Introduction

COVID-19 continues to exert an extraordinary global impact, with nearly 780 million confirmed cases and more than 7 million reported deaths to date¹⁰. Despite unprecedented efforts to vaccinate populations worldwide, the viral spread persists—primarily driven by the emergence and spread of variants capable of evading vaccine-induced immunity^11,12. This troubling trend reflects a confluence of factors: uneven vaccine distribution and uptake, relaxation of non-pharmaceutical interventions (such as mask mandates and physical distancing), and high rates of viral transmission in partially immunized communities. Under these conditions, SARS-CoV-2 variants that harbor mutations conferring both increased transmissibility and immune escape gain a selective advantage¹².

This evolutionary pressure is most pronounced in the viral spike (S) glycoprotein—the principal target of all vaccines authorized in the United States and European Union. The spike protein’s receptor-binding domain and adjacent epitopes are inherently hypermutable^6,11–13, permitting the rapid accrual of substitutions that diminish neutralizing antibody binding. Indeed, successive variants of concern—Delta, Omicron, XBB lineage, and most recently JN.1—have each accrued distinct constellations of spike mutations that enhance ACE2 receptor affinity and reduce vaccine-elicited antibody recognition, thereby increasing both infectivity and virulence^11,12.

In light of these challenges, combinatorial antiviral strategies have emerged as a promising approach to suppress the evolution of treatment-resistant strains. By applying multiple, independent selective pressures—analogous to the multidrug regimens that transformed the management of HIV¹⁴, HBV¹⁵, tuberculosis¹⁶, and certain cancers¹⁷—such therapies elevate the mutational barrier required for viral escape. For SARS-CoV-2, identifying viral targets that are both functionally indispensable and evolutionarily constrained is therefore a critical priority to complement existing vaccination campaigns.

One such conserved feature is the mechanism of discontinuous transcription, a hallmark of coronavirus genome replication¹⁸. This process hinges on a short Transcription Regulatory Sequence Leader (TRS-L) in the 5′-untranslated region (UTR) of the viral genome, which base-pairs with complementary TRS-body (TRS-B) sequences upstream of each open reading frame (ORF)^8,9,19–21. During negative-strand synthesis, the replicase complex switches templates from TRS-B sites to the TRS-L, generating a nested set of Subgenomic RNAs (sgRNAs) that are subsequently transcribed into individual viral mRNAs. Because every sgRNA—spanning structural and accessory genes—requires the same TRS-L segment for its synthesis, the 5′-UTR remains highly conserved across the Coronaviridae family, both in primary nucleotide sequence and in predicted secondary structure^8,18,21–23. Moreover, the non-structural viral proteins are translated from the polycistronic ORFs 1a, and 1b, which utilize the same 265 bp region as their 5′-UTR. The remaining transcripts also known as sub-genomic transcripts contain 5′-UTRs of varying lengths with the 5′ 1-75 nucleotides common to all⁹. Indeed, sequence alignments reveal minimal variation in the 5′-UTR among early arising SARS-CoV-2 variants (Supplementary Figure 1).

Despite the central role of 5′-UTRs in regulating mRNA translation^24–26, we know little about the regulatory influence of SARS-CoV-2 5′-UTR on translation. Secondary structure analyses^{21,23,27–29} predict a complex architecture of stem-loops and pseudoknots in the full-length 5′-UTR. Such complexity would ordinarily impede translation initiation, which relies on 5′-to-3′ scanning by the pre-initiation complex and ATP-dependent unwinding by the eIF4A helicase^30–32. In cellular mRNAs, extensive 5′-UTR structure correlates strongly with reduced translational efficiency³³; however, coronaviral mRNAs paradoxically achieve high translation rates and outcompete host transcripts for ribosomal engagement^22,34–36.

We reasoned that specific RNA modifications might underlie this apparent contradiction. N⁶-methyladenosine (m⁶A) is the most prevalent internal modification in eukaryotic and viral mRNAs³⁷, comprising approximately 99% of all known modifications. First detected in viral RNAs in 1974³⁸, m⁶A has only recently been linked to the life cycles of diverse viruses^39–44. The “writer” complex, centered on METTL3, deposits m⁶A at consensus DRACH motifs ([A/G][A/G]AC[U/A/C]) in a context-dependent manner^37,45,45 after which “reader” proteins (YTHDF1–3) interpret these marks to influence RNA splicing, stability, localization, and translation⁴⁶. Recent studies have mapped m⁶A sites in host transcriptomes during SARS-CoV-2 infection⁴⁷, as well as in the viral genome itself⁴⁸, but few have focused specifically on the 5′-UTR⁴⁹.

