YTHDC1 recognizes METTL16-dependent m6A on caRNAs and coordinates cotranscriptional splicing Authors: Zhong Zhang, Qi Yin, Weimin Lin, Qiwen Li, Rui Sheng, Shuang Jiang, Kexin Lei, Linfeng Liu, Lanxin Zhang, Chunlin Qian, Junru Wen, Zirui Wang, Chong Chen, Quan Yuan Published: 2026 ABSTRACT SignificanceThe presence of multiple m6A writers and readers strongly suggests undiscovered layers of functional specialization and unique modes of action, yet how these diverse components achieve precise, cell-type, or developmental stage-specific regulation remains largely unknown. We found loss of METTL3, considered to be the most dominant m6A writer, has no significant impact on limb development. Instead, METTL16 is implicated. METTL16 showed preference for intronic sequences and up to 10 K METTL16-dependent m6A sites on chromatin-associated RNAs were identified. Our findings further established a specific METTL16-m6A-YTHDC1 axis that regulates cotranscriptional splicing of development-related genes and function during limb organogenesis, providing evidence in RNA-epigenetic bridge between transcription and splicing—two essential processes in gene expression. N6-methyladenosine (m6A) plays essential roles in various RNA biological processes, including transcription, splicing, transporting, decaying, and R-loop formation (1–6). The m6A methyltransferase complex, composed of METTL3 and METTL14, has been thought to be the dominant m6A writer and participates in various biological processes (7–9). However, the proportion of m6A marks dependent of METTL3/METTL14 methyltransferase complex is limited (~60% in some cell types), implying the existence of additional m6A methyltransferases that catalyze METTL3/METTL14-independent m6A (10). METTL16 was recently identified as an m6A methyltransferase with reported catalytic activity to insert m6A on mRNAs and some noncoding RNAs (11–13). The YTH domain-containing protein, YTHDC1, is identified to bind RNAs in an m6A-dependent manner with various functions underlying RNA processing (6, 14–16). It is worth noting that YTHDC1 is sought to be the functional downstream of METTL3, as depletion of YTHDC1 generally caused a similar phenotype with METTL3 in ES, hematopoietic progenitor, and several cancer cell-lines (4, 17–20). However, YTHDC1 could also recognize METTL16-dependent m6A in a few studies (13, 21). These facts raised the questions below: 1) Whether YTHDC1 recognition of m6A methylated by METTL3 or METTL16 is a context-dependent mechanism according to different cell types, diseases, and developmental processes; and 2) Whether METTL3-m6A-YTHDC1 axis and METTL16-m6A-YTHDC1 axis differentially regulate RNA metabolism and depletion of which results in distinct cellular outcomes. Here, we find that YTHDC1 and METTL16—but not METTL3—form a functionally coupled axis essential for mammalian organogenesis. We demonstrate that METTL16-deposited m6A marks on chromatin-associated RNAs recruit YTHDC1 to transcriptionally active loci, where it orchestrates cotranscriptional splicing through LLPS-mediated assembly of cotranscriptional splicing microcompartments. This work establishes a paradigm for epitranscriptomic regulation and provides mechanistic insights into how RNA modifications coordinate nuclear organization during development. Results Depletion of Ythdc1 or Mettl16 Impairs Limb Organogenesis. YTHDC1 is identified as a major m6A reader that typically functions downstream of the canonical writer METTL3, with their depletion generally producing similar phenotypes (4, 17, 19, 20, 22–24). To investigate their roles in organogenesis, we generated limb mesenchymal progenitor-specific Mettl3 (Prrx1-Cre; Mettl3fl/fl) and Ythdc1(Prrx1-Cre; Ythdc1fl/fl) knockout mice. Ythdc1 depletion caused dramatically more severe limb malformations compared to Mettl3 depletion, with P0 mice displaying profound shortening stylopods, zeugopods, and autopods in both forelimbs and hindlimbs (Fig. 1A) (25). This unexpected phenotypic divergence suggests that YTHDC1 may function through an alternative regulatory mechanism during organogenesis. METTL16 or YTHDC1 depletion severely impairs limb organogenesis. (A) Gross images and whole-mount skeletal stain of P0 mice of indicated genotypes. (B) UMAP plot demonstrating the clustering profile of Single-cell RNA sequencing applied on E13.5 limb buds. Ctrl: Ythdc1fl/fl, KO: Prrx1-Cre; Ythdc1fl/fl. MPC: mesenchymal progenitor cell. RBC: red blood cell. (C) Proportion of different cell types in isolated OCL cluster. (D) UCell scoring of 6 gene modules as indicated in 3 clusters. To determine whether this phenotype requires YTHDC1’s m6A recognition activity, we generated another genetically modified mouse expressing a catalytically inactive YTHDC1W378A mutant (Prrx1-Cre; Ythdc1fl/(CDS-W378A), hereafter, W378A mice), which abolishes m6A capacity14,20. The W378A mice phenocopied Prrx1-Cre; Ythdc1fl/fl mice (Fig. 1A), demonstrating that YTHDC1’s role in limb development strictly depends on its m6A reader function. The phenotypic disparity between Ythdc1 and Mettl3 knockouts prompted us to investigate whether YTHDC1 preferentially recognizes m6A deposited by an alternative writer. Correlation analysis of gene expression patterns across multiple cell line datasets revealed that Ythdc1 expression showed strongest correlation with Mettl16 among all known m6A writers (SI Appendix, Fig. S1 A and B). To test whether METTL16 serves as the cognate writer for YTHDC1 in this developmental context, we generated Mettl16 conditional knockout mice (Prrx1-Cre; Mettl16fl/fl). Strikingly, Mettl16 deletion precisely phenocopied Ythdc1 loss, producing severely truncated limbs with