A quantitative model predicts how m6A reshapes the kinetic landscape of nucleic acid hybridization and conformational transitions

Liu, Bei; Shi, Honglue; Rangadurai, Atul; Nussbaumer, Felix; Chu, Chia-Chieh; Erharter, Kevin Andreas; Case, David A.; Kreutz, Christoph; Al-Hashimi, Hashim M.

doi:10.1038/s41467-021-25253-8

Download PDF

Article
Open access
Published: 31 August 2021

A quantitative model predicts how m⁶A reshapes the kinetic landscape of nucleic acid hybridization and conformational transitions

Nature Communications volume 12, Article number: 5201 (2021) Cite this article

4328 Accesses
15 Citations
7 Altmetric
Metrics details

Subjects

ABSTRACT

N⁶-methyladenosine (m⁶A) is a post-transcriptional modification that controls gene expression by recruiting proteins to RNA sites. The modification also slows biochemical processes through mechanisms that are not understood. Using temperature-dependent (20°C–65°C) NMR relaxation dispersion, we show that m⁶A pairs with uridine with the methylamino group in the anti conformation to form a Watson-Crick base pair that transiently exchanges on the millisecond timescale with a singly hydrogen-bonded low-populated (1%) mismatch-like conformation in which the methylamino group is syn. This ability to rapidly interchange between Watson-Crick or mismatch-like forms, combined with different syn:anti isomer preferences when paired (~1:100) versus unpaired (~10:1), explains how m⁶A robustly slows duplex annealing without affecting melting at elevated temperatures via two pathways in which isomerization occurs before or after duplex annealing. Our model quantitatively predicts how m⁶A reshapes the kinetic landscape of nucleic acid hybridization and conformational transitions, and provides an explanation for why the modification robustly slows diverse cellular processes.

DNA mismatches reveal conformational penalties in protein–DNA recognition

Article 21 October 2020

Interaction preferences between protein side chains and key epigenetic modifications 5-methylcytosine, 5-hydroxymethycytosine and N6-methyladenine

Article Open access 15 November 2022

The regulation and functions of DNA and RNA G-quadruplexes

Article 20 April 2020

Introduction

N⁶-methyladenosine (m⁶A) (Fig. 1a) is an abundant RNA modification^1,2 that helps control gene expression in a variety of physiological processes including cellular differentiation, stress response, viral infection, and cancer progression^3,4,5. m⁶A is also the most prevalent form of DNA methylation in prokaryotes where it is used to distinguish benign host DNA from potentially pathogenic nonhost DNA⁶. Although under debate⁷, there is also evidence for m⁶A in mammalian DNA where it is proposed to play roles in transcription suppression and gene silencing^8,9.

**Fig. 1: The *syn* and *anti* isomers of m6A.**

In RNAs, m⁶A is thought to primarily function by recruiting proteins to specific modified sites (reviewed in refs. ^3,4,5). However, there is also growing evidence that the modification can impact a range of biochemical processes by changing the behavior of the methylated RNAs^10,11. For example, by destabilizing canonical double-stranded RNA (dsRNA)¹², m⁶A has been shown to promote the binding of proteins to single-stranded regions of RNAs (ssRNAs)¹⁰. The modification has also been shown to slow biochemical processes that involve base pairing. For example, in messenger RNAs (mRNAs), m⁶A delays transfer RNA (tRNA) selection and reduces the translation efficiency in vitro¹³ and in vivo¹⁴ by 3–15-fold. In mRNA introns, m⁶A slows splicing and promotes alternative splicing in vivo¹⁵. Additionally, m⁶A reduces the rate of NTP incorporation during DNA replication¹⁶ and reverse transcription¹⁷ in vitro by 2–13-fold.