To fill this gap, we engineered a series of SARS-CoV-2 5′-UTR variants with targeted alterations at predicted m⁶A sites and assessed their impact on translation using luciferase reporter assays. Our data confirm that, despite its predicted intricate secondary structure, the SARS-CoV-2 5′-UTR is capable of driving robust translation. Strikingly, mutation of a single m⁶A site within the 5′-UTR dramatically reduces translational output, an outcome mimicked by pharmacological inhibition of METTL3. Contrary to our expectation, this effect is not mediated by YTHDF reader proteins; rather, m⁶A appears to facilitate translation likely by destabilizing local RNA structure. Based on these findings, we propose a model in which m⁶A modification of the SARS-CoV-2 5′-UTR fine-tunes viral translational efficiency, thereby highlighting the viral 5′-UTR as a potential target for future COVID-19 therapeutics.

Results

SARS-CoV-2 5′-UTR is unexpectedly efficient at driving protein translation

Computational RNA modeling algorithms predict a highly complex secondary structure for the SARS-CoV-2 5′-UTR. To place it in the broader landscape of 5′-UTR structural diversity, we compared it to a 4-fold concatemer of the human ß-globin (hbb) 5′-UTR (Figure 1A), which is an efficient promoter of translation, that has also served as the canonical leader sequence in many of the current mRNA vaccine platforms^50–52. We also employed genome-wide secondary structure prediction across all annotated human 5′-UTRs, computing a length-normalized stability metric (MFE per nucleotide) to correct for the intrinsic bias of longer sequences toward more negative free energy values⁵³. Focusing on the subset of human transcripts whose GC content (40–50 %) and length (150–275 nt) closely resemble those of the SARS-CoV-2 5′-UTR (GC = 44.5 %, length = 265 nt) and 4-fold Hbb 5′-UTR (GC = 44 %, length = 200 nt), we observed that the SARS-CoV-2 5′-UTR resides in the upper tail of the distribution for structural density (FDR < 0.05; Figure 1B). This indicates that, on a per-base basis, it forms a more densely packed network of secondary structures than over 95% of human transcripts of comparable composition.

Figure 1.

Given the well-established inverse relationship between 5′-UTR secondary structure complexity and translational efficiency—whereby those with more stable secondary structure and lower minimum free energy (MFE) typically impede ribosomal scanning and initiation—we predicted that the SARS-CoV-2 5′-UTR (MFE of –82.10 kcal/mol) would drive lower protein output than the relatively unstructured 4X Hbb 5′-UTR (MFE of –40.40 kcal/mol) leader³³. To test this hypothesis, we cloned each 5′-UTR directly upstream of a firefly luciferase open reading frame in a standardized expression vector and transfected these constructs into HEK293T cells. Quantitative luminescence assays performed 24 hours post-transfection revealed, to our surprise, no statistically significant difference in luciferase activity or mRNA levels between the SARS-CoV-2 and 4X Hbb 5′-UTR reporters (Figure 1C and Supplementary Figure 2A), suggesting that the viral leader harbors additional features—beyond its predicted thermodynamic stability—that enable robust translation in a cellular context.

SARS-CoV-2 5′-UTR m⁶A methylation enhances protein translation efficiency

Given the abundance of m⁶A modification in eukaryotic and viral transcripts³⁷and its destabilizing effect in regions of RNA that are base-paired^54,55, we reasoned that m⁶A methylation might underlie the unexpectedly high translational efficiency of the SARS-CoV-2 5′-UTR. To test this hypothesis, we first asked whether the viral leader is decorated with m⁶A marks. In silico scans for the consensus DRACH motif across the 265-nt 5′-UTR, coupled with established m⁶A site predictors, revealed a single high-confidence candidate at adenine 74, embedded within the essential TRS-L sequence in a base-paired region (Figure 2c)⁵⁶. Next, we performed m⁶A RNA immunoprecipitation (MeRIP) on total RNA harvested from HEK293T cells transfected with a SARS-CoV-2 5′-UTR–firefly luciferase reporter. Quantitative RT-PCR of the anti-m⁶A pulldown fraction showed a robust ∼52-fold enrichment of the viral leader relative to input (mean ± SEM: 52.0 ± 10.2; p < 0.001), confirming that the 5′-UTR is indeed methylated in these cells (Figure 2A, B).

Figure 2.