identical morphological defects (Fig. 1A), establishing a specific METTL16-YTHDC1 functional axis in limb organogenesis. To comprehensively characterize the cellular basis of these defects, we performed single-cell RNA sequencing (scRNA-seq) on E13.5 limb buds from control and Ythdc1 knockout embryos (Fig. 1A). Loss of Ythdc1 resulted in a marked reduction in differentiated cell populations, particularly chondrogenic cells and osteoblasts (Fig. 1C). Pseudotime trajectory analysis revealed impaired progression through chondrogenic and osteogenic differentiation programs (SI Appendix, Fig. S2 B and C). Differential gene expression analysis (SI Appendix, Fig. S2 C and D) and module scoring demonstrated widespread downregulation of developmental gene modules, including “skeletal system development” and “collagen fibril organization,” upon loss of YTHDC1 (Fig. 1D and SI Appendix, Fig. S2E). METTL16-Specific m6A on caRNAs Recruits YTHDC1 to Chromatin. To identify the site-resolved METTL16-specific and METTL3-specific m6As, we generated CRISPR/Cas9-mediated knock-out cell-lines (sgMettl16 and sgMettl3) and performed for m6A-seq on isolated caRNAs (SI Appendix, Fig. S3A). We identified a total of 28,416 m6A sites, comprising 12919 METTL16-dependent and 5166 METTL3-dependent m6A sites. Among these, we identified10545 METTL16-specific m6A sites and 2792 METTL3-specific m6A sites (Fig. 2A), with only 2,374 peaks overlapped between METTL3 and METTL16 targets, confirming their selectivity (Fig. 2A). Peak annotation of the different sets of m6A sites showed that METTL16-specific sites were predominantly enriched in introns, whereas METTL3-specific sites were primarily exonic (Fig. 2B). RIP-seq was also conducted in primary limb bud mesenchymal progenitor cells (LBMPCs) to demonstrate the transcript-binding pattern of YTHDC1, METTL3, and METTL16. This kind of cells were harvested from E12.5 embryonic limb buds and cultured in defined conditions (26). METTL3 exhibits the lowest proportion of intronic peaks (6.97%) and highest exonic enrichment (28.68%), consistent with its established role in postsplicing transcripts (SI Appendix, Fig. S3B). In contrast, YTHDC1 exhibited peak distribution patterns strikingly similar with both METTL16 and caRNA m6A (SI Appendix, Fig. S3B), with substantial intronic enrichment indicative of cotranscriptional function. Moreover, METTL16 knock-out has minor effect on SAM homeostasis or Mat2a mRNA expression level, which might potentially contribute to global methylation level (SI Appendix, Fig. S3 C and D). METTL16-dependent caRNA m6A associates YTHDC1 to chromatin. (A) Venn plot showing the overlapping peaks between METTL16-dependent m6A and METTL3-dependent m6A. (B) Annotations of peaks of caRNA m6A-seq of indicated sets. (C) Aggregation plot showing YTHDC1 CLIP-seq signals to caRNA m6A-seq peak center. (D) Homer de novo motif search on indicated peak sets. (E) Motif enrichment analysis of “GAAGA” motif on indicated peak sets. (F) GO enrichment analysis of genes associated by YTHDC1 and marked with METTL16-specific m6A. (G) heatmaps of signals of YTHDC1 CLIP-seq, METTL16 RIP-seq, and caRNA MeRIP-seq on YTHDC1 CLIP-seq peak centers. The regions were sorted by average reads density of YTHDC1 CLIP-seq. (H) Fractionation assay. LBMPCs were separated into cytosolic (Cy), nucleoplasmic (Nu), and chromatin (CH) fractions. The Cy and Nu fraction were mixed as soluble supernatant (SN) fraction. Indicated antibodies were used for immunoblotting. (I) Fractionation and WB analysis of YTHDC1 in LBMPCs upon different treatment as indicated. RNaseH+, RNase H was added to cells and incubated for 30 min at 37 °C. RNaseH-, cells were incubated for 30 min at 37 °C without RNaseH. (J–L) Boxplot showing the signals of YTHDC1 ChIP-seq on indicated genes. To elucidate the molecular basis for METTL16-YTHDC1 selectivity, we performed crosslinking-immunoprecipitation and high-throughput sequencing (CLIP-seq) for YTHDC1 (27). We identified 9795 YTHDC1 CLIP peaks. Combined analysis identified a colocalization pattern between YTHDC1, METTL16, and caRNA m6A (Fig. 2C and SI Appendix, Fig. S3E) and identified 2005 sites where METTL16-specific m6A modification coincided with YTHDC1 binding (SI Appendix, Fig. S3F). Motif analysis revealed a shared sequence contexts (GAAGA) of caRNA m6A, METTL16-specific m6A, METTL16 RIP-seq, and YTHDC1-bound sites, which is distinct from METTL3-specific m6A (GGAC) (Fig. 2D and SI Appendix, Fig. S3G). Moreover, genome wide motif search showed high enrichment score of GAAGA motif in METTL16-specific m6A sites and YTHDC1 CLIP peaks, but not in METTL3-specific m6A (Fig. 2E). Consistent with the previously reported METTL16 consensus motif UGAAGA (28), our data confirmed METTL16’s sequence-specific targeting of chromatin-bound substrates. Further, GO enrichment analysis of genes marked by METTL16-specific m6A and bound by YTHDC1 indicated their involvement in basic cellular processes such as replication and splicing, and several developmental processes such as osteoblast differentiation and cartilage development (Fig. 2F and SI Appendix, Fig. S3H). METTL16 and caRNA MeRIP reads were enriched inYTHDC1 CLIP-seq peaks (Fig. 2G), supporting a model where METTL16-deposited m6A marks on caRNAs serve as recruitment platforms for YTHDC1. Subcellular fractionation confirmed that YTHDC1 was readily detected in chromatin-enriched fractions (Fig. 2H). Critically, depletion of METTL16—but not METTL3 loss—caused YTHDC1 dissociation from chromatin (Fig. 2I). Furthermore, RNase H treatment, which degrades the RNA component of RNA–DNA hybrids, released YTHDC1 into the soluble fraction (Fig. 2I), confirming that