m⁶A RNA methylation orchestrates transcriptional dormancy during developmental pausing

Abstract

Embryos across metazoan lineages can enter reversible states of developmental pausing, or diapause, in response to adverse environmental conditions. The molecular mechanisms that underlie this remarkable dormant state remain largely unknown. Here we show that m⁶A RNA methylation by Mettl3 is required for developmental pausing in mice by maintaining dormancy of paused embryonic stem cells and blastocysts. Mettl3 enforces transcriptional dormancy via two interconnected mechanisms: i) it promotes global mRNA destabilization and ii) suppresses global nascent transcription by specifically destabilizing the mRNA of the transcriptional amplifier and oncogene N-Myc, which we identify as a critical anti-pausing factor. Our findings reveal Mettl3 as a key orchestrator of the crosstalk between transcriptomic and epitranscriptomic regulation during pausing, with implications for dormancy in stem cells and cancer.

Development is often described as a sequential unfolding of genetic programs towards increased complexity, occurring with stereotypical timing. However, many animal species can pause early embryonic development to improve survival under adverse environmental conditions by entering a dormant state called embryonic diapause(1, 2). In mammals, pausing manifests as a delayed implantation of the blastocyst, the source of pluripotent embryonic stem cells (ESCs)(3, 4). This state of paused pluripotency can be induced in mouse blastocysts and ESCs by the inhibition of mTOR, a conserved growth-promoting kinase, and is characterized by a drastic decrease in proliferation and biosynthetic activity, including transcription(5). However, a detailed characterization of the mechanisms regulating transcription in this context is still missing.

To gain insights into the regulation of transcriptional dormancy during pausing, we performed a comprehensive screen for chemical modifications of RNA in paused ESCs. Mass spectrometry revealed significantly increased levels of N⁶-methyladenosine (m⁶A) in paused ESCs, relative to control condition (Fig. 1A, fig. S1A). This modification is deposited on nascent RNA by a methyltransferase complex, with Mettl3 as the catalytically active subunit(6, 7). m⁶A plays essential roles during post-implantation development via mRNA destabilization of key cell fate regulators(8–10). We hypothesized that the increase in m⁶A may indicate an unexpected role for this mark during paused pluripotency.

Fig. 1. The m6A methyltransferase Mettl3 is essential for paused pluripotency.

(A) Screening of RNA modifications by mass spectrometry in paused ESCs, normalized to FBS-grown ESCs. Data are mean ± SD (n=3). Ratio paired Student’s t-tests **P < 0.01. (B) Dot blot showing that the increase of m⁶A in paused ESCs is abrogated in Mettl3^−/−. (C) Growth curves showing that Mettl3^−/− ESCs fail to suppress proliferation in paused conditions. Data are mean ± SEM (n=3). Linear regression test **P < 0.01, ***P < 0.001. (D) Mettl3 loss leads to the premature death of mouse blastocysts cultured ex vivo in paused conditions. Log-rank test **P < 0.01, ****P < 0.0001. (E) Quantification of recovered (live) embryos at E3.5 (control) and at Equivalent Days of Gestation (EDG) 8.5 following hormonal diapause, showing that Mettl3^TCP−/− embryos are impaired at undergoing hormonal diapause. χ2 test *P < 0.05.

Paused ESCs, induced by mTOR inhibition, are viable and pluripotent but proliferate very slowly(5). However, we found that paused Mettl3^−/− ESCs grow at a much faster rate than paused wildtype (Mettl3^+/+) ESCs, suggesting defective suppression of proliferation (Fig. 1B–C, fig. S1B–F). Interestingly, this faster proliferation rate of Mettl3^−/− ESCs is observed only in the paused state and not in control FBS conditions (Fig. 1C, fig. S1D,F), suggesting a specific role in developmental pausing. As we reported previously, inhibition of mTOR prolongs survival of blastocysts ex vivo for 1–2 weeks and induces a paused state(5), a finding reproduced here with Mettl3^+/+ embryos (Fig. 1D). By contrast, we found that Mettl3^−/− blastocysts are prematurely lost during ex vivo pausing (Fig. 1D and see model in fig. S1G–H). Mettl3^−/− embryos are also largely incompatible with hormonally induced diapause (Fig. 1E). Taken together, these findings reveal an essential role for Mettl3 in ESC and blastocyst pausing.

A central feature of paused pluripotency is a global reduction in transcriptional output, or hypotranscription, encompassing both coding and non-coding RNAs(5). In comparison with paused Mettl3^+/+ cells, paused Mettl3^−/− ESCs display increased levels of both total and nascent RNA per cell (Fig. 2A–C, fig. S2A). Levels of nascent RNA are also elevated in paused Mettl3^−/− blastocysts, as compared with Mettl3^+/− or Mettl3^+/+ embryos (Fig. 2D–E). We next performed cell number-normalized RNA-sequencing, which uses exogenous RNA spike-ins and allows for quantification of global shifts in transcriptional output(11) (fig. S2B, Table S1, and see Methods). In line with the global changes observed in RNA levels (Fig. 2A–C), paused Mettl3^−/− ESCs displayed a defective hypotranscriptional state (Fig. 2F, fig. S2C–E). Indeed, while 10,656 genes are downregulated in paused Mettl3^+/+ cells, only 5,916 genes (i.e. 55.5%) are downregulated in paused Mettl3^−/− cells (fig. S2D). This suppression of hypotranscription is particularly evident for pathways and functional categories whose silencing is a feature of developmental pausing, such as translation, ribosome biogenesis, mTOR signaling, Myc targets and energy metabolism(4, 5, 12–14) (Fig. 2G–H, fig. S2F–G, Table S2). These results reveal that Mettl3 contributes to the global transcriptional dormancy observed during developmental pausing.

