TET cardiomyocytes, in which the Meg3 DMR was hypomethylated and active now on both the parental chromosomes, there was biallelic CTCF binding to the Meg3 DMR (intron-1 region). Concordantly, in these differentiated cells the paternal chromosome acquired a sub-TAD structuration that was similar to that on the maternal chromosome, and this correlated with biallelic repression of Dlk1 and Rtl1. Our results thus demonstrated that the methylation level of the Meg3-DMR is both instructive for CTCF binding, Meg3 expression, sub-TADs structuration and the repression of paternally-expressed genes (Figure 6).

Notwithstanding the obtained insights, the Meg3-TET mESCs were sub-optimal since they still had some residual methylation at the Meg3-DMR. To set up the CRISPR-dCas9-SunTag-TET technology, we had initially demethylated the H19 ICR in our hybrid ESCs. Similarly as in other studies (38,61), we achieved significant demethylation at the H19 ICR, but also detected about 10% residual methylation (Supplementary Figure S5). Our data suggest the possibility that at the H19 ICR and the Meg3 DMR the achieved demethylated state was not fully stable, leading to a WT methylation pattern in a small fraction of the cells. In addition, following differentiation of the Meg3-TET ESCs into NPCs, we noted that the Meg3 DMR methylation level had become normal again, which prevented us from doing further studies on neural cells. In our system we could not determine whether the latter artefact was caused by re-acquisition of de novo methylation (61), or by positive selection of cells with normal Meg3 methylation during the differentiation process. Meg3 hypomethylation was stable upon differentiation into CMCs, however, suggesting that there had not been an advantage of a given methylation state in this embryonic lineage. More generally, our studies using the CRISPR-dCas9-SunTag-TET technology highlight the importance to use a culture medium that prevents acquisition of de novo DNA methylation in ESCs (34), and to ascertain that the achieved hypomethylation is stable in cell types of interest.

Combined, our different Meg3 mutants clarified some of the functional interplays between Meg3 expression, the 3D-organization of the locus and Dlk1 regulation. In differentiated neural PAS-A1 cells, which had reduced levels of the lncRNA due to premature transcription termination within the first intron, we noticed unaltered allelic CTCF binding at the maternal Meg3-DMR. Concordantly, the sub-TADs remained largely similar, with even increased occurrence of short-range structural interactions within the sub-TAD upstream of Meg3-DMR on the maternal chromosome (Supplementary Figure S7). This suggests that the Meg3 lncRNA—which is partly retained in cis and spatially co-localizes with the Dlk1 locus (22)—is largely dispensable for shaping the Dlk1-Meg3 sub-TAD. Yet, we observed defective Dlk1 repression in differentiated cells (Figure 1F), which may be linked to interactions between Meg3 RNA and components of PRC complexes, possibly through lncRNA-protein aggregates (3,54,65,66). We conclude that the accumulation of the Meg3 lncRNA is thus dispensable for CTCF binding at the Meg3-DMR and the maintenance of the maternal-specific sub-TADs. Further studies would be required to formally assessed if the RNA binding domains of CTCF—which appear important for some aspects of the higher order folding of chromatin (74,75) — are contributing to the maternal-specific sub-TADs insulation at the Dlk1-Dio3 domains. Yet, our results already show that Meg3 lncRNA interplay with CTCF differs from HOTTIP lncRNA, which enhances CTCF recruitment at a subset of TAD boundaries thereby increasing their insulation (76), and also differ from Jpx lncRNA whose recruitment to chromatin antagonises CTCF recruitment to specific binding sites (77).

Our data on the Meg3-TET cells provide also additional insights on the imprinted Rtl1 gene—an important regulator of placental and muscle development (40,78). In these cells, the CRISPR-induced expression of the Meg3 polycistron on the paternal chromosome lead to a complete loss of Rtl1 expression. This strongly suggests that—similarly as on the maternal chromosome—transcription of the polycistron across the Rtl1 no longer allowed this gene to be expressed. Besides such a transcriptional interference effect, the reduction of Rtl1 mRNA levels may also result from increased levels of miRNAs, produced downstream of Meg3, that target the Rtl1 mRNA (16,40,78).