To demonstrate that this enrichment reflects specific methylation at A74, we introduced a point mutation (C75G) that disrupts the DRACH consensus while preserving the adjacent sequence context (Figure 2C). Strikingly, the C75G mutant abolished MeRIP enrichment almost entirely (0.061 ± 0.005; WT vs. C75G p = 0.002), indicating that methylation at A74 is responsible for the observed signal. Consistent with this, siRNA-mediated knockdown of METTL3 reduced enrichment of the WT 5′-UTR from 52.04 ± 10.2 to 4.40 ± 0.49 (p = 0.003), whereas the C75G mutant remained unenriched regardless of METTL3 levels (Figure 2B). These results establish that A74 is a bona fide METTL3-dependent m⁶A site.

Having confirmed m⁶A modification at A74, we next assessed its functional impact on translation. HEK293T cells were transfected with luciferase reporter constructs bearing either the WT 5′-UTR or the C75G mutant. Luciferase assays revealed a dramatic ∼60% reduction in translation for the C75G mutant relative to WT (WT = 1.00 ± 0.043 RLU; C75G = 0.401 ± 0.107 RLU; p = 0.009; Figure 2D). To rule out the possibility that this effect arose from disruption of a predicted base-pair (G61–C75), we generated a compensatory double mutant (C75G + G61C) that restores the pairing but eliminates the DRACH motif. This double mutant exhibited an identical translational defect to C75G alone, confirming that loss of m⁶A, and not secondary-structure perturbation, underlies the diminished luciferase output (Figure 2D). We found no significant differences in mRNAs between the different conditions suggesting that translational, and not transcriptional, effects are responsible for the different luciferase levels (Supplementary Figure 2B).

In support of the critical role of m⁶A modification in translational regulation, we found that METTL3 siRNA knockdown phenocopied the effect of elimination of the DRACH motif by suppressing the ability of the WT SARS-CoV2 5′-UTR to drive translation (siControl = 1.00 ± 0.043 vs. siMETTL3 = 0.47 ± 0.064 RLU; p = 0.023) without causing further suppression of the C75G mutant 5′-UTR (Figure 2D). Similarly, pharmacological inhibition of METTL3 with the specific small-molecule STM2457 produced a dose-dependent suppression of WT 5′-UTR-driven translation, while leaving the mutant 5′-UTR^C75G-driven translation unchanged (Figure 2E). We found no change in luciferase reporter mRNA levels between any of the conditions, further indicating the observed differences in luciferase activity were a result of translational regulation (Supplementary Figure 2C). The lack of additive effect of DRACH mutation and pharmacological inhibition supports the conclusion that m⁶A modification at A74, and not at other potential methylation sites, accounts for the translational enhancement conferred by the viral leader. Furthermore, these results rule out any significant contribution from the effect of METTL3 knockdown on the host machinery.

The SARS-CoV-2 5′-UTR as typically defined spans nucleotides 1–265 (SL1-TSS) of the viral genome, terminating at the translational_start site (TSS) of ORF1. However, structural and biochemical probing studies have demonstrated that an additional 35-nt segment immediately downstream of the TSS participates in the formation of the fifth stem–loop (SL5), completing the native in vivo secondary structure of the UTR sequence^21,57–59. To determine whether inclusion of this SL5 extension alters methylation-dependent regulatory features, we cloned an “extended” 5′-UTR (nt 1–300) upstream of the luciferase reporter (SL1-SL5) and performed the same suite of translation assays. The SL1-SL5 leader recapitulated the translational regulation we observed with 1-265 leader (Figure 1C) and allowed for efficient translation of luciferase (Figure 2F). To test the role of m⁶A modification at A74, we mutated this site and the complimentary nucleotide on the opposite strand to disrupt the DRACH motif and maintain the integrity of the stem loop (SL1-SL5^A74T+T62A). The 5′-UTR^A74T+T62A mutant behaved similarly to the 5′-UTR^C75G+G61C mutant, showing impaired translation efficiency compared to the control SL1-SL5 leader and remained unresponsive to METTL3 knockdown (Figure 2F). Both the SL1-TSS and SL1-5 constructs produced similar levels of mRNA, ruling out the possibility that an unexpected transcriptional difference underlies our observations (Supplementary Figure 2D). Together, these data confirm that the canonical 265 nt 5′-UTR and C75G mutant faithfully represent the translational regulation of the extended SARS-CoV-2 5′-UTR—including the complete SL5—and loss of A74 methylation. For the remainder of the experiments, we used these two leader constructs interchangeably.

