Quantitative profiling of m⁶A at single base resolution across the life cycle of rice and Arabidopsis

Abstract

N⁶-methyladenosine (m⁶A) plays critical roles in regulating mRNA metabolism. However, comprehensive m⁶A methylomes in different plant tissues with single-base precision have yet to be reported. Here, we present transcriptome-wide m⁶A maps at single-base resolution in different tissues of rice and Arabidopsis using m⁶A-SAC-seq. Our analysis uncovers a total of 205,691 m⁶A sites distributed across 22,574 genes in rice, and 188,282 m⁶A sites across 19,984 genes in Arabidopsis. The evolutionarily conserved m⁶A sites in rice and Arabidopsis ortholog gene pairs are involved in controlling tissue development, photosynthesis and stress response. We observe an overall mRNA stabilization effect by 3’ UTR m⁶A sites in certain plant tissues. Like in mammals, a positive correlation between the m⁶A level and the length of internal exons is also observed in plant mRNA, except for the last exon. Our data suggest an active m⁶A deposition process occurring near the stop codon in plant mRNA. In addition, the MTA-installed plant mRNA m⁶A sites correlate with both translation promotion and translation suppression, depicting a more complicated regulatory picture. Our results therefore provide in-depth resources for relating single-base resolution m⁶A sites with functions in plants and uncover a suppression-activation model controlling m⁶A biogenesis across species.

Introduction

RNA modifications are critical regulators of mRNA processing and metabolism, which include splicing, 3′-end processing, nuclear export, translation, and decay. m⁶A is identified as the most abundant internal mRNA modification in mammals and plants^{1,2,3,4,5,6,7,8,9,10,11,12}. In Arabidopsis, two different m⁶A writer complexes have been suggested to install m⁶A on mRNA. The first methyltransferase complex is composed of five respective orthologs of the components of the mammalian m⁶A methyltransferase complex that include mRNA adenosine methylase (MTA), MTB, VIRILIZER (VIR), FKBP12 INTERACTING PROTEIN 37KD (FIP37), and an E3 ubiquitin ligase HAKAI^13,14,15,16. FIONA1 (FIO1), by contrast, is the Arabidopsis ortholog of the human methyltransferase METTL16, also depositing m⁶A modifications in U6 small nuclear RNA and a subset of mRNAs^17,18,19. Defects in the m⁶A writer protein complex, such as MTA failure in plants, cause an embryo-lethal phenotype^15,20,21,22 and stress responses^{23,24,25,26,27}. Studies of other components of the writer complex in Arabidopsis revealed that FIP37 regulates shoot stem cell fate, FIO1 regulates floral transition and chlorophyll homeostasis, and VIRILIZER (VIR) is critical to vascular development^{13,14,18,19,28}; whereas the defect of FIP37 in rice leads to early degeneration of microspores^29,30. The m⁶A methylation can be reversed^31,32. RNA m⁶A demethylases ALKBH10B and ALKBH9B, homologs of the human m⁶A demethylase ALKBH5³², affect floral transition³³ and viral infection³⁴ in Arabidopsis. In our recent study, overexpression of the mammalian m⁶A demethylase FTO in rice dramatically increased the biomass and yield of rice and potato³⁵, revealing that modulation of RNA m⁶A methylation could be a promising breeding or engineering strategy for crop improvement in the future. Although these observations indicate conserved mRNA m⁶A methylation regulators in plants as compared with animals, so far the high-resolution mRNA m⁶A maps in plants are still missing and the molecular level connection of m⁶A to plant development and other pathways are mostly unknown. We proceeded to fill this gap by using the newly developed high-resolution sequencing method^36,37.

Methylated RNA m⁶A immunoprecipitation sequencing (MeRIP-seq or m⁶A-seq)³⁸ has been widely used in identifying m⁶A-enriched transcripts in animals and plants. However, this method lacks single-base resolution and cannot quantify the extent of the modification. Variations of MeRIP-seq, including m⁶A individual-nucleotide-resolution crosslinking and immunoprecipitation (miCLIP)³⁹, have been developed to detect m⁶A sites at single-base resolution. Unfortunately, these methods typically display low efficiency of UV crosslinking and cannot assess modification stoichiometry. Antibody-independent single-base m⁶A profiling methods, such as m⁶A-REF-seq⁴⁰ or MAZTER-seq⁴¹, have been introduced. However, these methods can only identify RNA modifications in the ACA motif, and fail to identify other methylation motif. Nanopore direct RNA-seq (DRS) has been utilized to map m⁶A sites, but quantifying the differences in m⁶A modification levels with DRS is still challenging^42,43,44. We have recently reported m⁶A-selective allyl chemical labeling and sequencing (m⁶A-SAC-seq)^36,37 as a method capable of precisely mapping of individual m⁶A-modified sites in whole transcriptomes at single-nucleotide resolution. Although there are recently reported deamination-based methods that can also map m⁶A at base resolution, including one from us and collaborators^45,46, m⁶A-SAC-seq is capable of reading m⁶A as a positive mutation signal without subtraction, and the optimized protocol works with 2–5 ng of input RNA³⁶. We, therefore, decided to deploy m⁶A-SAC-seq to establish comprehensive maps of mRNA m⁶A at single-nucleotide resolution across various tissue types in two different plant species: Arabidopsis and rice.

A considerable number of high-confident m⁶A sites were identified spanning the entire life cycle of both rice and Arabidopsis. The evolutionarily conserved m⁶A mRNA modification sites across rice and Arabidopsis ortholog gene pairs play regulatory roles in tissue development, photosynthesis, and stress response. m⁶A levels are positively correlated with the length of the internal exon, but such correlation is missing in the last exon. Through comparative base-resolution m⁶A analysis across humans, rice and Arabidopsis, we unveil a distinct m⁶A distribution pattern that a suppression-activation dual model governs the m⁶A deposition in humans and plants. Using the robust method for comparison of m⁶A levels at single base resolution, we noticed that rice and Arabidopsis possess higher percentages of the overall m⁶A modifications in the 3’ UTR of their mRNAs than those of mammalian systems. These 3’ UTR m⁶A modifications generally stabilize mRNA and enhance translation, and these effects correlate well with the m⁶A fraction. These observations indicate that both fraction and position of m⁶A modification are critical for mRNA metabolism in plants. In addition, we noticed that the MTA-mediates m⁶A deposition in photosynthesis-related genes that transcribed from both nuclear and chloroplast genome in Arabidopsis. These m⁶A sites can either promote or reduce translation efficiency in a pathway-dependent manner. Altogether, our base-resolution and quantitative m⁶A sites across rice and Arabidopsis, provide a foundation for future studies to explore the regulatory roles of m⁶A in regulating plant development and evolution and for future plant engineering.