Recently, we developed and validated a nuclear magnetic resonance (NMR) relaxation–dispersion (RD)^18,19,20-based method to measure the hybridization kinetics in DNA and RNA duplexes²¹. Using this approach, we showed that m⁶A preferentially slows the apparent rate of RNA duplex annealing by ~5–10-fold while having little effect on the apparent rate of duplex melting²¹ (Fig. 1b). This impact of m⁶A on hybridization kinetics stands in contrast to mismatches that slow the rate of duplex annealing and also substantially increase the rate of duplex melting by up to ~100-fold^22,23,24. How m⁶A selectively slows duplex annealing remains unknown. The comparable m⁶A-induced slowdown observed for duplex annealing and a variety of biochemical processes indicates that a common mechanism might be at play^13,16,17.

It has been known for many decades that the methylamino group of the m⁶A nucleobase can form two rotational isomers that interconvert on the millisecond timescale^25,26 (Fig. 1a). The preferred syn isomer^12,25,26 cannot form a canonical Watson–Crick base pair (bp) with uridine as the methyl group impedes one of the hydrogen bonds (H-bonds) (Fig. 1a). Rather, when paired with uridine, the methylamino group rotates into the energetically disfavored anti isomer and forms a canonical m⁶A–U Watson–Crick bp that retains both (A)N1···H-N3(U) and (A)N6···H-O4(U) H-bonds (Fig. 1a). As isomerization is energetically disfavored, it has been proposed to explain how m⁶A destabilizes dsRNA via the so-called “spring-loading”¹² mechanism despite forming a canonical Watson–Crick m⁶A–U bp.

Kinetic mechanisms involving binding and conformational change can occur via pathways wherein the conformational change occurs prior to or post binding²⁷. We, therefore, hypothesized that m⁶A could slow hybridization via at least two pathways in which isomerization of the methylamino group occurs either before or following duplex formation (Fig. 1c). In the conformational selection (CS) pathway, hybridization proceeds via an unpaired intermediate (ssRNA^anti) with m⁶A in the energetically disfavored anti conformation (Fig. 1c). In the induced-fit (IF) pathway, the more populated ssRNA^syn species with m⁶A in the syn conformation initially hybridizes to form a double-stranded intermediate (dsRNA^syn) that entails the loss of at least one Watson–Crick H-bond between m⁶A and the partner uridine (Fig. 1a). This is then followed by isomerization to form the Watson–Crick bp (dsRNA^anti) with m⁶A in the anti conformation (Fig. 1c). To date, there has been no evidence for the dsRNA^syn intermediate.

Here, using NMR RD, we show that m⁶A with the methylamino group in the anti conformation forms a Watson–Crick bp with uridine that transiently exchanges on the millisecond timescale with an unusual singly hydrogen-bonded, low-populated (1%), and mismatch-like conformation through isomerization of the methylamino group to the syn conformation. This ability to rapidly interchange between Watson–Crick or mismatch forms, combined with different syn:anti isomers preferences when paired versus unpaired, explains how m⁶A robustly and selectively slows duplex annealing without affecting melting via two pathways in which isomerization occurs before or after duplex annealing. We develop a model that quantitatively predicts how m⁶A reshapes the kinetic landscape of nucleic acid hybridization, and that could explain why the modification robustly slows a variety of cellular processes. The model also predicts that m⁶A more substantially slows fast intramolecular RNA conformational transitions, and this prediction was verified experimentally by using NMR.

Results

Kinetics of m⁶A methylamino isomerization in ssRNA

The ssRNA^anti which is the intermediate along the CS pathway has been extensively characterized in the past, whereas there is no evidence for the dsRNA^syn IF intermediate. We therefore initially examined whether the CS pathway alone could explain how and why m⁶A reduces the rate of duplex annealing while not affecting the melting rate. We developed a CS model which assumes that the minor anti isomer of m⁶A hybridizes with apparent annealing (k_on) and melting (k_off) rate constants similar to those of the unmethylated RNA. This assumption is reasonable given that like unmethylated adenine, the anti isomer forms a canonical m⁶A–U Watson–Crick bp when paired with uridine^11,12,25,26. Since the syn isomer is incapable of Watson–Crick pairing with uridine, the model assumes that hybridization only proceeds via annealing of the single-strand containing the minor anti isomer (ssRNA^anti) through a CS-type pathway^27,28 (Fig. 2a). The apparent k_on would then be reduced relative to the unmethylated RNA because the methylamino group has to rotate from the major syn to the minor anti isomer prior to hybridization (Fig. 2a). However, because anti is the preferred isomer in the canonical duplex, and because hybridization is rate-limiting under our experimental conditions (see below), the apparent k_off would remain equivalent to that of the unmethylated duplex.