YTHDC1’s chromatin association is RNA-mediated and METTL16-dependent. Further ChIP-seq revealed that YTHDC1 preferred to locate at METTL16-associated genes compared to METTL3-associated targets (Fig. 2J). Moreover, YTHDC1 chromatin binding was preferentially enriched at loci containing m6A-modified caRNAs and METTL16-bound genes (Fig. 2 K and L), establishing a direct correspondence between METTL16-mediated m6A deposition and YTHDC1 recruitment to transcriptionally active chromatin. Chromatin-Bound YTHDC1 Affects Splicing. To investigate the functional consequences of YTHDC1 chromatin recruitment, we performed RNA-seq analysis on Ythdc1 knockout limb tissues. While differential expression analysis revealed downregulation of developmental genes consistent with the observed phenotypes (Fig. S4 A-F), a more striking finding emerged from alternative splicing analysis. We identified thousands of transcripts undergoing differential splicing events upon YTHDC1 loss, with exon skipping representing the most abundant category among all alternative splicing (AS) types (Fig. 3 A and B). YTHDC1 finetunes cotranscriptional splicing. (A) Scatter plot showing differentially spliced genes. (B) Calculation of differential splicing events in E11.5 forelimb RNA-seq. SE, skipped intron. A3SS, alternative 3’ splicing. A5SS, alternative 5’ splicing. MXE: mixed spliced exon. RI: retained intron. (C) Venn plot showing the overlap of YTHDC1 RIP-seq, caRNA MeRIP-seq, and DSGs (Left) and the GO enrichment analysis of overlapped genes (Right). DSGs, differentially spliced genes. (D–F) Top: schematic description of exon skipping events of 3 indicated transcripts and the PSI calculation. Bottom: Selected genomic views of ChIP-seq, CUT & Tag and ATAC-seq data for the indicated genes. (G) Venn plot showing the overlap of genes identified by METTL16 RIP-seq and DSGs. 67.58 % of DSGs were associated with METTL16. (H–K) Boxplots showing the reads density of ChIP-seq and RIP-seq of indicated targets on DSGs and steady genes. Remarkably, 738 differentially spliced genes (DSGs) were directly bound by YTHDC1 and carried m6A modifications, representing high-confidence targets of the METTL16-YTHDC1 regulatory axis (Fig. 3C). Gene ontology analysis of the 738 YTHDC1-regulated DSGs revealed enrichment for fundamental cellular processes essential for progenitor cell function, including DNA recombination, double-strand break repair, and cell cycle regulation (Fig. 3C). This functional profile suggests that YTHDC1-mediated splicing control is critical for maintaining genomic stability and proliferative capacity in developing mesenchymal progenitors. To examine specific splicing targets, we analyzed representative DSGs including Adar, Btbd7, and Usf2, all of which exhibited significant decreases in percent spliced-in (PSI) values due to enhanced exon skipping upon YTHDC1 depletion (Fig. 3 D–F). ChIP-seq analysis confirmed YTHDC1 occupancy near the transcription start sites of these genes, coinciding with active transcription marks including H3K4me3 and RNA polymerase II (Fig. 3 D-F), consistent with a cotranscriptional splicing regulatory mechanism (29). Moreover, only YTHDC1-WT, but not YTHDC1-W378A mutant could rescue the binding affinity of splicing factor to the genome loci (SI Appendix, Fig. S5 A and B) and the impaired splicing patterns (SI Appendix, Fig. S5 C–E), demonstrating the m6A-dependent recruitment mechanism. Genome-wide analysis revealed that 67.58% of DSGs (1,628 genes) were associated with METTL16 binding (Fig. 3G). Critically, both YTHDC1 and METTL16 demonstrated statistically significant enrichment at DSG loci relative to control gene sets (Fig. 3 H–J), and DSGs exhibited markedly higher caRNA m6A levels compared to nontargets (Fig. 3K). Taken together, these results demonstrated YTHDC1 regulates cotranscriptional splicing in mesenchymal progenitor cells. METTL16-m6A-YTHDC1 Axis Targets Active Transcription Genes. To determine the regulatory specificity of the METTL16-m6A-YTHDC1 axis, we analyzed the relationship between axis-dependent regulation and transcriptional activity. Similar to DSGs, DEGs showed preferential association with YTHDC1 and METTL16 binding and exhibited higher caRNA m6A level (SI Appendix, Fig. S6A). We identified 463 DEGs and 698 DSGs under the direct control of METTL16-m6A-YTHDC1 axis (Fig. 4A). To assess transcriptional selectivity, we stratified all genes into three equal-sized clusters based on expression levels (low, medium, and high). Both DEGs and DSGs were predominantly found among medium and high expression genes (Fig. 4B), with only 471 DEGs (24.9%) and 489 DSGs (20.3%) classified as low expression (Fig. 4B). This bias was more pronounced for direct axis targets: only 50 (10.8%) DEGs and 71 (10.2%) DSGs fell into low-expression genes (Fig. 4B). Genome-wide correlation analysis revealed that YTHDC1 and METTL16 RIP-seq signals, along with caRNA m6A levels, positively correlated with gene expression, mirroring the distribution patterns of RNA polymerase II (Pol II), chromatin accessibility (ATAC-seq), and key transcriptional initiation factors BRD4 and MED1 (Fig. 4C and SI Appendix, Fig. S6B) (30). On the other hand, the genome wide localization of BRD4, MED1, and Pol II showed similar patterns on YTHDC1 binding sites (Fig. 4D), and the genome-wide distribution patterns of BRD4, MED1, and Pol II showed strong concordance with YTHDC1 (SI Appendix, Fig. S6C). Consistent with these results, YTHDC1 ChIP-seq peaks were predominantly marked by active histone modifications, including H3K4me3 and H3K27ac (SI Appendix, Fig. S6D). Moreover, Pol II, BRD4, and chromatin accessibility signals were enriched at