Finally, because canonical m⁶A biology invokes YTH-domain “reader” proteins to mediate downstream effects, we tested whether two prominent cytoplasmic readers YTHDF1 and YTHDF2 are required for 5′-UTR activity. Surprisingly, siRNA knockdown of both readers in HEK293T cells led to a ∼1.6-fold increase in luciferase translation from the WT 5′-UTR (siControl = 1.00 ± 0.083 vs. DF1/2 siRNA = 1.58 ± 0.067 RLU; p = 0.002; Figure 2G), an effect that was lost when m⁶A was abrogated by either C75G mutation or METTL3 knockdown. We found no change in mRNA levels for any of the constructs following either METTL3 or DF1/DF2 knockdown (Supplementary Figure 2E). Our findings indicate that rather than acting as positive effectors of METTL3, YTHDF proteins appear to antagonize 5′-UTR-mediated translation in this context. This reveals an antagonistic interplay between m⁶A modification, which promotes translation, and reader engagement at A74, which appears to hinder translation. Furthermore, these results suggest that mere methylation of A74 by Mettle3 is sufficient to promote translation independently of YTHDF readers. While the action of the readers appears to be dependent on m⁶A modification at A74, we cannot at this point determine at which point in the process of translation they exert their inhibitory influence. We did not pursue the role of readers in this regulation any further in the current work.

Loss of SARS-CoV-2 5′-UTR methylation negatively affects association with polysomes

The number of ribosomes associated with a given transcript is a reliable measure of translational efficiency of that transcript; this is estimated by determining the position of the peak of gradient fractions containing that transcript in a polysome fractionation experiment. To build on our discovery that m⁶A enhances SARS-CoV-2 5′-UTR–mediated translation, we performed ribosome profiling to map the changes in ribosome association of our reporter mRNAs in HEK293T cells under conditions of low and high m⁶A modification.

Polysome profiles generated from total HEK293T cell extracts showed indistinguishable patterns of 40S, 60S, 80S, and polysome peaks in control versus METTL3-knockdown condition, indicating that global translation was largely unperturbed as a result of METTL3 depletion (Figure 3A, B). Consistent with this, the distribution of the endogenous, non-methylated hprt1 mRNA (internal control) remained constant across fractions regardless of METTL3 status (Figure 3C, lower panel). These findings indicate that global depletion of m⁶A modification does not grossly alter translation efficiency.

Figure 3.

In stark contrast, when we examined the SARS-CoV-2 5′-UTR reporter transcripts, we observed a pronounced redistribution of ribosome occupancy toward lighter gradient fractions upon METTL3 knockdown (Figure 3C, upper panel and Figure 3E). In control cells, the viral-UTR–driven mRNA predominantly co-sedimented with heavier polysome-associated fractions, reflecting efficient translation. On the other hand, METTL3 depletion shifted the peak toward 40S/60S-associated fractions, indicating a strong inhibition of translation (Figure 3C and E). We next asked whether removal of the single m⁶A site at A74 was sufficient to phenocopy METTL3 depletion. Indeed, the C75G mutant reporter displayed a virtually identical shift toward lighter fractions, both in the presence and absence of METTL3 knockdown (Figure 3D, upper panel and Figure 3F), suggesting that no additional methylation sites contribute to this effect. These results demonstrate that methylation of this site is a critical determinant of translational regulation exerted by SARS-CoV-2 5′-UTR.

m⁶A methylation promotes accessibility of SARS-COV-2 5′-UTR

Previous work has demonstrated that m⁶A modification can interfere with RNA secondary structure and destabilize local structures⁵⁵. This prompted us to explore whether the effect of m⁶A modification at A74 on translation is related to the effect of m⁶A on the secondary structure of the 5’-UTR. Modeling the influence of m⁶A on RNA folding⁶⁰ predicted that methylation at A74 energetically destabilizes the SARS-CoV-2 5′-UTR, particularly by disfavoring the closed conformation of stem–loop 3, SL3, (ΔG = 2.1 kcal/mol vs. 1.6 kcal/mol when methylated), while concomitantly stabilizing the open conformation (Supplementary Figure 3A, B). Consistent with this prediction, Becker et al. (2024) have reported that m⁶A methylation at A74 in the SARS-CoV-2 5′-UTR destabilizes SL3, thereby favoring the formation of TRS L:TRS B duplexes and promoting the discontinuous transcription of subgenomic mRNAs ⁴⁹. These findings together suggest that changes in SL3 stability could influence both local stability at A74 and interactions between SL3 and other structural elements in the 5’UTR. We, therefore, set out to examine the role of structural domains in the 5’UTR by engineering a series of truncated UTR variants: one spanning stem–loops 1 through 3 (SL1–3), another including SL1 through SL4.5 (SL1–4.5), and compared them to the longer constructs ending at the transcription start site (SL1–TSS) or the fifth stem–loop (SL1–5) (Figure 4A). Each fragment was cloned upstream of the luciferase reporter, and translation was assayed in HEK293T cells under conditions that either preserve or abrogate m⁶A (via METTL3 siRNA or the A74T point mutation). All of the constructs generated robust and consistent mRNA once transfected and qPCR did not reveal any alteration in mRNA levels as a result of METTL3 knockdown (Supplementary Figure 2F).