**Fig. 2: Testing a conformational selection kinetic model for m⁶A hybridization.**

To test this CS model, we first used NMR RD to measure the isomerization kinetics in a ssRNA containing the most abundant m⁶A consensus sequence^1,2 in eukaryotic mRNAs (${{{{{\mathrm{ssGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$; Fig. 2b). This was important given that prior kinetic measurements of isomerization were performed on the m⁶A nucleobase dissolved in organic solvents and the kinetics may differ in ssRNA under aqueous conditions²⁵.

To enable the RD measurements, we used organic synthesis (see “Methods”) to incorporate m⁶A ¹³C labeled at the base C2 and C8, or methyl C10 carbons (Supplementary Fig. 1) into ssGGACU. We then performed NMR chemical exchange saturation transfer (CEST)^29,30,31 and off-resonance spin relaxation in the rotating frame (R_1ρ) experiments^18,19,20 to measure the isomerization kinetics. NMR RD experiments can be used to characterize conformational exchange between a dominant ground-state (GS) and short-lived low-populated “excited-state” (ES). The R_1ρ experiment measures the line-broadening contribution (R_ex) to the transverse relaxation rate (R₂) during a relaxation period in which a continuous radiofrequency (RF) field is applied with variable power (ω_SL) and frequency (ω_RF). The RF field reduces the R_ex contribution in a manner dependent on ω_SL and ω_RF and the exchange parameters of interest (see below). The RD profiles are typically displayed by plotting the measured R₂ + R_ex as a function of ω_SL and ω_RF. For detectable exchange, a peak is observed centered at the difference between the chemical shift of the GS and ES (−Δω, assuming ω_GS = 0 and ω_ES = Δω). The CEST experiment measures the impact of conformational exchange on longitudinal GS magnetization during a relaxation period following application of a continuous RF field with variable power (ω_SL) and frequency (ω_RF). When applied on resonance with the ES, the RF field saturates the ES magnetization, and this saturation can be transferred via conformational exchange to the GS. This typically results in a reduced signal intensity for the GS and a minor dip centered at ω_ES = Δω when the RF is on resonance with ES. A major dip is also observed centered at ω_GS = 0 when the RF field is on resonance with the GS. The dependencies of R₂ + R_ex (R_1ρ) or the GS signal intensity (CEST) on ω_SL and ω_RF can be fit to the Bloch–McConnell (B–M) equations³² describing N-site exchange to determine exchange parameters of interest (see below). Together, R_1ρ and CEST, which are optimized for different nuclei and exchange kinetics, allowed robust characterization of chemical exchange between the major GS syn methylamino and the low-populated and short-lived ES³³ anti methylamino isomer in unpaired m⁶A.

Shown in Fig. 2c on the left is the CEST profile recorded for the m⁶A-C10 methyl carbon in ${{{{{\mathrm{ssGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ as a function of RF. As is typical for CEST profiles, a major dip is observed when the RF field is on-resonance with the GS chemical shift at Δω = 0. In addition, a minor dip was observed indicative of conformational exchange with a sparsely populated ES. The dip was observed at a chemical shift Δω_C10 = ω_ES − ω_GS = 3 p.p.m., which was in good agreement with the value predicted for the anti isomer (Δω_C10 = 3–5 p.p.m.) using density functional theory (DFT) calculations³⁴ (see “Methods”). Shown in Fig. 2c on the right is the R_1ρ profile measured for m⁶A-C2 in ${{{{{\mathrm{ssGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ as a function of RF field. A peak was observed at −Δω_C2 = 0.6 p.p.m. indicative of conformational exchange. A similar C2 RD was observed in methylated but not unmethylated AMP, as expected if the RD is reporting on isomerization (Supplementary Fig. 2a).