genes bound by METTL16, YTHDC1, or with m6A modifications on caRNAs (SI Appendix, Fig. S6 E and F), revealing strong functional coupling between the METTL16-m6A-YTHDC1 axis and the transcriptional machinery. The METTL16-m6A-YTHDC1 axis finetunes transcription through regulating Pol II pause-release. (A) Venn plot showing the overlap of indicated genes. (B) Distribution of indicated DEGs or DSGs on 3 clusters of genes identified according to their expression levels. (C) Aggregation plots of signals of indicated targets in TSS of 3 clusters of genes identified according to their expression levels. (D) heatmaps of signals of YTHDC1 ChIP-seq, BRD4 CUT & Tag, MED1 CUT & Tag, Pol II CUT & Tag and ATAC-seq on YTHDC1 ChIP-seq peak center in LBMPCs. The regions were sorted by average reads density of YTHDC1 ChIP-seq. (E) Boxplot showing the signals of Pol II CUT & Tag on indicated genes. (F) Aggregation plots of Pol II CUT & Tag signals in gene body in LBMPCs upon siRNA treatment. siScramble: untargeting siRNA control. siYTHDC1: siRNA to YTHDC1. (G) WB analysis of LBMPCs whole-cell lysate upon siRNA treatment. Pol II indicates antibody to both Hypo-phosphorylated and phosphorylated forms of Pol II. (H and I) Immunofluorescence of Pol II in LBMPCs upon siRNA treatment (H) and statistical analysis of intracellular intensity (I). HypoP: hypo-phosphorylated RNA Polymerase II. Ser2P: phosphorylated RNA Polymerase II. (n = 20) J, EU incorporation assay. LBMPCs were treated as indicated and EU was added to 1 μM to incorporate into nascent RNAs for 30 min. siM16: siRNA to METTL16. ActD: actinomycin D. DRB: benzimidazole. Right: EU intensity calculation of EU incorporation assay. YTHDC1 Depletion Results in Pol II Promoter Pausing and Transcription Arrest. This colocalization pattern prompted us to hypothesize that the METTL16-m6A-YTHDC1 axis directly regulates Pol II-mediated transcription. To test this hypothesis, we examined Pol II dynamics following YTHDC1 depletion. ChIP-seq analysis revealed increased robust accumulation of Pol II at TSS upon YTHDC1 deletion, particularly in actively transcription genes (Fig. 4F). YTHDC1 loss resulted in decreased levels of elongation-competent phospho-Ser2 Pol II and a concomitant increase in paused, hypo-phosphorylated Pol II (Fig. 4 G-I). The shift toward hypo-phosphorylated Pol II confirms genome-wide promoter pausing and suggests that YTHDC1 is required for efficient transition from transcriptional initiation to productive elongation. Consistent with impaired transcriptional elongation, nascent RNA synthesis was significantly reduced upon YTHDC1 depletion, as measured by 5-ethynyl uridine (EU) incorporation (Fig. 4J). Moreover, acute loss of YTHDC1 significantly increased R-loops (SI Appendix, Fig. S7 A and B), especially on YTHDC1-target loci (SI Appendix, Fig. S7 C and D), indicating the contribution of R-loop formation in transcription regulation. YTHDC1 Recruits Splicing Factors to Chromatin for Cotranscription Splicing. To elucidate how YTHDC1 regulates cotranscriptional splicing, we conducted CoIP-MS/MS. This analysis revealed hundreds of potential YTHDC1 interacting proteins, with most of them enriched in RNA processing pathways, especially RNA splicing (Fig. 5A and SI Appendix, Fig. S8A). We selected two representative splicing factors from its interactome, SRSF3 and SRSF7, for further investigation. Immunofluorescence staining confirmed nuclear colocalization between YTHDC1 and both SRSF3 and SRSF7 (Fig. 5B). Subcellular fractionation and CoIP validated these interactions and demonstrated their association with chromatin fractions (Fig. 5C and SI Appendix, Fig. S8B). Critically, deletion of either Ythdc1 or Mettl16 dissociated splicing factors from chromatin (Fig. 5D), and decreased number of SRSF7 foci in nucleus (Fig. 5E), indicating that the METTL16-m6A-YTHDC1 axis is essential for chromatin recruitment of splicing factors. YTHDC1 associates splicing factors to chromatin. (A) GO enrichment analysis of proteins identified in YTHDC1 interactome. (B) Immunofluorescence stain of YTHDC1 and SRSF3 or SRSF7 in LBMPCs (Left) and the colocalization analysis of the area selected by a white line (Right). (C) CoIP analysis of Flag-tagged YTHDC1 with HA-tagged SRSF3 or SRSF7. (D) WB analysis of SRSF7 and SRSF3 in LBMPCs upon different treatment as indicated. RNaseH+, RNase H was added to cells and incubated for 30 min at 37 °C. RNaseH-, cells were incubated for 30 min at 37 °C without RNaseH. (E) Immunofluorescence stain of SRSF7 in LBMPCs upon different treatment as indicated (Left) and analysis of number of SRSF7 condensates per cell (Right). (F) Aggregation plots showing the reads intensity of SRSF7 CUT & Tag on TSS of 3 clusters of genes classified according to expression levels. (G) Aggregation plots showing the reads intensity of SRSF7 CUT & Tag on YTHDC1-bound genes upon YTHDC1 depletion. (H) Selected genomic views of ChIP-seq, CUT & Tag and ATAC-seq data for the indicated genes. (I) Boxplot showing the signals of SRSF7 CUT & Tag on indicated genes. (J) Venn plot showing the overlap of genes identified by SRSF7 RIP-seq and METTL16-m6A-YTHDC1 axis-dependent DSGs. CUT & Tag analysis revealed that SRSF7 occupancy showed strong positive correlation with gene expression levels (Fig. 5F), mirroring Pol II distribution pattern (Fig. 4E) and supporting a cotranscriptional regulatory mechanism. The genomic localization patterns of SRSFs closely paralleled those of YTHDC1 (SI Appendix, Fig. S8C). Importantly, deletion of YTHDC1 leads to decreased SRSF7 occupancy near transcription start sites of YTHDC1-bound genes (Fig. 5 G and H). Calculation of reads density of SRSF7 CUT & Tag revealed that SRSF7 preferred to localize at genes bound by METTL16, YTHDC1, or with caRNAs m6A (SI Appendix, Fig. S8D). This enrichment was particularly pronounced at METTL16-m6A-YTHDC1 axis-dependent DSGs (Fig. 5I). Notably, DSGs showed strong association with both SRSF7 and RNA polymerase II signals (SI Appendix, Fig. S8 D and E), providing direct evidence for cotranscriptional splicing regulation. To comprehensively characterize the functional relationship between the METTL16-m6A-YTHDC1 axis and splicing factor recruitment, we conducted RIP-seq analysis on SRSF7. Genome-wide analysis of genes sorted by METTL16 binding intensity revealed striking colocalization patterns among SRSF7, METTL16, YTHDC1, and caRNA m6A signals (SI Appendix, Fig. S8F). There were 3156 (96.8%) METTL16-m6A-YTHDC1 axis-dependent genes or 677 (97.0%) METTL16-m6A-YTHDC1 axis-dependent DSGs associated with SRSF7, demonstrating a strong association between SRSF7 and METTL16-m6A-YTHDC1 axis (Fig. 5J and SI Appendix, Fig. S8G). YTHDC1 Coordinates Cotranscriptional Splicing Via LLPS. Given that YTHDC1 interacts with both the transcriptional machinery and splicing factors, we hypothesized that YTHDC1 coordinates these processes within the same temporal framework to fine-tune cotranscriptional splicing. Immunofluorescence of YTHDC1, SRSF3, SRSF7 showed that these proteins form subnuclear condensates (Fig. 5B), strongly suggesting that YTHDC1 coordinates cotranscriptional splicing via liquid–liquid phase separation (LLPS). To testify this hypothesis, we constructed an mCherry-YTHDC1 fusion protein and examined its condensation properties in vitro. Concentration-dependent droplet formation assays demonstrated robust, reversible concentration-dependent droplet formation by YTHDC1 (SI Appendix, Fig. S9 A–C). Using fluorescence labeled recombinant proteins (mCherry-YTHDC1, EGFP-MED1-IDR, and EGFP-SRSF3), we demonstrated that YTHDC1 incorporates and concentrates both MED1-IDR and SRSF3 in a salt concentration-dependent manner (Fig. 6 A and B and SI Appendix, Fig. S9D). Additionally, the in vivo expression pattern of YTHDC1 recapitulates the in vitro LLPS-dependent condensation activity (Fig. 6C and SI Appendix, Fig. S9E). YTHDC1 sequester Splicing factors to transcription complex in a LLPS-dependent manner. (A) mCherry-YTHDC1 droplets incorporate EGFP-MED1-IDR and EGFP-SRSF3 in vitro. The fluorescent fusion proteins were mixed at 10 μM each in buffer containing 5% PEG-8000 at different salt concentrations as indicated. (B) Calculation of relative intensity of sequestered EGFP-MED1-IDR (Upper) and EGFP-SRSF3 (Lower). (C) Immunofluorescence stain of YTHDC1 in E12.5 forelimb. Yellow triangles indicate subnuclear condensates formed by YTHDC1. (D) mCherry-YTHDC1 droplets incorporate EGFP-MED1-IDR and mTagBFP2-SRSF3 in vitro. The fluorescent fusion proteins were mixed at 10 μM each in buffer containing 5% PEG-8000 at 125 mM NaCl. (E and F) Statistical analysis of condensate size (E) and number of condensates (F) of sequestered EGFP-MED1-IDR in each group. (G) Top: PONDR scoring of YTHDC1 showing the disordered propensity of each region. Bottom: schematic description showing the construction of YTHDC1 truncated mutants and site-mutants. (H) in vitro droplet formation assay of each indicated protein at each buffering conditions. Critically, at physiological protein concentrations (10 μM), mCherry-YTHDC1 successfully incorporated both transcriptional and splicing components into shared condensates, while mCherry control protein failed to promote such copartitioning (Fig. 6D). Quantitative analysis revealed that YTHDC1 significantly increased condensate size compared to control conditions, demonstrating its enhanced phase separation capacity (Fig. 6 E and F). Computational analysis using disorder prediction algorithms identified two IDRs within YTHDC1: an N-terminal region (AA1-345) and a C-terminal region (AA508-736) (Fig. 6G). Sequence composition analysis revealed significant arginine enrichment in the C-terminal IDR (SI Appendix, Fig. S9 F and G). Given that arginine residues are known drivers of phase separation in transcriptional condensates (31), we investigated whether YTHDC1’s arginine content determines its LLPS activity. We noted that SRSF3, SRSF7, and the YTHDC1 C-terminal IDR share remarkably similar arginine content (24.39%, 24.37%, and 23.68%, respectively) and isoelectric points (pI: pH 11.95, 11.22, and 12.13, respectively), suggesting they readily coassemble under physiological conditions (SI Appendix, Fig. S9H). In contrast, full-length YTHDC1 contains only 12.38% arginine with a neutral pI (6.08), while the N-terminal region is acidic (8.43% arginine, pI 4.12) (SI Appendix, Fig. S9G). To test whether the arginine-rich C-terminal region drives YTHDC1’s phase separation activity, we generated a series of mCherry-tagged YTHDC1 truncation mutants (Fig. 6G). In vitro and cellular assays revealed that full-length YTHDC1, YTH domain mutants (W378A and W429A), ΔYTH domain, and C-terminal fragments (C228) all retained LLPS activity. However, the N-terminal fragment (N345) and isolated YTH domain failed to form droplets (Fig. 6H), confirming that the arginine-rich C-terminal IDR is both necessary and sufficient for YTHDC1’s phase separation behavior and likely essential for coordinating transcription-splicing complexes. Arginine Residues on YTHDC1 C-terminal Region Are Molecular Determinants of LLPS Activity. To define the molecular basis by which arginine residues in the C-terminal region determine the LLPS activity of YTHDC1, we compared the in vitro phase-separation activity of 4 proteins: mCherry, C228, and 2 mutants of C228, including