More generally, our study highlights the antagonism between the parental genomes at the Dlk1-Dio3 imprinted domain. A sperm-derived paternal DNA methylation imprint at the intergenic IG-DMR keeps this ICR inactive on the paternal chromosome, thus preventing activation of the close-by Meg3–Rian–Mirg polycistron on this parental chromosome (7,79). The non-activated paternal Meg3 promoter, consequently, becomes de novo methylated in the early embryo (7,10,80). On the maternal chromosome, in contrast, the intergenic ICR is unmethylated and transcriptionally active, producing enhancer RNAs (eRNAs) (8). The enhancer activates the maternal Meg3 promoter, leading to the expression of Meg3 lncRNA, Rian snoRNAs and the Mirg and other miRNAs produced by the polycistron. The conserved lncRNA Meg3 (81) itself is important for the repression of Dlk1 on the maternal chromosome, as shown in this study. Consequently, the transcriptional activation of Dlk1 during cell differentiation occurs on the paternal chromosome only. As an additional layer of antagonism, several of the miRNAs produced by the maternally expressed Meg3-Rian-Mirg polycistron reduce the protein levels of the paternally expressed genes (82). For instance, one miRNA of the Mirg locus (miR-329) targets the 3′ UTR of Dlk1 mRNA and inhibits its translation, thereby reducing DLK1 protein levels in neural cells (83).

In conclusion, different mechanisms evolved through which maternally-expressed non-coding RNAs antagonise paternally-expressed protein-coding genes, to limit their effects on growth, development and metabolism. Possibly affecting the same biological processes, several miRNAs of the domain antagonise paternally-expressed imprinted genes elsewhere in the genome (82). In neural cells, for instance, several of the Mirg cluster miRNAs reduce post-transcriptionally the expression of developmental paternally-expressed genes, including Igf2 at the Igf2-H19 domain, and the imprinted Plagl1 transcription factor gene (52,82).

Our data may help to understand the molecular etiology of human imprinting disorders that are associated with altered MEG3 expression. Particularly, we predict that in patients with Temple Syndrome (TS14) (29), and in some patients with the clinically overlapping Silver-Russell Syndrome (SRS) (24,84), their aberrant, biallelic MEG3 expression would lead to strongly reduced DLK1 in tissues in which this non-canonical Notch ligand is imprinted. In addition, we predict loss of RTL1 expression in muscle and other tissues where this retrotransposon-derived gene is expressed. In Kagami-Ogata Syndrome (KOS14), conversely, the loss of MEG3 expression would correlate with increased DLK1 and RTL1 expression, now from both the parental chromosomes. In mice, DLK1 deficiency leads to fetal growth restriction and to metabolic and endocrinal defects (85–87). RTL1 deficiency in mice affects placental development and function (88) and leads to distinct muscle abnormalities. DLK1 and RTL1 are therefore likely main contributors to TS14 and KOS14, and recent clinical studies have started to explore this question (57,89).

Notes

Present address: Sabina Farhadova, Department of Neuroscience, University of Wisconsin-Madison, Wisconsin, WI 53705, USA.

Present address: Amani Ghousein, Centre de Biologie Structurale and MIVEGEC, University of Montpellier, 34090 Montpellier, France.

Present address: Melisa Gomez-Velazques, Helmholtz Zentrum Muenchen, 85764 Neuherberg, Germany.

Data availability

All sequencing data (4C-seq, Capture-C, Capture Hi-C, RNA-seq and RRBS) are available from the European Nucleotide Archive (EMBL-EBI ENA) repository under accession number PRJEB57653. The CTCF ChIP data used in this study are available in the NCBI GEO database, under accession number GSE207166.

Supplementary data

Supplementary Data are available at NAR Online.

Funding

Agence National de Recherche [ANR-18-CE12-0022-02 IMP-REGULOME to R.F. and D.N., ANR-22-CE12-0016-03 IMP-DOMAIN to R.F. and D.N., ANR-21-CE12-0034-01 to D.N., ANR-16-IDEX-0006 to M.B.]; LabEx EPIGENMED—an ANR ‘Investissement d’avenir’ programme [ANR-10-LABX-12-01 to R.F.]; Fondation pour la Recherche Médicale (FRM) [EQU202103012763 to R.F., AJE202005011598 to M.B.]; PlanCancer [19CS145-00 to D.N.]; CNRS-INSERM ATIP-Avenir Programme (to M.B.); S.F. acknowledges PhD salary funding from the Azerbaijan National Academy of Sciences (ANAS); Ministry of Education of the Azeraijani Republic and the University of Montpellier (to U.M.); C.R. acknowledges PhD Fellowship funding from La Ligue Nationale Contre le Cancer. Funding for open access charge: ANR [ANR-22-CE12-0016-03].

Conflict of interest statement. None declared.