Arabidopsis nanopore direct RNA sequencing data for mutants in splicing, polyadenylation, and m6A methylation

Abstract

Co-transcriptional RNA processing, including alternative splicing and polyadenylation, generates diverse mRNA isoforms essential for development, differentiation, and stress responses. We present Oxford Nanopore Direct RNA Sequencing (DRS) data from Arabidopsis thaliana wild-type (Col-0) plants and mutants defective in RNA processing pathways, including polyadenylation, RNA methylation, and pre-mRNA splicing. We obtained full-length reads together with native poly(A) tail information. The provided resources include FAST5 and FASTQ files, poly(A) tail length estimates from Nanopolish, poly(A) composition data from Ninetails, and differential expression analyses generated with DESeq2, all aligned to the TAIR10 reference transcriptome. These data offer a comprehensive view of RNA processing at the single-molecule level and can be reused to explore alternative polyadenylation, RNA modification patterns, or indirectly, splicing alterations across various genetic backgrounds. This resource is intended to support studies of gene expression regulation and RNA metabolism in Arabidopsis thaliana and related plant systems.

Background & Summary

Regulation of RNA processing, including splicing, 3′ end formation, and polyadenylation, is central to defining transcriptome structure and gene expression programmes in plants^1–9. Components of the polyadenylation complex (PAC) and spliceosomal proteins cooperate to determine poly(A) site choice, affecting transcript stability and translation^2,5,7,10. Although its exact composition varies across eukaryotes, PAC typically includes conserved cleavage factors (CF) I and II, cleavage and polyadenylation specificity factor (CPSF), cleavage stimulation factor (CstF), as well as additional catalytic and scaffold proteins such as poly(A) polymerase, factor interacting with polymerase I (FIP1), symplekin, E3 ubiquitin-protein ligase RBBP6, and nuclear poly(A)-binding proteins^6,10–12 (Fig. 1a). Mechanistic understanding of PAC function comes largely from mammalian and yeast systems and provides a foundation for investigating related processes in plants, where 3′-end formation is tightly connected to splicing^2,5,10.

Fig. 1.

Overview of RNA-processing factors, mutant lines, and nanopore DRS workflow. (a) Schematic representation of human U1 snRNP and RNA 3′-end processing machinery, according to So et al.¹⁰. Orthologs of factors mutated in this study are highlighted in red and yellow. Key mRNA sequence motifs are shown in orange: 5′SS (5′ splice site) recognized by U1 snRNP components, USE (UGUA upstream sequence element) bound by the CFIm (mammalian cleavage factor I) complex, AAUAAA polyadenylation signal (PAS) recognized by the CPSF (cleavage and polyadenylation specificity factor) complex, and DSE (downstream G/U-rich element) interacting with the CstF (cleavage stimulation factor) complex. For clarity, some components involved in splicing and polyadenylation, such as serine-rich (SR) proteins, the Sm core of U1 snRNP, and additional cofactors contributing to cleavage and poly(A) tail formation, are not shown. (b) Arabidopsis genotypes sequenced in this study, including components of the AtCFI complex, U1 snRNP, and RNA methylation. (c) Representative phenotypes of plants included in this study, captured in three-week-old plants with a red 10 mm scale bar shown in each image. (d) Experimental workflow for nanopore direct RNA sequencing (DRS), including RNA extraction, library preparation, sequencing, basecalling, mapping, and poly(A) tail analysis, followed by potential downstream analyses such as differential expression, differential adenylation, non-adenosine profiling, and alternative polyadenylation.

Previous studies have reported roles for U1 snRNP and the CFI complex in regulating polyadenylation site selection^6–8 and for mRNA methylation in RNA stability and processing^13,14 in Arabidopsis thaliana. To explore these interactions, we established a collection of Oxford Nanopore Direct RNA Sequencing (DRS) data derived from wild type Col-0 plants and mutants defective in key components of the polyadenylation⁶, splicing⁷, and RNA methylation¹³ machinery. Three types of mutant plant lines are included: (I) those disrupted in the cleavage factor I complex (AtCFI): atcfi25a-1/atcfi25b-2, atcfi59-1, atcfi68-3, atcfi59-1/atcfi68-3, (II) knockdowns of the U1 snRNP components: amiR-u1-70k, amiR-u1-c, and (III) the mRNA adenosine methylase mutant: SALK_074069 T-DNA insertion with the ABI3MTA complementation cassette (Fig. 1b,c).