Figure 4.

Neither of the truncated SL1–3 and SL1–4.5 reporters exhibited any dependence on m⁶A: luciferase output remained unchanged regardless of METTL3 knockdown or A74T disruption (Figure 4B). In stark contrast, both the full-length (SL1–265) and extended (SL1–5) constructs recapitulated the robust m⁶A-dependent enhancement of translation, suggesting that the interplay between SL3 (which harbors A74) and distal elements in SL5 is required for the methylation-driven regulatory effect (Figure 4B).

These results suggested to us that a potential interaction between SL3 and SL5, when A74 is not methylated, may exert translation inhibition by hindering protein accessibility and slowing the ability of the helicase action prior to the start of translation. As an alternative to SHAPE (Selective 2′-Hydroxyl Acylation analyzed by Primer Extension)⁶¹, we used an RT-qPCR approach to estimate any potential interaction between m6A at A74 and other structural features in the SARS-CoV-2 5′-UTR that might influence RNA stability and thus act as hinderance to translation machinery. destabilizes. A similar method has recently been used to assay interference of small molecules with the tertiary structure of SARS-CoV-2⁶². Our approach takes advantage of a strand-displacement–deficient reverse transcriptase from Moloney murine leukemia virus (MMLV-RT)⁶³ that is prone to pausing at folded structural elements⁶⁴. By performing parallel RT reactions under denaturing (high-temperature) versus non-denaturing (low-temperature) conditions, template accessibility can be inferred: complex folds impede enzyme progression in the native setting, but impose minimal hinderance under denaturing conditions. As such, combining this reverse transcriptase assay with a qPCR strategy, we can quantify the efficiency of the RT reaction and predict which structural features correlate with more or less stable secondary structures. Following this approach, we found that under denaturing conditions (55°C), RT efficiency on the SARS-CoV-2 5′-UTR reporter was invariant across the truncated or full-length 5′-UTR reporters, and remained so after METTL3 knockdown or DRACH motif abrogation (Figure 4C-D, top). However, in non-denaturing reactions (37°C), METTL3 knockdown resulted in reduced cDNA synthesis from the longer SL1-TSS and SL1-SL5 5′-UTR RNA templates and both the C75G and A74T mutants showed a similar reduction in cDNA synthesis (Figure 4D, bottom). No further reduction occurred when combining METTL3 depletion with the point mutations, indicating that m⁶A modification at this site was the underlying cause. Additionally, truncated SARS-CoV-2 5′-UTRs containing only SL1-3 or SL1-4.5 did not show changes in reverse transcription efficiency in response to changes in m⁶A status (Figure 4D). These findings suggest that the destabilizing effect of m⁶A at A74 in SL3 can influence accessibility of protein factors such as reverse transcriptase to the full-length leader.

To further examine the effectiveness of our approach of using reverse transcription efficiency to correlate RNA secondary structure and accessibility with m⁶A modification, we turned to HEK293T cells and screened examples of human mRNAs with a single m⁶A modification exclusively in their 5’UTR⁶⁵. We chose two human mRNAs that fit this criterion: ACTA2 and COX8A. Prediction of the minimum free energy at 37°C for these two 5’UTRs compared to that of HPRT1 (control) indicates that ACTA2 has a more complex 5’UTR compared to HPRT1, while 5’UTR of COX8A appears less complex (Supplementary Figure 3C). As expected, at 55°C all three 5’UTRs are less stable with less negative ΔGs, yet maintaining the same trend (Supplementary Figure 3C). Consistently, we found that the efficiency of the RT reaction correlated with the complexity of 5’UTR for each respective gene and increasing temperature increased the success of the reaction (Supplementary Figure 3D). Strikingly, we found that under denaturing temperatures, all three mRNAs allowed for efficient reverse transcriptase activity regardless of the presence or absence of METTL3 activity; however, at 37°C knockdown of METTL3 significantly reduced the efficiency of the reverse transcriptase for both COX8A and ACTA2 but was unchanged for HPRT1 (Supplementary Figure 3C). These results further support our proposed framework for understanding the relationship between 5’UTR m⁶A modification and translational regulation and suggest that a similar process likely governs this relationship in humans.