arginine-to-lysine (R to K) mutant and arginine-to-alanine (R to A) mutant. Initial screening revealed that only C228 and the R→K mutant displayed phase-separating activity (SI Appendix, Fig. S10 A and B). Checkerboard assay confirmed that the C228 protein, as well as other mutants containing this region, exhibited concentration-dependent condensation kinetics similar to full-length YTHDC1 (Fig. 7 A and B and SI Appendix, Fig. S10 D–N). Furthermore, both wild-type C228 and the R→K mutant recruited SRSF7 in a concentration-dependent manner, while mCherry control and the R→A mutant lacked this activity (Fig. 7 C–E). Moreover, both mCherry-C228 and the R→K mutant successfully coincorporated mTagBFP2-SRSF3 and EGFP-MED1-IDR into shared condensates, while others failed to promote such multicomponent assembly (Fig. 7F). Quantitative analysis of condensate size confirmed that wild-type C228 and the R→K mutant formed significantly larger droplets when comixed with transcriptional and splicing factors, compared to mCherry control or the R→A mutant (Fig. 7 G and H). Alkalic residues in CTD IDR contribute to the phase-separation activity of YTHDC1. (A and B) The phase diagram (A) and relative intensity calculation (B) of sequestered recombinant mCherry-C228 proteins at various amounts and with different concentrations of NaCl as indicated under 5% PEG-8000. (C) Droplet formation assay of EGFP-SRSF7 with indicated mCherry proteins in vitro. The fluorescent fusion proteins were mixed at 10 μM each in buffer containing 5% PEG-8000 at each concentration as indicated. +PEG, LLPS buffer with 5% PEG-8000. +1,6-H: LLPS buffer with 5% 1,6-Hexanidol. (D) Calculation of relative intensity of sequestered EGFP-SRSF7. (E), Pearson’s correlation coefficient of EGFP-SRSF7 and indicated mCherry fusion proteins. (F) mCherry-C228 droplets incorporate EGFP-MED1-IDR and mTagBFP2-SRSF3 in vitro. The fluorescent fusion proteins were mixed at 10 μM each in buffer containing 5% PEG-8000 at 125 mM NaCl. (G and H) Statistical analysis of condensate size (G) and number of condensates (H) of sequestered EGFP-MED1-IDR in each group. (I) immunofluorescence of 3T3 cells transfected with indicated fluorescent fusion proteins (Top) and the intensity alongside indicated linear region of each group (Bottom). Red line represents the mCherry channel. Green line represents the EGFP channel. To validate these biochemical findings in living cells, we performed colocalization analysis in NIH-3T3 cells cotransfected with EGFP-SRSF3 and various mCherry-tagged YTHDC1 constructs. Strong nuclear colocalization was observed between SRSF3 and full-length YTHDC1, the W378A and W429A point mutants, the ΔYTH deletion mutant, and the C228 fragment. Conversely, mCherry control, the isolated YTH domain, the N345 fragment, and the C228 R→A mutant showed no significant colocalization with SRSF3, confirming that arginine-mediated phase separation drives protein–protein interactions in living cells. Interestingly, while the C228 R→K mutant retained phase separation activity in vitro, it showed imperfect colocalization with SRSF3 in cells, appearing in some subnuclear domains that lacked SRSF3 signal (Fig. 7I). Discussion Our study demonstrates that YTHDC1 specifically reads METTL16-deposited m6A marks on chromatin-associated RNAs and coordinates cotranscriptional splicing through LLPS during organogenesis. While YTHDC1 and METTL3 depletion produce similar phenotypes in embryonic stem cells, hematopoietic progenitors, and cancer cell lines (4, 17, 18, 20, 22–24, 32), biochemical studies suggested functional differences between METTL3 and METTL16 (11, 25, 33, 34). The phenotypic divergence between Mettl16, Ythdc1, and Mettl3 knockouts during limb organogenesis identified by our work provides the in vivo evidence that these writers operate through distinct regulatory axes. METTL16 recognized RNA substrates with specific secondary structure (stem-loop) and might not strictly rely on sequence context (12), which is revealed by research focusing on structural biology (11). We propose that METTL16 recognizes its substrates with 1) specific sequence context and 2) appropriate regional structure. Addressment of this issue requires future study with combinatorial tools and analysis on structure/molecular/cellular biology. Moreover, whether the thousands of METTL16-specific m6A sites are direct targets of METTL16 still remains to be investigated. Beyond the sequence-specific motifs, we propose that YTHDC1 discrimination of m6A sites is governed by the local “molecular context”. This state is defined by the convergence of regulatory factors, and coresident splicing machinery, which create a permissive, high-affinity niche for YTHDC1 occupancy. We suggest that YTHDC1 does not merely scan for individual m6A marks in isolation but rather functions as a central hub that interprets these marks within their unique subnuclear microenvironments. Consequently, the distinct m6A landscapes and YTHDC1-binding patterns observed across various cell types and developmental stages likely reflect these dynamic epigenetic states, allowing for the precise, context-dependent execution of cotranscriptional regulatory programs critical for processes such as limb organogenesis. Cotranscriptional splicing plays essential roles in various biological processes (35–40). Despite its importance, m6A-regulation of cotranscriptional splicing has received limited attention, even though m6A factors are known to influence both transcription initiation and pre-mRNA splicing (1, 3, 5, 15, 41). Our findings establish the METTL16-m6A-YTHDC1 axis as an important coordinator of transcription-splicing coupling. Previous