Poly(A) tails are essential in post-transcriptional regulation, yet their analysis remains challenging. This is because long homopolymer stretches are prone to biases in sequencing approaches relying on a cDNA proxy^15–17. DRS is an amplification-free, real-time method that overcomes these limitations, yielding full-length, strand-specific reads from individual RNA molecules^15,18–20. In DRS, the sequencer records characteristic shifts in ionic current (i.e. raw signal; squiggle), as given molecule passes through the pore^15,18. Based on the properties of the resulting squiggle, poly(A) tails can be identified and their lengths estimated^9,18,21. Recent studies have shown that poly(A) tails may also contain non-adenosine residues with possible regulatory functions, underscoring the need to assess their full nucleotide composition^17,22–27. To address this, we developed Ninetails, a neural network-based R package that leverages raw signals to detect and quantify non-adenosines within poly(A) tails¹⁶.

With DRS, we obtained full-length RNA reads with native poly(A) tails, complemented by Nanopolish-derived¹⁸ poly(A) length estimates, Ninetails-based¹⁶ poly(A) composition profiles, and DESeq2-generated²⁸ differential expression data (Fig. 1d). Together, these data offer a valuable resource for studying how mutations in RNA-processing factors shape transcriptome architecture, poly(A) tail length and composition, and the interplay among splicing, polyadenylation, and mRNA methylation. As the datasets were generated from total RNA rather than from a nucleolar-enriched fraction, they allow only indirect assessment of splicing changes. Nonetheless, they provide an important foundation for research on development, differentiation, fertility, vernalization, and stress responses^1,29–33 – processes in which post-transcriptional gene regulation plays a key role – by offering integrated RNA-level measurements that enable comparative analyses across multiple mutants affecting distinct steps of RNA maturation (Fig. 1).

Methods

Experimental plant models

The dataset was generated using Arabidopsis thaliana (L.) Heynh. lines in the Columbia-0 (Col-0) background representing mutations affecting (I) the AtCFI cleavage factor complex, (II) U1 snRNP components, and (III) mRNA adenosine methylation.

Mutants affecting the AtCFI complex included the T-DNA insertion lines atcfi25a-1 (SALK_032994; AT4G25550) and atcfi25b-2 (SAIL_367_F03; AT4G29820), previously characterized as null alleles⁸. These lines were crossed to generate the atcfi25a-1/atcfi25b-2 double mutant. In addition, the T-DNA insertion mutants atcfi59-1 (SALK_090357; AT1G13190) and atcfi68-3 (SALK_036546; AT3G66652), originally identified in the Salk Institute Genomic Analysis Laboratory collection, were included, together with the atcfi59-1/atcfi68-3 double mutant generated by crossing the respective single mutants⁶.

To assess the role of U1 snRNP components, artificial microRNA knockdown lines amiR-u1-70k (targeting U1-70K, AT3G50670) and amiR-u1-c (targeting U1C, AT4G03120) were used. These lines were kindly provided by Sascha Laubinger and were previously described and validated⁷.

The dataset also included a line affecting mRNA adenosine methylation, mta (AT4G10760), carrying the SALK_074069 T-DNA insertion together with the ABI3:MTA complementation cassette^13,14, originally generated and characterized previously¹³.

The parental SALK and SAIL insertion lines were obtained from the Arabidopsis Biological Resource Center collections. All plant genotypes were verified by RT-PCR-based genotyping⁶. The functionality of the AtCFI-related mutant lines and their molecular phenotypes were further validated in our previous work⁶, where independent molecular assays confirmed their impact on RNA 3′-end processing. Observed plant traits were coherent with literature data^6,8,13.

Plant growth conditions

All seeds were surface-sterilized and placed on solid half-strength Murashige and Skoog (MS) medium supplemented with sucrose (0.8% agar, B5 vitamins, 2.5 mM MES, and 1% sucrose, pH 5.7)³⁴, stratified for 3 days at 4 °C, and then transferred to 22 °C under a 16:8 h light/dark photoperiod at 70% relative humidity^6,8. Seedlings were harvested 14 days after germination, flash-frozen in liquid nitrogen, and stored at −80 °C until use⁶.

