The stability of m6A-marked transcripts is linked to cell identity in planarians Authors: Constantin Höhn, Andreas Pittroff, Anne-Sophie Gribling-Burrer, Lisa Hülsmann, Redmond P Smyth, Claus-D Kuhn Published: 2026 ABSTRACT Abstract N 6-methyladenosine (m6A) is a prevalent internal modification of eukaryotic mRNA that influences transcript fate, including mRNA stability and cell-type-specific gene expression. However, the mechanisms underlying m6A-mediated regulation remain poorly understood in many systems, including the highly regenerative planarian Schmidtea mediterranea. To address this, we generated a high-confidence atlas of ∼72 200 m6A sites across the planarian transcriptome using multiplexed direct RNA sequencing. The m6A sites follow a DRAYW consensus motif and are highly enriched near stop codons while being largely excluded from coding sequences. This pattern aligns with an exon and intron length-dependent variant of the exon junction complex (EJC)-mediated exclusion model, wherein the EJC restricts m6A deposition near splice sites. Knockdown of the m6A writer complex induced pronounced, cell-type-specific changes in transcript stability. Destabilized transcripts were enriched for intestinal markers, whereas stabilized transcripts were associated with neoblasts, the adult stem cells of planarians. Transcriptional shut-off experiments confirmed that m6A has opposing effects on mRNA decay depending on cellular context: it stabilizes transcripts in differentiated cells, while it promotes the degradation of mRNAs associated with neoblasts. Collectively, these results support a model in which cell-type-specific regulation of mRNA stability by m6A plays a crucial role in shaping cell identity in planarians. Introduction The interest in modified nucleobases in messenger RNAs (mRNAs) has seen a tempestuous renaissance in the past decade. In contrast to the omnipresence and functional importance of modified nucleobases for transfer RNAs (tRNAs) and ribosomal RNAs (rRNAs), modifications on mRNAs were only recently found to possess essential and diverse functions in almost all eukaryotic organisms [1, 2]. Out of >150 possible types of modifications detected so far, N6-methyladenosine (m6A) is by far the most prevalent, with 0.15%–0.6% of all adenosines being methylated in human cells [3]. m6A modifications require a sophisticated machinery for their installation and for mediating downstream functions. First, a large methyltransferase complex (MTC), which forms in the nucleus around the catalytic core of the METTL3–METTL14 heterodimer, installs m6A on mRNAs [4]. Second, a diverse set of both nuclear and cytoplasmic m6A reader proteins, predominantly from the YTH domain family of proteins, recognizes m6A-modified mRNAs and orchestrates downstream functions [5]. Lastly, m6A demethylases, such as ALKBH5 and FTO in humans, reverse m6A placement by actively removing it from mRNAs [6]. The molecular functions of m6A do not appear to be limited to a single mode of action. In contrast, they are diverse and range from affecting mRNA half-life [7, 8], at times in conjunction with translation [9], to altering mRNA splicing [10], promoter-proximal pause release [11], and translation efficiency [12]. The mode of action of m6A-modified adenosines is primarily determined by three factors: (i) the intracellular localization and identity of the m6A reader that mediates their effects; (ii) the precise position of the modification within individual mRNAs [7, 9]; and (iii) m6A-induced remodeling of RNA structure, which influences downstream regulatory processes and reshapes the RNA interactome [13, 14]. In Drosophila melanogaster, for example, m6A is enriched at the 5′ ends of transcripts. Consequently, the MTC, in conjunction with the reader YTHDC1, was found to regulate the release of paused RNA polymerase II (Pol II), thereby significantly affecting mRNA synthesis rate [11]. In contrast, human transcripts exhibit low—but nonetheless significant—levels of m6A across their entire open reading frame, with a characteristic enrichment of m6A around the stop codon [15]. In humans, the MTC installs m6A specifically on adenosines located within so-called DRACH motifs (where D = A/G/U, R = A/G, A = the methylated adenosine, C = cytosine, and H = A/C/U) [16]. However, despite the abundance of DRACH motifs in the human transcriptome, only a fraction of these is methylated. Selecting the adenosines that are to be methylated in coding sequences (CDSs) was found to depend on the exon junction complex (EJC), which serves as a physical obstacle to m6A placement close to exon–intron junctions in human cells. All other adenosines that are at the center of suitable DRACH motifs are equally likely methylated by the MTC [7]. However, it is unknown whether this model of m6A placement holds true for all higher organisms, in which mRNAs were found to carry m6A modifications. To tackle this question, we explored the influence of m6A on mRNA surveillance in the highly regenerative flatworm Schmidtea mediterranea, a popular model organism for studying whole-body regeneration fueled by pluripotent stem cells [17]. In S. mediterranea, roughly 24% of all transcripts carry at least one m6A modification, with 0.5% of adenosines being methylated [18, 19]. This modification plays a critical role in planarian regeneration, tissue homeostasis, stem cell differentiation—particularly in the intestinal lineage—and cell cycle regulation [18, 19]. Recent single-nucleotide resolution mapping using GLORI [15, 20] identified 19 328 high-confidence m6A sites (>10% methylation) in the planarian transcriptome [21]. While m6A marks are installed on adenosines within a degenerate sequence motif akin to the human DRACH motif, an expanded family of YTDHF m6A reader proteins forms a robust and redundant network critical for proper lineage specification in planarians [21]. Yet, despite progress in understanding the impact of m6A on stem cell differentiation and lineage commitment, its direct influence on mRNA stability and surveillance remains unexplored. This study addresses this knowledge gap, providing novel insights into the role of m6A in mRNA regulation in planarians. First, we examined the effects of MTC depletion on transcript levels and evaluated these changes considering the exact position and number of m6A sites per transcript. This analysis revealed cell type-specific gene expression changes, where loss of m6A was mainly linked to transcript downregulation, largely independent of where m6A sites were located. Second, we studied the impact of m6A on transcript stability after blocking planarian transcription by actinomycin D, which revealed that m6A exerts opposing effects on mRNA stability depending on cellular context. Whereas transcripts in cathepsin+, epidermal, and intestine cells were stabilized, we found mRNA decay to be promoted in neoblasts and muscle cells. Importantly, we based our results on precisely mapped m6A sites using Nanopore direct RNA sequencing (dRNA-seq). This approach allows for the determination of m6A sites with single-nucleotide precision. Moreover, as dRNA-seq does not rely on the chemical conversion of non-modified adenosines and additional reverse transcription steps, as does GLORI [21], the data presented in this study constitute both the most complete and the most robust dataset of planarian m6A sites available to date. Using this comprehensive dataset in combination with EJC depletion in planarians, we demonstrate that the EJC restricts MTC access to m6A sites proximal to exon-exon junctions, as previously shown for human cells [7]. However, in planarians we found the exclusion zone for m6A placement to depend on exon and adjacent intron length and to be shifted slightly downstream. Our analysis further revealed a pronounced enrichment of m6A sites near stop codons, while only few sites are present throughout the coding regions of planarian transcripts, in contrast to comparably higher levels of CDS-linked sites in humans. Collectively, our results reveal that m6A is an essential, direct actor in determining the life cycle of marked transcripts in planarians. Together with the recent study on the effect of m6A reader protein on proper lineage specification [21], this study paves the way to decipher the interconnection between m6A-marked transcripts, m6A reader proteins and proper stem cell differentiation. Materials and methods Animal husbandry The asexual S. mediterranea (clonal line CIW4; [22]) was maintained in darkness at 20°C in 1× Montjuic medium (1.6 mM NaCl, 1 mM CaCl2, 1 mM MgSO4, 0.1 mM MgCl2, 0.1 mM KCl, and 1.2 mM NaHCO3). Animals were fed once a week with homogenized calf liver. For experimental procedures, planarians were kept in small boxes filled with 1× Montjuic medium supplemented with 50 µg/ml gentamycin sulfate and maintained under the same conditions (20°C, darkness). Prior to experiments, worms were starved for a minimum of 7 days. RNAi knockdowns Inserts carrying the target sequences for mettl3, mettl14, eIF4A3, and y14 knockdowns were amplified from planarian cDNA and cloned into the pPR-T4P vector via Gibson assembly (Supplementary Table S1) [23]. Double-stranded RNA (dsRNA) for RNA interference (RNAi) was generated as previously described [24]. Briefly, the templates for in vitro transcription (IVT) were amplified by Taq-based PCR from plasmids containing the cloned target sequence. PCR products were purified with the QIAquick PCR purification Kit (Qiagen) according to the manufacturer’s protocol. dsRNA for knockdown experiments was generated by T7 RNA polymerase-mediated in vitro run-off transcription [25, 26]. The two T7 promoters flanking the fragment of interest enable the production of dsRNA. IVT reactions were prepared in 150 µl and incubated at 37°C at 600 rpm for at least 4 h. Following transcription, 17.1 µl of DNase I incubation buffer and 1.5 µl of DNase I were added to each reaction, and the mixture was incubated for 15 min at 37°C and 800 rpm. Reactions were terminated by adding 19.2 µl of 500 mM EDTA (pH 8.0) to a final concentration of 50 mM, followed by vortexing. Product dsRNA was purified by precipitation with three volumes of 100% ethanol and a final concentration of 0.3 M sodium acetate. After resuspension in RNAse-free ddH2O, dsRNA was formed by heating to 90°C for 3 min, followed by gradual cooling to 25°C at a rate of 0.1°C/s. Residual nucleotides were removed using Microspin G50 columns (GE Healthcare) following the manufacturer’s instructions. For RNAi experiments, worms were starved for at least 7 days prior to treatment and fed twice a week with calf liver mixed with a total of 3 µg dsRNA per animal. When multiple dsRNAs were fed, the amount of each individual dsRNA was adjusted accordingly. Control dsRNA (unc22) was included in knockdown experiments to monitor non-specific effects of the RNAi treatment. In vitro transcription of spike-in (luciferase) RNA The DNA template for in vitro transcription of the luciferase spike-in RNA was generated by PCR from a pUC18-luciferase plasmid. The in vitro transcription was carried out as described in the RNAi knockdown chapter with the following alteration: Following transcription and precipitation, the RNA was purified by preparative 10% urea PAGE (run in 1× TBE buffer: 89 mM Tris base, 89 mM boric acid, and 2 mM EDTA) to remove unincorporated NTPs and IVT by-products; target RNA bands were visualized by UV shadowing, excised, and eluted from the gel by passive diffusion in DEPC-treated water (crush-and-soak method). The eluate was concentrated using a 10 kDa Amicon centrifugal filter and further purified by size-exclusion chromatography (Superose 6 Increase 10/300) in SEC buffer (50 mM Na-HEPES pH 7.4, 150 mM NaCl); peak fractions containing full-length RNA were identified by analytical urea PAGE, pooled, and concentrated. RNA concentration and integrity were verified by UV absorbance (A260) and analytical 8% urea PAGE. The purified luciferase RNA was used as an exogenous spike-in control for normalization in the actinomycin D stability assay. Actinomycin D assay Knockdowns targeting mettl14 and unc22, along with wild-type (WT) controls, were established and maintained until 22 dpff. For each condition, 40 worms were fed twice a week with either dsRNA-mixed calf liver (experimental groups) or calf liver alone (WT controls). At 22 dpff, worms were transferred to Petri dishes containing 1× Montjiuc medium supplemented with actinomycin D (final concentration: 200 µg/ml). At 0, 6, and 12 h post-actinomycin D treatment, three worms per biological replicate were collected (four replicates per time point). Worms were transferred to 1.5 ml tubes containing 100 µl of TRIzol, snap-frozen in liquid nitrogen and stored at −80°C until RNA extraction and library preparation. Time points were selected based on preliminary experiments indicating detectable transcript decay after 6 h (not shown). RNA extraction for qPCR samples To extract RNA for qPCR, one or three planarians were homogenized in 100 µl of TRIzol using a douncer. The volume was adjusted to 500 µl with additional TRIzol. Hundred microliters of chloroform was added per 500 µl of TRIzol, and the mixture was mixed by inverting it. After incubating for 3 min at room temperature, samples were centrifuged for 15 min at 4°C with 12 000 × g. The upper aqueous phase, containing RNA, was transferred to a new tube. Subsequently, 1:1 volume of 100% ethanol was added to the aqueous phase, and the mixture was applied to a Zymo-Spin™ column. The column was centrifuged at 16 000 × g for 30 s and the flow-through was discarded. The column was washed with 400 µl of RNA Wash Buffer and centrifuged again for 30 s at 16 000 × g. For on-column DNase treatment, 0.5 μl of DNase I (10 U/μl) and 39.5 μl of 1× DNA Digestion Buffer (10× DNA Digestion buffer: 100 mM Tris–HCl pH 9.5, 25 mM MgCl2, 1 mM CaCl2; 1× DNA Digestion buffer: 100 µl of 10× DNA Digestion buffer, 700 µl of 100% ethanol, 200 µl of DEPC-treated ddH2O) were added to the column. The mixture was incubated at room temperature for 15 min. After incubation, the column was centrifuged at 16 000 × g for 90 s, and the flow through was discarded. The column was then washed with 400 μl of RNA Preparation Buffer, centrifuged at 16 000 × g for 30 s, and the flow-through discarded. Next, 700 μl of RNA Wash Buffer was added, the column was centrifuged at 16 000 × g for 2 min, and the flow-through was discarded. This wash step was repeated with 400 μl of RNA Wash Buffer and centrifuged for 1 min. To remove residual ethanol, the column was centrifuged again at 16 000 × g for 1 min. RNA