Results

m⁶A-SAC-seq identifies m⁶A modification sites in rice and Arabidopsis

m⁶A-SAC-seq utilizes the dimethyltransferase MjDim1 to introduce an allyl group to m⁶A, which upon chemical-induced cyclization could be read as mutation signals during reverse transcription (Supplementary Fig. 1a)^36,37,47. We extracted total RNAs from nine Arabidopsis tissues (seedling, shoot, root, rosette leaf, cauline leaf, stem, flower, silique and seed) as well as eight rice tissues (plumule dark, plumule light, seedling at 8 days, seedling at 2 weeks, panicle, flag leaf at 10 days after anthesis, endosperm at 10 days after anthesis, and embryo at 10 days after anthesis) with two biological replicates for each sample (Fig. 1a, b). PolyA-tailed RNA of each biological replicate was purified and subjected to LC-MS/MS to measure the m⁶A/A ratio. The ratio of m⁶A/A in polyA-tailed RNA from these different tissues varied within the range of 0.36–0.75% in Arabidopsis (Fig. 1c) and in the range of 0.52–0.67% in rice (Fig. 1d). The remaining polyA-tailed RNAs were then processed following the m⁶A-SAC-seq library construction protocol^36,37 to map m⁶A sites at the base resolution.

Fig. 1: m⁶A-SAC-seq effectively identifies m⁶A sites across plant tissues.

a, b, Tissues from eight different rice organs (a) and nine different Arabidopsis organs (b), collected throughout their respective life cycles, were subjected to m⁶A-SAC-seq. c, d, mRNA m⁶A levels in the harvested samples were quantified using LC-MS/MS for both rice (c) and Arabidopsis (d). The m⁶A-to-A ratio was determined using calibration standards. Data are means ± SD, n = 6. e, Venn diagram showing m⁶A sites detected by SAC-seq overlapping with those identified by miCLIP and DRS, with the ±5 nt sliding window around each m⁶A site. f, A comparison of the percentages of overlapping m⁶A sites identified by different methods. g, h, The number of m⁶A sites and their motif distribution in rice (g) and Arabidopsis (h) tissues were shown. Motif sequences were separated into three groups, RAC, GAT and others. Base “R” denotes either A or G. 8D represents 8 days and 2 W represents 2 weeks. Source data are provided as a Source Data file.

To reduce background noise and eliminate any potential batch effect, we added 2% spike-in calibration probes into each sample. These calibration probes contained varying fractions of m⁶A within the NNm⁶ANN motif. By determining the mutation rates of both A and m⁶A within different motifs in these probes, we could best determine the m⁶A modification fraction at individual sites in each sample. Our initial results showed that the labeling efficiency of the calibration probes in plant cells was very high, comparable to that observed in mammalian cells (Supplementary Fig. 1b). Additionally, the average background noise of unmodified A sites was as low as 0.49%, which is 10 times lower than the cutoff (<5%) used for m⁶A site detection, indicating high m⁶A-SAC-seq data quality obtained from these plant samples. Furthermore, we assessed the relative conversion ratio and background noise in each sample. The relative conversion ratio ranged from 0.92 to 1.11 in Arabidopsis samples (Supplementary Fig. 1c) and from 0.96 to 1.07 in rice samples (Supplementary Fig. 1d), showing consistent efficiency of m⁶A labeling across samples. By carefully calibrating the differences observed in each sample, we ensured a fair and accurate comparison of m⁶A levels among different samples. This calibration step is crucial for obtaining reliable and meaningful results, enabling us to effectively compare the m⁶A modification among various plant samples.

Before proceeding to downstream analysis, we compared our m⁶A-SAC-seq data with previously published m⁶A sites profiled by DRS⁴⁸ and miCLIP³⁹ in Arabidopsis. Only sites with a sufficient number of sequence coverage among all tissues (depth> 10) were selected for further analysis. A total of 188,282 m⁶A sites were obtained when we combined the m⁶A sites detected by SAC-seq in this study, with ~42% and ~17% of the m⁶A sites measured by DRS and miCLIP overlapping with SAC-seq m⁶A sites, respectively (Supplementary Fig. 1e). With a sliding window of ±5 nt around the m⁶A sites, many more previously detected m⁶A sites overlap with the m⁶A-SAC-seq sites (Fig. 1e, f). As expected, a much higher percentage of m⁶A sites overlap between m⁶A-SAC-seq and DRS than that between m⁶A-SAC-seq and miCLIP (Fig. 1f). This might suggest the low accuracy of miCLIP in identifying m⁶A site, which could be caused by low crosslinking efficiency in plant tissues.

Principal component analysis (PCA) revealed a distinct clustering based on the m⁶A fractions of the different tissues (Supplementary Fig. 2a–d). An average of 49,791 m⁶A sites from 12,652 genes with at least 20 reads were identified in Arabidopsis libraries. In rice, an average of 67,173 m⁶A sites were detected from 15,138 genes. The number of high-confidence m⁶A sites from eight different rice tissues varied within the range of 52,646 in seedling 2 W to 82,157 in flag leaf (Fig. 1g). While the number of m⁶A sites among nine Arabidopsis tissues ranged from 25,990 in seed to 74,259 in silique (Fig. 1h). In comparison with previously published Arabidopsis MeRIP data of seedling (7,489 m⁶A peaks), we identified approximately 59,212 m⁶A sites within 14,180 genes in the seedling datasets, demonstrating high sensitivity of the SAC-seq method. Therefore, Arabidopsis seedling transcriptome contains ~4.2 m⁶A sites per gene, which is four-fold higher than that observed in the MeRIP data⁴⁹. Interestingly, ~4.4 m⁶A sites per gene were observed among rice tissues, suggesting a likely conserved distribution density across different plant species. Consistent with the findings in mammals^36,37 and plants^9,50,38,51, the RAC (R = A or G) motif displayed the highest frequency among methylated motifs in both rice and Arabidopsis (Fig. 1g, h).

