Exon architecture controls mRNA m⁶A suppression and gene expression

Abstract

N⁶–methyladenosine (m⁶A) is the most abundant mRNA modification and plays crucial roles in diverse physiological processes. Utilizing a Massively Parallel Assay for m⁶A (MPm⁶A), we discover that m⁶A specificity is globally regulated by “suppressors” that prevent m⁶A deposition in unmethylated transcriptome regions. We identify Exon Junction Complexes (EJCs) as m⁶A suppressors that protect exon junction-proximal RNA within coding sequences from methylation and regulate mRNA stability through m⁶A suppression. EJC suppression of m⁶A underlies multiple global characteristics of mRNA m⁶A specificity, with the local range of EJC protection sufficient to suppress m⁶A deposition in average-length internal exons, but not in long internal and terminal exons. EJC-suppressed methylation sites co-localize with EJC-suppressed splice sites, suggesting that exon architecture broadly determines local mRNA accessibility to regulatory complexes.

N⁶–methyladenosine (m⁶A), the most prevalent mRNA modification in mammals, influences wide-ranging aspects of gene expression in diverse physiological and pathophysiological processes (1–3). The METTL3-METTL14 methyltransferase complex installs m⁶A methylation on mRNA in a common DRACH (D = A, G, or U; R= A or G; H= A, C, or U) sequence motif, but only a fraction of DRACH sequences (~5%) in a subset of cellular transcripts are selected for methylation (4). Additionally, m⁶A exhibits a marked regional bias in its transcriptomic distribution, being strongly enriched in unusually long internal exons and near stop codons (5, 6). Despite the central importance of specific m⁶A deposition in m⁶A-mediated gene regulation, the mechanistic basis for m⁶A specificity has remained poorly understood.

In this study, we discover the existence of prevalent regulatory mechanisms that restrict m⁶A methylation to specific transcript regions through targeted suppression of m⁶A in unmethylated regions. We find that pre-mRNA splicing selectively suppresses m⁶A deposition in average-length internal exons, but not in longer exons. We identify Exon Junction Complexes (EJC) as major m⁶A suppressors that mediate this effect and control several key characteristics of global m⁶A specificity. EJC depletion results in pervasive aberrant methylation of mRNAs and m⁶A-mediated transcript destabilization. EJCs, together with interacting proteins, package and protect long stretches of proximal RNA from cellular methylation deposition, which may represent a general mechanism by which exon architecture and EJC positioning determine local mRNA accessibility to regulatory machineries.

Massively Parallel Assay for m⁶A

The extent to which global m⁶A specificity is controlled has important implications for m⁶A regulation but is poorly understood (4). We approached this problem by asking: is the local sequence surrounding an m⁶A methylated site, when uncoupled from its endogenous context, sufficient to specify methylation at that site? Conversely, is the local sequence surrounding an unmethylated DRACH site, when uncoupled from its endogenous context, sufficient to prevent methylation at that site?

To assess this systematically on an epitranscriptome-wide scale, we developed a Massively Parallel Reporter Assay (MPRA) that enables high-throughput assessment of the m⁶A methylation status of thousands of designed sequences, which we termed Massively Parallel assay for m⁶A (MPm⁶A) (Fig. 1A and fig. S1A). In the MPm⁶A workflow, thousands of endogenously methylated m⁶A sites and endogenously unmethylated DRACH sites, with 102 nucleotides of sequence surrounding each site, were synthesized and cloned into the 3′UTR of a plasmid-based intronless GFP transgene. For each sequence, we also designed a corresponding negative control sequence in which all DRACH motifs were mutated to prevent methylation. The sequences were expressed and then m⁶A methylated through transfection into cells, or, when specified, through in vitro transcription and in vitro m⁶A methylation. The methylation status of each individual sequence was assessed by its enrichment following m⁶A-immunoprecipitation (IP) of mRNA. We selected 6,897 HeLa m⁶A sites and 3,058 unmethylated DRACH sites to assay in HeLa cells and validated the assay’s precision and accuracy (fig. S1, B to D).

Fig. 1. MPm⁶A reveals suppression of thousands of m⁶A sites in unmethylated transcriptome regions.