was eluted in 10–15 μl of DEPC-treated water by incubating the column for 1 min at room temperature, followed by centrifugation at 16 000 × g for 90 s. RNA extraction for actinomycin D qPCR measurements RNA extraction was performed as described before using four worms per replicate. In deviating from the above protocol, the aqueous phase was transferred to a new tube, and an equal volume of 100% isopropanol was added. After incubation for 10 min at room temperature, the precipitate was pelleted by centrifugation for 10 min at 4°C and 12 000 × g for 10 min. The RNA pellet was subsequently washed with 1 ml of 70% ethanol and resuspended in 43 µl of DEPC-treated water. DNAse treatment [43 µl of RNA solution, 5 µl of 10× DNAse buffer, and 2 µl of DNAse I (10 U/µl)] was performed for 25 min at 37°C. For binding the RNA to the column, 100 µl of 2× RNA Binding buffer (Zymo-Research) and 150 µl of 100% ethanol were added to the DNase-treated RNA. The mixture was applied to a Zymo-Spin™ IC column, and cleanup was performed as described above. RNA was eluted in 15 µl of DEPC-treated water. Large-scale RNA extraction for Nanopore dRNA-seq For large-scale RNA extraction, 35 fully grown worms (MTC RNAi experiment; animals had been fed twice weekly for 4 weeks) or 20 fully grown worms [EJC RNAi experiment; animals had been fed once (6 dpff) or twice (8 dpff)] were picked and transferred into a medium-sized homogenizer tube. Ten milliliters of TRIzol was added, and the mixture was homogenized using a douncer. An additional 10 ml of TRIzol was added, and the mixture was homogenized again. The lysate was divided equally between two 20 ml of centrifuge tubes. To maximize RNA yield, another 10 ml of TRIzol was added to the homogenizer tube, and the residual lysate was recovered by pipetting up and down. This additional volume was split between the two centrifuge tubes. The mixture was incubated for 5 min at room temperature. Chloroform (3 ml) was added to each tube, and the contents were mixed by pipetting up and down. After incubation for 3 min at room temperature, the samples were centrifuged for 15 min at 4°C with 12 000 × g. The upper aqueous phase was transferred to a 50 ml falcon tube. Extracted RNA was purified using the Zymo RNA Clean & Concentrator Kit 25. The volume of extracted RNA solution was measured using a pipette controller. One volume of 100% ethanol was added to one volume of extracted RNA. The mixture was divided across 16 Zymo-Spin™ columns (each in a 2 ml collection tube). Due to the column’s maximum load volume of 800 µl, each column was loaded three times. On-column DNAse treatment and purification were performed as described above, but the DNAse reaction mix was scaled up twofold. Each column received 80 µl of the DNase mixture. RNA was eluted in 32 µl of DEPC-treated water. RNA extraction of actinomycin D assay samples for RNA-seq libraries To extract RNA from actinomycin D-treated samples, three worms per replicate were homogenized as described above. TRIzol (900 µl) was added to each sample, and the mixture was incubated at room temperature for 5 min. Chloroform (200 µl) was added, and the mixture was incubated again. Samples were centrifuged for 15 min at 4°C and 12 000 × g. The aqueous phase was transferred to a new tube, and an equal volume of 100% isopropanol was added. After incubation for 10 min at room temperature, the precipitate was pelleted by centrifugation for 10 min at 4°C and 12 000 × g for 10 min. The RNA pellet was subsequently washed with 1 ml of 70% ethanol and resuspended in 43 µl of DEPC-treated water. DNAse treatment [43 µl of RNA solution, 5 µl of 10× DNAse buffer, 2 µl of DNAse I (10 U/µl)] was performed for 25 min at 37°C. To size-select for RNAs >200 nt, 100 µl of 2× RNA Binding buffer (Zymo-Research) and 50 µl of 100% ethanol were added to the DNase-treated RNA. The mixture was applied to a Zymo-Spin™ IICR column, and the cleanup was performed as described above. RNA was eluted in 30 µl of DEPC-treated water. cDNA synthesis and real-time qPCR cDNA was synthesized from 1 to 5 µg total RNA using SuperScript IV Reverse Transcriptase, according to the manufacturer’s instructions. 3 pg of in vitro transcribed spike-in RNA were added for the validation of actinomycin D efficacy utilizing intronic real-time quantitative PCR (RT-qPCR). The efficiency of knockdown experiments and the reduction of target transcripts were assessed by real-time quantitative PCR (RT-qPCR). Amplification was performed using a CFX Connect™ Real-Time System, and data were analyzed with Bio-Rad CFX Manager™ software (v2.1). Primers used for qPCR are listed in Supplementary Table S1. RT-qPCR reactions were performed in technical triplicates. Relative expression was calculated using the Pfaffl method [27]. Relative transcript abundance was calculated using the comparative Ct method. Ct values of the target genes were first normalized to the luciferase spike-in (ΔCt = Cttarget − Ctluciferase). ΔCt values were then normalized to the mean ΔCt of the 0 h control samples (ΔΔCt), and relative RNA levels were calculated as 2−ΔΔCt. Transcript abundance at each time point was expressed relative to the 0 h control [28]. mRNA isolation Poly(A) mRNA was enriched using the NEBNext® Poly(A) mRNA Isolation Module for m6A specific mass spectrometry. mRNA isolation was performed according to the manufacturer’s instructions. The buffers [2× Binding buffer: 0.5 M LiCl, 0.1 M Tris–HCl (pH 8.0), 0.01 M EDTA, 1% (w/v) Lithium dodecylsulfate, 5 mM DTT, and Washing buffer: 0.15 M LiCl, 0.01 M Tris–HCl (pH 8.0), 1 mM EDTA, and 0.1% (w/v) SDS] were prepared as described [29]. For Nanopore sequencing, the NEBNext® High Input Poly(A) mRNA Isolation Module was used according to manufacturer’s instructions. In deviation from the standard protocol, elutions were performed at 70°C to preserve the integrity of longer RNA transcripts. rRNA depletion rRNA depletion was performed using the siTOOLs Biotech riboPOOL system, which is based on prior work from our lab [30]. To 14 µl of RNA, 1 µl of riboPOOL and 5 µl of Hybridization Buffer (siTOOLs) were added. The mixture was incubated at 68°C for 10 min in a heating block, followed by 30 min at 37 °C. Meanwhile, the streptavidin-coated magnetic beads were prepared. For each sample, 90 µl of bead suspension was transferred to a reaction tube and placed on a magnetic rack until the solution cleared. The supernatant was removed, and the beads were resuspended in 120 µl of Depletion Buffer per sample. The bead suspension was split into two tubes (tube 1 and tube 2). After placing the tubes on the magnetic rack and removing the supernatants, the beads in tube 1 were resuspended in 220 µl of Depletion buffer, and the beads of tube 2 were resuspended in 120 µl of depletion buffer per sample. To capture rRNA–DNA hybrids, 220 µl of beads from tube 1 were added to the hybridized RNA sample and incubated for 15 min at 37°C. Tubes were tapped every 3 min to keep the beads in suspension. The tube was then placed on the magnetic rack, and the supernatant was transferred to a new tube. An additional 120 µl of beads from tube 2 were added to the supernatant, followed by another incubation for 15 min at 37°C. After placing the tube on the magnetic rack, the supernatant was mixed with 360 µl of 2× RNA Binding buffer (Zymo-Research) and 360 µl of 100% ethanol. The entire binding mixture was loaded onto a Zymo-IC column and RNA purification was performed as described previously (RNA extraction for qPCR samples). In deviation from the standard protocol, the on-column DNAse treatment mixture was performed using 1 µl of DNase I (recombinant, RNase-free; Roche, 10 U/µl), 3 µl of 10× DNase I Buffer (Roche) and 26 µl of RNA Wash Buffer (Zymo-Research). Mass spectrometric analysis To quantify alterations in m6A levels on mRNAs following knockdown experiments, poly(A) selected RNA was analyzed by Cathrine Broberg Vågbø, PhD. at the Proteomics and Modomics Experimental Core Facility (PROMEC) of the Norwegian University of Science and Technology (NTNU). Relative m6A levels were calculated as the number of m6A-modified adenosines divided by the total number of adenosines, as determined by mass spectrometry. Percentage reductions upon knockdown were calculated compared to the m6A level in WT samples. Statistical analyses of the resulting data were performed using the Tukey-HSD test with P-values adjusted for multiple comparisons. RNA library preparation and NovaSeq sequencing Library preparation and sequencing of the actinomycin D assay was performed from rRNA depleted RNA by Genewiz Germany GmbH. RNA samples were quantified using Qubit 4.0 Fluorometer (Life Technologies, Carlsbad, CA, USA), and RNA integrity was checked with RNA Kit on Agilent 5300 Fragment Analyzer (Agilent Technologies, Palo Alto, CA, USA). RNA sequencing libraries were prepared using the NEBNext Ultra II Directional RNA Library Prep Kit for Illumina (NEB, Ipswich, MA, USA), according to the manufacturer’s general workflow. Briefly, ribo-depleted RNA was taken directly into the fragmentation and first-strand synthesis module. First-strand cDNA was generated using random priming, and second-strand synthesis was performed in the presence of dUTP to enable strand specificity through downstream degradation of the second strand during amplification. Following double-stranded cDNA generation, fragments underwent 3′ end-repair and adenylation. Adapters containing unique molecular identifiers (UMIs) were ligated to the A-tailed cDNA fragments. After adapter ligation, libraries were amplified using limited-cycle PCR to enrich adapter-ligated molecules and selectively degrade the dUTP-containing second strand, preserving library directionality. Completed libraries were assessed for fragment size distribution using an NGS fragment analysis system (e.g. Agilent 5300 Fragment Analyzer; Agilent Technologies, Palo Alto, CA, USA) and quantified using a fluorescence-based assay (e.g. Qubit 4.0 Fluorometer; Invitrogen, Carlsbad, CA, USA). The sequencing libraries were multiplexed and loaded on the flow cell on the Illumina NovaSeq X Plus instrument according to manufacturer’s instructions. The samples were sequenced using a 2 × 150 Pair-End (PE) configuration v1.5. Image analysis and base calling were conducted by the NovaSeq Control Software on the NovaSeq instrument. Raw sequence data (.bcl files) generated from Illumina NovaSeq was converted into fastq files and de-multiplexed using Illumina bcl2fastq program version 2.20. One mismatch was allowed for index sequence identification. After investigating the quality of the raw data, sequence reads were trimmed to remove possible adapter sequences. Nanopore direct RNA sequencing and base calling Nanopore direct RNA sequencing was performed following the WarpDemuX adapter-barcoding and demultiplexing strategy as described [31]. Briefly, custom reverse transcription adapters (RTA) (10 µM forward and 10 µM reverse) were annealed in 30 mM HEPES-KOH (pH 7.5), 100 mM K-acetate by incubation for 1 min at 95°C followed by a slow cooling to 25°C at a rate of 0.5°C/min. The annealed reverse transcription adapters (RTA) were then diluted to 0.7 µM and stored at −20°C until use. For each sample, 100 ng of polyA-selected RNA were ligated to 1 µl of 0.7 µM annealed RTA in the presence of 2 µl of 5× NEBNext Quick Ligation Buffer (NEB), 0.5 µl of T4 DNA Ligase high concentration (NEB), and 0.2 µl of RNAsin (Promega) in a final volume of 10 µl by incubation at 22°C for 15 min. To stop the ligation, 2 µl of 0.5 M EDTA was added and samples were pooled and purified with 0.5 volumes of SPRI beads (Mag-Bind TotalPure NGS, Omega Bio-tek), washed twice with 200 µl of fresh 80% EtOH, and eluted in 10 µl of RNase-free H2O. Reverse transcription was performed by adding 8 µl of 5X Induro buffer, 2 µl of 10 mM dNTP, 2 µl of 10 µM random hexamer oligonucleotides, 2 µl of Induro RT (NEB), and 0.5 µl of RNasin in a final volume of 40 µl and incubated for 2 min at 22°C (annealing), 15 min at 60°C (reverse transcription), and 10 min at 70°C (heat-inactivation). The reaction was purified with 1.4 volumes of SPRI beads, washed twice with 200 µl of fresh 80% EtOH, and eluted in 10 µl of RNase-free H2O. The sequencing adapter ligation was realized using the SQK-RNA004 library kit (ONT) according to the manufacturer’s instruction. Briefly, the 10 µl purified library was mixed with 6 µl of 5× NEBNext Quick Ligation Buffer, 5 µl of RLA sequencing adapter (ONT SQK-RNA004), and 3 µl of T4 DNA Ligase high concentration in a final volume of 30 µl and incubated 15 min at 22°C. The reaction was purified with 0.5 volumes of SPRI beads and washed twice with 100 µl of RNA wash buffer (WSB, ONT SQK-RNA004). The library was then eluted in 32 µl of RNA elution buffer (REB, ONT SQK-RNA004). The library was loaded onto a Promethion (FLO-PRO004RA) flow cell for sequencing. Data acquisition was performed using MinKNOW version 24.11. Basecalling was performed with dorado v0.8.3 using the model parameters sup v5.0.0 m6A with poly(A) length estimation. Demultiplexing was performed using the demux command of WarpDemuX v0.4.5. Processing, mapping, and m6A calling of Nanopore dRNA-seq data All genome and transcriptome reference files (version schMedS3) used in this study were retrieved from [32]. Sequences from demultiplexed Nanopore dRNA-Seq libraries were extracted using samtools fastq [33]. Prior to mapping, reads were aligned to a combined reference database comprising S. mediterranea rRNA sequences [30] and tRNA sequences (GtRNAdb; http://gtrnadb.ucsc.edu). Unmapped reads were filtered using the samtools flag -f 0 × 4, and the resulting rRNA- and tRNA-depleted reads were then mapped to the planarian genome. Both mapping steps were performed using minimap2 (-ax map-ont -uf -k 14; -ax splice -uf -k 14 respectively [34, 35]. Mapped reads were filtered using the samtools flag -F 0 × 4 to retain primary and secondary alignments for downstream analysis. m6A modification probabilities and levels were assessed using modkit’ sample-probs function (Modkit, 2023/2026). Relative m6A levels were calculated as the number of m6A-modified adenosines divided by the total number of adenosines as identified by modkit [36]. Percentage reductions upon knockdown were calculated in comparison to the relative m6A level in WT samples. To construct bedMethyl tables, genome-mapped reads were processed with modkit’s call-mods and pileup utilities (modification threshold = a:0.99; filter-threshold 0.9; Modkit, 2023/2026). The resulting bedMethyl tables were filtered to retain m6A calls covered by ≥10 reads and ≥15% methylation