Fig. 2. Mettl3 regulates hypotranscription in paused pluripotency.

(A) Total RNA per cell in Mettl3^+/+ and Mettl3^−/− ESCs, in FBS and paused conditions. Data are mean ± SD, (n=3). Paired Student’s t-tests **P < 0.01. (B,C) Nascent transcription by EU incorporation, quantified by median fluorescence intensity (MFI) relative to Mettl3^+/+ FBS in each experiment. Data are mean ± SD (n=4). Paired Student’s t-tests **P < 0.01. (D,E) Immunofluorescence and nuclear signal quantification of EU incorporation in ex vivo paused blastocysts, showing increased nascent transcription in Mettl3^TCP−/−. Data are mean ± SD. One-way ANOVA with Dunnett’s multiple comparison test ****P < 0.0001. (F) Heatmap of gene expression for all genes expressed in ESCs, showing defective hypotranscription in paused Mettl3^−/− ESCs. (G,H) Gene set enrichment analysis (GSEA) of gene expression changes in paused ESCs, using the “GO biological processes” (G) and “hallmarks” collections (H). Scatter plots of the normalized enrichment scores (NES), with Spearman correlation coefficient (ρ) with representative pathways showing defective hypotranscription in Mettl3^−/− (red dots) highlighted. pre-ranked gene set enrichment analysis with FDR correction *P < 0.05, **P < 0.01, ****P < 0.0001.

To probe the mechanism by which Mettl3 regulates pausing, we mapped m⁶A modifications by methylated RNA immunoprecipitation followed by sequencing (MeRIP-seq), using a cell number-normalization approach (fig. S3A–E, and see Methods). We identified 15,046 m⁶A peaks within 7,095 genes, which represents 48% of all genes expressed in control or paused ESCs (Table S3). Principal Component Analysis (PCA) revealed that paused ESCs are in a distinct state with regards to the m⁶A RNA profile (Fig. 3A). Consistent with the quantitative mass spectrometry analysis (Fig. 1A), MeRIP-seq showed a global increase in m⁶A in paused ESCs, with 1,562 peaks significantly hypermethylated versus only 249 regions being hypomethylated relative to control ESCs (Fig. 3B–C, fig. S3F–G).

Fig. 3. Mettl3-mediated m6A promotes RNA instability during pausing.

(A) PCA plot showing that paused ESCs have distinct m⁶A profiles (n=3). (B,C) MeRIP-seq shows increased m⁶A in paused ESCs (n=3, gain/loss: adjusted P < 0.05 and absolute fold-change >1.5). (D) Heatmaps of Mettl3 ChIP-seq signal in FBS-grown and paused ESCs, showing increased Mettl3 binding in paused ESCs (n=2). (E) Overlap between targets of m⁶A and Mettl3, identifying all target genes or genes with increased levels of m⁶A and Mettl3 (log₂FC > 0) in paused ESCs. P-value by hypergeometric test. (F) Metagene profiles of m⁶A and Mettl3 peaks. (G) RNAs with increased m⁶A in pausing, as defined in (A), are significantly more downregulated than RNAs with decreased m⁶A in paused Mettl3^+/+ (but not Mettl3^−/−) ESCs. Student’s t-tests **P < 0.01. (H) Differences in expression (log₂FC paused/FBS) between exonic and intronic RNA-seq data indicate a global decrease in RNA stability in Mettl3^+/+ (but not Mettl3^−/−) ESCs upon pausing. P-value by paired Student’s t-test.

To understand the gain in m⁶A in paused ESCs, we investigated the levels of Mettl3 protein. Interestingly, despite no change in whole cell levels, Mettl3 is increased in the chromatin compartment upon transition to the paused state (fig. S4A–B). Mettl3 has previously been shown to bind to chromatin in ESCs and cancer cells, where it deposits m⁶A co-transcriptionally(15, 16). We therefore mapped the genome-wide distribution of Mettl3 by chromatin immunoprecipitation-sequencing (ChIP-seq) (fig. S3C–D). As anticipated, we observed higher levels of Mettl3 occupancy in paused ESCs relative to control ESCs (Fig. 3D). Although Mettl3 binds extensively throughout the genome, it is more abundant over expressed genes, particularly if their mRNAs are also m⁶A methylated (fig. S4E). The majority of m⁶A methylated RNAs (4,922/7,095, 69.4%) are transcribed from genes occupied by Mettl3 in ESCs, and transcripts gaining m⁶A during pausing largely arise from genes with elevated Mettl3 binding in paused ESCs (3,505/5,886, 59.6%, Fig. 3E). In agreement with recent reports(15, 17), Mettl3 localizes mainly to the transcriptional start site (TSS), while m⁶A is enriched near the stop codon and 3’ untranslated region (UTR) of coding genes in ESCs (Fig. 3F). Overall, these results indicate that an increased chromatin recruitment of Mettl3 underlies the gains of m⁶A RNA methylation during paused pluripotency, although we cannot exclude that Mettl3 may also exert other functions at chromatin.