(A) Schematic of the MPm⁶A workflow. (B) MPm⁶A enrichment scores (experimental IP/input – negative control IP/input) for endogenously methylated (n = 6,095) and unmethylated (n = 2,716) sequences, mean ± SD, four biological replicates. (C) Metagenes of endogenously methylated and unmethylated sequences that are significantly methylated in MPm⁶A. (D) Exon lengths of endogenously methylated and unmethylated sequences that are significantly methylated in MPm⁶A. Median, and IQR, Wilcoxon rank sum test. Sample size for each violin plot from left to right is: n = 175, n = 22, n = 696, n = 519, n = 3,539, and n = 1,328. (E) Distribution of internal exon lengths in the human genome. Black lines indicate 10^th percentile (left, 60 nt) and 90^th percentile (right, 246 nt). Blue and red lines indicate median internal exon length for MPm⁶A endogenously methylated (915 nt) and unmethylated (167 nt) sequences, respectively.

Widespread mRNA m⁶A suppression controls m⁶A epitranscriptome specificity

When we compared the methylation levels of the endogenously methylated sequences to their negative control sequences, we found that 92.8% of the sequences exhibited significant methylation in this reporter assay (Fig. 1B and fig. S1E), indicating that most endogenously methylated sites do not strictly require their larger surrounding native context for methylation. Unexpectedly, 90.2% of endogenously unmethylated sequences also exhibited significant methylation (Fig. 1B and fig. S1E). The MPm⁶A enrichment scores of the endogenously unmethylated group were similar to the endogenously methylated group, despite their diverging endogenous methylation states (Fig. 1B). We observed similar results when the sequences were in vitro transcribed and methylated with recombinant METTL3-METTL14 (fig. S2). Thus, thousands of endogenously unmethylated DRACH sites became methylated when they were uncoupled from their endogenous contexts and expressed in an artificial reporter context. We term these identified sites “suppressed m⁶A sites”. We validated these results for three selected sequences (fig. S1F), and confirmed that methylation was not notably influenced by the CMV promoter of the MPm⁶A plasmid (fig. S3). We observed similar results when sequences were expressed within CDS or 5′UTR, though m⁶A enrichment was significantly lower for many sequences when placed in the CDS or 5′UTR versus in the 3′UTR (fig. S4, A to D). This suggests that 5′ regions are generally less conducive for m⁶A methylation than 3′ regions (fig. S4E). Collectively, these results reveal the existence of thousands of suppressed m⁶A sites that are silenced by unknown mechanisms.

We noted that suppressed m⁶A sites were enriched in the CDS and 3′UTR and were depleted near the stop codon, which is the reverse of endogenous m⁶A site enrichment (Fig. 1C and fig. S5A). Further, suppressed m⁶A sites in internal exons reside within much shorter exons (median = 167 nt) than endogenous m⁶A sites (median = 915 nt) (Fig. 1D). These observations suggest that endogenous m⁶A enrichment in long internal exons may be a consequence of suppression of m⁶A deposition in shorter internal exons, which comprise most exons (90% of internal exons are < 246 nt) (Fig. 1E). 942 genes containing suppressed m⁶A sites did not contain any endogenous m⁶A peaks on their transcripts (fig. S5B). Suppression of these sites appears to involve suppression of m⁶A deposition rather than active demethylation, as binding sites for RBM15, a METTL3-METTL14 methyltransferase complex accessory subunit, were highly enriched near endogenous m⁶A sites compared to suppressed m⁶A sites (fig. S6A), while FTO and ALKBH5 binding sites were not significantly enriched near suppressed sites and exhibited little binding near suppressed sites overall (fig. S6B). These results were unexpected as previous reports on m⁶A specificity had mainly focused on activating mechanisms (7–12). Our MPm⁶A assay suggests the existence of unknown m⁶A “suppressors” that govern global m⁶A specificity by suppressing m⁶A deposition.