rate. Calls were compared across replicates and only those fulfilling these criteria in at least two of three replicates were retained for further analysis (Supplementary Table S2). Motif assessment and m6A distribution Metaplots of m6A distribution were generated using the R package metaplotR [37]. GenePred files were derived from the planarian trancriptome (version schMedS3; [32]) using UCSC’s gtftoGenePred utility (https://github.com/BioQueue/ucsc-tools). To investigate the sequence context of m6A installation, previously identified m6A sites were expanded by two nucleotides in each direction using bedtools slop (-b 2). The resulting 5-mer sequences were extracted with bedtools getfasta [38]. Nucleotide distributions within these 5-mer sequences were assessed, and nucleotides with a contribution exceeding 20% at any individual position were considered as part of the m6A installation consensus. Sequence motifs were visualized in R using the Biostrings [39], seqLogo [40], and DiffLogo [41] packages. The S. mediterranea genome was scanned for all possible 5-mer combinations of nucleotides within the consensus motif using patman [42]. Genic motif occurrences were intersected with bedMethyl tables using bedtools intersect [38] to assess the distribution of methylation rates for each individual 5-mer. Comparison of Nanopore dRNA-seq with GLORI data Yesharim et al. (2026) provided a file of GLORI detected m6A sites lifted to the S. mediterranea genome version schMedS3. In brief, identified m6A sites were expanded by 5 nt in both directions, and the resulting sequences were mapped to schMedS3 [32]. We compared m6A sites identified by Nanopore dRNA-Seq with those from GLORI using two approaches: first, all GLORI sites as reported in Yesharim et al. (defined by the highest methylation rate in any of the three replicates), second, with a strict filter reflecting the criteria used for Nanopore dRNA-Seq data (methylation rate threshold (10%) met in at least two of three replicates). Overlap visualization was performed using the eulerr package in R [43]. To assess the agreement in methylation rate measurements between Nanopore and GLORI for shared sites, we calculated the Pearson correlation coefficient for mean methylation rates across biological replicates. For each shared site, mean methylation rates were computed for both methods, and their correlation was tested using cor.test in R (method = “pearson”). Inter-replicate variability was quantified using the coefficient of variation (CV), defined as: Differences in CVs between methods were assessed using a one-sided Wilcoxon rank-sum test (wilcox.test, alternative = “less”) to test whether Nanopore dRNA-Seq exhibits lower variability than GLORI. Differential gene expression analysis Nanopore dRNA-Seq reads were mapped to the planarian transcriptome [32] using minimap2 (-ax map-ont -k 14 -N 10; [34, 35]) and quantified with salmon (salmon quant –ont –noLengthCorrection -l A; [44]). Resulting transcript level abundance estimates were summarized to gene level using tximport [45]. Differential gene expression analysis was performed using DESeq2 [46] with the design formula ∼condition. All possible comparisons were calculated: metll14 (RNAi) versus WT, unc22 (RNAi) versus WT, and mettl14 (RNAi) versus unc22 (RNAi). Genes with a log2 fold change exceeding ±0.5 and an adjusted P-value < 0.05 were considered significantly deregulated. Log2 fold changes were shrunk using DESeq2’s lfcShrink function (type = “normal”). Volcano plots were generated using the EnhancedVolcano package [47]. Additionally, we calculated the log2 fold change of gene-wise m6A site abundance to reflect differential gene expression with respect to respective m6A changes: Cell identity mapping for genes of interest Single-cell expression data were downloaded from the planarian single-cell atlas [48]. To generate a reference dataset of cell-type specific markers, we selected transcripts associated with only one distinct cell type. Transcripts with ambiguous or multi-tissue associations were excluded from the reference dataset. Since the planarian single-cell atlas is based on an older version of the planarian transcriptome version, we converted dd_v6 transcript IDs to schMedS3 transcript IDs to enable comparison with our dataset. To achieve this, dd_v6 transcripts were mapped to the schMedS3 transcriptome using minimap2 (-x splice -k 14 -N 10; [34, 35]). The resulting PAF file was filtered for sense hits with ≥80% match rate. Secondary alignments were retained only for isoforms of the primary alignment. This process allowed us to assign dd_v6 IDs to ∼50.78% of schMedS3 transcripts. Where possible, schMedS3 transcripts IDs were converted to dd_v6 IDs for genes of interest and compared to the reference database of cell-type-specific markers. Overrepresentation and depletion of cell types was determined using an odds-ratio, with trends categorized as biologically meaningful if the relative change in representation exceeded 0.3. Methylation frequency at exon–intron junctions Exons in the S. mediterranea transcriptome were classified as single, first, internal, or last exons based on annotated transcripts, accounting for multiple definitions of the same exon across different isoforms. m6A sites and motif occurrences were overlapped with these exon annotations using bedtools intersect [38]. m6A-harboring exons were then stratified into four length classes based on exon length quartiles. The distance of m6A sites and motifs to the respective exon boundaries was calculated in bins of 10 nt. To model methylation frequencies as a function of distance to internal exon starts and ends, as well as last exon starts, generalized additive models (GAMs) with a binomial distribution and logit link function were fitted separately for each exon length class using the gam() function from the R package mgcv [49]. The models had the following form (in R formula notation): Here, Sites represents the number of observed m6A sites per 10-nt bin, Motifs represents the total number of detected motif occurrences in the same bin (as described in section Motif assessment and m6A distribution), and Distance is the distance from the exon junction of the corresponding bin (each consisting of 10 nucleotides). The response term cbind(Sites, Motifs − Sites) encodes the number of successes (methylated motifs) and failures (unmethylated motifs) for the binomial model. The smooth term s(Distance, k = 6) allows a flexible, non-linear relationship between methylation probability and distance from the exon boundary while limiting the spline complexity to six basis functions. For calculation of continuous GAMs, we merged the nucleotide bins up- and downstream of the exon splice junction, creating a total of 100 (−50 to +50) bins of 10 nucleotides each. Models for human data were modeled in the same way. Bayesian mixture model for the effect of m6A position on differential gene expression upon m6A loss We fitted a Bayesian hierarchical mixture-of-experts model to transcript-level log2 fold change estimates of gene expression upon m6A loss, with three components representing negative, null, and positive regulatory effects. Observations were modeled using Student-t likelihoods with component-specific means and a scale parameter given by sqrt(se^2 + sigma^2), combining known measurement error with an estimated residual variance term. The Student-t degrees of freedom parameter was estimated from the data, allowing tail heaviness to adapt flexibly. The negative and positive component means were parameterized as signed softplus-transformed offsets beyond a minimum effect size of 0.5, applied to linear predictors in m6A position, ensuring directional consistency while allowing effect magnitude to vary smoothly with m6A position along transcripts. A hierarchical prior linked the positive and negative component magnitude parameters through a shared location-scale structure, providing partial pooling while preserving biologically interpretable sign constraints. The null component mean was strongly regularized around zero. Mixture probabilities were modeled through a multinomial logistic gating function with the null component as the reference category, allowing regulatory state probabilities to vary with m6A position. The slopes governing position dependence of the gating probabilities were strongly regularized to reduce confounding between changes in mixture membership and changes in component-specific effect magnitudes. We included all transcripts that lost at least 80% of their m6A sites, defined as those with a methylation rate of at least 15% before and 0% after METTL14 knockdown (n = 3966). For each transcript, m6A position was defined as the average position of all lost m6A sites within that transcript. Poly(A) tail analysis from Nanopore dRNA-seq data Poly(A) tail status and length was assessed by parsing the pt:i: flag from Nanopore dRNA-Seq reads. The pt:i: flag reports the number of nucleotides in the poly(A) tail, as estimated by Nanopore’s basecalling algorithm. To analyze poly(A) tailing in distinct gene clusters, reads originating from these features were identified using bedtools intersect [38]. Read-IDs from these overlaps were extracted and used to select the corresponding Nanopore dRNA-Seq reads for further analysis. Transcript-level poly(A) tail length quantification and differential analysis To assign each read to a specific transcript isoform, reads were separately aligned to the canonical transcriptome, and for each read the alignment with the highest mapping quality (MAPQ) was retained as its best mapping (samtools view -q 1); read IDs, best-mapping transcript IDs, and poly(A) tail lengths were merged into a single per-read table for each of the nine libraries (WT, unc22 (RNAi), and mettl14 (RNAi); three replicates each). For each transcript, mean poly(A) tail length was first calculated per individual library, retaining only transcripts with a minimum of five supporting reads in that library. Per-condition matrices were then built from the resulting per-library means, requiring complete data (3/3 replicates) for a transcript to be included. Differential poly(A) tail length between conditions was assessed using limma, fitting a linear model to the per-library mean poly(A) length matrix with condition as the design factor. Mean-variance trend fitting (eBayes(…, trend = TRUE)) was applied based on inspection of the sigma-versus-average-expression relationship (limma::plotSA), which showed increasing residual variance with mean poly(A) tail length. Pairwise contrasts [mettl14 (RNAi) versus WT, mettl14 (RNAi) versus unc22 (RNAi), and unc22 (RNAi) versus WT) were first evaluated to confirm that WT and unc22 (RNAi) did not differ significantly at the transcript level. As no significant differences in poly(A) tail length were found, this justified their pooling into a single control group to increase statistical power for the primary contrast [mettl14 (RNAi) versus pooled WT/unc22 (RNAi)]. Transcripts with a Benjamini–Hochberg adjusted P-value ≤ 0.05 in this contrast were considered to have a significant difference in poly(A) tail length. Overlap of poly(A)-switch transcripts with m6A status (WT) was assessed by direct intersection. m6A position and poly(A) tail length The position of each m6A site relative to transcript structure was determined using metaPlotR [37], which assigns each site a metagene coordinate scaled such that the start codon is at 1 and the stop codon at 2 (5′UTR: 0–1; CDS: 1–2; 3′ UTR: 2–3). For each m6A-marked transcript with a complete gene model and sufficient read coverage in all libraries (n = 10 702; see above), the minimal absolute distance between any m6A site and the stop codon was calculated and related to that transcript’s poly(A) tail length difference [mettl14 (RNAi) versus pooled controls] using linear regression. To test whether this effect differed depending on whether the closest site was located upstream (CDS) or downstream (3′ UTR) of the stop codon, mean poly(A) length differences were compared between the two groups using a Welch’s t-test, and an interaction model (poly(A) length difference ∼ distance × region) was fitted to test whether the slope of the distance-dependent effect differed between regions. Alternative polyadenylation site usage To assess alternative polyadenylation site usage, reads were aligned to the genome, and the 3′-most mapped reference coordinate of each read (strand-corrected: the rightmost coordinate for + strand genes, the leftmost for − strand genes) was taken as a proxy for the read’s polyadenylation site. Read-level genomic mappings were linked to transcript identity via the same best-mapping assignment used for the poly(A) length analysis, requiring agreement between the chromosome of the genomic alignment and that of the annotated transcript to exclude reads mapping to redundant, unannotated copies of duplicated genomic sequence present in the assembly. Reads implying a cleavage site >20 kb from the annotated transcript end were additionally excluded as likely residual mapping artifacts (paralogous/duplicated loci; affecting < 5% of chromosome-matched reads). For each transcript, the mean distance between the read-derived and annotated 3′ end was calculated per library (minimum 5 supporting reads) and then per condition (requiring complete 3/3 replicate data) and compared between mettl14 (RNAi) and pooled WT/unc22 (RNAi) controls. Transcripts were classified according to whether they carried an m6A site in the 3′ UTR (metagene coordinate ≥ 2) in WT, and whether this mark was retained or lost in mettl14 (RNAi). Differences in poly(A) site position between groups (loss versus retention of 3′ UTR m6A; presence versus absence of 3′ UTR m6A in WT) were tested using Wilcoxon rank-sum tests; effect sizes were additionally calculated (Cohen’s d) to assess the biological magnitude of any detected differences given the large sample sizes involved. Processing and mapping of actinomycin D assay sequencing data Adapters were trimmed from paired-end reads using trimmomatic in paired-end mode to retain synchronized reads (ILLUMINACLIP:2:30:10, SLIDINGWINDOW:5:20; [50]). rRNA and tRNA reads were filtered using SortMeRNA and bowtie2, respectively ([30, 51–53]; http://gtrnadb.ucsc.edu). Filtered reads were mapped to the planarian transcriptome (version schMedS3; [32]) using bowtie2 (–no-unal, –very-sensitive, –no-mixed; [53]). PCR duplicates introduced during library preparation were collapsed using umicollapse [54]. Transcriptome-mapped and UMI-collapsed reads were quantified using salmon [44]. Analysis of