To dissect the function of m⁶A RNA methylation during pausing, we asked how changes in m⁶A impact mRNA levels during ESC pausing. In the context of global hypotranscription in wildtype ESCs upon pausing ((5) and Fig. 2F), RNAs with increased m⁶A are significantly more downregulated than RNAs with decreased m⁶A (Fig. 3G). However, this association is lost in Mettl3^−/− cells, when analyzing the same genes (Fig. 3G). These data suggest that Mettl3-mediated m⁶A RNA methylation may contribute to transcript destabilization during pausing. In order to further explore this possibility, we re-analyzed the RNA-seq data from control and paused Mettl3^+/+ and Mettl3^−/− ES cells to assess post-transcriptional regulation (Fig. 2F and fig. S2). Exonic reads reflect steady-state mature mRNAs, whereas intronic reads mostly represent pre-mRNAs, and therefore comparing the difference between these has been shown to effectively quantify post-transcriptional regulation of gene expression(18) (fig. S5A–C). This analysis pointed to a global destabilization of the transcriptome when Mettl3^+/+ ESCs are induced to the paused state (Fig. 3H). In contrast, this global destabilization effect of pausing is entirely lost in Mettl3^−/− ESCs (Fig. 3H). Taken together, these results indicate that Mettl3-dependent m⁶A methylation is responsible for a global destabilization of the transcriptome in paused ESCs.

Our findings to this point indicate that the transcriptionally dormant state of paused cells is imparted by a combination of reduced nascent transcription and increased transcript destabilization, effects that are muted in Mettl3^−/− paused ESCs (Fig. 2C, fig. S5D–E). We hypothesized that m⁶A may contribute to destabilizing mRNAs encoding putative “anti-pausing” factors. To identify such factors, we mined the RNA-seq and MeRIP-seq data for genes that i) gain m⁶A in paused ESCs; ii) are downregulated upon pausing but to a lesser extent in Mettl3^−/− ESCs; iii) are expressed at least 2× higher in paused Mettl3^−/− ESCs relative to control paused ESCs; and iv) are destabilized in paused ESCs in a Mettl3-dependent manner. This analysis identified 953 candidate anti-pausing factors (fig. S5F, see Methods). We then took advantage of published data from early mouse embryos(4) to rank the candidates by their correlation with an expression signature of the m⁶A machinery. We reasoned that, if these candidate genes are regulated by m⁶A in vivo, their expression should be anti-correlated with expression of the methyltransferase complex and correlated with expression of the m⁶A demethylation machinery (Fig. 4A, Table S5, see Methods for details). Remarkably, the top-ranked candidate from this analysis is Mycn, which codes for the N-myc proto-oncogene (N-Myc) and is highly expressed in both ESCs and embryos (fig. S6A). The Myc-regulated set of genes is a major module of the ESC pluripotency network and it is downregulated in diapause(4, 13). Importantly, Myc family members can act as global transcriptional amplifiers in the context of development and cancer(19, 20). We therefore investigated in-depth the regulation of Mycn by m⁶A RNA methylation and its potential impact in paused ESCs.

Fig. 4. Mettl3 regulates pausing via m6A-mediated destabilization of Mycn mRNA.

(A) Top putative anti-pausing factors ranked by Spearman correlation (ρ) with the m⁶A machinery. (B) Increased N-Myc in Mettl3^−/− ESCs by RNA-seq (n=3) and western blot (n=4). (C) Immunofluorescence and nuclear signal quantification of N-Myc in ex vivo paused blastocysts. (D) Median log₂ fold-changes for “hallmark” gene sets in Myc/Mycn DKO ESCs versus diapaused embryos or paused ESCs, with spearman correlation (ρ). (E) Blocking Myc signaling partially restored the decreases in proliferation (n=4) and total RNA (n=5) in paused Mettl3^−/− ESCs. (F) Increased Mycn mRNA stability in paused Mettl3^−/− ESCs (n=3). t_1/2: half-life. (G) Mycn m⁶A site reduces transcript stability in paused ESCs (n=5). (H-K) Site-specific demethylation of Mycn leads to increased Mycn mRNA stability (H, n=3). increased expression (I, n=7 by RT-qPCR and n=3 by western blot), increased total RNA (J, n=6) and higher proliferation (K, n=5). (L) Model for the regulation of paused pluripotency by m⁶A RNA methylation. All data are mean ± SD, except times series which are mean ± SEM. P-values by two-tailed paired Student’s t-tests (B,C,G), linear regression test (E left,F,H,K), one-way ANOVA with Dunnett’s multiple comparison tests (E right,I,J). *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

We found that N-Myc expression is elevated at the RNA and protein level in paused Mettl3^−/− ESCs and blastocysts (Fig. 4B–C, fig. S6B–D), further supporting its status as an m⁶A target. Additionally, ESCs depleted for both c-Myc and N-Myc (Myc DKO) partially recapitulate gene expression changes in diapaused embryos and paused ESCs(5, 13), but this relationship is entirely abolished in paused Mettl3^−/− ESCs (Fig. 4D). In agreement with this result, the downregulation of Myc target genes that occurs upon pausing is partially suppressed in Mettl3^−/− ESCs (Fig. 2H). We thus wondered whether elevated Myc signaling contributes to the defective pausing observed in Mettl3^−/− ESCs. Accordingly, treatment with the Myc inhibitor 10058-F4 restored the decreases in proliferation and total RNA content in paused Mettl3^−/− ESCs to levels equivalent to paused Mettl3^+/+ cells (Fig. 4E, fig. S6E). The downregulation of Myc target genes in pausing is imparted via reduced nascent transcription, rather than by being directly targeted by m⁶A themselves (fig. S6F–G), consistent with the established role of Myc as a transcriptional activator(19, 20). Taken together, these results indicate that N-Myc levels are downregulated in an m⁶A-dependent manner in paused ESCs, and that the subsequent decrease in Myc signaling results in reduced transcriptional output and proliferation.