Base-resolution mRNA m⁶A maps from different tissues of rice and Arabidopsis

We next analyzed the distribution of m⁶A in the whole transcriptome for both Arabidopsis and rice. All the identified m⁶A sites in Arabidopsis and rice are accessible under the GEO numbers GSE245738 and GSE243722, respectively. As observed in the metagene profile, most m⁶A sites are highly enriched within 3′-untranslated region (3′ UTR), followed by coding DNA sequence (CDS) and 5′-untranslated region (5′ UTR) in both Arabidopsis (Fig. 2a) and rice (Supplementary Fig. 3a). Although both mammalian and plant mRNAs highly enrich m⁶A in the 3’ UTR (Fig. 2b), rice and Arabidopsis mRNAs harbor noticeably higher percentages of the overall m⁶A modifications in the 3’ UTR compared with that of human HeLa cells³⁶ (Fig. 2b). In addition to m⁶A sites in the regions of 3’ UTRs and CDSs, we also observed a considerable number of m⁶A sites in the intronic, and 5’ UTRs regions in both rice and Arabidopsis, which are consist with past results observed in mammals³⁷ (Supplementary Fig. 3b, c). The average m⁶A fraction is notably higher in intronic regions than in the 5’ UTR and CDS, but lower than that in the 3’ UTR (Supplementary Fig. 3b, c).

Fig. 2: Comprehensive base-resolution maps of m⁶A sites in rice and Arabidopsis.

a Metagene profiles showing the m⁶A site distribution across transcripts in nine Arabidopsis tissues. Each transcript is segmented into three regions: 5′ UTR, CDS and 3′ UTR. The black dashed line represents the average m⁶A fraction across the nine tissues. The m⁶A site number (N) is indicated in the figure. The percentage (%) of overall m⁶A modifications distributed in 5′ UTR, CDS, and 3′ UTR regions within different tissues were shown. b Percentages of total m⁶A modifications distributed in 5′ UTR, CDS, and 3′ UTR regions in Hela cells, rice and Arabidopsis. We combined m⁶A sites from all the rice tissues and Arabidopsis tissues, respectively, to calculate percentages of total m⁶A fractions within 5′ UTR, CDS, and 3′ UTR regions. c, d Bar plot showing the number of tissue-unique and tissue-common m⁶A sites in eight rice tissues (c) and nine Arabidopsis tissues (d). The m⁶A site number is the overlapped m⁶A sites between two biological replicates for each tissue. e Metagene profiles showing common m⁶A sites among nine Arabidopsis tissues distributed across transcripts. f Metagene profiles of tissue-unique m⁶A sites among nine Arabidopsis tissues. g Metagene profiles showing the rice reproductive unique-, vegetative unique- and common m⁶A sites distributed across transcripts. Tissues of flower, seed and silique were grouped as reproductive tissues, while the remaining tissues were considered vegetative. Each transcript is divided into three regions: 5′ UTR, CDS and 3′ UTR. For e.g., the numbers of m⁶A sites are indicated in the figures.

Both the m⁶A/A ratio and m⁶A site number vary among different tissues, suggesting the presence of both tissue-shared and tissue-specific deposition of m⁶A methylation in plants. We, therefore, analyzed tissue-specific and commonly shared m⁶A modification sites among all the tissues in rice and Arabidopsis, and identified 14,953 and 6,815 shared m⁶A sites among all rice and Arabidopsis tissues, respectively (Fig. 2c, d). These shared m⁶A modification sites were predominately enriched in the 3′ UTR region relative to this tissue-specific m⁶A both in rice (Supplementary Fig. 3d, e) and Arabidopsis (Fig. 2e, f), which may suggest that 3′ UTR m⁶A sites, rather than the CDS or 5′ UTR, play general roles in maintaining plant transcriptome metabolism across the entire life cycle. To find the biological difference between reproductive- and vegetative unique m⁶A sites, we further divided the different tissues into reproductive tissue and vegetative tissue (see Method). The reproductive unique and vegetative unique m⁶A sites were identified in both rice and Arabidopsis, respectively (Supplementary Data 1). The metagene profile revealed that reproductive unique m⁶A sites show increased distribution in the 3′ UTR region compared to the vegetative unique one, implying the significance of m⁶A regulation through 3′ UTR in the reproductive phase (Fig. 2g and Supplementary Fig. 3f). GO enrichment analysis showed that genes containing reproductive unique m⁶A sites in both rice and Arabidopsis are enriched in similar pathways, such as reproductive structure development, embryo development, immune response, photosynthesis, and chloroplast organization (Supplementary Fig. 3g). Meanwhile genes containing vegetative unique m⁶A modifications are enriched in stimulus response, such as genes of ARF1, ARF7 and ARF9 involved in response to hormone stimulus in rice (Supplementary Fig. 3h), although these pathways are not significant enriched in Arabidopsis. This likely suggests tissue-specific m⁶A methylations play regulatory roles in plant growth regulation.

Evolutionary conservation and variability of m⁶A regulation across rice and Arabidopsis

Next, we investigated the evolutionary conservation of mRNA m⁶A modification in Arabidopsis and rice orthologous gene pairs. We found a total of 12,359 pairs of conserved m⁶A sites (Supplementary Data 2) in orthologous genes, while 108,856 and 226,673 m⁶A sites were only identified in Arabidopsis and rice, respectively (Fig. 3a). Interestingly, the average m⁶A fractions of unique m⁶A sites are much higher than the rice-Arabidopsis conserved sites in rice (Supplementary Fig. 4a) and Arabidopsis (Supplementary Fig. 4b). Amongst the conserved m⁶A site pairs, 7,734 pairs of conserved m⁶A sites exist within the same motif sequence (Supplementary Data 3, Supplementary Fig. 4c and Fig. 3b). The fractions of the conserved m⁶A sites within homologous genes are weakly correlated (Fig. 3c) and vary among different tissues (Fig. 3d and Supplementary Fig. 4d), suggesting that the modification levels of conserved m⁶A sites are tissue dependent even though they are universal among all the tissues. Since the presence of m⁶A is critical for normal plant development^19,20,21,52, we explored functional insights about the rice-Arabidopsis conserved m⁶A sites. GO analysis showed that genes with conserved m⁶A sites are significantly enriched in stimulus response and plant development-related pathways, such as the chloroplast, photosynthesis, photomorphogenesis, embryo development, shoot morphogenesis, flower development, leaf development, root development, and ovule development (Fig. 3e). For example, the light harvest related genes, CAB3, Lhca5, LHCA3, and LHB1B2; root epidermal cell differentiation and root hair cell differentiation-related genes, POM1, GN, UBC36, GCS1, UBC35, GEM, SCN1, and MRH1; flower development genes PFT1, PS1, RDR6, DCL4, ARF8, and MET1 are all conserved in their m⁶A methylation sites between rice and Arabidopsis. Overall, these results provide a foundation for future studies to explore the potential roles of m⁶A under evolutionary pressure in plants.