Pre-mRNA splicing suppresses m⁶A methylation proximal to splice sites

We next examined the relative enrichment of binding sites for 120 RBPs near endogenously methylated versus suppressed m⁶A sites to identify candidate suppressors (13). Several spliceosome components (BUD13, SF3B4, EFTUD2) were significantly enriched near suppressed sites, suggesting that splicing may suppress m⁶A (fig. S6A). Because suppressed m⁶A sites in CDS primarily reside within average-length internal exons, we hypothesized that the splicing of average-length internal exons may suppress m⁶A methylation. To test this, we cloned a suppressed m⁶A site from an average-length internal exon in the CRY1 gene (fig. S7A) into a rabbit beta-globin minigene reporter (BG), as well as a version with the introns removed (BG Δi1,i2). First, we cloned the suppressed CRY1 site and 50/51 nt of flanking sequence into the internal exon, or in the last exon of these constructs. Notably, the spliced construct strongly suppressed methylation of the sequence placed within the internal exon, but not within the last exon (fig. S7B). Removal of either intron (BG Δi1 CRY1 102, BG Δi2 CRY1 102) resulted in partial loss of suppression, indicating that splicing of both introns contributed to methylation suppression (Fig. 2A). Consistent with this notion, deletion of all splice sites also resulted in a decrease in m⁶A suppression (fig. S7C). Cloning in 912 nt of the CRY1 exonic sequence surrounding the suppressed site into the internal exon (BG CRY1 912), forming a long internal exon, resulted in a loss of suppression (Fig. 2A). We hypothesized that the suppression is dependent on the proximity of the m⁶A site, located within the center of the exon, to splice sites. Expanding the length of the BG CRY1 102 internal exon by cloning in larger amounts of CRY1 flanking exonic sequence resulted in a progressive loss of suppression, with a > 476 nt internal exon unable to suppress m⁶A (Fig. 2B and fig. S7D). These results reveal a causal role for pre-mRNA splicing in m⁶A regulation.

Fig. 2. Pre-mRNA splicing suppresses m⁶A methylation in average-length exons.

(A and B) Left: schematic of specified BG CRY1 constructs. Blue regions indicate sequences derived from the CRY1 endogenous sequence, gray regions indicate sequences derived from rabbit beta-globin (BG). Number following CRY1 refers to the number of nucleotides of exonic sequence surrounding the CRY1 suppressed m⁶A site in the CRY1 endogenous mRNA that was cloned into the BG construct. Grey dot in the blue region denotes the suppressed m⁶A site; the number at the left and right of the m⁶A site shows the distance (nt) between the m⁶A site and the 3′ and 5′ splice site, respectively; the number next to the TSS shows the distance (nt) between the m⁶A site and the promoter. Δ denotes deletion of the specified intron(s). Details of each construct are described in the supplementary method. Right: m⁶A enrichment at a CRY1 suppressed m⁶A site. Primers amplifying a 62 nt-fragment containing the CRY1 suppressed m⁶A site. m⁶A enrichment was calculated as IP/input normalized to m⁶A-marked Gaussia luciferase RNA spike-in IP/input. Mean ± SEM, two-tailed t-test, *P < 0.05; **P < 0.01, ***P < 0.001. Three biological replicates.

Exon junction complexes control m⁶A epitranscriptome specificity

We next sought to understand the mechanism by which splicing suppresses m⁶A deposition. Exon junction complexes (EJCs) are deposited by spliceosomes onto mRNA ~24 nt upstream of exon-exon junctions and plays multifaceted roles in gene expression regulation (14, 15). Notably, two recent studies reported that EJCs efficiently block splicing at proximal aberrant splice sites (16, 17). Additionally, EJCs, together with interacting serine and arginine-rich (SR) proteins, package and compact mRNA and can protect long stretches of proximal RNA from nuclease accessibility in vitro, and also block 5′ to 3′ exonuclease degradation in vivo (18, 19). We reasoned that suppressed m⁶A sites within average-length internal exons are within relatively close proximity to both an upstream and downstream EJC. Conversely, m⁶A sites within long internal exons and near stop codons (which generally reside in long last exons) can be hundreds of nucleotides away from the nearest EJC. We therefore hypothesized that EJCs could mediate the splice site-proximal suppression of m⁶A we observed.

We knocked down (KD) the core EJC factor EIF4A3 in HeLa cells and assessed the effect on m⁶A deposition transcriptome-wide using m⁶A-MeRIP-seq. Notably, 24,350 regions were significantly hypermethylated upon EIF4A3 KD, while 3,140 regions were hypomethylated (Fig. 3A). 39% of these hypermethylated regions exhibited a greater than 8-fold increase in m⁶A enrichment compared to the non-targeting siRNA control. We knocked down RBM8A, another core EJC factor (20), and observed similar, though relatively milder, transcriptome-wide m⁶A changes, with 14,034 significantly hypermethylated regions observed, of which 57% overlapped with hypermethylated regions observed in EIF4A3 KD (Fig. 3A and fig. S8, A and B). The relatively milder m⁶A changes upon RBM8A KD may result from relatively lower KD efficiency (table S1) or may indicate a stronger requirement of EIF4A3 for suppression. Concordant with these transcriptome-wide m⁶A changes, using UHPLC-QQQ-MS/MS, we found that EIF4A3 KD increased global levels of m⁶A in polyadenylated RNA by two-fold, while RBM8A KD resulted in a ~25% increase (fig. S8C).