We sought to further probe the direct regulation of Mycn mRNA by methylation. The m⁶A mark can affect mRNA metabolism through binding of reader proteins, including the YTH domain family and HNRNP proteins(21). In particular, the reader Ythdf2 can mediate destabilization of m⁶A methylated mRNAs, including in ESCs. In agreement with this, Mycn mRNA is bound by both Mettl3 and Ythdf2 in paused ESCs, and this binding is abolished by knockout of Mettl3 (fig. S7A). Moreover, the half-life of Mycn mRNA is significantly increased in Mettl3^−/− ESCs (Fig. 4F), specific to Mycn among the Myc family members (fig. S7B–C). These results corroborate that m⁶A methylation regulates Mycn mRNA stability in paused ESCs. After identifying a hypermethylated m⁶A site near the stop codon of the Mycn mRNA (fig. S8A–B), we found that this site confers transcript destabilization in paused ESCs in a luciferase reporter assay, in a manner dependent on the integrity of the m⁶A site as an A to C mutation nullifies this effect (Fig. 4G).

Finally, we explored how RNA methylation impacts Mycn transcript stability. We performed site-specific RNA demethylation using dCasRx-conjugated ALKBH5(22) (fig. S8C–G). We found that targeted demethylation of the 3’ end of Mycn mRNA in Mettl3^+/+ paused ESCs stabilizes the transcript, leading to higher steady-state levels of N-Myc mRNA and protein (Fig. 4H–I, fig. S8H). Remarkably, this targeted loss of m⁶A at the Mycn mRNA is sufficient to increase levels of total RNA and proliferation in paused ESCs (Fig. 4J–K). Thus, m⁶A demethylation of Mycn mRNA in otherwise wildtype ESCs recapitulates the suppression of pausing observed in Mettl3^−/− ESCs.

In summary, we show that Mettl3-dependent m⁶A RNA methylation is required for developmental pausing by maintaining transcriptional dormancy (Fig. 4L). Our findings shed light on the molecular mechanisms that underlie mammalian developmental pausing and reveal Mettl3 as a key integrator between transcriptomic and epitranscriptomic levels of gene regulation. m⁶A RNA methylation was recently reported to modulate the transcriptional state of ESCs by destabilizing chromosome-associated RNAs and transposon-derived RNAs(23–25) and by promoting the recruitment of heterochromatin regulators(17). Our results support the notion that Mettl3 acts at chromatin and methylates RNA co-transcriptionally(15, 16). Surprisingly, even though Mettl3 methylates thousands of transcripts, we found that one target, Mycn mRNA, is key for its function in maintaining the suppressed transcriptional state of paused cells. Future studies may uncover additional functions of other m⁶A-regulated putative anti-pausing factors identified here.

We anticipate that the regulatory relationship between m⁶A RNA methylation and cellular dormancy will have implications extending well beyond embryonic diapause. Modulations of mTOR signaling have been implicated in the control of stem cell dormancy in various embryonic and adult tissues(26). Moreover, we and others have shown that cancer cells can enter a dormant state molecularly and functionally similar to diapause to survive chemotherapy(27, 28). The insights gained here provide exciting new opportunities to explore the biology of m⁶A RNA methylation in the fields of developmental biology, reproductive health, regenerative medicine, and cancer.

Materials and Methods

Mouse embryonic stem cell culture

E14 ESCs were grown on gelatin-coated plates in standard serum/LIF medium: DMEM GlutaMAX with Na Pyruvate, 15% FBS (Atlanta Biologicals), 0.1 mM Nonessential amino acids, 50U/ml Penicillin/Streptomycin, 0.1mM EmbryoMax 2-Mercaptoethanol, and 1000 U/ml ESGRO LIF. For FBS/2i culture, the medium was supplemented with 1μM PD0325901 and 3μM CHIR99021. Pausing was induced by adding 200nM INK128 to the medium, as described(5). Unless otherwise stated, ESCs were paused for at least 5 days before use, and for exactly 2 weeks for all sequencing experiments.

Mettl3^−/− ESC models

Mettl3^−/− and control Mettl3^+/+ ESCs were kindly provided by J. Hanna(8) and used for most experiments, except where indicated. An independent ESC line mutant for Mettl3 (Mettl3^{−/− #2}) cells was generated via CRISPR/Cas9 for validation of key results. Cloning was performed by annealing oligos targeting Mettl3 into pSpCas9(BB)-2A-GFP (PX458), a gift from Feng Zhang (Addgene plasmid #48138; RRID:Addgene_48138)(29). E14 were transfected with lipofectamine 2000, isolated by FACS, clonally expanded and validated for Mettl3 loss.