Fig. 3: Evolutionary conservation and divergence in m⁶A regulation between rice and Arabidopsis.

a Number of conserved and species-unique m⁶A sites between rice and Arabidopsis. b Motif of RAC is the most prevalent one among the conserved m⁶A sites. Motif sequences were divided into three groups, including RAC, GAT and others. Base “R” denotes either A or G. c Correlation analysis of the m⁶A fraction in the ortholog gene pair between rice and Arabidopsis. The R² value is labeled in the figure. d Principal component analysis (PCA) of the m⁶A fraction in those conserved m⁶A sites across all tissues in rice and Arabidopsis. The modification levels of these m⁶A sites were normalized based on its tissue rankings within the respective species. e Gene ontology (GO) analysis of genes containing conserved m⁶A sites. One-sided Fisher’s exact test. Adjusted P values using the linear step-up method. f, g The ratios of standard deviation to mean for m⁶A levels across different tissues in rice (f) and Arabidopsis (g) are presented for various transcript regions: 5′ UTR, CDS and 3′ UTR. Generally, both species exhibit reduced variance from the 5′ UTR to 3′ UTR. Notably, rice showed a gradually reduced variance across the gene structure, while Arabidopsis showed rather stable variance in the CDS region followed by a sharp reduction in the 3′ UTR. Source data are provided as a Source Data file.

Despite the evolutionarily conserved m⁶A modification sites across rice and Arabidopsis, the metagene profile showed some differences, especially regarding the distribution of the m⁶A sites in the 3′UTR region. To illustrate these differences, we combined all the m⁶A sites within different tissues in rice and Arabidopsis, respectively. We calculated the ratio of standard deviation to the mean of m⁶A site counts. We observed that in general, the reduced variance from 5′ UTR to 3′ UTR in rice (Fig. 3f) and Arabidopsis (Fig. 3g), while the variance in the 5′ UTR region is fluctuant (Fig. 3f, g). Despite the similarity, rice showed gradually reduced variance across the gene structure, while Arabidopsis showed rather stable variance in the CDS region followed by a sharp reduction in variance in the 3′ UTR (Fig. 3f, g). Together these results may suggest differential m⁶A deposition regulations, especially in the CDS and 3′ UTR regions between rice and Arabidopsis.

Divergent paradigms governing m⁶A deposition in plant genomes

Earlier research has elucidated the distribution patterns of m⁶A in both mammalian and plant genomes^38,49,53,54, suggesting that m⁶A is predominantly enriched in the last exon and long internal exons^53,54. This distribution is additionally shaped by the underlying exon architecture and is regulated by the Exon Junction Complex (EJC)^55,56,57. We categorize these unique distribution patterns into three basic rules: the “long exon,” “last exon,” and “exon structure” rules. Using these quantitative m⁶A sites at single-base resolution, we investigated whether these rules are conserved across the plant kingdom. It’s worth noting that exons in the human genome are generally longer than those in rice and Arabidopsis (Supplementary Fig. 5a-c), which could lead to differences in m⁶A distribution both per exon and per sliding window. To improve the accuracy of our measurements, we introduced two metrics: “m⁶A density,” which normalizes the total m⁶A level within each exon by its length, and “m⁶A likelihood,” which normalizes the m⁶A level within each sliding window by the pileup coverage of the exons in that window (Supplementary Fig. 5d).

We re-analyzed single-base mRNA m⁶A data from mammalian samples, specifically from the HeLa cell line³⁶, and compared them with data from rice and Arabidopsis. We found that the m⁶A level per exon generally increases with internal exon length when that length is under 1000 nucleotides (nt) in both human and plant genomes (Fig. 4a-c and Supplementary Fig. 5e-g). This pattern does not hold for internal exons longer than 1000 nt in humans. Rice and Arabidopsis do not display this trend, largely because they have extremely rare internal exons exceeding 1000 nt (Fig. 4a-c). Examining “m⁶A density,” we observed an inverse correlation with internal exon length in both humans and plants (Fig. 4d-f and Supplementary Fig 5h-j). This suggests that m⁶A modifications accumulate more slowly than the exon length increases. Interestingly, overall m⁶A density is higher in plants than in humans, with a peak at around 100 nt in exon length. This indicates that excessively long or short exons could reduce m⁶A modification more effectively in plants.

Fig. 4: Divergent rules governing m⁶A deposition in plant and mammalian genomes.

a-c Internal exons of human (a), rice (b), and Arabidopsis (c) transcripts were grouped into 100 bins of equal size based on their length, and the average m⁶A level for each bin was plotted against exon length, represented by blue dots. d-f ‘m⁶A density’ of each bin of the internal exons of human (d), rice (e), and Arabidopsis (f) transcripts were shown against exon length, represented by yellow dots. m⁶A density was calculated as the m⁶A level within each exon, normalized by its length and multiplied by 1,000. g-i All internal exons were aligned at their internal exon junction sites in human (g), rice (h), and Arabidopsis (i) genomes, and the overall m⁶A level per sliding window in the flanking regions was shown against the distance to exon junction sites, represented by brown dots. j-l Distribution of ‘m⁶A likelihood’ near the internal exon junction in human (j), rice (k), and Arabidopsis (l) transcripts were shown in dark green line, with 95% confidential intervals shadowed. m-o Similar to panel (a-c) but the average m⁶A level per exon in the last exons of human (m), rice (n), and Arabidopsis (o) genomes were shown. p-r Similar to panel (d-f) ‘m⁶A density’ in the last exons of human (p), rice (q), and Arabidopsis (r) genomes. s-u Similar to panel (g-i) m⁶A level per sliding window flanking last exon junction site in human (s), rice (t), and Arabidopsis (u) genomes. v-x Similar to panel (j-l) Distribution of ‘m⁶A likelihood’ near the last exon junction in human (v), rice (w), and Arabidopsis (x) genomes. y, Diagram showing the inhibition mode in humans and activation mode in plants contributes to distinct m⁶A distribution pattern. For j-l and v-x data are presented as median values.