Discussion

We demonstrate that despite its predicted complex secondary structure, the SARS-CoV-2 5′-UTR is an efficient leader sequence. A single m⁶A modification at A74 accounts for this anomaly, as its mutation or inhibition of METTL3 significantly impair the ability of SARS-CoV-2 5′-UTR to drive translation. Our experiments indicate that this m⁶A modification is particularly relevant to the efficiency of the full-length leader, responsible for facilitating translation of the polycistronic ORFs 1a and 1b when the highly complex stem loops SL5 is present. As such, the translational efficiency of the truncated versions of the leader, in the absence of SL5, do not show sensitivity to A74 mutation or inhibition of METTL3. Our work provides a mechanistic framework for understanding the epitranscriptomic control of coronavirus replication and highlights the viral 5′-UTR as a potential target for next-generation antiviral strategies against COVID-19⁴⁹.

Recent studies have highlighted the pervasive role of m⁶A methylation in regulating the life cycle of diverse RNA viruses. On the one hand, in the case of enterovirus 71 (EV71), influenza A virus (IAV), human immunodeficiency virus (HIV), hepatitis B virus (HBV), and simian virus 40 (SV40) m⁶A methylation generally promotes replication and infectivity^40,41,66,67. On the other, m⁶A appears to exert inhibitory effects⁶⁶ on other viruses such as hepatitis C virus (HCV) and Zika virus (ZIKV)^40,66,68. In the case of SARS-CoV-2, early reports yielded conflicting insights into the net impact of m⁶A: some groups observed that m⁶A deposition on viral RNA restricts replication and that depletion of METTL3, its cofactors, or YTH-domain reader proteins exacerbates infection⁴⁸; by contrast, other work found that pharmacological blockade of m⁶A erasers (ALKBH5, FTO) suppresses viral load, whereas METTL3 knockdown or inhibition reduces infectivity^47,69. Additional clinical and genetic studies have linked variation in m⁶A machinery components to COVID-19 disease severity and risk, further underscoring the complexity of host–virus epitranscriptomic interactions^70–75. Our findings reconcile these seemingly discordant observations by pinpointing an essential, METTL3-dependent m⁶A mark at A74 within the viral 5′-UTR that specifically enhances translation initiation. Unlike bulk viral replication assays, our targeted approach—combining precise mutagenesis of the DRACH motif, ribosome-profiling, and RNA-structure probing—reveals that loss of A74 methylation uniformly impairs reporter-based translation without altering global host translation. Furthermore, our findings indicate an unexpected relationship between YTH cytoplasmic m⁶A readers and METTL3. While we find m⁶A modification enhances translation, the same site is used by the readers to dampen it. This suggests that while the readers require the presence of m⁶A moiety at A74 to exert their inhibitory action, methylation of A74 by METTL3 independently of YTH readers is sufficient for enhancing translation initiation, This reader-independent mechanism offers a molecular explanation for why different experimental systems, cell types, or timing of m⁶A perturbation may yield divergent outcomes: the balance between pro-viral effects of site-specific m⁶A modification and anti-viral m⁶A reader-mediated actions likely varies depending on cellular context, the ensemble of methylated sites across host and viral transcripts, and the interplay of writers, readers, and erasers in each setting.

Despite the growing catalog of m⁶A maps in SARS-CoV-2 genomes^47,48,76 the A74 site within the TRS-L has been detected only rarely^76,77. This underrepresentation likely stems from technical biases—such as 3′ end capture preference in sequencing workflows, platform-specific limitations of Nanopore and PacBio sequencing, and the formidable secondary-structure complexity of the 5′-UTR that impedes reverse-transcription and library preparation⁷⁸. Moreover, A74 lies adjacent to the cap structure, raising the possibility that internal m⁶A could be misclassified as cap-proximal m⁶Am (cap-adjacent adenosine dimethylation)^79,80. Finally, the field’s emphasis on 3′-UTR–enriched m⁶A signals in host mRNAs has historically overshadowed functional investigation of 5′-end methylation³⁷. In this work, by focusing exclusively on the viral 5′-UTR and employing complementary biochemical and genetic assays, we provide definitive evidence that A74 methylation, at a site universally maintained among SARS-CoV-2 variants (Supplementary Figure 1A), confers a crucial translational advantage.