Site-specific m⁶A demethylation

Lentivirus was produced by transfecting HEK293T cells with pMSCV-dCasRx-ALKBH5-PURO and viral packaging/envelope vectors pMD2.G and psPax2, gifts from Qi Xie and Didier Trono (Addgene plasmid #175582, #12259 and #12260; RRID:Addgene_175582; RRID:Addgene_12259, RRID:Addgene_12260)(22). E14 cells were infected with pMSCV-dCasRx-ALKBH5-PURO lentivirus, and after 48h were selected with 2μg/mL puromycin. Clonal lines were expanded and selected for expression of dCasRx-ALKBH5 by western blot. Guide RNAs were cloned using lenti-sgRNA-BSD, a gift from Qi Xie (Addgene plasmid #175583; RRID:Addgene_175583)(22). Each guide plasmid (1μg) was transfected with Lipofectamine 2000 into dCasRx-ALKBH5-expressing cells in a 6-well plate. Cells were selected with 8μg/mL blasticidin for 3 days before use. See Table S6 for primer sequences.

Mouse models.

The mouse C57BL/6NCrl-Mettl3^em1(IMPC)Tcp allele was generated as part of the Knockout Mouse Phenotyping Program (KOMP2) project at The Centre for Phenogenomics (TCP) by Cas9-mediated deletion of a 142 bp region (Chr14:52299764 to 52299905 in ENSMUSE00001224053, GRCm38), causing a frameshift and early truncation (I171Mfs*4). Mice were purchased from the Canadian Mouse Mutant Repository through TCP. Heterozygotes mice (referred to as “Mettl3^TCP+/−”) were maintained on a hybrid C57BL/6×CD-1 background in order to increase litter size and ex vivo pausing efficiency. Genotyping of mice was performed by Transnetyx. All mice were housed at TCP in Toronto. All procedures involving animals were performed according to the Animals for Research Act of Ontario and the Guidelines of the Canadian Council on Animal Care. All procedures conducted on animals were approved by The Animal Care Committee at TCP (Protocol 22–0331). Sample size was not predetermined.

For all embryo experiments, 6 to 12-week-old Mettl3^{TCP +/−} females were mated with 6-week- to 8-month-old Mettl3^{TCP +/−} males. Collection and ex vivo culture of blastocysts was performed as previously described(5), with pausing induced by flushing blastocysts at E3.5 and culturing them at 5% O2, 5% CO2 at 37 °C in KSOM^AA, with addition of 200nM RapaLink-1 on the day after flushing. Blastocysts with collapsed blastocoel were considered non-viable and collected for genotyping every day. Hormonal diapause was induced as previously described(5), and blastocysts were collected at EDG8.5 and genotyped. For genotyping of embryos, DNA was extracted from individual blastocysts using the Red Extract-N-Amp kit (Sigma), in 36μl final volume. Mettl3 status was assessed by PCR using 1μl DNA extract in 15μl total volume reaction with Phire Green Hot Start II PCR Master Mix (Thermo Fisher). Cycling conditions: 98°C for 30 s; 35 cycles of 98°C for 5s, 57°C for 5s, 72°C for 5s; 72°C for 1 min. See Table S6 for primer sequences.

Cell number-normalized (CNN) RNA analyses

RNA was extracted from equal number of ESCs, typically ~2×10⁵ cells, with RNeasy Micro Kit with on-column DNase I digestion (QIAGEN). RNA concentration was measured by Qubit^™ RNA High Sensitivity to calculate the amount of total RNA per cell. cDNAs were generated using equal volumes of extracted RNAs with the SuperScript IV VILO Master Mix (Thermo Fisher) and qPCR data were acquired on a QuantStudio 5 (Thermo Fisher Scientific). Unless otherwise stated, gene expression was normalized according to cell number(19). See Table S6 for primer sequences.

CNN RNA-sequencing and data analysis

RNA extracted from equal number of ESCs was spiked with synthetic RNAs from the External RNAs Control Consortium (ERCC) Spike-in Mix1 (Thermo Fisher), by adding 2μl of 1:100 ERCC dilution to 10μl of RNA (equivalent to ~1–2μg). Library preparation was done using the NEBNext Ultra II Directional Library Prep Kit for Illumina with the mRNA Magnetic Isolation Module from 1μg RNA, per manufacturer’s instructions (NEB). Library quality was assessed by Fragment Analyzer NGS (Agilent). Sequencing was performed on a NextSeq500 (Illumina) with 75bp single-end reads at the Lunenfeld-Tanenbaum Research Institute Sequencing Facility.

Libraries were trimmed of adaptors and quality-checked using Trim Galore! v0.4.0, and then aligned to the mm10 transcriptome with ERCC sequences using TopHat2 v2.0.13. Gene counts were obtained from the featureCounts function of subread (v1.5.0) with options: -t exon -g gene_id. Raw counts were imported into R, and normalized to ERCCs with edgeR (v3.32.1) as previously described(19, 30). Data were further analyzed using tidyverse v1.3.0 and plotted using ggplot2 v3.3.5. The threshold for significant differential expression was adjusted P < 0.05 and absolute fold-change > 1.5. Normalized counts were mean-centered per batch and log-transformed for PCA and heatmaps. GSEA was performed using the fGSEA package v1.16.0, pre-ranking genes by t-values from the differential analysis (paused/FBS). Gene set collections were downloaded from the Molecular Signatures Database v7.5.1 (http://www.gsea-msigdb.org/gsea/msigdb/index.jsp). For intron analysis, RefSeq-annotated intronic regions were shortened by 50bp on each side to limit overlap with exonic regions, then counts were obtained with featureCounts and analyzed in R as for exons.