research indicated nascent RNA structure formed along with m6A insertion and reader binding, which is likely to influence cotranscriptional splicing (11, 42, 43). As closely connected processes, disruption of either transcription or splicing would influence another process (39, 44). We demonstrated that m6A modifications on nascent transcripts serve as sequence-specific recruitment platforms for chromatin-resident YTHDC1, which then assembles the splicing machinery. Perturbing this axis impacts both transcription and splicing. This aligns with observations that splicing factor depletion leads to promoter-proximal Pol II pausing and transcription arrest, highlighting the mutual influence between Pol II elongation rate and cotranscriptional splicing (11, 45–50). Rather than a linear sequence, our data support a “Bridging Model” where YTHDC1 orchestrates the physical and functional integration of these machineries. YTHDC1 is predominantly enriched at the TSS (ChIP-seq) and binds nascent intronic m6A (bfCLIP-seq). This indicates that YTHDC1 acts at the onset of the transcription cycle, prior to large-scale splicing. By forming subnuclear condensates, YTHDC1 creates a “reaction hub” that cocompartmentalizes Pol II and splicing factors. This biophysical mechanism ensures that elongation and splicing are mutually reinforcing, rather than strictly sequential. The role of YTHDC1 in cotranscriptional splicing is like a “midfielder”—passing the “ball” (nascent RNAs) between “defenders” (transcription complex) and “forwards” (splicing machinery). RNA processing, including transcription and splicing, is driven by LLPS-dependent multivalent interactions between protein–protein and protein–RNA interacting domains/motifs (31, 51). These interactions, often facilitated by intrinsically disordered regions (IDRs), lead to the formation of biomolecular condensates that compartmentalize biochemical reactions (52). Previous research has found that transcription condensate and splicing condensate are spatial-temporally associated, and phosphorylation of Pol II translocates the transcription apparatus from transcription initiation condensates to splicing condensates (29, 31, 53–55). Our study demonstrates the biophysical basis of how YTHDC1 brings splicing factors (SRSF3, SRSF7) and transcription mediators (BRD4-IDR, MED1-IDR) into proximity to form LLPS-dependent condensates. This promotes the compartmentalization and concentration of cotranscriptional splicing components, aligning with previous work showing YTHDC1 phase-separates with BRD4 to facilitate transcription condensate formation at enhancers (55). Furthermore, while arginine (R) and lysine (K) residues both contribute to phase separation due to their positive charges (30, 31), their distinct cellular expression patterns despite similar in vitro droplet-forming activities (Fig. 7 F and I) reveal how amino acid sequence precisely controls protein function and interaction. This suggests a more complex “amino acid code” determining LLPS activity, where subtle sequence differences can lead to distinct cellular outcomes. Future research is needed to fully unravel this code and explore whether other m6A readers sequester RNA processing factors in an LLPS-dependent manner, and how m6A itself shifts RBP–RNA interaction patterns. In conclusion, our findings 1) demonstrate the functional selectivity of METTL16 and YTHDC1; 2) identify a set of genes under the control of METTL16-m6A-YTHDC1 axis that affects basic cellular processes; 3) reveal the LLPS-dependent regulatory role of YTHDC1 on cotranscriptional splicing. Importantly, the METTL16-m6A-YTHDC1 axis finetunes cotranscriptional splicing events and guards the expression of development-related genes during embryonic organogenesis. Our findings provide compelling evidence in RNA-epigenetic links between transcription and splicing—two important processes in gene expression, opening prospects for future breakthroughs, specifically in developmental biology and potential diseases, and emphasizing the importance of m6A-epigenetics in gene expression regulation. Materials and Methods Generation of Conditional Knockout Mice. For conditional knockout, Mettl3fl/+, Mettl16fl/+, Ythdc1fl/+, and Ythdc1W378A/+ mice with C57BL6/J background were generated using CRISPR-Cas9 systems. Prrx1-Cre transgenic mice were purchased from The Jackson Laboratory. The design and construction of METTL3 and METTL16-conditional knock-out mouse model was reported (25, 33). The YTHDC1-floxed allele was designed with the LoxP flanked exon 5-exon 7. The primers used in this study are listed (SI Appendix, Table S3). Tail genomic DNA PCR and sequencing were performed to identify genotype. All the mice were bred and maintained under specific-pathogen-free conditions. All studies performed were approved by the Subcommittee on Research and Animal Care (SRAC) of Sichuan University. The detailed mouse construction was described in SI Appendix. Skeletal Preparation. Mice at P0 were dissected in PBS. The mice were then skinned, eviscerated, and fixed in 95% ethanol. Skeletal preparations were then performed as described (26). Extraction of Limb Bud Mesenchymal Progenitor Cell. Mouse primary limb bud mesenchymal progenitor cells were derived from 12.5 d post coitum (E12.5) from C57/BL mice. The forelimb and hindlimbs were isolated from the E12.5 embryos in Hanks Balanced Salt Solution (HBSS, Sigma) and digested in 0.5% collagenase, 0.25% dispase in HBSS at 37 °C for 10 min. The cells were disassociated by repetitive pupating with a 1 ml micropipette, spun down, resuspended in DMEM (High glucose) containing 10% FBS and 1× Antibiotics/Antimycotics (Invitrogen), and