Based on a two-state fit of the m⁶A-C10 and m⁶A-C2 RD data (Fig. 2c), the population of the ssRNA^anti isomer in ${{{{{\mathrm{ssGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ was ~9% and the exchange rate for isomerization (k_ex = k₁ + k₋₁, where k₁ and k₋₁ are the forward and backward rate constants, respectively) was ~600 s⁻¹ at T = 25 °C (Supplementary Table 1). The population was ~2-fold higher than the value measured in the nucleobase in organic solvent (Fig. 1a)²⁵ while the exchange rate was ~20-fold faster, and in better agreement with values reported recently for ssDNA³⁵ (at T = 45 °C; Supplementary Table 1). Similar syn–anti isomerization kinetics were obtained for another different sequence (Supplementary Fig. 2b).

m⁶A(anti)–U and A–U have similar thermodynamic stabilities in dsRNA

Before testing whether the CS model can predict the hybridization kinetics of methylated duplexes, we tested a thermodynamic prediction made by our model, namely that the energetics of annealing a single-strand containing the anti isomer of m⁶A should be similar to the energetics of annealing the unmethylated control. In this scenario, m⁶A destabilizes a duplex¹² primarily due to the conformational penalty (${{\Delta}} {G}_{{{{{{\mathrm{iso}}}}}},{{{{{\mathrm{ss}}}}}}}^{^\circ }$) accompanying syn to anti isomerization in the ssRNA, which we have measured here for ${{{{{\mathrm{ssGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ using NMR RD.

To test this prediction, we decomposed (Fig. 2a) the overall annealing energetics (${{\Delta}} G_{{{{{{\mathrm{anneal}}}}}},{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}}^{^\circ }$ = −6.5 ± 0.1 kcal/mol) of methylated ${{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ (Fig. 2e) measured previously using melting experiments²¹ into the sum of ${{\Delta}} {G}_{{{{{{\mathrm{iso}}}}}},{{{{{\mathrm{ss}}}}}}}^{^\circ }$ = 1.6 ± 0.2 kcal/mol plus the desired annealing energetics (${{\Delta}} {G}_{{{{{{\mathrm{anneal}}}}}},anti}^{^\circ }$) of m⁶A when it adopts the anti isomer,

$${{\Delta}} {G}_{{{{{{\mathrm{anneal}}}}}},{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}}^{^\circ }={{\Delta}} {G}_{{{{{{\mathrm{iso}}}}}},{{{{{\mathrm{ss}}}}}}}^{^\circ }+{{\Delta}} {G}_{{{{{{\mathrm{anneal}}}}}},anti}^{^\circ }$$

Indeed, we find that ${{\Delta}} {G}_{{{{{{\mathrm{anneal}}}}}},anti}^{^\circ }$ = −8.1 ± 0.2 kcal/mol is similar to that measured for the unmethylated RNA ${{\Delta}} {G}_{{{{{{\mathrm{anneal}}}}}},A}^{^\circ }$ = −7.6 ± 0.1 kcal/mol, with the methyl group being only slightly stabilizing within error by 0.5 ± 0.2 kcal/mol. A similar result was obtained for a different duplex (Fig. 2d) and a similar conclusion was also reached previously using the isomerization energetics measured in the nucleobase^25,26. Therefore, with respect to the thermodynamics of annealing canonical duplexes, m⁶A in the anti isomer behaves similarly (within <0.5 kcal/mol) to unmethylated adenine and m⁶A primarily destabilizes dsRNA due to the conformational penalty accompanying isomerization, consistent with the previously proposed “spring-loading” mechanism¹². Consistent with this interpretation, RD measurements on the m⁶A monomer reveal that 3 mM Mg²⁺ stabilizes the anti relative to syn isomer by ~0.5 kcal/mol³⁶, and correspondingly, the destabilizing effects of m⁶A on RNA duplexes is reduced by ~0.2 kcal/mol in the presence of 3 mM Mg²⁺ relative to the absence of Mg²⁺ (Supplementary Table 1).