RNA extraction

For each RNA isolation, 60 mg of plant material from pooled seedlings ground in liquid nitrogen was used. Total RNA was extracted with the RNeasy Plant Mini Kit (Qiagen) according to the manufacturer’s instructions. Subsequently, the RNA quality was assessed using the Agilent RNA 6000 Nano Kit (Agilent). Only samples with an RNA integrity number (RIN) > 8 were further processed. Individual RNA preparations were then pooled to obtain 100 µg of total RNA per biological replicate⁶. Each genotype was represented by 3 biological replicates (samples). Each sample was further processed separately.

mRNA enrichment with GST-eIF4E^K119A protein

To support efficient sequencing, it is necessary to obtain highly enriched mRNA while limiting bias from total RNA. Accordingly, before library preparation, we performed mRNA enrichment using the mutated recombinant initiation factor GST-eIF4E^K119A which binds strongly to the 5′-cap structure. This strategy has been proven to be one of the most effective methods for mRNA enrichment^35,36. Glutathione Sepharose 4B resin was incubated with GST-eIF4E^K119A in sterile PBS (200 µl resin per 200 µg protein) for 1 h at room temperature with rotation. The resin was then washed twice with PBS and three times with buffer B (10 mM potassium phosphate buffer, pH 8.0, 100 mM KCl, 2 mM EDTA, 5% glycerol, 0.005% Triton X-100, 6 mM DTT, and 20 U/mL RiboLock). Denatured arabidopsis total RNA (100 µg; 10 min at 70 °C) was added to the prepared resin and incubated for 1 h at room temperature with rotation. The resin was washed three times with buffer B, twice with buffer B containing 0.5 mM GDP, and twice with buffer B lacking GDP. RNA was then recovered by acid phenol:chloroform extraction followed by ethanol/sodium acetate precipitation with GlycoBlue co-precipitant (Thermo Scientific)³⁵.

Oxford Nanopore direct RNA sequencing

Libraries for direct RNA sequencing were prepared using the SQK-RNA002 Direct RNA Sequencing Kit (Oxford Nanopore Technologies, ONT). For each biological replicate, 2–5 µg of cap-enriched plant RNA was mixed with 100–150 ng of Saccharomyces cerevisiae oligo(dT)-enriched RNA, 2–2.5 µg of murine cap-enriched RNA, and 50 ng of in vitro-transcribed poly(A) standards^35,37. All subsequent steps, including reverse transcription, were carried out according to the manufacturer’s instructions. Sequencing was performed on the flow cell type R9.4.1 RevD with the MinION device (ONT) controlled by MinKNOW software (ONT)^16,35. If sequencing yield for a given sample was insufficient, the sample was resequenced, and the resulting technical replicates were pooled and treated as a single dataset. All sequenced samples are listed in Table 1 uploaded in BioStudies³⁸ and in Zenodo³⁹.

Data processing

Obtained DRS reads were basecalled using Guppy 5.0.11 (ONT). Depending on the downstream analysis, reads were aligned either to the Arabidopsis reference transcriptome (TAIR10) for poly(A) profiling or to the genome for differential expression analysis using minimap2⁴⁰ 2.17. Alignment to the transcriptome was performed with the parameters: -k 14 -ax map-ont–secondary = no. Alignment to the reference genome used the parameters: -k 14 -ax splice -uf–secondary = no. Resulting alignments were processed with samtools⁴¹ to remove unmapped reads and supplementary alignments. In case of transcriptome-mapped data, reads aligned to the reverse strand were also removed (samtools view -b -F 2320).

Data analysis

Differential expression analysis

Read distribution across genes was summarized for each sample using featureCounts from the Subread package⁴² (-L–fracOverlap 0.5–fracOverlapFeature 0.5 -s 1). Reads overlapping multiple features and multimapping reads were excluded. Differential expression analysis was performed using DESeq2²⁸ with default settings. For clearer visualization, we recommend applying log₂ fold-change shrinkage using the apeglm⁴³ method.