Fig. 3. EJCs protect exon junction-proximal RNA in average-length exons within CDS regions from m⁶A methylation.

(A) Differentially methylated regions upon EIF4A3 KD (left) and RBM8A KD (right) in HeLa cells (FDR<.1, |log₂FC|>1). Three biological replicates. Gray and red dots indicate differentially methylated regions that overlap and do not overlap m⁶A peaks in the control cells, respectively. (B) Input and m⁶A-IP read coverage at FUS and SRSF6 in EIF4A3 KD, RBM8A KD, and control HeLa cells. (C) Numbers of EIF4A3 KD hypermethylated regions (left) and m⁶A peaks in control cells (right) that reside within first, internal or last exons (D) Exon lengths for m⁶A peaks residing within internal exons in control KD, EIF4A3 KD and RBM8A KD cells, and exon lengths of hypermethylated regions residing within internal exons in EIF4A3 and RBM8A KD cells. Dot and bar represent median and interquartile range, Wilcoxon rank sum test of indicated group vs. siC all m⁶A peaks. Sample size for each violin plot from left to right is: n = 3166, n = 6659, n = 8438, n = 3817, and n = 3827. (E) Metagenes of m⁶A peaks and significantly hypermethylated m⁶A regions (and top quartile) in EIF4A3 KD HeLa cells, and m⁶A peaks in control cells.

94% of hypermethylated regions from EIF4A3 KD and 82% of hypermethylated regions from RBM8A KD did not overlap with m⁶A peaks identified under the non-targeting siRNA control conditions, suggesting that these regions contain newly methylated suppressed m⁶A sites (Fig. 3, A and B, and fig. S8D). Indeed, out of 1,024 CDS sequences identified by MPm⁶A to contain suppressed m⁶A sites, 46% become methylated upon EIF4A3 and/or RBM8A KD, including the CRY1 suppressed site (fig. S8E), with three selected suppressed sites validated (fig. S8, F and G) (21). Furthermore, EIF4A3 KD substantially alleviated the previously observed m⁶A suppression within the internal exon of BG CRY1 102 (fig. S8H).

Consistent with our model, newly methylated and hypermethylated regions were highly enriched in average-length internal exons within CDSs (Fig. 3, C to E, and fig. S8, I and J), with transcriptome-wide increases in m⁶A enrichment in exon junction-proximal regions observed (fig.S9, A and B) upon EIF4A3 or RBM8A KD. EIF4A3 KD disrupted m⁶A epitranscriptome specificity globally, resulting in substantial loss of enrichment of m⁶A peaks in long internal exons and increased density of m⁶A in the CDS relative to the stop codon (Fig. 3, D and E). It was previously reported that the peak of m⁶A density near stop codons on metagene plots can be more precisely visualized as an increased enrichment 150 nt past the start of last exons (6). EIF4A3 KD resulted in a global increase in m⁶A enrichment < 150 nt past the start of last exons (fig. S9, A to C), indicating that EJC suppression of methylation proximal to last exon-exon junctions is responsible for the characteristic m⁶A peak density near stop codons. While most transcripts exhibited hypermethylation and contained one or more endogenous m⁶A peaks upon EIF4A3 KD, over a thousand transcripts that ordinarily lack endogenous m⁶A peaks also gained aberrant m⁶A methylation upon EIF4A3 KD, revealing a major role for EJCs in suppressing m⁶A deposition on the subset of transcripts that ordinarily are not subject to m⁶A regulation (fig. S9, D to F).