transcript decay rates in actinomycin D assay Salmon quantifications of transcriptome-mapped sequencing data were imported and summarized to gene-level counts using tximport [45]. Counts were normalized to untransformed counts per million (CPM) using edgeR [55]. Since mettl14 dsRNA fragments were detected in the sequencing data, mettl14 gene counts were removed prior to normalization. Normalized gene counts were filtered to retain only m6A-harboring genes. Linear models of gene abundance over the time course of the actinomycin D assay were computed for each condition separately (gene expression ∼ time). Log2 fold changes between the 0-h and 6-h or 12-h time points were calculated for each gene. P-values from the linear models were adjusted for multiple testing using the Benjamini–Höchberg procedure. To reduce noise, genes displaying a positive slope in gene abundance (P-adj ≤ 0.05) and a log2 fold change ≥ 0.2 over the time course were excluded from further analysis. CPM values were centered to the 0-hour time point for each individual gene using the following formula: where CPM is the counts per million at the respective timepoint, α is a pseudocount (0.0001) to avoid division by 0, CPMref = CPM0h + α, resulting in relative CPM values. Using maSigPro [56], we defined a quadratic regression model (degree = 2) for the time course experiment to capture non-linear decay dynamics over time. We identified significant genes using the p.vector function (Q = 0.01, MT.adjust = “BH”, min.obs = 20). After removing influential genes as defined by the T.fit function, stepwise regression was performed using the two-ways forward method at a significance level of 0.05. The optimal number of clusters for the differential abundance profile time-series analysis was estimated using within-cluster sum of squares (WCSS), silhouette analysis, and principal component analysis (PCA). Time-series analysis was conducted, and genes with significantly differential gene abundance profiles between unc22 (RNAi) and mettl14 (RNAi) conditions were selected using maSigPro’s see.genes function (show.fit = T, cluster.method = “hclust”, cluster.data = 1, k = 2; [56]) Results A high-confidence planarian m6A atlas derived from multiplexed Nanopore direct RNA sequencing When analyzing the most recent version of the planarian transcriptome [32] for components of the m6A machinery, we identified core components of the MTC, as well as numerous m6A reader proteins of the YTH and HNRNP families via BLAST, in line with previous reports [19, 21] (Supplementary Table S3). Notably, and as is the case in D. melanogaster [57], we found no evidence of m6A demethylases (“erasers”) in the planarian transcriptome. While this suggests that m6A marks may not be actively removed in planarians, we cannot exclude the possibility that uncharacterized or divergent erasers exist. Previous studies have demonstrated that the YTHDF reader family is expanded in planarians [21], which, together with the bona fide MTC (with METTL3 and METTL14 at its core), suggests that m6A is installed on planarian mRNAs and subsequently read by a diverse set of m6A readers to exert numerous biological functions [18, 19, 21]. To establish the basis for investigating the direct effect of m6A on planarian transcripts, we applied dRNA-seq to polyA-selected mRNA isolated from whole animals. We did so to detect and characterize all m6A sites that are robustly methylated in the transcriptome of S. mediterranea with single nucleotide precision. To minimize batch effects and other sources of error, we utilized a novel protocol that allows multiplexing of up to six Nanopore dRNA-seq libraries into one sequencing run (Fig. 1A) [31]. Using this method, we obtained three dRNA-seq datasets averaging ∼5.9M reads with a median read length of 464 nucleotides (nt) and N50 value of 687 nt (Supplementary Table S4). Transcriptome coverage was extensive, as we were able to detect corresponding direct RNA reads for 18 095 (84.6%) of 21 401 genes in the planarian genome in at least one replicate, and for 14 809 (69.2%) in all three replicates. Oxford Nanopore’s Dorado software was used to directly detect N6-methyladenosines during base calling. This procedure resulted in a robust base modification profile, which allowed us to use a stringent cutoff of 99% base modification probability for the calculation of bed-methyl tables with modkit [36] (Fig. 1B). Next, we analyzed the methylation rate of individual m6A sites, which revealed that most sites exhibit methylation levels below 15%, a finding that becomes even more pronounced at base modification probability thresholds of 60% or 80% (Fig. 1C). Consequently, we excluded these lowly methylated sites from all subsequent analysis. Moreover, because low methylation rates could severely impede connecting observed transcript changes to m6A methylation, we decided to only consider sites with methylation rates ≥15% and required every m6A site to meet this threshold in at least two out of three replicates for further analysis (red dashed bar in Fig. 1C). This threshold mimics the cutoff applied to human GLORI data [15] and is more stringent than presented for planarians before [21]. At a methylation threshold of 15%, our analysis identified 72 237 high confidence m6A sites distributed over 28 221 transcripts and 12 583 genes respectively, rendering this dataset the most comprehensive m6A resource for planarians to date (Supplementary Table S2 lists genomic coordinates of all sites meeting this threshold with their respective methylation levels). Identification and characterization of high confidence m6A sites in S. mediterranea. (A) Poly(A) RNA is ligated to a barcoded reverse transcription adapter, followed by attachment of sequencing adapters and a motor protein. Individual RNA molecules are sequenced directly as they translocate through a Nanopore, generating characteristic current signals that are used for base calling and demultiplexing, yielding high-quality RNA sequencing data and enabling the detection of RNA modifications. The integration of barcoded adapters allows for multiplexing, which enables high-quality transcriptome profiling with minimized batch effects and detection of m6A directly from raw signal [31]. (B) m6A modification probability profile in Nanopore dRNA-Seq reads; the red dashed line represents the applied cutoff (99%), green bars represent the modifications that fulfill this threshold. (C) Methylation rates of individual m6A sites applying 60%, 80%, and 99% modification probability as cutoffs. Lowering the modification probability threshold increased the fraction of sites with very low apparent methylation rates (≤10%), whereas a 99% cutoff minimized low-confidence calls and was therefore used for high-confidence m6A site definition. (D) Metagene analysis reveals strong enrichment of m6A sites (15% methylation rate in ≥ 2/3 replicates) at the CDS–3′ UTR junction, with depletion from the 5′ UTR and the CDS. (E) Applying a ≥20% contribution threshold at each position within ±2 bp of m6A sites yielded a DRAYW consensus motif. (F) Individual 5-mer motifs with an AC core exhibit substantially higher methylation rates than AU-core motifs. (G) Metagene plots of individual 5-mer consensus motifs reveal preferential enrichment of weakly methylated variants in UTRs, in contrast to the more uniform distribution of highly methylated motifs. Line color represents individual methylation rates as identified in (F). We detected a median of 7, 5, and 3 m6A sites per planarian gene at methylation rate cutoffs of 15%, 30%, and 60%, respectively (Supplementary Fig. S1A). To validate our Nanopore dRNA-Seq-based m6A sites, we compared our dataset to the recently published GLORI-seq dataset [21]. In total, 18 694 GLORI sites were lifted to the S. mediterranea genome version schMedS3 ([32]; see Materials and methods), of which 12 453 (66.61%) were captured in our Nanopore dataset. When applying the same consistency filter of sites having to meet the methylation threshold in at least two of three replicates, 8574 GLORI sites remained of which 7334 (85.5%) overlapped with our Nanopore-based sites (Supplementary Fig. S1B). Both methods showed strong agreement in mean methylation rates of shared sites (Pearson’s r = 0.825; 95% CI 0.818–0.833; P < 2.2 × 10⁻¹⁶; Supplementary Fig. S1C). To assess inter-replicate consistency of both methods, we calculated the coefficient of variation (CV) for all shared sites (n = 7334; see Materials and methods). Nanopore CVs were significantly lower than GLORI CVs (P < 2.2 × 10⁻¹⁶, Wilcoxon rank-sum test), indicating that our Nanopore data demonstrate greater reproducibility than GLORI, most likely because Nanopore direct RNA-seq omits the chemical-conversion step that likely introduces variable, sequence-dependent biases in the GLORI method [15]. Specifically, 85.1% (n = 6 239) of Nanopore measurements had a CV below 15%, while only 48.8% (n = 3580) of sites cleared the same CV threshold in GLORI (Supplementary Fig. S1D). A subsequent metagene analysis of our Nanopore sites across planarian transcripts reveals them to be markedly enriched around stop codons and depleted from the 5′ UTR and CDS (Fig. 1D). This is in line with previous reports [19, 21] and confirms that m6A sites in planarians are positioned across transcripts in a manner similar to that observed in humans and other vertebrates [15, 58]. In contrast, this pattern differs from that described for D. melanogaster, where, in addition to a prominent peak of m6A in the 3′ UTR, m6A is also found in the 5′ UTR of transcripts [10, 59]. Next, we used the dRNA-seq data to identify the consensus motif for m6A placement in planarians. To this end, we extended called m6A sites by ±2 nucleotides and extracted the sequences of these 5-mers transcriptome-wide. We considered bases with a contribution of ≥20% at individual positions for the consensus motif. This approach revealed that, in planarians, m6A is predominantly installed in a DRAYW sequence context (or DRACW when using 30% and 60% cutoffs for motif methylation rate; with D = not C, R = A or G, Y = C or T and W = A or T) (Fig. 1E). The DRAYW motif in planarians is reminiscent of the consensus motifs previously described for vertebrates (DRACH, with H = not G) [15, 58] and D. melanogaster (RRACH) [10]. The planarian-specific divergence can likely be attributed to the high AT content of the planarian genome [60], and an adaptation of the writer machinery’s motif recognition from H to W at the +2 position of the m6A sequence motif context. Having identified a consensus motif for m6A placement, we next asked which of the individual 5-mers is most efficiently methylated. To that end, we plotted the methylation rate as a function of the methylated 5-mer (Fig. 1F). Interestingly, this analysis revealed the critical importance of a + 1 cytosine for efficient methylation by the planarian MTC, as corroborated by GLORI data [21]. All 5-mers containing a + 1 cytosine (+1C) exhibited average methylation levels exceeding 45%, while the inclusion of a + 1 uridine (+1U) resulted in a clear decrease in the average methylation rate to about 25% to 35%. This preference is also evident when comparing motif-specific methylation rates with the occurrence of individual 5-mers (Supplementary Fig. S1E). To confirm that it is solely the interaction of the MTC with the motif that determines the methylation rate and to exclude that the rate is influenced by the location of the individual sites within transcripts, we created a metaplot of individual 5-mer occurrences along planarian transcripts (Fig. 1G). Interestingly, this analysis revealed the significant enrichment of inefficiently methylated motifs in both 5′ and 3′ UTRs. In contrast, strongly methylated motifs are distributed more evenly along transcripts. These observations allow us to draw the following conclusions: First, in the CDS and 3′ UTR of planarian transcripts m6A motif occurrence can qualitatively explain the presence of methylated adenosines (Fig. 1D and G). Second, in planarians, unlike in D. melanogaster [10, 11], a mechanism must exists that prevents the MTC from accessing the 5′ UTR of transcripts. To corroborate our findings in a better-studied system, we plotted the occurrence of all permutations of the DRACH consensus motif in the human transcriptome (Supplementary Fig. S1F). Intriguingly, and in contrast to the situation in planarians, human transcripts exhibit elevated levels of CDS-based motifs, resulting in a relatively even distribution of DRACH consensus motifs along transcripts. This pattern suggests that, in humans, the MTC needs to be specifically recruited to the STOP codon of transcripts, e.g. by its VIRMA/KIAA1429 subunit [61]. Planarians also possess a KIAA1429 homolog [19, 21], which may contribute to MTC recruitment to the stop codon, as suggested by the slight offset between motif occurrence and the peak in m6A sites (Fig. 1D and G). Taken together, multiplexed Nanopore dRNA-seq enabled us to generate a high-confidence atlas of ∼72 200 m6A sites across the planarian transcriptome, which—owing to its stringency and robustness—provides a valuable reference resource for the field. Most planarian genes harbor one to five m6A sites, strongly enriched near stop codons and depleted from 5′ UTRs and CDS regions. Together with the identification of a planarian-specific DRAYW consensus motif that shows strong dependence on a +1 cytosine for efficient methylation, these results suggest that the m6A profiles of planarians are more similar to those of vertebrates than to those of Drosophila [15, 57–59]. Differential methylation frequency of m6A motifs at exon starts and ends reveals an altered exclusion zone in planarians To assess whether the EJC-mediated exclusion zone model of m6A placement applies to planarians [7], we analyzed the distribution of m6A sites across different length classes of planarian exons. Specifically, we used flexible (see Materials and methods) to model a metric that we termed methylation frequency—defined as the proportion of consensus DRAYW motifs that are methylated—across a ±500 nt window surrounding splice junctions for internal exon starts, internal exon ends, and last exons (Fig. 2A). Because N6-methylated adenosines are directly identified from Nanopore raw signal, we could not omit chemical treatment to derive this metric, as has been done for GLORI data [15, 21]. Instead, we defined methylation frequency as the ratio of the abundance of methylated consensus DRAYW motifs to the abundance of possible permutations of the motif across the planarian transcriptome (Fig. 1F