Global m6A quantification

For quantification of m⁶A in poly(A) RNA, nucleoside digestion was performed as previously described(31). Separation was accomplished by reversed phase chromatography with an Acquity UPLC HSS T3 (Waters) on a Vanquish^™ Flex Quaternary UHPLC system (Thermo Fisher Scientific). Mass spectrometry was performed on a Quantiva^™ triple quadrupole mass spectrometer interfaced with an H-ESI electrospray source (Thermo Fisher Scientific). Data were analyzed with Tracefinder 4.1 (Thermo Fisher Scientific) and Qual browser of Xcalibur 3.0. The mass transitions (precursor → product) for m⁶A were 282 → 94, 282 → 123 and 282 → 150.

Changes in m⁶A were also measured by dot blot from 50ng of poly(A) RNA. Blotting was performed as previously described for 5-hydromethylcytosine(32, 33), except that Diagenode C15410208 (1:400) was used as primary antibody.

m⁶A MeRIP-seq

ESCs were spiked with 2% of human cells (HeLa, CVCL_0030), then poly(A) RNA was extracted using the Magnetic mRNA Isolation Kit (NEB). Immunoprecipitation of methylated RNA (MeRIP) was done using the EpiMark^® N6-Methyladenosine Enrichment Kit with 4μg spiked poly(A) RNA. Specificity of the IP was tested by spiking samples with exogenous RNA controls (m6A modified and unmodified), per manufacturer’s instructions. The cell number normalization approach with human cells was validated beforehand by spiking ESCs with 1, 2 or 4% human cells to simulate global changes in methylation (see fig. S3b). m⁶A enrichment was measured by RT-qPCR in 3 mouse mRNAs (Neurod1, Nr5a2, Sox1) known to be methylated in ESCs(34) and normalized to the average levels of 5 highly-expressed and methylated human mRNAs (HSBP1, PCNX3, GBA2, ITMB2, PCBP1)(35). MeRIP libraries were constructed from 0.5–1ng of input or IP RNA and prepared using the SMARTR-seq RNA library prep v2 kit (TakaraBio), per manufacturer’s recommendations. Library quality was assessed with the High Sensitivity DNA Assay on an Agilent 2100 Bioanalyzer (Agilent Technologies), and samples were sequenced on a HiSeq 4000 using single-end 50 bp reads at the UCSF Center for Advanced Technology.

Pre-processing of sequencing data was performed similarly to RNA-seq, but without ERCCs and with reads unmapped to mm10 being aligned to hg19. For input samples, gene expression was normalized as for CNN RNA-seq, except that the ratio of hg19/mm10 reads was used for normalization instead of ERCCs. For m⁶A RIP samples, peaks were called with MACS2 (using IP samples and their input counterpart as controls and q<0.01). Peaks were annotated by intersecting center positions with RefSeq annotations. Peak analysis was performed using DiffBind v3.0.15, with the options minOverlap=2, score=DBA_SCORE_READS. MeRIP peaks were then first normalized using the ratio of hg19/mm10 reads in each sample for normalization, then adjusted by dividing values by the ratio Input_sample/Input_average of the corresponding gene to consider expression changes. In the following differential analysis with edgeR, these normalized m⁶A levels were protected from further re-scaling by fixing the library size for all samples as lib.size = rep(10^6, 6) in the voom function. The threshold for significant differential methylation was adjusted P < 0.05 and absolute fold-change >1.5. For motif analysis, all peaks were limited to 100bp surrounding the center and submitted to DREME of the Meme-suite (http://meme-suite.org). Bigwig files were generated by Deeptools v3.3.0 and visualized in Integrated Genome Viewer (IGV v2.9.4), with the vertical scale being adjusted to consider expression changes individually for each gene (similar to the m⁶A value adjustment performed before the differential analysis).

Site-specific quantification of m⁶A

We identified a putative m⁶A site within the Mycn MeRIP peak using the m⁶A-Atlas database (http://www.xjtlu.edu.cn/biologicalsciences/atlas)(36). We then measured m⁶A by RT-qPCR, following a method previously described(37, 38), which takes advantage of the diminished capacity of Bst enzymes to retrotranscribe m⁶A residues compared to MRT control enzyme, and RT primers targeting just before or after the site (primer + or −). Each cDNA was generated with ~100ng of total RNA, 100nM primer (+ or −), 50μM dNTPs and 0.1U of Bst3.0 (NEB) or 0.8U of MRT (ThermoScientific). The cycling conditions were 50°C for 15min, 85°C for 3min, then 4°C. RT-qPCR data were then acquired on a QuantStudio 5 (Thermo Fisher Scientific) and normalized as [2^−(Ct_Bst− − Ct_MRT−) − 2^-(Ct_Bst+ − Ct_MRT+)] / 2^−(Ct_Bst− − Ct_MRT−). Negative values were considered below the detection threshold and rounded to 0.