The widespread suppression of m⁶A by the EJCs also implies that many m⁶A are deposited following splicing, which we confirmed using pulse-chase metabolic labeling experiments and UHPLC-QQQ-MS/MS (supplementary text and fig. S10). Two genes used in gene therapies for mucopolysaccharidosis type II and spinal muscular atrophy, IDS and SMN, contain EJC-suppressed m⁶A sites in their mRNAs, respectively. As expected, when these mRNAs were expressed from cDNA constructs, and thus not bound by EJCs, they were significantly hypermethylated relative to the corresponding endogenous mRNAs (fig. S11). Further, lncRNAs that contain three or more exons globally exhibit EJC suppression of m⁶A in internal exons, while those with two or less do not (supplementary text and fig. S12). We depleted EIF4A3 with a different siRNA in HeLa cells, and knocked down EIF4A3 in HEK293T cells as well as in a glioblastoma cancer cell line (U87) that is sensitive to EIF4A3 perturbation (22), and observed similar transcriptome-wide m⁶A changes in each case (figs. S13 to S15). Altogether, our results indicate that spliceosomes widely suppress m⁶A methylation via deposition of EJCs that protect proximal RNA from methylation.

EJCs regulate mRNA stability by suppressing m⁶A methylation

m⁶A is known to mainly accelerate mRNA degradation via the reader protein YTHDF2 (23, 24). Accordingly, we observed globally reduced mRNA half-life of hypermethylated transcripts (~90%) upon EIF4A3 KD (Fig. 4, A and B). Consistently, we observed generally increased YTHDF2 binding on hypermethylated mRNAs, accompanied with the decreased mRNA half-life (Fig. 4C). YTHDF2 KD could rescue accelerated degradation of YTHDF2 target transcripts upon EIF4A3 KD (fig. S16). Further, the density of EJC-loading on transcripts (estimated by the number of exons within CDS regions per 1 kb) correlated with transcriptome-wide mRNA stability (fig. S17). Higher EJC density on transcripts tended to correlate with reduced m⁶A methylation and higher mRNA stability, and the strength of this correlation was diminished by Mettl3 KO (fig. S17, A and B).

Fig. 4. mRNA m⁶A hypermethylation upon EJC depletion destabilizes mRNAs.

(A) Correlation between fold changes in mRNA half-life and m⁶A level upon EIF4A3 KD in HeLa cells (n = 3840). (B) Boxplots showing half-life fold changes of hypermethylated mRNAs upon EIF4A3 KD in HeLa cells. mRNAs were categorized into three groups according to their methylation changes upon EIF4A3 KD in HeLa cells. P values from Wilcoxon rank sum test. Sample size for each boxplot plot from left to right is: n = 1887, n = 1201, and n = 752. (C) Left: heatmap showing fold changes in m⁶A level, mRNA half-life, and YTHDF2 binding upon EIF4A3 KD in HeLa cells. Right: scatter plots showing the correlation among fold changes in m⁶A level, mRNA half-life, and YTHDF2 binding upon EIF4A3 KD in HeLa cells. The hypermethylated mRNAs (m⁶A log₂FC > 0; n = 3424) were categorized into 100 bins based on ranked fold change of m⁶A level upon EIF4A3 KD. For (A) and (C), PCC and P values are shown.

We also found that METTL3 depletion could generally reduce the expression level changes of hypermethylated genes upon EJC depletion in HeLa cells (supplementary text and fig. S18), indicating that these EJC-dependent gene expression changes are at least in part mediated by m⁶A methylation.

While the vast majority of hypermethylated transcripts were destabilized by EIF4A3 KD, a small subset of hypermethylated transcripts were stabilized (Fig. 4A). One example is p53, which mediates neurodevelopmental defects in mouse models of EJC haploinsufficiency (25). The TP53 transcript was hypermethylated but also up-regulated upon EIF4A3 KD. Mechanistically, we observed increased binding to TP53 mRNA by IGF2BP proteins, which are known to stabilize methylated transcripts (supplementary text and fig. S19). In summary, while the predominant effect of EJC-mediated m⁶A suppression is to stabilize mRNAs by preventing the YTHDF2-mediated decay, in a minority of instances hypermethylated transcripts can be stabilized by other mechanisms, such as binding by IGF2BPs (26).

Consistent with a general role for m⁶A in promoting translation (12, 27), EIF4A3 KD led to slightly increased translation efficiency of hypermethylated transcripts, with more highly hypermethylated transcripts exhibiting greater increases in translation efficiency (fig. S20), although the impact was modest relative to the effects observed on mRNA stability.