Testing the CS kinetic model

Next, we tested whether the CS kinetic model could explain the impact of m⁶A on the hybridization kinetics of the ${{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ RNA measured recently using NMR RD²¹. These experiments were performed at T = 65 °C under conditions in which the duplex was the GS, and the ssRNA comprising two species in rapid equilibrium (ssRNA^syn ⇌ ssRNA^anti) was the ES with a population of ~25%. The CEST experiments were performed at high temperatures because at 37 °C, the ssRNA is too lowly populated (<0.1%) and the hybridization is too slow (<50 s⁻¹) to be effectively characterized by RD. Based on a two-state fit (dsRNA ⇌ ssRNA) of the m⁶A6-C2 and m⁶A6-C8 RD data (Supplementary Fig. 3a), m⁶A reduced the apparent rate of ${{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ annealing (${k}_{{{{{{\mathrm{on}}}}}},{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}}^{{{{{{\mathrm{app}}}}}}}$) relative to the unmethylated control (${k}_{{{{{{\mathrm{on}}}}}}}$) by 5-fold while having little impact on the melting rate (${k}_{{{{{{\mathrm{off}}}}}},{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}}^{{{{{{\mathrm{app}}}}}}}\approx {k}_{{{{{{\mathrm{off}}}}}}}$)²¹.

We used the three-state CS model to simulate the m⁶A6-C8 and m⁶A6-C2 RD profiles measured for the methylated ${{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ duplex. The exchange parameters for the first isomerization step (ssRNA^syn ⇌ ssRNA^anti) were fixed to the values determined independently from RD measurements on ${{{{{\mathrm{ssGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ (Supplementary Fig. 2c). ${k}_{{{{{{\mathrm{off}}}}}},anti}$ was assumed to be equal to ${k}_{{{{{{\mathrm{off}}}}}}}$ measured for the unmethylated dsGGACU. This assumption is reasonable considering that hybridization is rate-limiting under our experimental conditions, and given the similarity between the experimentally measured ${k}_{{{{{{\mathrm{off}}}}}}}$ for methylated and unmethylated duplexes²¹. The value of ${k}_{{{{{{\mathrm{on}}}}}},anti}$ was slightly adjusted relative to ${k}_{{{{{{\mathrm{on}}}}}}}\,$ of the unmethylated control (${k}_{{{{{{\mathrm{on}}}}}},anti}\approx 2\times {k}_{{{{{{\mathrm{on}}}}}}}$) to take into account small differences in their annealing energetics (Fig. 2a). The remaining NMR exchange parameters (Δω, R₁, and R₂ of GS and two ESs) for the hybridization and isomerization steps were fixed to the values obtained from the two-state fit of the RD data measured for ${{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ and ${{{{{\mathrm{ssGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ (see “Methods”).

Interestingly, this simulation with no adjustable parameters satisfactorily reproduced the RD data with ${\chi }_{{{{{{\mathrm{red}}}}}}}^{2}$ = 6.8. This can be compared with ${\chi }_{{{{{{\mathrm{red}}}}}}}^{2}$ = 3.3 (Supplementary Fig. 3a) obtained from a two-state fit of the RD data with six adjustable parameters. As a negative control, the agreement deteriorated considerably (${\chi }_{{{{{{\mathrm{red}}}}}}}^{2}$ = 51.5) (Supplementary Fig. 3b) when decreasing the exchange rate by 20-fold to mimic values observed for the nucleobase in organic solvents²⁵. A constrained three-state fit to the RD data using the CS model in which the exchange parameters were allowed to vary within experimental error by 1 s.d., and in which the ratio (but not absolute magnitude) of ${k}_{{{{{{\mathrm{on}}}}}},anti}$ and ${k}_{{{{{{\mathrm{off}}}}}},anti}$ was constrained to preserve the free energy of the hybridization step improved the agreement to ${\chi }_{{{{{{\mathrm{red}}}}}}}^{2}$ = 3.0 (see “Methods” and Fig. 2f) and yielded ${k}_{{{{{{\mathrm{on}}}}}},anti}$ $\approx$ $2\times {k}_{{{{{{\mathrm{on}}}}}}}$ and ${k}_{{{{{{\mathrm{off}}}}}},{anti}}\approx {k}_{{{{{{\mathrm{off}}}}}}}$ (Supplementary Table 2). Therefore, even when it to comes to hybridization kinetics, m⁶A in the anti isomer behaves similarly to unmethylated adenine.