and G). To examine exon length-dependent trends, we stratified planarian m6A-marked exons into quartiles. For internal exons, these classes comprised exons <242 nt (shortest), 242–415 nt (short), 416–727 nt (long), and >727 nt (longest); for last exons, the corresponding ranges were <251 nt (shortest), 251–385 nt (short), 386–640 nt (long), and >640 nt (longest) (Fig. 2A). Planarians display an altered m6A exclusion zone at exon–exon junctions. (A) GAMs of methylation frequency at exon boundaries reveal a pronounced depletion of m6A at exon starts, whereas exon ends show no detectable depletion in short exons. GAMs were fitted for exon length classes defined by length quartiles. The exon length classes cover the following length ranges: <242 nt (shortest), 242–415 nt (short), 416–727 nt (long), and >727 nt (longest) for internal exons; for last exons, the corresponding ranges are <251 nt (shortest), 251–385 nt (short), 386–640 nt (long), and >640 nt (longest). Shaded areas indicate 95% confidence intervals of the model fit; methylation frequency was tested in bins of 10 nt. The number of exons and transcripts on which this analysis is based is listed in Supplementary Table S5. (B) GAMs across splice junctions uncover a shifted m6A exclusion zone ∼20 nt downstream of exon–intron junctions, which is narrowed in short exons and widens with increasing exon length. GAMs were fitted separately for exon length classes defined by length quartiles. Shaded areas denote 95% confidence intervals of the model fit; methylation frequency was tested in bins of 10 nt. (C) Comparison of exon–intron architecture highlights major differences between planarian and human genes: planarian genes exhibit a prominent peak of short introns accompanied by a broad distribution of long introns, whereas human genes are characterized predominantly by long introns. We observed a depletion of m6A at the start of both internal and last exons across all length classes (Fig. 2A). In contrast, the extent of m6A depletion at internal exon ends varied with exon length. In the shortest exon length class (<242 nt) we did not observe any depletion; however, a shallow exclusion zone up to ∼200 nt from the exon-intron junction became apparent in the two long exon quartiles (Fig. 2A). To obtain a comprehensive view of regions flanking splice junctions, we combined methylation frequency profiles from internal exon ends and starts and fitted a continuous GAM across ± 500 nt. Apart from the longest exon quartile, m6A depletion peaked approximately 20–50 nt downstream of the splice junction in all exon length classes (Fig. 2B). Moreover, with decreasing exon length we found the resulting exclusion zone to become narrower. While its spans ∼100 nt in short exons, it is expanded to 200 nt in long exons, a width comparable to what was observed in humans (Supplementary Fig. S2A and B). To further evaluate the reduced methylation frequency at exon–intron junctions, we repeated the motif occurrence and m6A site analyses focusing exclusively on single-exon transcripts. Because these transcripts are not spliced, they do not recruit the EJC and thus provide a direct contrast to multi-exon transcripts, in which EJC deposition could influence m6A placement (Fig. 2A). Although motif occurrence did not differ between single-exon and spliced transcripts (Fig. 1G and Supplementary Fig. S2C), m6A deposition patterns were clearly distinct. In both transcript classes, 5′ UTR motifs remain largely unmethylated and a prominent 3′ UTR peak dominates the overall m6A distribution. However, single-exon transcripts exhibited a notably higher proportion of CDS-localized m6A sites (compare Fig. 1D and Supplementary Fig. S2D). This suggests that the EJC likely restricts m6A deposition within the CDS of multi-exon transcripts. Nevertheless, because CDS motif density is comparatively low in planarians (Fig. 1G), this effect may be less pronounced than in other species. In summary, the analysis of CDS-m6A methylation patterns in planarians revealed a distribution that is consistent with an EJC-associated exclusion model [7]. However, we found the planarian exclusion zone to be dependent on exon length (Fig. 2A and B). Notably, 48.1% of all planarian CDS-linked m6A sites reside in exons shorter than 400 nt (methylation rate ≥ 15%; Supplementary Table S5). These exons are excluded from receiving any methylation under the human-derived EJC exclusion model [7]. A proper interpretation of our data thus requires the consideration of planarian gene architecture, which closely resembles that of Drosophila and is characterized by a prominent peak of short introns (45–85 bp) alongside a broad distribution of longer introns [62] (Fig. 2C). Given this gene architecture, we found 78.5% of all internal exons in S. mediterranea to be enclosed by at least one intron shorter than 250 nt (Supplementary Fig. S2E), compared to 85% in Drosophila and 25% in humans [63]. These short introns likely result in intron-defined splicing [63], which in turn might limit the efficient placement of the EJC in the adjacent short exons, allowing for the deposition of m6A by the MTC close to these exon boundaries (see Discussion). EJC depletion partially alters the m6A methylome in planarians To clarify the potential role of the EJC in m6A placement in planarians, we next performed RNAi-induced knockdowns of the planarian EJC components elF4A3 and Y14. In line with prior work that had found the planarian EJC to be essential for planarian regeneration by contributing to stem and progenitor cell maintenance [64], both eIF4A3 (RNAi) or y14 (RNAi) resulted in severe phenotypical changes 10 days post first feeding (dpff), which subsequently led to lethality (Supplementary Fig. S3A and B). More specifically, RNAi-treated worms showed massive deformations, accompanied by severely restricted mobility. However, when we used mass spectrometry to measure the influence of eIF4A3 (RNAi) and y14 (RNAi) on global m6A levels in these animals, we found no changes as compared to WT levels (Supplementary Fig. S3C). Next, we performed Nanopore dRNA-seq at 6 dpff and 8 dpff, following efficient eIF4A3 (RNAi) and y14 (RNAi) knockdowns (Supplementary Fig. S3D). Importantly, both time points precede the onset of phenotypic changes, and they both did not lead to significant alterations in global m6A levels, as assessed by both mass spectrometry and Nanopore m6A base calling (Fig. 3A). If the EJC were to block access to suitable m6A motifs close to splice junctions, its loss should lead to increased m6A levels in these regions. However, EJC depletion did not significantly change overall m6A methylation levels but rather resulted in WT-like methylation patterns (Fig. 3B and Supplementary Fig. S3E). Because CDS-linked m6A deposition is rare in planarians—as compared to humans—and the exclusion zone is narrow around splice junctions in short planarian exons, we reasoned that any effect that the EJC might have on m6A deposition, will be limited to these small areas without causing global changes to the methylation landscape. The EJC is responsible for the narrow and shifted exclusion zone in planarians. (A) Global m6A levels were quantified by mass spectrometry (left) and Nanopore direct RNA sequencing (right) following RNAi-mediated knockdown of the indicated genes. Concordant measurements across both approaches show that EJC depletion does not lead to a global increase in m6A levels. Bars represent mean ± SE. Statistical significance was assessed by ANOVA followed by Tukey’s HSD test, with P-values corrected for multiple testing based on the studentized range distribution (adjusted P ≤ 0.05). (B) Methylation rates of individual m6A sites show no marked differences between WT and EJC depleted samples. (C) GAMs of methylation frequency in eIF4A3 (RNAi) and y14 (RNAi) animals across splice junctions at 8 dpff reveal no considerable changes in m6A deposition compared to WT (Fig. 2B). GAMs were fitted for exon length classes, stratified by quartiles of exon length in WT samples. Shaded areas indicate 95% confidence intervals of the model fit; methylation frequency was tested in bins of 10 nt. (D) GAMs of methylation frequency ratios between WT and eIF4A3 (RNAi) or y14 (RNAi), respectively, at 8 dpff uncovers a localized increase in m6A deposition at exon starts. GAMs were fitted for exon length classes defined by length quartiles in WT. Shaded areas denote 95% confidence intervals of the model fit; methylation frequency ratios were tested in bins of 10 nt. To assess the impact of EJC depletion on m6A deposition near splice junctions, we analyzed methylation frequency as a function of distance following eIF4A3 (RNAi) and y14 (RNAi) using GAMs. At the level of absolute methylation frequency, this analysis did not reveal considerable differences relative to WT (Fig. 3C; Supplementary Figs S3F, G, and S4A). To increase sensitivity for detecting more subtle, spatially restricted effects, we additionally utilized GAMs to examine ratios of methylation frequency in the knockdown conditions relative to WT [7]. This comparative analysis uncovered an increase in m6A methylation frequency toward splice junctions at exon starts in both knockdown conditions, whereas no comparable changes were observed at internal exon ends (Fig. 3D and Supplementary Fig. S4B). Because this increase in methylation frequency at exon starts following an EJC knockdown is consistent with our observations in WT animals (Fig. 2A and C), our EJC knockdown data confirm that the m6A exclusion zone is indeed the result of post-splicing deposition of the EJC on planarian mRNAs. To test whether the exon-length-dependent change in m6A exclusion upon EJC-competent knockdown is further modulated by the length of the immediately flanking intron, we stratified exons by flanking intron length (short: <200 nt; long: >250 nt; exons with a flanking intron in the 200–250 nt transition zone range [63] were excluded). To strengthen the statistical power of the model, we combined the two shorter exon length quartiles and the two longer quartiles into one respective short exon and long exon class, yielding four total classes per junction type (short exon – short intron, short exon – long intron, long exon – short intron, and long exon – long intron). For each class we utilized GAMs to survey ratios of methylation frequency in knockdown conditions relative to WT as described above. If the exon-length dependence of EJC-mediated m6A exclusion (Fig. 3C and D) reflects, at least in part, reduced EJC deposition efficiency at short introns rather than exon length per se, its effect upon EJC-component knockdown should differ between short- and long-intron exon classes. Indeed, we found changes in m6A deposition upon knockdown of the EJC at exon starts, particularly in the long-intron classes (Supplementary Fig. S4C). Taken together, we show that the EJC exclusion model near splice junctions [7] is conserved from humans to planarians, albeit with a shifted exclusion zone dependent on exon and flanking intron length (see Discussion). Loss of m6A alters gene expression patterns in planarians To evaluate the impact of m6A on transcript fate, we depleted the MTC components METTL3 and METTL14 by RNAi for 24 days, which led to significantly lower transcript levels relative to unc22 (RNAi) controls (Supplementary Fig. S5A and B). We selected the 24-day time point for our analyses because phenotypic abnormalities—such as lysis spots and head deformation—that culminated in lethality at 41 dpff, only became apparent at 27 dpff (Supplementary Fig. S5C). We reasoned that sampling prior to the onset of these overt phenotypes would also help limit the extent of secondary, indirect effects on the transcriptome, such as shifts in cell-type composition (see Discussion). To confirm that both mettl3 and mettl14 (RNAi) knockdowns lead to a significant reduction in m6A levels in planarian mRNA, we next analyzed poly(A)-selected RNA from mettl3 (RNAi) and mettl14 (RNAi) worms via mass spectrometry. Given that we find no evidence for m6A erasers in planarians (Supplementary Table S3), only reduced MTC activity should lead to a reduction in overall m6A levels on mRNAs. In line with this assumption, both knockdowns resulted in a substantial depletion of m6A from the planarian transcriptome, e.g. relative methylation levels were reduced to 34% in mettl14 (RNAi) worms and to 44% in mettl3 (RNAi) animals (Fig. 4A, see Materials and methods section for details). Because of the more substantial reduction of m6A levels following mettl14 (RNAi), we focused on this knockdown strategy for further analyses and generated dRNA-seq libraries for mettl14 (RNAi) and unc22 (RNAi) controls. Whereas mettl14 (RNAi) libraries averaged ∼4.4M reads (median read length: 631 nt, N50: 936 nt), unc22 (RNAi) libraries averaged ∼3.9M reads (median read length: 584 nt, N50: 1008 nt) (Supplementary Table S4). Consistent with our mass spectrometry results, m6A detection using Oxford’s Dorado software revealed an ∼73.2% reduction in global m6A levels in mettl14 (RNAi) animals (Fig. 4B). Furthermore, methylation at individual sites dropped considerably, with most of the remaining sites showing methylation rates of <25% (Fig. 4C). Thus, within the limitations of RNAi-mediated knockdowns in planarians, we achieved a significant reduction of m6A in the planarian transcriptome at a time point where phenotypical changes to the animals were not yet detectable. We next defined high confidence m6A sites in mettl14 (RNAi) data as previously described (Supplementary Table S2), using methylation rate cutoffs of 15%, 30%, and 60% in at least two out of three replicates as threshold. In applying these criteria, we observed a significant reduction of m6A sites to 20 852 for sites methylated ≥15% (down from 72 237 in WT samples), to 7966 for 30% sites (down from 46 520), and to 1172 for 60% sites (down from 24 968) (Fig. 4D). Depletion of the MTC by mettl14(RNAi) reveals a significant remodeling of the planarian transcriptome. (A) Mass spectrometric quantification of global m6A levels following RNAi-mediated knockdown of METTL3 and METTL14. Abundance of m6A, normalized to the total of adenine nucleotides (relative m6A level), was measured at the indicated time points. Both knockdowns led to a progressive and significant reduction in m6A levels compared to controls, with METTL14 depletion showing the strongest effect at 24 dpff, thereby identifying it as the optimal target for subsequent analyses. Bars represent mean ± SE. Statistical significance was assessed by ANOVA followed by Tukey’s HSD test, which corrects