Mettl3 ChIP-seq

ESCs were spiked with 2% of human cells (HeLa), then cross-linked in 1% formaldehyde/PBS for 10min at room temperature. After quenching with 125mM glycine for 5min at room temperature, followed by 15min at 4°C, cells were washed in cold PBS and stored at −80 °C. Cells were diluted at 5 million cells per 100μl in shearing buffer (1% SDS, 10mM EDTA, 50mM Tris-HCl pH8.0, 5mM NaF, Halt^™ Protease Inhibitor Cocktail (Thermo Fisher), 1mM PMSF), rotating at 4°C for 30min, then 100μl of lysate was passed into a microTUBE AFA Fiber Snap-Cap (Covaris). Chromatin was sheared to 200–500 bp fragments on a Covaris E220 with settings PIP 175, Duty 10%, CPB 200, for 7min. Immunoprecipitation was performed overnight using 200μl of each lysate (~chromatin from 10 million cells) and 5μg of antibody (Proteintech 15073–1-AP), following the iDeal ChIP-seq kit for Transcription Factors (Diagenode) protocol. Elution, de-crosslinking, and DNA purification were performed per manufacturer’s instructions. Libraries were constructed from ~2ng DNA and prepared using the NEBNext Ultra II DNA Library Prep Kit for Illumina (NEB). Library quality was assessed by Fragment Analyzer NGS. Samples were sequenced on a NextSeq500 (Illumina) with 75bp single-end reads at the Lunenfeld-Tanenbaum Research Institute Sequencing Facility.

Reads that passed quality control were trimmed of adaptors using Trim Galore! v0.4.0 and aligned to mm10 using bowtie2 v2.2.5131. Unmapped reads were then mapped to hg19. SAM files were converted to BAM files, sorted, and indexed with samtools v1.9. Bam files were deduplicated using MarkDuplicates (picard v2.18.14). Peaks were called with MACS2 (using IP samples and their input as controls) with the options --gsize 3.0e9 -q 0.05 --nomodel --broad and annotated by intersecting center positions with RefSeq annotations. The most upstream and downstream annotated TSS and TES, respectively, were considered for each gene. Peak analysis was performed using DiffBind v3.0.15 with score=DBA_SCORE_TMM_READS_FULL_CPM. Normalization was done with edgeR, using the ratio of mm10/hg19 reads (relative to respective input sample). For TSS analysis, a 1kb window surrounding the TSS of every RefSeq gene was used. Bigwig files were generated with Deeptools v3.3.0 using –scaleFactor for normalisation and visualized in Integrated Genome Viewer (IGV v2.9.4).

Screening for m⁶A targets

First, we selected all genes that gain m⁶A in pausing (MeRIP-seq: log₂FC > 0). Second, we looked for genes whose downregulation is suppressed in Mettl3^−/− ESCs (RNA-seq: logFC_Mettl3+/+ < 0 & log₂FC_Mettl3+/+ < log₂FC_{Mettl3−/−}). Third, we looked for genes whose expression is at least 2× higher in paused Mettl3^−/− ESCs (RNA-seq: CPM_{Mettl3−/−} > 2× CPM_Mettl3+/+). Finally, we selected genes whose RNA is more stable in paused Mettl3^−/− ESCs (RNA-seq: log₂FC_exon > 1.5 × log₂FC_intron). We then used published RNA-seq data from early mouse embryos(3), with expression data first transformed into z-scores. To establish a quantitative signature of the m⁶A machinery, we averaged the z-scores of the writers (Mettl3, Mettl14, Wtap) and the z-scores of the erasers (Fto, Alkbh5) multiplied by −1. We then ranked all targets by their Spearman correlation coefficients with the m⁶A machinery signature, focusing on genes with negative correlations, reflecting a potential destabilization of mRNAs in presence of the m⁶A machinery in embryos.

RNA immunoprecipitation

For RIP experiments, 2μg anti-Mettl3, Ythdf2 or control IgG antibodies were pre-bound to 20μL Protein A Dynabeads (Thermo Fisher), rotating for 3h at 4°C. Beads were next collected on a DynaMag (Thermo Fisher) and resuspended in RIP buffer (150mM KCl, 25mM Tris pH7.4, 5mM EDTA, 0.5mM DTT, 0.5% NP40, protease and RNase inhibitors) containing 500ng/mL tRNA (Thermo Fisher) and 1mg/mL BSA to block for 30 min. During incubation time, ESCs were collected and lysed in RIP buffer on ice for 20min. Supernatants (500μl, equivalent to 10 million cells) were pre-cleared with 20μL Protein A Dynabeads, rotating for 30min at 4°C. Cleared lysates were incubated together with antibody-bound blocked beads overnight at 4°C. The next day, lysates were washed 5 times in RIP buffer, and RNA was extracted using Direct-zol RNA Kits (Zymo Research), before analyzing by RT-qPCR.

Western blot analysis

Whole-cell and chromatin extract we prepared according to methods described previously(39, 40). Denatured samples were separated on 4–15% Mini-Protean TGX SDS-PAGE gels (Bio-Rad) and transferred to nitrocellulose membranes by wet transfer using Trans-Blot^® Turbo^™ Mini Nitrocellulose Transfer Packs (Bio-Rad). Membranes were blocked in 5% milk/TBS-T and incubated with indicated primary antibodies overnight at 4 °C. HRP-conjugated anti-mouse/rabbit secondary antibodies, were incubated for 1h at room temperature. Proteins were detected by ECL (Pierce) or Clarity Max (Bio-Rad). Quantification of bands was performed using ImageJ. See Table S7 for antibodies and dilutions.

mRNA stability assay

Cells were treated with 5μg/ml actinomycin D (Sigma-Aldrich) for 0, 1, 2 or 4h before cell collection. RNA level was measured by RT-qPCR and normalized to Actb, a highly stable transcript. Expression values (relative to 0h) were fitted to an exponential decay model using linear regression in R.