Differential m⁶A methylation across tissues and species through EJC suppression

Our model suggests that the cellular EJC levels may impact global mRNA m⁶A deposition in different tissues. Indeed, we observed a negative correlation between EIF4A3 expression level and global mRNA m⁶A modification level in 25 different human tissues with available transcriptome-wide m⁶A profiles (fig. S21A) (28). We examined the top 10% of genes with the strongest correlations and found that the majority (> 70%) exhibited a negative correlation between m⁶A and EIF4A3 levels in different tissues. Further, m⁶A levels of these genes also negatively correlated with their transcript abundances (fig. S21B). Similar trends were also observed in mouse tissues (fig. S21C). These results further support m⁶A suppression by EJCs and subsequently mRNA stability regulation in mammalian tissues.

Notably, we observed the lowest EIF4A3 expressions in brain tissues, which exhibited the highest overall mRNA m⁶A levels (fig. S21A). We further compared the methylome of the human cerebellum (lowest EIF4A3 level and highest overall mRNA m⁶A) with that of the heart (higher EIF4A3 level and lower overall mRNA m⁶A). Regions that are hypermethylated in the cerebellum (compared to heart) reside within short internal exons (fig. S21D), suggesting reduced m⁶A suppression due to low EIF4A3 expression in cerebellum. This association between high m⁶A level and low EIF4A3 expression in cerebellum was attenuated upon depletion of METTL3 (fig. S21C). These observations further indicate the widespread suppression by EJCs contributes to tissue-specific m⁶A deposition. We also found that a subset of EJC-suppressed m⁶A sites physiologically escape suppression in certain tissues via methylation of alternative transcript isoforms. These isoforms contain longer exons and thus altered EJC positioning; methylation of these isoforms generates tissue-specific m⁶A patterns (supplementary text and fig. S22).

Lastly, the effect of exon-intron architecture on mRNA stability may have co-evolved with YTHDF2 in vertebrates. The strong correlation between EJC loading, represented by the number of exons, and mRNA level across tissues is maintained across humans, mice, and zebrafish, but not fly and worm, which lack YTHDF2 orthologs (supplementary text and fig. S23).

EJCs and peripheral EJC factor RNPS1 protect exon junction-proximal RNA regions from aberrant mRNA processing

We did not observe interactions between the methyltransferase complex and EJC complexes (fig. S24), suggesting that steric hindrance from EJCs, rather than a specific inhibitory interaction, accounts for methylation suppression. Nuclear EJCs bound with the peripheral EJC factor RNPS1 multimerize and associate with wide variety of SR and SR-like proteins to package and compact mRNA into higher-order, megadalton-scale mRNPs that ensheathe proximal RNA well beyond the canonical EJC deposition sites (18, 29, 30). Tens to hundreds of nucleotides of proximal RNA could be protected by this mega-complex from nuclease digestion due to this packaging (18, 31). To examine whether the mRNA packaging function of the EJC-mediates suppression of proximal methylation, we isolated EJCs/EJC-bound RNA from cellular extracts, digested away physically accessible RNA with in vitro nuclease treatment, and then measured m⁶A levels on the EJC-protected RNA footprints (fig. S25, A and B). EJC-protected footprints were strongly depleted of m⁶A, indicating that these inaccessible RNA regions are largely protected from m⁶A deposition within cells (fig. S25C). EJCs also protected these footprints from in vitro methylation by recombinant METTL3-METTL14 (fig. S25D). This was not due to general inhibition of methyltransferase activity or lack of methylatable sites on the EJC footprints, as free, unmethylated RNA spiked into the methylation reaction as well as deproteinized footprints were both robustly methylated (fig. S25, D and E). Therefore, EJCs suppress local m⁶A deposition by packaging proximal RNA.

We next asked whether the peripheral EJC factor RNPS1, which associates with high molecular weight EJCs in these highly packaged mRNP structures (29), plays a role. RNPS1 knockdown led to substantial transcript m⁶A hypermethylation within average-length internal exons and CDS regions (Fig. 5, A to C, and fig. S26, A to C). We detected fewer hypermethylated regions overall compared to depletion of the core EJC factors; however, we did observe high overlap (45%) between siRNPS1 hypermethylated regions and siEIF4A3/siRBM8A hypermethylated regions (Fig. 5C and fig. S26C). In contrast, depletion of UPF1, a central NMD factor that interacts with the EJC in the cytoplasm, did not result in m⁶A changes similar to those of the core EJC (fig. S27).

Fig. 5. EJCs and RNPS1 protect proximal RNA regions from aberrant mRNA processing.