Poly(A) tail length estimation

Poly(A) tail lengths and coordinates within the raw signals were determined using nanopolish 0.13.2 polya function¹⁸ with default settings. The software assigns a quality tag to each tail estimate, classifying reads into five categories. In most cases, reads tagged as PASS or SUFFCLIP are recommended, as they are technically accurate and consistent with the reference^16,17. For differential adenylation analysis, only reads tagged as PASS were considered. Exploratory data analysis and statistical inference was performed in RStudio 4.2.1 with the NanoTail³⁷ package. For transcripts supported by at least 10 reads, statistical significance was assessed using the two-sided Wilcoxon signed-rank test (α = 0.05), and p-values were corrected for multiple testing with the Benjamini-Hochberg adjustment. A transcript was considered to show a meaningful change in poly(A) tail length if its adjusted p-value was below 0.05 and the tail-length difference between genotypes exceeded 5 nt. To help distinguish genuine biological shifts in tail length from statistically significant results driven primarily by large read counts, we additionally evaluated effect sizes using Cohen’s d. The results were visualized with NanoTail built-in plotting functions, based on the ggplot2⁴⁴ and dplyr⁴⁵ packages.

Poly(A) tail nucleotide content prediction

We analyzed raw DRS data from A. thaliana mutants and wild-type plants to identify non-adenosine residues within poly(A) tails using the Ninetails¹⁶ tool (see Code Availability). The analysis was performed using the check_tails function with the following parameters: pass_only = FALSE and qc = TRUE. Results generated by the Ninetails pipeline are provided in TSV format in BioStudies³⁸ (https://identifiers.org/biostudies:S-BSST3126). The same dataset is also available on Zenodo³⁹ (https://doi.org/10.5281/zenodo.17711220).

Data Records

Here, we present previously unpublished Oxford Nanopore DRS sequencing data (24 samples in total, sequenced across 28 sequencing runs) together with accompanying technical validation reports, differential expression and poly(A) profiling results. The raw sequencing reads in ONT native FAST5 format have been deposited in the European Nucleotide Archive (ENA) under the accession numbers PRJEB75356⁴⁶ (secondary accession: ERP159930), PRJEB80123⁴⁷ (secondary accession: ERP164163), and PRJEB103075⁴⁸ (secondary accession: ERP184427). The identifiers of the individual samples are listed in Table 1 obtainable from repositories indicated below. It should be noted that the samples were mixed with additional biological and synthetic material (e.g., yeast and synthetic spike-in controls), sequenced at different time points and deposited as part of multiple distinct sequencing projects.

The remaining data, including all tables containing raw read counts for each sample and basic quality metric, results of differential expression analysis, and poly(A) tail profiling, have been deposited in BioStudies³⁸ (https://identifiers.org/biostudies:S-BSST3126) and Zenodo³⁹ (https://doi.org/10.5281/zenodo.17711220). This dataset provides a previously unpublished extension to prior work, providing new insights into RNA metabolism in Arabidopsis⁶.

The metadata facilitating downstream analyses and the results of differential adenylation and expression analyses are provided in XLSX files named with Table prefix. The tables themselves are not included within the manuscript and are available through the record stored in abovementioned repositories.

Table 1 (Table1_sample_metadata.xlsx) contains essential metadata for all DRS samples, as well as summaries of poly(A)-tail length distributions for each replicate (sample) and genotype (condition). The table includes key information required for unambiguous sample identification and subsequent data processing.

Table 2 (Table2_polyadenylation_nanopolish.xlsx) contains differential adenylation results. It is organized into seven sheets, each comprising eighteen columns. These include: the transcript identifier in Ensembl format (ensembl_transcript); p-value, representing statistical significance calculated using a two-sided Wilcoxon rank-sum test with alpha = 0.05; cohen-d, reporting effect size; and stats_code, indicating data quality and whether both conditions contained sufficient reads for reliable inference. Additional columns include those with _counts suffix, report the number of mapped reads in control and mutant samples, and those with _polya_gm_mean suffix, which provide the geometric mean of poly(A) tail length in each condition. The length_diff column provides the measure of the change in poly(A) tail length between control and mutant samples, while fold_change indicates the magnitude of this shift. The padj column contains p-values adjusted for multiple testing with the Benjamini–Hochberg method to control the false discovery rate, which makes this metric more robust than unadjusted estimates⁴⁹, effect_size gives a descriptive measure of the magnitude of the change, and significance states whether the observed difference is statistically significant. The table furthermore reports the gene symbol (symbol), chromosome (chr), Ensembl gene identifier (ensembl_gene_id), transcript biotype (transcript_biotype), and a short functional summary of each gene (description). The corresponding raw nanopolish polya estimations are provided as a TAR archive file named Nanopolish_output.