How does m⁶A at A74 promote translation? Previous work has suggested that the destabilizing effect of A74 m⁶A on SL3 may influence interaction between remote regions of mRNA⁴⁹. Consistently, we find that A74 mutation curtails the ability of a strand-displacement–deficient reverse transcriptase to efficiently access the full-length leader sequence, suggesting that in the absence of m6A SL3 and SL5 may interact to increase the complexity of the secondary structure. From this a model emerges that a potential interaction between SL3 and SL5 likely underlies the differences we observe in translation-reporter activity. In the absence of m⁶A, the viral leader likely adopts a more rigid conformation that sterically hinders pre-initiation complex progression, thereby diminishing translation initiation. Our findings are consistent with this model: in the presence of SL5, m⁶A at A74 is critical for translation efficiency; however, when SL5 is removed, m⁶A does not influence the efficiency of translation. At the same time, mere presence of SL5 does not inhibit translation when SL3 is destabilized by m⁶A modification at A74. By linking m⁶A modification to secondary-structure dynamics and translational regulation, this work extends our foundational knowledge of m⁶A biology and points to a potential new therapeutic design to tackle SARS-CoV-2.

Materials and methods

Genomic Sequences

The complete genome sequence of the SARS-CoV-2 viral genome was obtained from NCBI’s Genbank reference sequence NC_045512. The 5′-UTR region (1-265 nt) was selected based on Genbank annotation. The complete mRNA sequence of Homo sapiens hemoglobin subunit beta (Hbb) was obtained from Genbank reference sequence NM_000518. The 5′-UTR Region (1-50 nt) was selected based on Genbank annotation. Genomic sequences for human 5′-UTRs were obtained from UCSC Genome Browser database, assembly GRCh37/hg19 for m⁶A annotation, and GRCh38/hg38 otherwise. 5′-UTRs were selected using the extractor functions from R package GenomicFeatures 1.48.1.

m⁶A Site Prediction and Annotation

m⁶A sites in the SARS-CoV-2 and Hbb 5′-UTRs were predicted by BLAST search for DRACH Motifs (i.e. [AGU][AG]AC[ACU]), and confirmed using SRAMP⁸¹.

RNA Secondary Structure and Minimum Free Energy Prediction

RNA secondary structure and minimum free energy was predicted using RNAFold 2.4.18 with default settings⁸². For structure prediction at denaturing temperature (Figure 4C), RNAFold 2.4.18 was used with temperature parameter set at 55°C. For m⁶A-dependent structure prediction, RNAStructure 6.4 with the following options: --alphabet m⁶A -m 100 -p 100⁶⁰. R2R and Illustrator were used to visualize the predicted structures. Predicted MFE was length normalized using Trotta’s formula to compute adjusted MFE⁵³:

Where L is the length of the 5′-UTR and is the expected MFE for an RNA sequence of length L and GC fraction of 0.5.

Luciferase Activity Assays

Luciferase activity assays were performed as per kit manufacturer instructions (Promega, #E1500). Briefly, cell media was aspirated and HEK293T cells washed briefly in PBS. Cells were then lysed in 1X Lysis Buffer (20 µL/well for 96-well plates or 900 µL/100mm dish). Lysate were then removed and briefly centrifuged. 20 µL of supernatant was mixed with 100 µL of Luciferase Assay Reagent and then measured on a plate reader.

Plasmid Construction and Transfection

pGL3-HBB 5′-UTR-4X-fLuc and pGL3-SARS-CoV2-5′-UTR-fLuc plasmids were obtained as a generous gift from Ruggero Lab, UCSF. For SARS-CoV-2 5′-UTR^C75G and SARS-CoV-2 5′-UTR^{C75G + G61C}, reporter plasmids, gBlocks™ DNA fragments (Integrated DNA Technologies) containing 1-265 bp of SARS-CoV-2 genome with indicated mutations were flanked by restriction sites and inserted immediately upstream of fLuc into pGL3-fLuc plasmids (Promega, #E1751) using standard restriction cloning. For construction of complete and truncated reporter constructs, SARS-CoV-2 5′-UTR 1-300 bp WT and A74T or A74T+T61A mutants joined with a 6xGly flexible linker⁸³ were obtained as gBlocks™ DNA fragments from which SARS-CoV-2 5′-UTR SL1-3 (1-82 bp), SL1-4.5 (1-147 bp) constructs were amplified and flanked by restriction through PCR, and then inserted in pGL3-fLuc vector using standard restriction cloning. For the SL1-5 constructs only, the initiating methionine of luciferase was mutated to lysine to ensure that only the CoV-2 TSS initiated ORF was present. Cells were transfected as per manufacturer protocol (Polyplus, #101000046). Briefly, HEK293T cells were seeded at ∼3x10⁵ cells/cm² and transfected with ∼300pM plasmid DNA (0.1µg/well in 96-Well plate or 10ug/100mm dish), and 25nM siRNA (Dharmacon, #L-018095-02, #L-021009-02, and #D-001810-10) or treated with STM2457 in DMSO (Selleckchem, # S9870) after 24 hours. Gene expression and Luciferase activity were assayed 24 and 48 hours respectively after transfection.