To estimate the probability of m⁶A modifications, we aligned all exons at their junction sites and calculated the m⁶A levels per sliding window for both humans and plants. Regions closer to these junction sites typically have higher coverage of exons, resulting in increased m⁶A levels (Fig. 4g-i). However, when normalized by the coverage (Supplementary Fig. 5q-s), “m⁶A likelihood” showed suppression behaviors at these junction sites in the HeLa cell line, consistent with prior mammalian studies. In contrast, this pattern was not observed in rice and Arabidopsis; instead, we found a slight inverse trend. This divergence suggests that the EJC complex may passively suppress m⁶A deposition in mammalian cells but not in plants. The opposite trend raises the intriguing question of whether the “exon structure” rule governing m⁶A distribution is universally conserved in plants. Active m⁶A deposition pathways may also shape mRNA m⁶A distribution in certain plants.

Regarding the last exons, a clear correlation exists between increasing m⁶A levels per exon and exon length in the HeLa cell line, even when the exon length exceeds 1000 nt (Fig. 4m and Supplementary Fig. 5k). This correlation is notably absent in rice and Arabidopsis (Fig. 4n, o and Supplementary Fig. 5l, m). Furthermore, an inverse correlation between ‘m⁶A density’ and exon length was observed for the last exon in both humans and plants (Fig. 4p-r and Supplementary Fig. 5n-p). In contrast to internal exons, m⁶A is more condensed in exons around 100 nt in both humans and plants. Remarkably, the ‘m⁶A density’ in the last exons of rice and Arabidopsis is higher than in humans, even though the overall m⁶A level is lower. These findings suggest that while the general pattern of m⁶A enrichment in the last exon is evolutionarily conserved, significant differences exist in how this modification correlates with last exon length across species.

In terms of ‘m⁶A likelihood’ near the last exon junction sites, we observed an asymmetric peak downstream from these sites in both mammals and plants (Fig. 4s-u and Supplementary Fig 5t-v). Similar to internal exons, ‘m⁶A likelihood’ decreases as it approaches the last exon junctions in mammals (Fig. 4v). Notably, specific breakpoints in this distribution pattern occur right at the exon junctions (Fig. 4j, v). This suggests that in mammals, the EJC complex plays a significant role in suppressing m⁶A deposition and that overall m⁶A distribution is regulated by passive processes⁴⁰. In contrast, rice and Arabidopsis display a pronounced peak about 300 nt downstream from last exon junction sites, followed by a decrease (Fig. 4w, x). This may indicate an active process driving m⁶A deposition in plants, suggesting enhanced recruitment of m⁶A methyltransferases to the peak region. Given that most stop codons are situated within the last exon, we aligned the exons based on their distance to the stop codon for a more nuanced analysis. We observed a pronounced breakpoint immediately adjacent to the stop codons of rice and Arabidopsis, and found that the peak of m⁶A likelihood is closer to the stop codon than to the starting point of the last exon (Supplementary Fig. 5w-y). These findings suggest that the active deposition of m⁶A in plants may be influenced by specific genomic features near the stop codon (Fig. 4y). Collectively, these insights point to potentially significant differences in m⁶A modification mechanisms between plant and mammalian kingdoms, indicating that there may be novel mechanisms for plant mRNA m⁶A deposition yet to be explored.

m⁶A modification enhances mRNA stability and translation mainly through 3′ UTR in Arabidopsis seedling

m⁶A modification promotes mRNA turnover³⁷, mainly through 3′ UTR sites in mammals^58,59,60. In contrast to mammals, the effects of m⁶A are less clear in plants, while several reports demonstrated that m⁶A modification stabilizes modified mRNAs in Arabidopsis^{21,27,61,62,63,64}. It seems that the role of m⁶A modification in plants differs from that of mammals. To obtain a more accurate correlation between m⁶A level and mRNA turnover transcriptome-wide, we retrieved the public RNA lifetime data in Arabidopsis seedlings⁶⁵ and found that m⁶A-modified transcripts tend to have a longer lifetime than the unmodified transcripts⁶³ (Supplementary Fig. 6a). Plants have redundant ECT proteins that bind preferentially to m⁶A-modified mRNAs. We next studied the effects on mRNA degradation regulated by the m⁶A reader of ECT2 with the publicly available ECT2 CLIP-seq data⁶¹ and mRNA lifetime data⁶⁵ of Arabidopsis seedlings. We observed considerable overlap between the m⁶A sites and ECT2 targets (Fig. 5a), confirming m⁶A binding by ECT2. ECT2 target genes containing m⁶A sites display higher lifetime than those not bound by ECT2 (Supplementary Fig. 6b), indicating that m⁶A stabilizes mRNA and that m⁶A readers could enhance mRNA stability in Arabidopsis seedlings.