Table 3 (Table3_differential_expression_deseq2.xlsx) contains the results of the DESeq2 analysis of differential gene expression. The file consists of seven sheets, each containing sixteen columns. The column ensembl_transcript lists transcript identifiers in Ensembl format. The column baseMean reports the mean normalized read count across all samples for each transcript, whereas baseMean_1 and baseMean_2 provide the mean normalized read counts calculated separately for condition 1 and condition 2. The column log2FoldChange represents the log-transformed fold change (base 2) between conditions, where positive values indicate increased expression in the second condition and negative values indicate a decrease. The column lfcSE contains the standard error associated with the log2FoldChange estimate, and stat represents the Wald statistic, calculated as the ratio of log2FoldChange to its standard error. The column foldChange shows the fold change between conditions on the original (non-logarithmic) scale. Raw p-values from the Wald test are reported under pvalue, while padj provides p-values adjusted for multiple testing using the Benjamini-Hochberg procedure. Statistically significant results are indicated in the sig column according to the threshold adopted for this study. Additional annotation fields are included for biological interpretation. The column symbol lists gene symbols, chr indicates the chromosome on which each gene is located, and ensembl_gene_id contains the corresponding Ensembl gene identifier. The column transcript_biotype categorizes the transcript type, for example protein-coding or non-coding RNA. Finally, description gives a concise summary of gene function or properties. Collectively, this dataset enables quantitative evaluation of expression changes and supports downstream functional analyses of transcripts in sequenced Arabidopsis genotypes.

The corresponding raw read to gene feature assignment summary and mapping obtained with featureCounts used for differential gene expression analysis with DESeq2 is provided as a TAR archive file named FeatureCounts_output.

The Ninetails_output TAR archive contains classification results in TSV format. For each analyzed sample, Ninetails produces two output tables. Files with the suffix class_data contain read classification results, where each row corresponds to a single read. These files indicate whether the poly(A) tail meets quality criteria and whether it is decorated with non-adenosine residues or is blank.

Files with the suffix residue _data provide details on individual non-adenosine occurrences. Each row represents one instance, specifying the residue type, its position within the tail, and its length. A comprehensive description of Ninetails output is available in the Wiki (https://github.com/LRB-IIMCB/ninetails/wiki).

Output files are named according to the metadata table. Data generated by the main analysis module (check_tails function) can be further processed using Ninetails built-in functions. For example, summarizing results, converting between wide and long formats, or visualizing with the Ninetails graphics module. All these features are documented in the built-in help files and the Wiki under Code Availability.

In tabular data, we avoid using spaces and special characters to ensure compatibility and simplify analysis with tools across different programming languages.

The repository also contains the TAR.GZ archive file containing reads mapping to Arabidopsis TAIR10 reference in FASTQ format and PycoQC_output TAR archive, containing interactive quality reports for each sequencing run included in this study.

Technical Validation

Genetic validation of plant lines and mutant phenotypes

The genetic identity and functional relevance of the plant lines used in this study were verified using several complementary approaches. T-DNA insertion mutants affecting components of the AtCFI complex: atcfi25a-1, atcfi25b-2, atcfi59-1, and atcfi68-3, were obtained from SALK and SAIL collections and combined by genetic crosses to generate the atcfi25a/atcfi25b and atcfi59/atcfi68 double mutants. U1 snRNP knockdown lines (amiR-u1-70k and amiR-u1-c) were previously developed and validated by Mangilet et al.⁷, while the mta mutant affecting mRNA adenosine methylation was originally described by Bodi et al.^13,14. In the present work, plant genotypes were confirmed by RT-PCR followed by agarose gel electrophoresis⁶.