for multiple testing based on the studentized range distribution (*** ≤ 0.001; ** ≤ 0.01; * ≤ 0.05; • ≤ 0.1). (B) Relative m6A levels quantified by mass spectrometry (left) and Nanopore direct RNA sequencing (right) confirm efficient depletion of m6A upon knockdown of the MTC core component METTL14. Both approaches consistently demonstrate a significant reduction in global m6A levels. Bars represent mean ± SE. Statistical significance was assessed by ANOVA followed by Tukey’s HSD test, with P-values corrected for multiple testing based on the studentized range distribution (*** ≤ 0.001; * ≤ 0.05). (C) Distribution of methylation rates for individual m⁶A sites in mettl14 (RNAi) compared to WT, revealing a pronounced decrease in site-specific methylation rates. (D) Number of m6A sites identified in WT, mettl14 (RNAi), and unc22 (RNAi) conditions at methylation rate cutoffs of 15%, 30%, and 60%. At all methylation rate cutoffs, mettl14 (RNAi) consistently showed a substantial reduction in identified m6A sites. (E) Volcano plot showing gene abundance changes between mettl14 (RNAi) and unc22 (RNAi) conditions. Genes with an adjusted P-value ≤ 0.05 and a log2 fold change ≤ −0.5 or ≥ 0.5 were classified as significantly deregulated; top: red dots). The corresponding log2 fold change in m6A site abundance between mettl14 (RNAi) and unc22 (RNAi) is indicated by dot color (bottom). (F) Distribution of cell-type-specific marker genes across the planarian single-cell atlas. Genes that were not significantly deregulated upon mettl14 (RNAi) showed no cell-type enrichment. Among significantly downregulated genes, markers for intestine, cathepsin⁺, and parapharyngeal cells were overrepresented, whereas neural, pharyngeal, protonephridial, and neoblast markers were depleted. Conversely, in significantly upregulated genes cathepsin⁺, epidermal, intestine, and neoblast markers were overrepresented, while protonephridial and muscle markers were depleted. (G) Scatter plot of log2 fold changes in m6A site abundance versus log2 fold changes in gene abundance shows no direct relationship between changes in m6A and the magnitude or direction of expression changes. Red dashed lines indicate log2 fold change thresholds of −0.5 and 0.5; dot color denotes the adjusted P-value for differential expression. (H) Bayesian mixture modeling indicates that m6A is more likely to be associated with transcript downregulation than upregulation, independent of its position along the transcript. Similarly, the magnitude of deregulation is largely position independent. The unregulated transcript class (gray line) shows the highest probability and centers around log₂ fold changes of ∼0. Shaded areas represent 80% credibility intervals. Next, we investigated the impact of m6A depletion on the planarian transcriptome. To achieve this, we removed the few rRNA and tRNA reads present, mapped the remaining reads to the planarian transcriptome, and quantified gene expression using Salmon [44]. We then employed DESeq2 [46] to investigate differential gene expression between mettl14 (RNAi) and unc22 (RNAi) following m6A depletion (Fig. 4E). Importantly, this approach minimizes confounding RNAi-induced effects, which were apparent when comparing mettl14 (RNAi) versus WT and unc22 (RNAi) versus WT (Supplementary Fig. S5D). In total, 1059 genes are significantly deregulated upon mettl14 (RNAi), including 412 upregulated and 647 downregulated genes (Fig. 4E and Supplementary Table S6). Given previous reports of altered cell type composition in S. mediterranea following kiaa1429 (RNAi), a component of the MTC [19], we asked whether specific cell types were preferentially affected in our data set. To that end we leveraged expression data from the planarian single-cell atlas [48] and extracted genes uniquely associated with individual cell types to generate a reference set of cell-type-specific markers (Materials and methods). Mapping mettl14 (RNAi)-deregulated genes onto this marker set revealed that upregulated genes were enriched for markers of cathepsin⁺, epidermal, and intestine cells, as well as for neoblasts, while being depleted for markers of protonephridia and muscle cells, which were not represented at all (Fig. 4F). In contrast, downregulated genes were dominated by intestine cell markers and further showed enrichment for cathepsin⁺ and parapharyngeal markers, alongside the depletion of neural, pharyngeal, protonephridial, and neoblast markers, with the latter two cell types being almost completely absent. Notably, genes that were not significantly deregulated exhibited no bias toward any cell-type-specific marker set, highlighting the specificity of cell type identity patterns observed for deregulated genes. These findings are consistent with previous studies showing that disruption of the m6A pathway primarily impairs lineage differentiation rather than global stem cell maintenance [18, 19, 21]. Loss of intestine cells due to depletion of intestine progenitors, accompanied by accumulation of neural-like progenitors, has been reported following m6A depletion [19]. Defects in protonephridia and muscle lineages, as well as impaired neural regeneration, were also observed upon m6A loss [18]. Similarly, knockdown of YTHDF family m6A readers predominantly affected differentiation of parenchymal, cathepsin⁺, and neural lineages while leaving the neoblast pool largely intact [21]. Together, these studies support a model in which m6A regulates lineage-specific differentiation programs, consistent with the cell-type-dependent stability effects identified here. However, as with any bulk transcriptomic approach applied to an entire animal, our data does not distinguish between two non-exclusive contributions to the observed cell-type-specific expression patterns: direct, transcript-intrinsic effects of m6A loss, and secondary shifts in the relative proportion of cell types, such as the reduction of intestinal cells independently reported for mettl14 (RNAi) and ythdc-1 (RNAi) by imaging-based quantification [19]. We therefore interpret the cell-type-specific patterns in Fig. 4F to most likely reflect a combination of these effects rather than direct transcriptional regulation alone (see Discussion). Notably, 90.5% of all deregulated genes (958 of 1059) carried m6A marks. This represents a significant enrichment over the genome-wide proportion of m6A-marked genes (Fisher’s exact test, odds ratio: 7.11, 95% CI: 5.78–8.84, P < 2.2 × 10⁻¹⁶) and is consistent with a substantial contribution of m6A status to the observed expression changes. To explore this further, we analyzed log2 fold changes of m6A sites with methylation rates ≥15%. This revealed no direct correlation between loss or gain of m6A sites and either the direction or magnitude of gene expression changes (Fig. 4G). Furthermore, these results indicate that m6A broadly influences the fate of modified transcripts but does not act uniformly to stabilize or destabilize them, likely reflecting context-dependent effects mediated by distinct m6A reader proteins. Last, prompted by the skewed distribution of m6A sites along planarian transcripts, with strong enrichment near the stop codon (Fig. 1D), we hypothesized that the position of an m6A mark within a transcript modulates its effect on transcript fate. To test this, we employed a Bayesian mixture model to assess whether the average position of m6A sites influences probability, direction and magnitude of the log2 fold change of transcript abundance following mettl14 (RNAi) (Materials and methods). For this analysis, only transcripts losing ≥80% of their m6A sites were considered (n = 3966), with m6A site loss being defined as the reduction in methylation rate per sites from ≥15% to 0% following mettl14 (RNAi) (Supplementary Fig. S5E). Consistent with previous observations (Fig. 4E and G), we found m6A to be associated with both stabilizing and destabilizing effects on transcripts. However, the model indicates that most transcripts (∼95.5%; 80% credibility interval: 94.82%–96.10%) exhibit no substantial change in abundance following mettl14 (RNAi) (Fig. 4H). This agrees with results from other organisms such as humans and Drosophila [13, 65, 66]. Downregulation has a probability of 3.97% (80% credibility interval: 3.39%–4.62%). Importantly, for these downregulated transcripts, the magnitude of the log₂ fold change did not substantially vary with the mean position of m6A sites along the transcript. In contrast, upregulation was rare (probability: 0.53%; 0.38%–0.70%). While the model captures the overall distribution of effects, particularly on the downregulated side, posterior predictive checks indicate an overestimation of strong positive effects (>0.5) of the model. Accordingly, conclusions regarding the probability and magnitude of such effects should be interpreted with caution, and that such effects may even be less likely than indicated by the model. Overall, the results suggest that m6A more often stabilizes than destabilizes planarian transcripts. Importantly, the magnitude of downregulation is largely independent of m6A position, which argues against a functionally enhanced significance of the frequently occurring sites near the stop codons of transcripts (Fig. 1G). Poly(A) tail shortening in a METTL14 knockdown occurs independently of m6A status Nanopore dRNA-Seq offers the unique possibility to directly measure poly(A) tail length. This benefits our study as both m6A sites and poly(A) tail length are involved in pre-mRNA surveillance. Moreover, prior evidence has begun to connect the m6A machinery to cleavage and (alternative) polyadenylation [61, 67, 68]. When we analyzed the length of poly(A) tails in our data, we found that they were significantly shorter in mettl14 (RNAi) animals (mean: 75 nt) as compared to WT and unc22 (RNAi) animals (mean: 85 and 87 nt, respectively) (Fig. 5A). To assess whether this effect was caused by m6A, we classified the planarian WT transcriptome into two groups: transcripts that harbor m6A sites and transcripts that do not. In all conditions analyzed [WT, unc22 (RNAi), and mettl14 (RNAi)], m6A-marked transcripts exhibited considerably longer poly(A) tails compared to non-m6A transcripts (Fig. 5B). However, we observed that poly(A) tails were uniformly shorter in mettl14 (RNAi) animals, regardless of their m6A status in WT conditions. To further examine a potential relationship between m6A and poly(A) tail length, we grouped genes according to previously calculated m6A log₂ fold changes between mettl14 (RNAi) and unc22 (RNAi) conditions. Genes were assigned into four categories: transcripts that lose m6A sites (LFC < −0.5), transcripts that gain m6A sites (LFC > 0.5), transcripts with unchanged m6A levels (−0.5 ≤ LFC ≤ 0.5), and transcripts that do not carry detectable m6A. Surprisingly, we observed poly(A) tail shortening across all gene classes upon mettl14 (RNAi), including for transcripts that lack m6A (Fig. 5C). This finding was unexpected, as m6A-marked transcripts display longer poly(A) tails under WT conditions (Fig. 5B). A METTL14 Knockdown leads to globally shortened poly(A) tails. (A) Distribution of poly(A) tail length in WT, mettl14 (RNAi), and unc22 (RNAi) conditions. dRNA-Seq reads of mettl14 (RNAi) displayed significantly shorter poly(A) tails than those of unc22 (RNAi) or WT (adjusted P-value = 0). (B) Poly(A) tail length in WT, mettl14 (RNAi), and unc22 (RNAi) stratified by m6A status in WT. Split violin plots display the poly(A) tail length distribution of transcript classes; left: length distributions of both classes, right: differences in length distributions [dusky pink = poly(A) tail length distribution of m6A-marked transcripts (WT); dark gray = poly(A) tail length distribution of transcripts without m6A detected (WT)]. m6A transcripts had longer poly(A) tails than transcripts without m6A. Poly(A) tail length in mettl14 (RNAi) conditions were shorter irrespective of m6A status. (C) Genes were clustered by m6A log2 fold change (m6A loss: LFC < −0.5; m6A gain: LFC > 0.5, No m6A: no methylation in either condition; No change: LFC ≥ −0.5 and ≤ 0.5). Split violin plots display the poly(A) tail length distribution in mettl14 (RNAi) and unc22 (RNAi) conditions; left half: length distributions in both conditions, right half: differences in length distributions [dusky pink = poly(A) tail length distribution in unc22 (RNAi); dark gray = poly(A) tail length distribution in mettl14 (RNAi)]. All four clusters displayed shorter poly(A) tails in metll14 (RNAi) than in unc22 (RNAi). (D) Density plot of poly(A) tail length differences (mettl14 (RNAi) versus control) among the 215 transcripts with a statistically significant difference in poly(A) tail length (moderated t-test, BH-adjusted P ≤ 0.05; see Materials and methods). Negative values indicate shorter, positive values longer poly(A) tails in mettl14 (RNAi); the dotted line marks no change (Δ = 0 nt). Of these, 195 showed shortening and 20 showed lengthening of their poly(A) tails. To resolve this contradiction, we modeled poly(A) tail length changes at single-transcript resolution rather than in bulk. This identified 215 transcripts with significant poly(A) length differences between mettl14 (RNAi) and pooled WT/unc22 (RNAi) controls (see Materials and methods), of which 181 carried m6A in WT and 34 did not. Of these, 195 showed poly(A) shortening and 20 showed lengthening in mettl14 (RNAi) relative to control (Fig. 5D). Given a prior report linking m6A sites near the stop codon and 3′ UTR to cleavage and alternative polyadenylation [61], we tested whether m6A position relative to the stop codon influenced shortening magnitude. For each m6A-marked transcript with sufficient coverage (at least 5 reads, n = 10 702, see Materials and methods), we calculated the minimal distance between an m6A site and the stop codon and related this to poly(A) length change derived from dRNA-Seq reads. A linear regression revealed a significant, distance-dependent effect (P = 9.95 × 10-7), with shortening strongest near the stop codon and attenuating with distance, though this explained only a small fraction of the variance (R² = 0.00235), indicating additional context-dependent factors also contribute. We then asked whether this effect differed between m6A sites upstream (CDS) versus downstream (3′ UTR) of the stop codon. Neither a Welch’s t-test comparing mean shortening (CDS: −4.42 nt versus 3′ UTR: −4.45 nt, P = 0.82) nor a linear model testing for a distance–region interaction (CDS slope: 2.54 nt/unit; 3′ UTR slope: 2.04 nt/unit; interaction P = 0.62) found a significant difference, indicating