Fig. 5: Impact of m⁶A modifications on mRNA stability and translation.

a Density plot displaying the distance between the peak centers of ECT2 targets identified by CLIP-seq and m⁶A sites identified in Arabidopsis seedlings by m⁶A-SAC-seq. b Lifetime difference of transcripts with 3′ UTR-only m⁶A modification and non-3′ UTR m⁶A modification compared to transcripts without (w/o) m⁶A modifications. 3′ UTR, n = 1,813; non-3′ UTR, n = 2,645; w/o m⁶A, n = 3,319. c Cumulative curves and box plots showing the mRNA lifetime distribution for transcripts with 3′ UTR-only m⁶A modification. Transcripts were grouped into five categories (0,0.2); (0.2, 0.4); (0.4, 0.6); (0.6, 0.8); and (0.8, 1)) based on the sum of their m⁶A fractions. (0,0.2), n = 62; (0.2, 0.4), n = 228; (0.4, 0.6), n = 353; (0.6, 0.8), n = 483; (0.8, 1), n = 612. d Cumulative curves and box plots showing the mRNA lifetime distribution for transcripts with m⁶A modification outside the 3′ UTR. Transcripts were grouped into five categories (0,0.2); (0.2, 0.4); (0.4, 0.6); (0.6, 0.8); and (0.8, 1)) based on the sum of their m⁶A fractions. (0,0.2), n = 466; (0.2, 0.4), n = 981; (0.4, 0.6), n = 566; (0.6, 0.8), n = 280; (0.8, 1), n = 294. e GO enrichment analysis for genes associated with 3′ UTR-only m⁶A sites. f GO enrichment analysis of non-3′ UTR m⁶A-associated mRNAs. For e, f one-sided Fisher’s exact test. Adjusted P values using the linear step-up method. g Transcripts with 3′ UTR-only m⁶A sites exhibit strong positive correlations with translation efficiency. Transcripts were grouped into five categories (0,0.2); (0.2, 0.4); (0.4, 0.6); (0.6, 0.8); and (0.8, 1)) based on the sum of their m⁶A fractions. (0,0.2), n = 88; (0.2, 0.4), n = 293; (0.4, 0.6), n = 489; (0.6, 0.8), n = 620; (0.8, 1), n = 749. For c, d and g, the Arabidopsis seedling lifetime data set GSE86361 was used for mRNA decay analysis and Arabidopsis seedling translation efficiency data set GSE206292 was used for translation efficiency analysis. For b-d and g the p-value was determined by a one-tailed Wilcoxon rank-sum test. In box plots, the center line represents the median, and the red dot represents the mean. Upper and lower quartiles were the box limits. Source data are provided as a Source Data file.

We then asked whether the position of m⁶A modification could underlie mRNA stability differences as that in mammals^59,60. To answer this, we first clustered the m⁶A sites into 3′ UTR-only m⁶A and non- 3′ UTR m⁶A, and found that 3′ UTR-only m⁶A significantly stabilizes mRNA, and only a slight increased lifetime was observed with genes carrying non- 3′ UTR m⁶A compared to genes without (w/o) m⁶A modification (Fig. 5b). This likely suggests a more predominant role of 3′ UTR-only m⁶A in regulating mRNA stability, especially stabilizing the modified transcripts. To further explore the correlation between m⁶A fraction levels and mRNA stability, we divided mRNA carrying the 3′ UTR-only and non- 3′ UTR m⁶A sites into five groups based on m⁶A levels, and found that higher m⁶A fractions are associated with higher mRNA stability in those genes bearing the 3′ UTR-only m⁶A sites (Fig. 5c), while no significant correlations were observed in genes containing non-3′ UTR m⁶A sites (Fig. 5d and Supplementary Fig. 6c). The above results were further confirmed with the mRNA metabolic data from Sorenson, R. S et al. ⁶⁶ (Supplementary Fig. 6d, e). Although different reader proteins can recognize m⁶A in different regions to exert either stabilization or decay function, given that a majority of mRNA m⁶A modification enriches in the 3′ UTR (Fig. 2a, b), this observation may suggest an overall mRNA stabilization effect by m⁶A in plants.

The distinct effects of m⁶A position in controlling mRNA stability prompted us to investigate whether m⁶A position underlies biological function differences. The GO enrichment analysis showed that genes associated with the 3′ UTR-only m⁶A are significantly enriched in general biological pathways like gene expression, RNA processing, and ribosome biogenesis (Fig. 5e), while non-3′ UTR m⁶A-associated mRNAs were enriched in more specific pathways, such as response to hormone, response to fungus, response to salt stress, leaf development, and plant organ senescence (Fig. 5f). In addition, we also observed a positive correlation between m⁶A levels and translation efficiency in transcripts modified with 3′ UTR-only m⁶A (Fig. 5g), but no significant correlations were noticed in transcripts bearing non-3′ UTR m⁶A sites (Supplementary Fig. 6f). These observations suggest more diverse effects of m⁶A in plant, affected by its location and downstream binding proteins.

m⁶A installed by MTA in the chloroplast transcriptome in Arabidopsis seedlings

We next profiled the m⁶A sites in Arabidopsis mta mutant seedlings (Supplementary Fig. 6g, h) using m⁶A-SAC-seq and compared the m⁶A sites with those of WT (col) seedlings. A total of 14,125 MTA-dependent m⁶A sites within 2,894 RNAs were identified in WT. The methylation levels of these m⁶A sites were noticeably reduced (10,505 m⁶A sites) or completely abolished (3,621 m⁶A sites) in mta compared with WT; these m⁶A sites were hereafter defined as MTA-dependent m⁶A sites. We noticed that the MTA-dependent m⁶A sites tend to be more preferentially located in the 3’ UTR region than CDS and 5’ UTR regions (Fig. 6a). GO enrichment analysis showed that mRNAs containing these m⁶A sites are mostly enriched in stimulus response, chloroplast, photosynthesis membrane and post-embryonic development (Fig. 6b). This is consistent with previous studies in Arabidopsis that MTA is involved in response to salt stress²⁶, blue light response²¹ and embryonic development¹⁵. Notably, m⁶A levels of chloroplast (538 genes) and photosynthesis membrane (59 genes) related transcripts were significantly reduced in mta, among which twenty m⁶A sites showed dramatically reduced m⁶A levels in the chloroplast transcriptome (Fig. 6c). The methylation levels of the chloroplast encoded transcripts vary in different tissues (Fig. 6c), suggesting dynamic MTA-dependent m⁶A modifications in the chloroplast transcriptome across Arabidopsis life cycle.

Fig. 6: The MTA-mediated m⁶A modification in chloroplast transcriptome of Arabidopsis seedlings.

a, Distribution of MTA-dependent m⁶A sites along transcripts in Arabidopsis seedlings. b, GO enrichment analysis of mRNAs containing MTA-dependent m⁶A sites. One-sided Fisher’s exact test. Adjusted P values using the linear step-up method. c, Methylation level heatmap of the chloroplast-encoded genes across different tissues. The position of m⁶A site on chloroplast-encoded genes was shown. m⁶A levels and fold change were also shown. Source data are provided as a Source Data file.