In addition, the molecular phenotypes of the AtCFI-related mutants were independently validated in our previous study⁶. That work combined PAT-seq analyses of poly(A) site usage with protein–protein interaction assays, including co-immunoprecipitation coupled to mass spectrometry, demonstrating that the affected proteins interact within the CFI complex and that disruption of these components leads to consistent and overlapping alterations in RNA 3′-end processing. Together, these complementary genetic and molecular validations support the attribution of the observed transcriptome and polyadenylation changes to the targeted mutations analyzed in this dataset.

Quality assessment and consistency of the DRS datasets

For the 24 DRS datasets, read counts vary between samples (Fig. 2a, left panel), largely due to factors such as the number of active pores within a given flow cell and amount of starting material, as DRS lacks amplification. Sequencing runs were monitored in real time using the MinKNOW software, which provides continuous quality control during nanopore sequencing. After completion of each run, sequencing performance was further evaluated using pycoQC⁵⁰, and interactive reports summarizing quality metrics (including N50 and read length distributions) were generated. These reports for all sequencing runs are provided together with the dataset on Zenodo.

Fig. 2.

Technical validation of the data: read quality, mapping summary, and sample-level variation across nanopore DRS libraries. (a) Read count (left) and read frequency (right) distributions for all samples across genotypes. Nanopolish quality-control tags are shown. (b) Summary of read assignment produced with FeatureCounts (Subread). Bars show the proportion of reads classified as respective type. (c) Principal component analysis (PCA) of log₂-scaled count data (DESeq2), illustrating sample clustering and variance structure across genotypes and replicates. (d) Transcript count correlation plot. Correlation analyses of read counts/transcripts were performed using Spearman’s rank correlation coefficient. Mapped reads fulfilling nanopolish quality criteria are included.

Nevertheless, this variation does not affect transcriptome profiles, as enrichment ratios remain consistent across biological replicates. Read quality tag distributions are also similar among the replicates (Fig. 2a, right panel). Read summarization performed after aligning sequencing reads to the reference TAIR10 genome showed that, in all DRS datasets, the majority of reads correctly mapped to genomic features (Fig. 2b). Accurate read alignment and counting are essential, as they directly influence the reliability of differential expression analysis²⁸. To further assess samples similarity, principal component analysis (PCA) was calculated based on differential expression profiles. All samples from the same group cluster together on the plots, which indicates a high degree of homogeneity among the replicates (Fig. 2c). Spearman’s rank correlation analysis of read counts per transcripts further demonstrated that the data are coherent and highly correlated between samples (Fig. 2d).

Poly(A) tail length and composition

Poly(A) tail length distribution is consistent within groups representing individual mutants. We observed a global elongation of poly(A) tails in the amir-u1-70k mutant, whereas SALK-074069 and the double mutant atcfi59-1/atcfi68-3 showed global shortening of poly(A) tail lengths (Fig. 3a). The effect size of the poly(A) tail length difference between control and mutant samples varies across distinct groups of transcript sets enriched for specific gene ontology (GO) terms (Fig. 3b).

Fig. 3.

Distributions of poly(A)-tail lengths across Arabidopsis genotypes. (a) Poly(A)-tail length distributions in all datasets, shown either by individual sample (left) or by genotype group (right). Vertical dashed lines indicate median poly(A)-tail lengths for each genotype. (b) Poly(A)-tail length distributions for transcripts belonging to selected Gene Ontology (GO) categories, grouped by genotype. Shown are transcripts annotated with extracellular region (GO:0005576), extracellular matrix (GO:0031012), mRNA 3′-end processing (GO:0031124), mRNA modification (GO:0016556), spliceosome assembly (GO:0000245), and cis-splicing (GO:0045292). Median poly(A)-tail lengths for each genotype are indicated by vertical dashed lines.

To characterise the poly(A) tail composition across 24 DRS datasets, reads were classified into five categories according to their quality and conformity to the criteria used for detecting non-adenosine residues within poly(A) tails. For subsequent analyses, only reads exhibiting a significant shift between consecutive k-mers (a so-called ‘move’) and displaying a characteristic anomaly in the raw signal were retained (Fig. 4a). Using a neural network, the Ninetails assigns the shape of each anomaly within a poly(A) tail to one of three categories, corresponding to the presence of cytidine, guanosine, or uridine¹⁶.