the effect is symmetric around the stop codon rather than 3′ UTR-intrinsic. Finally, we tested whether loss of 3′ UTR m6A sites in mettl14 (RNAi) animals shifted alternative polyadenylation site usage, as reported for the mammalian m6A machinery [61]. A Wilcoxon test detected a nominally significant association (P = 0.017), but the effect size was negligible (Cohen’s d = 0.007; mean difference ∼1.8 nt against a within-group SD of ∼255–260 nt), indicating no biologically meaningful shift in polyadenylation site choice transcriptome-wide upon mettl14 depletion. Thus, together, these results suggest that a mettl14 (RNAi) knockdown broadly disrupts poly(A) tail homeostasis and mRNA surveillance, likely through both direct effects on m6A deposition and indirect effects on global poly(A) tail regulation (see Discussion). While m6A position relative to the stop codon exerts a small, symmetric influence on the magnitude of poly(A) shortening, we find no evidence that m6A loss shifts polyadenylation site choice in planarians, suggesting that these effects act primarily through altered tail-length regulation rather than through re-routing of cleavage and polyadenylation. m6A has opposing effects on transcript decay depending on cellular context Our previous observations revealed pronounced changes in gene expression as well as a global reduction in poly(A) tail length upon m6A depletion (Figs 4E,F and 5A,B), strongly implicating m6A in being involved in the regulation of mRNA stability in planarians. These findings furthermore suggest that altered transcript decay dynamics may underlie a substantial fraction of the observed expression changes. To directly test this hypothesis and to quantify the impact of m6A on transcript stability, we next employed an actinomycin D assay to monitor mRNA decay kinetics over time. Actinomycin D is a well-established transcriptional inhibitor in numerous experimental systems that enables measurement of mRNA stability by acutely blocking transcription and allowing decay of pre-existing transcripts to be monitored over time [69, 70]. To demonstrate the efficacy of the drug in S. mediterranea, we monitored the pre-mRNA levels of three exemplary transcripts (h1SMcG0009165, h1SMcG0014608, and h1SMcG0021563) by intronic qPCR in actinomycin D- versus dimethyl sulfoxide (DMSO)-treated WT worms over a 12 h time course. To directly assess ongoing transcription, we used intron-spanning primers to specifically amplify unspliced transcripts, which—owing to the tight kinetic coupling between transcription and splicing—serve as a direct readout of nascent transcriptional activity [71–73]. To avoid confounding effects from endogenous reference genes and for proper normalization, a control RNA was used as a spike-in control. Actinomycin D treatment resulted in a clear decline in intronic signal for all three targets, reaching 40%–60% reduction by 6–12 h, whereas DMSO-treated controls remained largely stable (Supplementary Fig. S6A). These results confirm that actinomycin D blocks transcription under the conditions used in this study. Next, we repeated the mettl14 (RNAi) knockdown as previously described (Supplementary Fig. S5A, and Materials and methods) and treated worms with actinomycin D at 22 dpff. RNA from WT, unc22 (RNAi), and mettl14 (RNAi) animals was collected 0, 6, and 12 h after this treatment, ribo-depleted, and subjected to Illumina paired-end sequencing (Materials and methods). Importantly, to minimize potential biases in mRNA enrichment caused by shortened poly(A) tails (Fig. 5A and B), we used ribo-depleted RNA rather than poly(A)-selected RNA for this analysis. To identify genes displaying differential abundance trajectories during transcriptional arrest, the resulting data were then subjected to a time-series analysis with maSigPro [56]. To focus on m6A-dependent effects on transcript stability, we restricted this analysis to transcripts harboring at least one m6A site with a methylation rate ≥15% under WT conditions (based on our prior methylome analysis, Supplementary Fig. S1A and Supplementary Table S2). To reduce noise, we excluded genes whose transcript abundance increased over time, likely reflecting the activation of compensatory pathways such as drug efflux mechanisms [74], stress-response programs triggered by actinomycin D treatment itself, sequence-composition-dependent variation in actinomycin D efficacy, or a baseline rate of false-positive slope calls expected at this scale of testing. We also did not find evidence to support incomplete drug penetrance as an explanation for this subset of genes. In more detail, gene-wise linear models were fitted separately for each condition [WT, unc22 (RNAi), or mettl14 (RNAi)], and genes with a significantly positive slope and a log2 fold change ≥0.2 in any condition were removed from the analysis. Subsequently, all remaining genes were centered to their respective 0 h time points and analyzed using maSigPro. Following filtering, quadratic regression identified 2989 genes whose abundance profiles differed significantly between conditions over time (FDR = 0.0027). Subsequent stepwise regression (“two-way forward” method) refined this set to 503 genes that showed significant differences in decay behavior in mettl14 (RNAi) animals as compared to both WT and unc22 (RNAi) control conditions. To further classify these genes based on their decay dynamics, we determined k = 2 as the optimal number of clusters using PCA and within-cluster sum-of-squares metrics (Fig. 6A and Supplementary Fig. S6B and C, and Materials and methods). In the first cluster (n = 187), genes displayed significantly faster transcript decay upon mettl14 (RNAi), indicating that m6A might serve to stabilize these mRNAs. In contrast, genes in the second cluster (n = 316) exhibited slower decay rates in mettl14 (RNAi) as compared to WT and unc22 (RNAi) animals, suggesting that m6A promotes transcript decay for this subset (Supplementary Table S7). Time-series analysis of actinomycin D assay reveals distinct mRNA stability clusters. (A) Time-series analysis of transcript stability following mettl14 (RNAi) identified two distinct gene clusters with opposing decay behaviors. Cluster 1 transcripts (n = 187) exhibited accelerated decay upon m6A loss, indicating a stabilizing role of m6A, whereas Cluster 2 transcripts (n = 316) showed increased stability, consistent with m6A-mediated destabilization. Solid lines represent the mean expression values, while dashed lines indicate the regression fit curves generated by the model. (B) Genes assigned to either stability cluster harbored significantly more m6A sites than non-clustered genes in both unc22 (RNAi) and WT conditions (adjusted P-values = 4.752 × 10⁻²/9.392 × 10⁻4 for Cluster 1; 1.065 × 10⁻¹³/6.797 × 10⁻¹¹ for Cluster 2, respectively). In unc22 (RNAi) animals, Cluster 2 genes additionally had significantly more m6A sites than Cluster 1 genes (padj = 9.598 × 10⁻4). Dots indicate medians and whiskers the first and third quartiles. Statistical significance was assessed using the Kruskal–Wallis test with Bonferroni correction. (C) Poly(A) tail length was reduced in both clusters upon mettl14 (RNAi), indicating that differential stability outcomes are not explained by opposing poly(A) tail dynamics. (D) Metagene analysis of m6A site distribution revealed persistent enrichment toward the CDS–3′ UTR junction in both clusters. Notably, Cluster 1 genes displayed slightly higher m6A density within the CDS compared to Cluster 2 genes. (E) Cell-type marker analysis using the planarian single-cell atlas [48] showed distinct lineage associations for each cluster. In Cluster 1, intestine, epidermal, and cathepsin⁺ markers were overrepresented and neoblast, neural, pharyngeal, and parapharyngeal markers were depleted. In contrast, muscle and neoblast markers were overrepresented, while protonephridia, intestine, and parapharyngeal markers were depleted in Cluster 2. In building on these findings, we sought to identify determinants underlying the observed gene-specific differences in mRNA decay. We first examined the m6A status of the clustered genes and found that they carried significantly more m6A marks than non-clustered genes, supporting that the observed stability effects are m6A-dependent (Fig. 6B). Notably, genes in cluster 2, in which m6A promotes transcript destabilization, displayed significantly higher overall methylation levels than genes in cluster 1, which we found to be stabilized by m6A (Fig. 6B). Moreover, genes within stability clusters exhibited no significant difference in log₂ fold change of m6A site abundance compared to non-clustered genes. This observation underscores that the influence of m6A on transcript stability is mediated by gene-specific regulatory mechanisms, rather than a uniform response to overall m6A depletion (Supplementary Fig. S6D). We next assessed whether differences in poly(A) tail length could account for the opposing decay behaviors. Poly(A) tail analysis revealed a general shortening of poly(A) tails in mettl14 (RNAi) animals compared to unc22 (RNAi) controls, irrespective of cluster assignment (Fig. 6C). This indicates that differential mRNA stability between clusters is not primarily driven by poly(A) tail modulation, but likely reflect alternative mechanisms, such as context-dependent recognition of m6A sites by distinct m6A reader proteins [75–78] (see Discussion). Finally, we examined the positional distribution of m6A marks within transcripts. In both clusters, m6A sites were predominantly enriched in the vicinity of the stop codon, in line with the metagene plot for overall m6A distribution in planarian transcripts (Fig. 1D). However, transcripts in cluster 1, which are stabilized by m6A, exhibited modestly higher m6A levels within the CDS compared to transcripts in cluster 2 (Fig. 6D). This pattern contrasts with observations in human cells, where CDS-localized m6A sites were reported to promote transcript destabilization via translation-dependent decay [9]. Taken together with the very low overall frequency of m6A sites within the CDS of planarian transcripts (Fig. 1D), our data thus argue against a major role for translation-dependent m6A-mediated mRNA decay in planarians. As a final step in our analyses, we asked whether the effect of m6A on transcript stability depends on the cellular context. To address this, we leveraged the planarian single-cell atlas [48] and, as described previously (Fig. 4F, and Materials and methods), mapped genes from the identified clusters onto a reference set of cell-type-specific marker genes to assess cell-type enrichment (Fig. 6E). Genes from cluster 1, in which m6A stabilizes transcripts, were overrepresented in markers of cathepsin⁺, epidermal, and intestine cells, while being depleted among markers of neoblasts, neural, pharyngeal, and parapharyngeal cells, with the latter two cell types being completely absent from this cluster. In contrast, genes from cluster 2, in which m6A promotes transcript decay, were significantly enriched for muscle and neoblast markers but depleted for markers of protonephridia, intestine, and parapharyngeal cells. Notably, these enrichment patterns are largely reciprocal between the two clusters, indicating that m6A-mediated transcript stabilization and destabilization preferentially occur in distinct cellular contexts rather than affecting the same cell types in opposing ways. More specifically, our Actinomycin D experiments suggest that m6A predominantly exerts stabilizing effects in cathepsin⁺, epidermal, and intestine cells, whereas it preferentially promotes transcript decay in neoblasts and muscle cells. Notably, previous studies have reported that depletion of the m6A writer complex primarily affects intestine, muscle, protonephridia, neural, and specific neoblast subpopulations [18, 19], consistent with our observations. Collectively, our results indicate that the impact of m6A on mRNA stability is strongly cell-type dependent and likely reflects differences in the expression or activity of m6A reader proteins, rather than intrinsic properties of the modified transcripts themselves. Discussion Principles of m6A deposition in the planarian transcriptome In this study, we delineate the m6A methylation landscape of the planarian S. mediterranea and uncover both conserved and cell type-specific features of m6A regulation. Leveraging direct RNA sequencing, we quantified absolute m6A levels, mapped modification sites at single-nucleotide resolution, and assessed transcript abundance and poly(A) tail dynamics in a single experiment. We show that the core components of the MTC are conserved from humans to planarians and that m6A is a prevalent internal modification of planarian mRNAs. Notably, we find no evidence for m6A demethylases, indicating that m6A marks are not actively removed once deposited. This suggests a comparatively stable methylation landscape in which regulatory specificity is likely primarily achieved through controlled deposition and context-dependent reader engagement rather than dynamic erasure, a model further supported by our stability analyses. Planarian m6A sites conform to a DRAYW consensus motif closely related to vertebrate (and even insect) motifs, with modest deviations likely reflecting the high AT content of the planarian genome. Because protein-CDS is constrained by purifying selection on codon identity, DRAYW-like motifs may be enriched outside the CDS in planarians, independent of any methylation-specific deposition preference (Fig. 1G). At the transcript level, m6A is strongly enriched near stop codons and within 3′ UTRs, closely resembling vertebrate profiles and contrasting elevated m6A levels in 5′ and 3′ UTRs as observed in Drosophila [59]. Despite this overall conservation, m6A placement relative to splice junctions confirms that the canonical EJC exclusion model [7] operates in planarians, albeit with an exclusion zone shifted downstream of the splice junction and dependent on exon and intron length. While m6A deposition is consistently depleted at exon starts in an EJC-dependent manner, its effect at exon ends varies with exon length: in shorter exons, we observe no exclusion zone at exon ends, while in longer exons, the exclusion zone widens to ∼200 nt from the junction. Previous studies have shown that the transition from exon- to intron-defined splicing is dependent on intron length (with a transition zone between 200 and 250 nt [63]) and that the exon–intron architecture in Drosophila is dominated by relatively short introns (<100 nt), resulting in intron-defined splicing [79, 80]. In intron-defined