These results provide a plausible explanation for the unique impact of m⁶A on RNA hybridization kinetics at T = 65 °C. m⁶A does not impact the apparent melting rate because the dominant isomer in the duplex is anti and it melts at a rate comparable to that of the unmethylated RNA. On the other hand, m⁶A slows the apparent annealing rate by ~5-fold due to the ~10-fold lower equilibrium population of the ssRNA^anti intermediate relative to the unmethylated ssRNA control and because the ssRNA^anti intermediate anneals at a 2-fold faster rate relative to its unmethylated counterpart.

An additional hybridization intermediate at T = 55 °C

Although we did not observe any evidence for the IF dsRNA^syn intermediate, simulations indicate that its RD contribution was probably masked by the larger RD contribution from the ssRNA with a population ~22%. We therefore repeated the CEST measurements at a slightly lower temperature T = 55 °C. This reduced the ssRNA population to ~5%, but it remained large enough to permit accurate measurements of hybridization kinetics using NMR RD. Repeating the measurements at a different temperature also allowed us to test the robustness of the CS model. Based on a two-state fit of the adenine C8 RD data, which only reports on a two-state hybridization process (Supplementary Fig. 4a), m⁶A reduced the apparent annealing rate by 20-fold while minimally (~1.6-fold) impacting the apparent melting rate under these conditions (Supplementary Fig. 4b).

Interestingly, we observed evidence for an additional ES, which manifested as a second minor dip in the m⁶A-C2 CEST profile (Fig. 3a). This ES dip at Δω_C2 ~2 p.p.m. was also observed at lower temperatures in another dsRNA (${{{{{\mathrm{dsA6RNA}}}}}}^{{{{{{\mathrm{m}}}}}}^{6}{{{{{\mathrm{A}}}}}}}$) sequence context (Supplementary Fig. 5 and Supplementary Table 1). The fact that this ES was not observed in ${{{{{\mathrm{ssGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ indicated that it very likely was a dsRNA conformation. The ES was likely not observed at higher temperature T = 65 °C (Fig. 2f)²¹ because it was masked by the higher RD contribution from the more populated ssRNA ES.

**Fig. 3: A hybridization intermediate for ${{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}6}$ at T = 55 °C.**

The m⁶A-C2 RD data (Fig. 3a) could be satisfactorily fit to a three-state model that includes dsRNA, ssRNA, and the additional ES. Among several three-state topologies tested³⁷ (see Supplementary Fig. 4d), the best agreement was obtained with models that place the ES on-pathway between the dsRNA and ssRNA (Fig. 3b). Therefore, these results provide direct evidence for a dsRNA on-pathway hybridization intermediate and the CS pathway alone cannot fully explain the hybridization kinetics at T = 55 °C. Indeed, simulations using the CS model did not reproduce the m⁶A-C2 RD data at T = 55 °C (${\chi }_{{{{{{\mathrm{red}}}}}}}^{2}$ ~ 600) (Supplementary Fig. 4c) and neither did a constrained three-state fit to the CS model (${\chi }_{{{{{{\mathrm{red}}}}}}}^{2}$ ~ 43.3) (Fig. 3c) because the model fails to account for the RD contribution from the additional ES.