Next, we observed overall reduced translation efficiency in the mta mutant compared to that of WT using previously published datasets²⁶ (Supplementary Fig. 6i). However, similar numbers of genes with upregulated translation efficiency (332 genes, fold change >2, p < 0.05) and downregulated translation efficiency (257 genes, fold change <0.5, p < 0.05) were observed in mta mutant relative to WT control. The effect of m⁶A installed by MTA on translation efficiency appear to be heterogeneous in Arabidopsis seedlings, resembling to that observed in mammals⁶⁷. Perhaps consistently, the above differentially translated genes are also enriched in distinct GO terms. Genes with upregulated translation efficiency are mainly enriched in general pathways like the ribosome and nucleolus (Supplementary Fig. 6j), while genes with downregulated translation efficiency are specifically enriched in the chloroplast and photosynthesis membrane (Supplementary Fig. 6k), such as genes of FIBRILLIN 4 (FIB4) and SMO2⁶⁸. Collectively, MTA deposits m⁶A modifications both in the nuclear and chloroplast transcriptomes, which regulate photosynthesis.

Light-induced feedback regulation of the circadian clock through m⁶A

m⁶A methylation of mRNAs regulates the circadian clock in both plants²¹ and mammals⁶⁹. To further probe the light effect on rice m⁶A methylation, we germinated rice seeds under dark (24 h of dark per 24 h) and light conditions (16 h of light per 24 h), respectively. Plumules under dark and light conditions at 3 days after germination were studied, respectively, by using m⁶A-SAC-seq (Fig. 7a). Our data revealed a pervasive increase of m⁶A methylation levels (23,253 hypermethylated and 2,607 hypomethylated m⁶A sites) under light compared to dark conditions (Supplementary Fig. 7a-c). The light-induced m⁶A sites are largely outside of 3′ UTR (Fig. 7b). GO enrichment analysis revealed that genes containing hypermethylated m⁶A sites are highly clustered in stimulus response, including light and hormone stimulus (Supplementary Fig. 7d). Noticeably, light significantly increased the m⁶A modification levels of photoreceptor transcripts, for example, PHYA (Chr3: 29172686; CDS; AGATA), PHYB (Chr3: 11021272; CDS; AGATA), PHYC, (Chr3:31007707;CDS; GGACA), CRY1a (Chr2: 21976854;5’ UTR; AGAGC), and CRY2 (Chr2:24921916; 3’ UTR; AAACT). Thus, m⁶A methylation levels of circadian clock genes within their transcripts were notably updated by light in rice, as the case observed in mammals⁷⁰. Moreover, in the aforementioned mta mutant of Arabidopsis, a decreased m⁶A fraction of CRY1 in the 3′ UTR (Chr4:5727183; AAACA; 3′ UTR) was observed, potentially resulting in reduced translation efficiency, suggesting that MTA regulates CRY1 translation through m⁶A deposition (Fig. 7c, d). Given that light-induced phase separation of CRYs modulates MTA activity in plants⁶⁹, and MTA controls m⁶A modification on CRY transcripts to regulate CRY translation, there appears to be a feedback loop of epitranscriptome-translation regulation of the circadian clock in plants.

Fig. 7: Light responsive regulation through m⁶A in rice and Arabidopsis.

a Light increased the m⁶A fractions in the plumule of rice. Data are means ± SD, n = 2. The p-value was determined by a one-tailed Wilcoxon rank-sum test. b Relative distribution of m⁶A sites across 5′ UTR, CDS and 3′ UTR regions. c Reduced m⁶A fraction in CRY1 was observed in mta mutant. Data are means, n = 2. d Decreased translation efficiency of CRY1 in the mta mutant. Translation efficiency data set GSE206292 was used for the analysis. Data are means, n = 2. Student t-test was used to determine the statistic difference. Source data are provided as a Source Data file.

Discussion

m⁶A methylation of mRNA plays critical roles in both plant and mammal development as well as signaling and stimulation responses^{15,20,38,71,72}. Previous RNA m⁶A studies in plants lack base-resolution, precision and modification stoichiometry information^5,23,48,63. Benefit from the development of m⁶A sequencing at single-base resolution in mammalian transcriptome using the m⁶A-SAC-seq method for the first time^36,37, we report here comprehensive maps of m⁶A at single-base precision with stoichiometry information in eight rice tissues and nine Arabidopsis tissues spanning their life cycle. We uncovered high-confidence, single-base resolution m⁶A sites across rice and Arabidopsis tissues, providing in-depth resources for future investigations of m⁶A functions in rice and Arabidopsis.

Evolutionarily conserved m⁶A modification sites in orthologous gene pairs of Arabidopsis and rice were identified. Orthologous genes containing conserved m⁶A sites are significant for tissue development, photosynthesis and stimulus response, and might be selected under evolutionary pressure. Despite these conserved m⁶A sites in orthologous genes, we observed different m⁶A distribution patterns in 3′ UTR between rice and Arabidopsis. Furthermore, although a positive correlation between the total level of m⁶A and the internal exon length was observed, different from that in mammals, no such correlation was seen regarding to the last exon in plants. While the mRNA m⁶A distribution in mammals appears to be shaped by repressive pathways mediated through the exon junction complexes or other protein factors^55,56,57, plants, by contrast, also rely on active installation to control m⁶A deposition. In particular, our results suggest an active m⁶A deposition process occurring near the stop codon in plant mRNAs. This suppression-activation dual deposition regulation potentially depicts m⁶A distribution patterns across species. We and others have shown that m⁶A suppression is mediated by EJC in mammals^55,56,57, the suppression-activation model predicts RNA-binding proteins that may recruit m⁶A writers and direct m⁶A deposition near the stop codon in plant mRNAs. These observations suggest an as-yet-unknown mechanism that regulates m⁶A distribution in plant transcriptomes, which requires future in-depth investigations.