splicing, splice sites are recognized within individual introns (e.g. U1 snRNP binding to the 5′ splice site and U2AF binding to the 3′ splice site of the same intron). This contrasts with exon-defined splicing in vertebrates, where splice sites are recognized across exons [81]. An intron-defined splicing reaction (e.g. in Drosophila) results in rapid splicing kinetics, which in turn does not allow for efficient EJC positioning post splicing and thus results in inefficient nonsense-mediated decay [82, 83]. In contrast, in exon-defined splicing, the EJC is typically deposited ∼20–24 nt upstream of exon–exon junctions [84]. Because our analysis of planarian exon–intron architecture demonstrates that it closely resembles the situation in Drosophila (Fig. 2C and Supplementary Fig. S2E), we hypothesize that intron-defined splicing dominates in planarians. Together, the compact, largely intron-defined architecture of the planarian genome (Supplementary Fig. S2E) and the intron-length-dependent response to EJC-component knockdown (Supplementary Fig. S4C) support a model in which flanking intron length modulates EJC-dependent m6A exclusion in addition to exon length itself, consistent with reduced EJC deposition efficiency at short introns (Fig. 2A and B). To determine the precise position of the EJC on planarian transcripts further experiments, such as CLIP (crosslinking and immunoprecipitation) on the planarian EJC, which comprised all components found in vertebrates, are needed. In contrast to vertebrate systems, where dynamic m6A erasure enables rapid transcriptomic remodeling [85], planarians may achieve regulatory specificity of m6A placement primarily through elevated mRNA turnover, e.g. high intracellular RNase activity [86] and cell-type-specific m6A reader activity (see separate Discussion paragraph and [21]). Such regulatory strategy is well suited to planarian biology, where cell division is restricted to neoblasts and regeneration depends on the robust and sustained execution of lineage-specific gene expression programs. The conservation of a vertebrate-like m6A distribution in an invertebrate species further suggests strong evolutionary pressure to maintain post-transcriptional control mechanisms that support complex tissue organization and cellular plasticity. Future comparative analyses of m6A motifs and reader proteins across regenerative and non-regenerative flatworms may help identify features linked to regenerative capacity. Moreover, integrating single-cell transcriptomics with m6A mapping approaches [87] could further resolve how cell-type-specific m6A deposition and reader engagement shape gene regulatory programs during regeneration. Context matters: m6A directs lineage-specific mRNA decay Furthermore, our findings provide a mechanistic framework that reconciles previously reported phenotypes of m6A writer complex depletion with the cell-type-specific effects observed here [18, 19, 21]. While bulk transcriptomic approaches cannot fully disentangle direct transcript-intrinsic effects from secondary shifts in cell-type composition, the short duration (12 h) of our actinomycin D time series minimizes compositional confounds, as fractional decay kinetics are largely independent of starting cell proportions. Time-resolved actinomycin D assays following mettl14(RNAi) identified two distinct gene clusters with opposing stability responses to m6A loss: one in which m6A stabilizes transcripts and another in which it promotes decay. Importantly, these effects are not uniformly distributed across the transcriptome but are strongly cell-type dependent. Transcripts stabilized by m6A are enriched in intestine, epidermal, and cathepsin⁺ cells, whereas transcripts whose decay is promoted by m6A are preferentially associated with neoblasts and muscle cells (Fig. 6E). This context-dependent regulation provides a compelling explanation for prior observations that knockdown of m6A writer components such as kiaa1429 or wtap does not ablate the general neoblast pool but instead leads to the accumulation of X2 progenitors, particularly neural progenitor-like cells, alongside depletion of intestine and muscle progenitors and their differentiated descendants. Our data suggest that loss of m6A destabilizes lineage-specific gene programs in intestine, epidermal, and cathepsin⁺ cells, impairing differentiation and maintenance of these tissues. Conversely, in neoblasts, where m6A normally promotes transcript turnover, its depletion leads to aberrant stabilization of progenitor-associated transcripts, potentially trapping cells in an undifferentiated or lineage-biased progenitor state. This model explains how neoblast identity can remain largely intact while differentiation trajectories are selectively disrupted. Notably, poly(A) tail shortening upon m6A depletion occurred across both gene clusters, indicating that differential transcript stability is not primarily driven by poly(A) tail modulation. Instead, our results point to m6A reader proteins as key mediators of the opposing stability outcomes. We speculate that lineage-specific expression or activity of m6A readers underlies the contrasting effects of m6A on transcript fate across cell types. m6A exerts its opposing effects on mRNA stability independently of poly(A) tail length Our poly(A) tail analyses (Fig. 5) indicate that m6A might contribute to mRNA surveillance through both direct and indirect mechanisms. Directly, loss of m6A from transcripts that normally carry the modification abolishes the stabilizing influence of m6A-bound readers, leading to their accelerated deadenylation or their impaired polyadenylation. This interpretation is supported by the longer poly(A) tails observed on m6A-marked transcripts. These results are also consistent with emerging evidence linking m6A to protection of poly(A) tails and regulation of deadenylation dynamics [67, 68]. Indirectly, however, METTL14 depletion causes a global shortening of poly(A) tails, including in transcripts that do not normally harbor m6A. This unexpected effect suggests that disruption of the MTC is likely to broadly perturb mRNA surveillance pathways. Alternatively, a METTL14 knockdown may elicit a stress response that globally accelerates deadenylation independently of the m6A status of specific transcripts. The observation that even transcripts classified as m6A-gaining (LFC > 0.5) exhibit shortened poly(A) tails further supports the idea that compensatory or ectopic methylation under METTL14 depletion is insufficient to restore reader-mediated stabilization (Fig. 5C). Together, these findings indicate that intact m6A writer activity is required globally for proper poly(A) tail homeostasis. VIRMA-dependent m6A deposition near the stop codon and 3′ UTR was linked to alternative cleavage and polyadenylation in mammalian cells [61], prompting us to test at single-transcript resolution whether m6A position relative to the stop codon shapes poly(A) dynamics in planarians. The magnitude of poly(A) shortening upon mettl14 (RNAi) was modestly but significantly predicted by proximity of a transcript’s m6A site(s) to the stop codon, advocating for a functional link between m6A placement and 3′ end processing that is at least partially conserved with mammals. Notably, this effect was symmetric around the stop codon rather than confined to the 3′ UTR, suggesting that the relevant determinant in planarians is proximity to the stop codon itself, not a 3′ UTR-specific mechanism. Furthermore, although planarians possess a conserved VIRMA/kiaa1429 ortholog as a core MTC component [19], we found no evidence that loss of 3′ UTR m6A shifts polyadenylation/cleavage site choice. Unlike the alternative-polyadenylation mechanism described for mammalian VIRMA, the effect of m6A on poly(A) length in planarians thus appears independent of cleavage site selection. Rather than altering 3′ end processing, our data point towards a general modulation of tail length by m6A placement, e.g. by changing deadenylase or poly(A) polymerase activity near the stop codon. Direct interrogation of planarian VIRMA/kiaa1429 substrate binding (e.g. via CLIP) would help to clarify whether this reflects genuine functional divergence from mammalian VIRMA or redundancy among factors for cleavage-site choice in planarians. Cell-type-specific m6A readers dictate divergent mRNA stability programs The opposing effects of m6A on transcript stability across planarian cell types are likely mediated by selective recruitment of stabilizing or decay-promoting machinery by distinct m6A readers. A recent characterization of the expanded ythdf family in planarians revealed pronounced tissue-specific expression patterns, with ythdf-A enriched in the intestine, ythdf-B in the brain, and ythdf-C in the lining of the intestine and brain, alongside broader but overlapping expression domains [21]. Strikingly, these patterns align closely with our actinomycin D stability data: m6A stabilizes transcripts in epidermal, intestine, and cathepsin⁺ cells with the latter two tissues enriched for ythdf-A (ythdf-A, -B, and -C were found to be enriched in the epidermis), whereas cell types enriched for ythdf-B and ythdf-C, including neural, muscle, and pharyngeal cells, are overrepresented among genes for which m6A promotes transcript decay. This correlation supports a model in which ythdf-A predominantly confers transcript stabilization, while ythdf-B and ythdf-C favor decay, reminiscent of mammalian systems in which YTHDF proteins recruit distinct effector complexes to regulate mRNA fate. Consistent with this view, mammalian YTHDFs promote mRNA decay via the CCR4–NOT complex [88], whereas IGF2BP proteins stabilize transcripts by protecting them from deadenylation or by enhancing their translation [89]. Notably, we identified additional planarian homologs of m6A readers, including IGF2BP, YTHDC1, and HNRNP family members, raising the possibility that stabilizing effects observed in intestine and cathepsin⁺ cells may involve readers beyond the YTHDF family (Supplementary Table S3). At the same time, the broad co-expression of ythdf genes suggests potential functional redundancy or combinatorial regulation. Together, these observations indicate that m6A-dependent control of transcript stability in planarians might be dictated primarily by the local reader landscape, underscoring the need for future studies to dissect reader-specific functions, redundancy, and downstream effector pathways in a cell-type-resolved manner. Supplementary Material Acknowledgements We thank Miklas Hafke, Melinda Liebrecht, Jule Lengsfeld, and Simone Keß for conducting preliminary experiments. We thank Wiep van der Toorn for advice and support with demultiplexing Nanopore libraries using WarpDemuX. We thank Catherine Broberg Vågbø from the Proteomics and Modomics Experimental Core Facility (PROMEC) of the Norwegian University of Science and Technology (NTNU) for carrying out mass spectrometry measurements and advice in result evaluation. We thank Omri Wurtzel for kindly providing us with a lift over table of his study’s m6A sites [21] to the genome/transcriptome-build used in our study. We thank Vladyslava Gorbovytska for advice regarding the graphical abstract and for critically reading the manuscript. Author contributions: C.H., A.P., and C.-D.K. conceived and designed all experiments. C.H. carried out all experiments, except for dRNA-seq library preparations. A.P. performed all computational data analysis with advice from R.P.S. on dRNA-seq data analysis, except for the Bayesian hierarchical mixture-of-experts model. A.-S.G.-B. performed all dRNA-seq library preparation. Nanopore sequencing was carried out by R.P.S. and A.-S.G.-B. GAM analysis was established by L.H. and subsequently carried out by A.P. L.H. developed and applied the Bayesian hierarchical mixture-of-experts model. A.P., C.-D.K., and C.H. drafted the manuscript with input from all authors. Contributor Information Constantin Höhn, RNA Biochemistry, Medical Campus Upper Franconia (MCO), University of Bayreuth, 95447 Bayreuth, Germany. Andreas Pittroff, RNA Biochemistry, Medical Campus Upper Franconia (MCO), University of Bayreuth, 95447 Bayreuth, Germany. Anne-Sophie Gribling-Burrer, Genome Biology of Viruses, Architecture et Réactivitié de l’ARN, Université de Strasbourg, CNRS, Institute of Molecular and Cellular Biology (IBMC), 67000 Strasbourg, France. Lisa Hülsmann, Ecosystem Analysis and Simulation (EASI) Lab, Bayreuth Center of Ecology and Environmental Research (BayCEER), University of Bayreuth, 95447 Bayreuth, Germany. Redmond P Smyth, Genome Biology of Viruses, Architecture et Réactivitié de l’ARN, Université de Strasbourg, CNRS, Institute of Molecular and Cellular Biology (IBMC), 67000 Strasbourg, France. Claus-D Kuhn, RNA Biochemistry, Medical Campus Upper Franconia (MCO), University of Bayreuth, 95447 Bayreuth, Germany. Supplementary data Supplementary data is available at NAR online. Conflict of interest The authors declare no conflicts of interest. Funding This study was primarily funded by the German Research Foundation (DFG), Grant KU 3514/5-1 to C.-D.K. C.-D.K. also acknowledges support from the Heisenberg Program of the DFG (KU 3514/4-1), by DFG Grant KU 3514/1-2, and by the University of Bayreuth. R.P.S. acknowledges LabEx NetRNA and the interdisciplinary Thematic Institute IMCBio+, as part of the ITI 2021–2028 program of the University of Strasbourg, CNRS and Inserm, IdEx Unistra (ANR-10-IDEX-0002), SFRI-STRAT’US (ANR 20-SFRI-0012), and EUR IMCBio (ANR-17-EURE-0023) under the framework of the French Investments of the France 2030 Program. Funding to pay the Open Access publication charges for this article was provided by the DFG and University of Bayreuth. Data availability All sequencing data (Nanopore and short read) generated in this study have been deposited in the NCBI Gene Expression Omnibus (GEO) database under accession code GSE325813. All code for the Bayesian Mixture model analysis is available at GitHub under https://github.com/lisahuelsmann/m6abayesianmixture and Zenodo at https://zenodo.org/records/22082929 (https://doi.org/10.5281/zenodo.22082929). References Associated Data Supplementary Materials Data Availability Statement All sequencing data (Nanopore and short read) generated in this study have been deposited in the NCBI Gene Expression Omnibus (GEO) database under accession code GSE325813. All code for the Bayesian Mixture model analysis is available at GitHub under https://github.com/lisahuelsmann/m6abayesianmixture and Zenodo at https://zenodo.org/records/22082929 (https://doi.org/10.5281/zenodo.22082929).