Fig. 4.

Nucleotide composition of poly(A) tails. (a) Read classification generated by Ninetails for poly(A) tail composition analysis: left – detailed classification output, right – only decorated (containing non-adenosines) reads shown. (b) Frequency of occurrence of respective non-adenosines (cytidine, guanosine, uridine) in poly(A) tails reported by residue. In this reporting mode, each individual non-adenosine nucleotide contributes separately to the count; thus, a read containing three uridines contributes three events, while a read containing two cytidines, one guanosine, and one uridine contributes four events. This approach reflects the total abundance of non-adenosine nucleotides within poly(A) tails rather than the number of reads containing them. Only reads classified by the neural network (decorated and blank) were included. The methodology for non-adenosine reporting has been described in detail in our previous work¹⁶. (c) Examples of poly(A) tail signals: left – blank (i.e. with adenosines exclusively), right – decorated tails (i.e. containing non-adenosines). Reads were randomly sampled from blank and decorated classes in the Col-0 rep1 sample.

Non-adenosine abundance can be reported using two complementary modes implemented in the Ninetails framework: by read or by residue, allowing the user to tailor the analysis to the specific biological question. In the by read mode (by_read = TRUE), each read contributes once for each type of non-adenosine residue it contains^16,17,27. For example, a read containing three uridines is counted once, whereas a read containing cytidine, guanosine, and uridine is counted three times (once per residue type). In contrast, in the by residue mode (by_read = FALSE), each individual non-adenosine nucleotide contributes separately to the count. Thus, a read containing three uridines contributes three events, while a read containing two cytidines, one guanosine, and one uridine contributes four events. These two reporting strategies allow the analysis to emphasize either the diversity of non-adenosine residue types across poly(A) tails or their overall abundance within the dataset. Here, non-adenosine frequency was reported by residue (Fig. 4b), thereby reflecting the total abundance of cytidine, guanosine, and uridine nucleotides detected within poly(A) tails across the dataset.

These findings can be further verified with the raw signal – the examples of such visual inspection of signals are shown in Fig. 4c.

Author contributions

A.J. and A.D. acquired funding and together directed the study. Ł.S. cultured plants, performed crosses, genotyped the established plant lines, and isolated RNA. M.B. cultured plants and provided phenotype documentation. W.O., S.J. and S.M. prepared the DRS libraries, and W.O. with S.J. performed the sequencing. N.G. curated and analyzed the sequencing data, and prepared the figures with feedback from A.C. N.G. and A.C. wrote the manuscript with input from the remaining contributors. All authors have read and approved the final version of the manuscript.

Funding

This work was supported by grant from Polish National Science Centre to (UMO-2018/31/F/NZ2/03740 4 to A.J.) and European Research Council (ERC AdG 101097317 to A.D.).

Data availability

The raw FAST5 DRS data generated in this study have been deposited in the European Nucleotide Archive (ENA) database under following accession numbers: PRJEB75356⁴⁶ (secondary accession: ERP159930) (https://identifiers.org/ena.embl:ERP159930), PRJEB80123⁴⁷ (secondary accession: ERP164163) (https://identifiers.org/ena.embl:ERP164163) and PRJEB103075³⁹ (secondary accession: ERP184427) (https://identifiers.org/ena.embl:ERP184427). The remaining data (FASTQ, excel tabular files (Tables 1–3), figures, interactive QC HTML reports and raw outputs of Ninetails, nanopolish and featureCounts) are publicly available in BioStudies³⁸ (https://identifiers.org/biostudies:S-BSST3126). The same dataset is also deposited in Zenodo³⁹ (https://doi.org/10.5281/zenodo.17711220).

Code availability

The Ninetails software, designed for poly(A) composition profiling, was presented and evaluated in detail in our previous studies^16,17,27. The source code is publicly available on GitHub (https://github.com/LRB-IIMCB/ninetails). The version accompanying this publication is archived on Zenodo (https://doi.org/10.5281/zenodo.13309819). Supplementary processing scripts and the Docker build are provided via GitHub (https://github.com/LRB-IIMCB/ninetails_processing) and are additionally archived on Zenodo (https://zenodo.org/doi/10.5281/zenodo.13310034).

Competing interests

The authors declare no competing interests.