(A) Differentially methylated regions upon RNPS1 KD in HeLa cells (FDR < 0.1, |log₂FC| > 1), three biological replicates. Gray and red dots indicate differentially methylated regions that overlap and do not overlap with m⁶A peaks in the control KD cells, respectively. (B) Metagenes of significantly m⁶A hypermethylated regions (and top quartile) upon RNPS1 KD in HeLa cells in comparison with that of all m⁶A peaks in control cells. (C) Input and m⁶A-IP read coverage at CDH24 and NRAS upon RNPS1 KD and EIF4A3 KD, respectively, as well as corresponding controls in HeLa cells. (D) Enrichment of suppressed m⁶A sites (identified from MPm⁶A) at EIF4A3-suppressed splice sites (left) and enrichment of EIF4A3 KD hypermethylated regions at EIF4A3-suppressed splice sites (right). Fisher’s exact test, dot and bar represent odds ratio and 95% confidence interval. (E) Input and m⁶A-IP read coverage at SENP3 and HNRNPH1 in EIF4A3 KD and control HeLa cells. (F) Schematic model depicting that EJCs and RNPS1 (and potentially other EJC-associated proteins) protect exon junction-proximal RNA from m⁶A deposition through local mRNA packaging. For (C) and (E), red bracket indicates EIF4A3-suppressed splice variant, with ends of bracket indicating the suppressed splice junctions.

The ability of EJCs to protect proximal RNA regions from methylation resembles the recently characterized EJC- and RNPS1-mediated suppression of proximal aberrant splice sites and recursive splicing (16, 17). Transcriptome-wide, EJC-suppressed splice sites significantly colocalize with EJC-suppressed m⁶A sites (supplementary text; Fig. 4, C to E; fig. S26, D to F; and table S2). Altogether, these results suggest that RNPS1-associated EJCs suppress both local cellular m⁶A methylation and splicing through packaging of proximal RNA and point to exon architecture as an important determinant of local RNA accessibility to regulatory machineries. Additionally, beyond components of the m⁶A methyltransferase complex, a number of other RBPs also exhibit preferential binding at long internal exons, suggesting that EJCs may regulate mRNA accessibility to a broader range of mRNA regulators in addition to the splicing and m⁶A methylation machineries through their mRNA packaging function (supplementary text and fig. S28).

Discussion

Previously identified m⁶A effector proteins fall broadly into three categories according to their activities: “writers”, which catalyze m⁶A methylation, “readers”, which preferentially bind m⁶A, and “erasers”, which reverse m⁶A methylation. Here we establish the EJCs as a member of a new class of m⁶A regulators: “suppressors”, which broadly suppress the deposition of m⁶A (fig. S28). EJCs appear to be a major regulator of m⁶A deposition that mediate multiple key aspects of global m⁶A epitranscriptome specificity, including enrichment of m⁶A in long internal exons, depletion of m⁶A in CDSs and enrichment of m⁶A in last exons near stop codons, and methylation selectivity for transcripts possessing long internal exons. This mechanism may also explain the high abundance of m⁶A on certain non-coding RNAs, such as LINE-1 elements that are generally unspliced and thus not bound by the EJCs (32, 33). Further, our systematic analysis of m⁶A determinants using MPm⁶A may suggest the existence of additional m⁶A suppressing pathways, including m⁶A suppression within the CDS, as EIF4A3 KD does not appear to completely restore methylation to unspliced levels (supplementary text, fig. S8H, and fig. S29).

Our results point to exon length within transcripts as a functionally relevant element for post-transcriptional gene expression regulation. Mammalian EJCs stably bind the vast majority of pre-translational mRNAs in the transcriptome at closely spaced intervals. Long internal exons and terminal exons, which usually encode UTRs, are notably free of EJCs. This widespread binding, in conjunction with their mRNA packaging function, appears to uniquely position EJCs to broadly determine mRNA accessibility to regulatory machineries, such as the m⁶A methylation and splicing machineries (fig. S30). Our work has relevance for the use of cDNA expression constructs in research studies and gene therapies, as loss of endogenous mRNA exon architecture and EJC protection results in m⁶A hypermethylation (fig. S11), which could modulate gene expression outcome. Finally, our study also suggests that exon length and architecture co-evolved with mRNA processing steps as an additional regulatory layer of gene expression.

Data and materials availability:

Raw and processed data can be found at NCBI GEO accession GSE162199. Custom scripts available on Zenodo (34). All other data are available in the manuscript or the supplementary materials.