A majority of plant m⁶A modification resides in 3′ UTR from our analyses. Arising evidence in plants suggested that m⁶A modification stabilizes mRNA²¹ although mammalian m⁶A modification tends to destabilize the modified mRNA³⁷. Based on our single-base resolution results in Arabidopsis seedlings, we observed a positive correlation between m⁶A methylation level and mRNA stability for 3’ UTR m⁶A sites, confirming an overall mRNA stabilization effect by m⁶A in certain plant tissues. Consistently, the m⁶A sites bound by its reader protein ECT2 exhibit significantly elevated half lifetime as compared with transcripts not bound by ECT2, indicating the presence of m⁶A reader protein to enhance mRNA stability in plants^61,62. In addition, m⁶A methylation installed by MTA can either promote or reduce translation efficiency in a pathway-dependent manner, resembling observations made in mammals⁶⁰. Therefore, m⁶A modification stabilizes modified transcripts, with reader proteins stabilizing the bound mRNA in Arabidopsis seedlings. Translation effects can be complex and context-dependent but an overall translation promotion effect in Arabidopsis seedling was suggested from our data.

In conclusion, these base-resolution and quantitative m⁶A modification maps across rice and Arabidopsis’ life cycle have filled a pronounced gap in plant research. The comparative analysis of single-base m⁶A maps between humans and plants reveals a suppression-activation dual regulation model in shaping m⁶A distribution patterns in different species.

Methods

Plant material

Col-0 accession of Arabidopsis thaliana and japonica rice (Oryza sativa) cultivar Nipponbare were used in this study. Arabidopsis, plants were grown at 22 °C with 16 h of light per 24 h. Arabidopsis seedlings were harvested after growing on 1/2 Murashige and Skoog medium (MS) plates for 7 d. Arabidopsis shoots and roots were harvested after growing on 1/2 Murashige and Skoog medium (MS) plates for 14 d. Arabidopsis Rosetta leaves were harvested after growing in soil for 30 d. Arabidopsis cauline leaf, flower, stem and silique were harvested after flowering. Arabidopsis seeds were collected after the seeds is totally dry. Seeds of Arabidopsis mta mutant (ABI3::MTA/mta)^20,21 were sowed in the MS plates, and the mta seedlings were harvested at 8-day-old at 22 °C with 16 h of light per 24 h. For rice, plants were grown at 28 °C with 14 h of light per 24 h and 8-day-old, 2-week-old seedlings were harvested. The heading panicles, flag leaf at 10 days after anthesis, endosperms and embryos at 10 days after anthesis were harvested. The plumules under dark and light conditions at 3 days after germination were sampled. The tissue was flash-frozen in liquid nitrogen, ground using a mortar and pestle and stored at −80 °C. Total RNA were extracted using TRIzol™ Reagent (Catalog number: 15596026) according to the manufacturer′s instructions. All the plants were planted in the greenhouse of The Chinese University of Hong Kong.

mRNA capture from the extracted total RNA

A total of 50 μg total RNA for each of the two biological replicates was used for mRNA capture (Dynabeads mRNA DIRECT Purification Kit (Invitrogen)) following the manufacturer′s instructions with modification. Briefly, 50 ug total RNA was diluted with H₂O in 100 ul volume and then denatured under 65 °C for 2 min, and immediately put on the ice for exactly 2 min. Next, a total of 100 ul Dynabeads were washed twice with 200 ul lysis/binding buffer provided in the above kit. The washed beads were then eluted in 100 ul lysis/binding buffer, which was further mixed with the denatured total RNA. Then, the sample was put on the rotor to bind for 15 min at room temperature. After binding, wash buffer B was used to wash the beads for twice. 30 ul H₂O was used to elute the beads and immediately put on the 75 °C for 2 min. The eluted mRNA was collected after magnetic separation. The mRNA capture process was repeated as described above to obtain the more purified mRNA.

Quantification of m⁶A in RNA by LC–MS/MS

50 ng mRNAs were digested into nucleosides, and the amount of m⁶A was measured by using Agilent 6460 Triple Quad MS–MS with a 1290 UHPLC supplied with a ZORBAX Eclipse XDB-C18 column (UHPLC–QQQ–MS/MS) and calculated based on the standard curve generated by pure standards. For each sample, RNA was digested by using nuclease P1 (NEB) at 37 °C for 2 h. Then, 1 μl of Shrimp Alkaline Phosphatase (rSAP) and 3 μl of 10× rCutsmart buffer (NEB) were added, and the reaction was incubated at 37 °C for 2 h. Samples were then filtered using a 0.22-μm filter (Millipore) and injected into LC–MS/MS. The nucleosides were quantified by using the nucleoside-to-base ion mass transitions of 282 to 150 (m⁶A), and 268 to 136 (A). Quantification was performed in comparison to the standard curve obtained from pure nucleoside standards run on the same batch of samples. The ratio of m⁶A to A was calculated based on the calibrated concentrations.

m⁶A-SAC-seq library construction

50 ng mRNAs of each replicate were used for the library construction. All these libraries were constructed exactly following the previously published protocols^36,37. The constructed libraries were sequenced on the Illumina HiSeq sequencing platform in pair-end mode with 150 bp per read.

m⁶A-SAC-seq data processing

After sequencing, the m⁶A sites were detected using the method of (https://github.com/y9c/m6A-SACseq)^36,37. The analysis utilized the reference genome downloaded from the Ensemble database, with assembly versions TAIR10 and IRGSP-1.0 employed for Arabidopsis and rice respectively.

RNA lifetime profiling and translation data analysis

RNA lifetime and translation efficiency data of Arabidopsis Col seedling was downloaded from data set GSE206292 and GSE118462. The translation efficiency data of mta mutant in Arabidopsis was also retrieved from GSE206292.

Conserved m⁶A sites in ortholog genes between rice and Arabidopsis

The one-to-one ortholog genes between rice and Arabidopsis were first identified. The ortholog genes were then pairwise aligned to obtain the sites on the homolog positions which showed consistent flanking sequence (±1 nt) centered on A sites. The above A sites were defined as the conserved A sites. While the m⁶A modification on conserved A sites for both species are defined as conserved m⁶A sites.

Gene Ontology (GO) analysis

Functional GO enrichment analysis was performed by web-based toolkit for the agricultural community agriGO v2.038 (http://systemsbiology.cau.edu.cn/agriGOv2/). GO terms with a false discovery rate (FDR) < 0.05 were considered significantly enriched.

Statistics and reproducibility

All experiments were repeated independently at least twice and showed similar results. GraphPad Prism v.9 and R studio were deployed for the figure plotting.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Source data

Source Data