RNA Immuno-precipitation

m⁶A RNA-IP enrichment was performed using kit manufacturer protocol (New England BioLabs, #E1610S). Briefly, Total RNA from HEK293T cells was extracted using TRIzol and fragmented using NEBNext® Magnesium RNA Fragmentation Module (New England BioLabs, E6150S). 250µg of fragmented RNA was incubated with 250 µL of m⁶A Ab conjugated to protein G magnetic beads Reaction Buffer for 1 hour at 4°C. Samples were then briefly centrifuged, supernatant was discarded, and RNA was eluted in 150 µL of clean up binding buffer and purified for downstream RT-qPCR.

Reverse Transcription and Quantitative PCR

For cDNA production for standard qPCR, 1 µg of purified RNA was reverse transcribed using iScript cDNA synthesis Kit (BioRad, #1708891) as per manufacturer instructions where first strand cDNA synthesis was primed with a mix of poly-dT and random hexamers. For secondary structure sensitive reverse transcription reactions (Figure 4D, Supplementary Figure 3D), the same kit was used with the following modifications: the 70°C 10-minute denaturing incubation was omitted, reaction mix was supplemented with MgCl to a final concentration of 5 mM to ameliorate reduced binding of RT primers at higher temperatures, and first strand synthesis was primed with oligo(dT)₂₀ primers at a final concentration of 1 µM (IDT, #51-01-15-01).

Quantitative PCR was performed using GoTaq qPCR Master Mix (Promega, #A6001) with primers:

Ribosome Profiling

HEK293T cells were seeded at ∼3x10⁵ cells/cm² in a 100 mm dish and transfected as previously described. After 48 hours, Cycloheximide was added to culture media to a concentration of 50 µg/mL. After a brief 20-minute incubation, cells were scraped, pelleted, and ∼10⁷ cells were resuspended in a lysis buffer of 20 mM Tris, 10 mM MgCl₂, 250 mM NaCl, 100 µg/mL Cycloheximide, and 0.4% NP-40 supplemented with protease inhibitor (Thermofisher, #A32963). Cells were sheared in a Dounce homogenizer, and the resultant lysate was centrifuged briefly to remove debris. Absorbance at 260nm was then measured to estimate RNA concentration in supernatant. 500 µL of 25 OD₂₆₀ cleared lysate was carefully layered atop 20-60% sucrose gradient buffered in lysis buffer. Finally, gradients were centrifuged at 180,000 x g for 110 minutes at 4°C and subsequently fractionated.

SARS-CoV-2 5′-UTR Mutation Analysis

Recurrent mutations track in SARS-CoV-2 were obtained from Nextstrain (accessed August 4, 2022) and viewed using UCSC Genome browser track hub⁸⁴. Consensus sequences for SARS-CoV-2 5′-UTRs were derived from the NIH SARS-CoV-2 datahub. Briefly, all Complete Nucleotide SARS-CoV-2 sequences from obtained and grouped by WHO lineage classification, based on Pango lineage metadata. Nucleotides 1-300 from each sequence were then aligned to the SARS-CoV-2 reference 5′-UTR (NC_045512) using MAFFT 7.490 using options --6merpair --maxambiguous 0.05 –addfragments –keeplength⁸⁵. Consensus sequences for each multiple alignment (excluding gaps) were then determined using R Biostrings 2.64.0 package⁸⁶. Consensus secondary structure was determined using RNAFold 2.4.18, and structure diagrams were prepared using R2R. Multiple alignment figures were prepared using ESPript 3.0⁸⁷.

Supplementary figures

Supplementary Figure 1.

Supplementary Figure 2.

Supplementary Figure 3.

Data availability

We have not generated any large data sets. We will provide our biochemical data upon request.

Acknowledgements

We thank the Ruggero laboratory (UCSF) for kindly providing the pGL3-hbb-fLuc plasmids; M. Mori for help with the polysome preparations; and M. Moujahidine for administrative support. This work was supported by an NIH grant (R01NS082793) to APH.

Additional information

Author contributions

A.A. and A.P.H. designed the experiments. A.A., M.C., and G.S. performed the experiments. A.A. and A.P.H. prepared the manuscript. A.A., G.S. and A.P.H. edited the manuscript.

Materials and correspondence

Correspondence and requests for materials should be addressed to Pejmun Haghighi.

Funding

HHS | NIH | National Institute of Neurological Disorders and Stroke (NINDS) (NS082793)

• Pejmun Haghighi