plated in 6-well plates at 2.5 × 105 cells per well. In Vitro Droplet Formation. In vitro droplet formation was performed according to previous procedures (56). mCherry-tagged and EGFP-tagged fusion proteins were cloned into pET28a and transformed into Stbl3 E.Coli strain. Purified recombinant proteins were mixed in the in vitro phase separation buffer (20 mM Tris-HCl pH 7.5, 0.01% NP-40 and 5% PEG-8000) with indicated concentration of NaCl. The mixture of proteins was placed on a slip-embedded glass. After incubation in a 37 °C humidified chamber for 30 min, the images were obtained by Olympus confocal microscope. The particle sizes of the condensates (Square Pixel) were quantified by ImageJ. Cell Culture. NIH 3T3, C3H10T1/2, and HEK 293 T were cultured in high-glucose Dulbecco’s modified Eagle’s medium (DMEM, HyClone, SH30022.01), 10% fetal bovine serum (FBS, Front Biomedical, OPT500), 1 × penicillin-streptomycin (Hyclone, SV30010). All cell lines were cultured at 37 °C in a humidified atmosphere containing 5% CO2. Mycoplasma detection tests were conducted routinely to ensure Mycoplasma-free conditions throughout the study. For EU incorporation assay, the cells were incubated with 1 μM EU for 30 min at 37 °C. Fractionation. Fractionation was performed to obtain cytosolic and nuclear fractions from WT, YTHDC1 KD, METTL3 KD, and METTL16 KD cells (57). Mesenchymal progenitor cells were transfected transiently with control siRNA (siScramble) or siRNA against YTHDC1, METTL3, or METTL16. 24 h after transfection, cells were manually collected and applied to fractionation assay. The cytosolic fraction and nucleoplasm fractions were mixed as “SN.” Antibodies. METTL3 (Abcam, #ab195352), METTL16 (Abcam, #ab252420), H3K4me3 (CST, #9751), YTHDC1 (CST, #77422), YTHDC1 (Bethyl, #A305-096A), YTHDC1 (Abcam, #122340), BRD4 (MCE, #HY-P80976), SRSF3 (Abcam, #ab198291), SRSF7 (Thermo, #303-772A-T), HA-tag (CST, #3724S), Flag-tag (Sigma, #F3165), Pol II(CST, #2629S), Pol II Ser2P (Abcam, #193468), Pol II HypoP (Abcam, #252854), Mouse IgG (Santa Cruz, #2025). Molecular Cloning. Coding sequence for full-length mouse YTHDC1, SRSF3, and SRSF7 was amplified from a mesenchymal progenitor cell cDNA library by reverse transcription-PCR. The PCR product was then cloned into a mammalian modified pCAG plasmid allowing the production of tagged proteins. Plasmids coding for BRD4-IDR and MED1-IDR were purchased from Addgene. Plasmids coding for site-mutant or truncated mutant forms of YTHDC1 were obtained by site-directed mutagenesis following manufacturer’s instructions (Q5 Site-Directed Mutagenesis Kit, NEB). Transfection of Nucleic Acids. siRNAs or plasmids were transfected using Lipofectamine RNAiMAX or 3000 (Thermofisher) following the manufacturer’s instructions. Often, 500,000 cells cultured in 6-well plates were transfected with 40 nM siRNA or 0.5 mg of plasmids, and two rounds of transfection were often conducted to increase the knockdown efficiency. For most of the experiments, 24 to 36 h after transfection, cells were harvested and used for the experiments to avoid secondary effects. CoIP-MS/MS. Mesenchymal progenitor cells from 4 10-cm plates were lysed and the soluble total extract was recovered after centrifugation at 20,000 g for 10 min at 4 C after nuclease digestion. Magnetic beads coupled to protein A were used for immunoprecipitation of YTHDC1 and the associated proteins. MS raw data were analyzed with MaxQuant (2.2.0.0) using the Andromeda database search algorithm. Spectra files were searched against the Mouse protein sequence database downloaded from Uniprot. The identified peptides and proteins were listed (SI Appendix, Table S1). Coimmunoprecipitation. HEK 293 T cells were plated onto a 6-cm dish and cotransfected with Flag-tagged plasmids encoding YTHDC1 and HA-tagged plasmids encoding SRSF3 or SRSF7. Transfected cells were collected after 48 h and further lysed by sonication with low energy. The supernatant was collected and used for immuno-precipitation. The sample was incubated with 2 μg anti-Flag antibody (Sigma, F3165). Two micrograms of mouse IgG (Santa Cruz, #2025) were applied as a negative control. Western Blot. Total protein extracts were prepared using RIPA buffer (Cell signaling, 9806) containing phosphatase inhibitors and cOmplete. 10 μg of total protein was resolved by electrophoresis on Precast Gel (Yeason), transferred onto a 0.45 μm pore PVDF membrane (Millipore). Images were captured with Chemidoc (Bio-Rad). Colocalization Analysis. Fiji plot profile software was applied to assess colocalization of fluorescent signals. The colocalization plugins were used for calculating coefficient of colocalization. Prediction of Intrinsic Disordered Domain of YTHDC1. Prediction of IDR was conducted using the VSL2 predictor in Predictor of Natural Disordered Regions (http://www.pondr.com). High-Throughput Sequencing Library Preparation and Data Processing. The library preparation and data processing of RNA-seq, scRNA-seq, ChIP-seq, RIP-seq, CUT & Tag and ATAC-seq were conducted using standard protocol. The bfCLIP-seq library was prepared according to published protocol with modifications (27). The data processing was conducted with public-available software. The detailed experimental procedures and data analysis are described in SI Appendix, Materials and Methods. Statistical Analysis. All data were expressed as mean ± SD. Unpaired two-tailed Student’s t test was used for comparison between two groups. Data between multiple groups were compared with one-way ANOVA (ANOVA) followed by the Tukey’s post hoc test. P < 0.05 was considered as statistically significant. Supplementary Material Appendix 01 (PDF)