Luciferase reporters of RNA stability

The region surrounding the stop codon of Mycn mRNA, including the identified m⁶A site (wildtype sequence or A-to-C mutation) was cloned into the pmirGLO Dual-Luciferase miRNA Target Expression Vector (Promega), per manufacturer’s instructions. Vector (500ng) was transfected in ESCs with lipofectamine 2000. After 48h, cells were lysed and Firefly luciferase signal was measured with a luminometer and normalized to Renilla luciferase activity, using Dual-Glo^® Luciferase Assay System.

Nascent transcription assays in ESCs

To assess global transcriptional output, cells were treated with 1mM 5-ethynyl uridine (EU) for 45min, collected by trypsinization and prepared following the Click-iT^™ RNA Alexa Fluor 488 Imaging Kit (Thermo Fisher) instructions. Data were collected by flow cytometry using the Beckman Coulter Gallios and analyzed using Kaluza. Fluorescence values were plotted as median fluorescence intensity (MFI) per sample, relative to FBS Mettl3^+/+ cells.

For nascent RNA capture, EU incubation was performed in ESCs as above. Cells were collected by trypsinization, counted, and 2×10⁵ were used to extract RNA. Biotinylated nascent RNA was captured according to protocols within the Click-iT Nascent RNA Capture Kit (Invitrogen) and used in RT-qPCR.

Embryo immunofluorescence

Ex vivo paused embryos were labelled in their culture medium for 45min with 1mM EU for nascent transcription, then fixed in 4% paraformaldehyde for 15 min. Permeabilization was done with 0.5% Triton X-100 in PBS + 5% FBS for 15 min. After blocking in PBS, 2.5% BSA, 5% donkey serum for 1h, embryos were incubated overnight at 4°C with the primary antibodies (Mettl3 Abcam ab195352 1/200; N-Myc Cell Signaling Technology D1V2A 1/200). EU fluorescence coupling was performed following the manufacturer’s instructions for Click-iT^™ RNA Alexa Fluor 488 Imaging Kit. Embryos were incubated with fluorescence-conjugated secondary antibodies for 1h at room temperature. Embryos were stained with DAPI in fresh blocking, washed twice, and transferred into a drop of M2 media (~5μl).

Images were captured using a Leica DMI 6000 Spinning Disc Confocal microscope, and embryos were genotyped as before. Quantification was performed with ImageJ, with 10 consecutive image planes (imaged at 10μm intervals) stacked by “average intensity” projection. This was repeated 4 times (totaling 40 image planes used per embryo). Then, individual nuclei were quantified using the ROI Manager, and the background (captured outside of the embryo) was subtracted to values within each projection. To avoid batch effects, values were also normalized to the average of Mettl3^{TCP +/+} embryos within each litter.

Statistics and reproducibility

Statistical analyses were performed in GraphPad Prism v9.3.1 or R v4.0.3. Data are presented as mean ± SD or SEM, except where otherwise indicated. Box plots present center lines as medians, with box limits as upper and lower quartiles and whiskers as 1.5×IQR. Two-tailed Student’s t-test and one- or two-way ANOVA with Dunnett’s multiple comparison tests were used when normal distribution could be assumed. Time series were modeled by linear regression on log₂-transformed y values, with P-values extracted from the interaction between time and the categorical variable of interest. GSEA was performed with fGSEA in R, with the adjusted P-value as indicated. Correlation was measured by ρ Spearman’s rank correlation coefficient. All replicates for in vitro data are derived from independent experiments.

Sample size, number of replicates, errors bars and statistical tests were chosen based on experience and variability of in vitro studies and stated in each figure legend. Unless otherwise indicated, all experiments included technical replicates and were repeated at least three independent times. No statistical methods were used to predetermine sample sizes. All replicates for in vivo data are derived from at least 3 embryos per genotype and 2 separate litters. No statistical method was used to predetermine sample size. No data were excluded from the analyses. No randomization was required for design of in vivo experiments, as embryos were harvested, cultured and treated together and only later identified by genotype (Mettl3^+/+, Mettl3^+/− or Mettl3^−/−). No blinding was applied in this study, as all data produced derived from objective quantitative methods. To ensure consistent experimental conditions, all control and treatment samples were processed in parallel.

Materials availability

Plasmids used in this study were generated from vectors commercially available from Addgene (#48138, #175582, #12259, #12260, #175583) and Promega (pmirGLO Dual-Luciferase miRNA Target Expression Vector). All sequences used for cloning are listed at Table S6.

Data and code availability

Sequencing data have been deposited on the NCBI Gene Expression Omnibus repository (GEO, http://ncbi.nlm.nih.gov/geo) and will be accessible upon publication. Published RNA-seq data used in this study are available under the accession numbers E-MTAB-2958 (early mouse embryos) and E-MTAB-3386 (Myc/Mycn DKO ESCs). The authors declare that all other data supporting the findings of this study are available within the paper and its supplementary information files. Code supporting this study are available at a dedicated Github repository [https://github.com/EvelyneCollignon/Mettl3_pausing].

Data and materials availability:

Sequencing data have been deposited on the NCBI Gene Expression Omnibus repository (GEO, http://ncbi.nlm.nih.gov/geo) and will be accessible upon publication. Published RNA-seq data used in this study are available under the accession numbers E-MTAB-2958 (early mouse embryos) and E-MTAB-3386 (Myc/Mycn DKO ESCs). All data are available in the main text or the supplementary materials.