The dsRNA hybridization intermediate features an m⁶(syn)A···U stabilized by a single H-bond

Understanding how m⁶A selectively slows annealing of dsGGACU at T = 55 °C by 20-fold without affecting the melting rate requires that we characterize the structure of the intermediate, which can be part of a separate hybridization pathway distinct from the CS pathway.

Although never observed previously, one possibility is that the intermediate is a dsRNA conformation in which the methylamino group rotates into the syn conformation. Such a conformation is predicted to be highly energetically disfavored in dsRNA, given the loss of at least one Watson–Crick H-bond. However, this loss in energetic stability would be partly compensated for by a gain in the stability of ~−1.5 kcal/mol from restoring the energetically favored syn isomer. Such an intermediate would allow for an IF-type hybridization pathway, in which isomerization of the methylamino group occurs following and not before initial duplex formation (Fig. 1c).

To test this proposed conformation for the ES, we performed an array of NMR RD experiments using a stable hairpin variant of ${{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ (${{{{{\mathrm{hpGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$; Fig. 4a) with a much higher melting temperature (T_m is predicted to be ~80 °C), designed to eliminate any background RD contribution from the ssRNA across a range of temperatures. Interestingly, we observed two-state RD for both m⁶A-C10 (Fig. 4b) and m⁶A-C2 (Supplementary Fig. 6a) at T = 55 °C. A global fit of the data yielded an ES population (~1%), k_ex (~500 s⁻¹), and Δω_C2 = 2.5 p.p.m. that were in very good agreement with the values (Supplementary Table 3) measured for the on-pathway ES hybridization intermediate in ${{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$. The Δω_C10 and Δω_C2 values were also in very good agreement with values predicted for m⁶(syn)A···U based on DFT calculations (Fig. 4g). Additional support that in the ES the methylamino group is syn comes from the kinetic rate constants of interconversion (Supplementary Note 1).

**Fig. 4: Characterizing the conformation of the ES intermediate.**

To gauge the nature of the Watson–Crick (m⁶A)N1···H3-N3(U) H-bond in the ES, we performed additional RD experiments targeting the N3 and H3 atoms of the partner uridine. We observed ¹⁵N (Fig. 4c) and ¹H (Fig. 4d) RD only for the uridine partner of m⁶A (Supplementary Fig. 6a), and the two-state fit of the data yielded exchange parameters similar to those obtained from the carbon C2/C10 data (Supplementary Fig. 6a), indicating that they are reporting on the same ES. The Δω_N3 = −4.8 p.p.m. and Δω_H3 = −3 p.p.m. values indicated a substantial weakening of the remaining H-bond in the ES³⁸ (Fig. 4e). Indeed, a structural model for the m⁶(syn)A···U ES conformation that predicts the ES chemical shifts well based on DFT (Fig. 4g), and features a slightly (by 0.4 Å) elongated (m⁶A)N1···H3-N3(U) H-bond (Supplementary Fig. 6b). Note that while a minor peak was not observed in the ¹H CEST profile for U17-H3 in ${{{{{\mathrm{hpGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$, simulations indicate that this could be due to the 2-fold lower ES population (Supplementary Fig. 6c and Supplementary Table 1).

These results establish that the m⁶A methylamino group can also isomerize even in the context of a duplex m⁶(anti)A–U Watson–Crick bp and show that the preferences for the syn:anti isomers is inverted from ~10:1 in the unpaired single-strand to ~1:100 in the paired dsRNA.

Chemical-shift fingerprinting the m⁶(syn)A···U ES using m⁶ ₂A

To further verify the unusual m⁶(syn)A···U conformation proposed for the ES, we stabilized this species and rendered it the dominant conformation by replacing the m⁶A amino proton with a second methyl group so as to eliminate the GS Watson–Crick H-bond (Fig. 4e). This N⁶,N⁶-dimethyl adenine (m⁶₂A) modification (Fig. 4e) is also a naturally occurring RNA modification³⁹.

Comparison of NMR spectra of dsGGACU with and without m⁶₂A showed that the modification primarily affected the methylated bp while minimally impacting other neighboring bps (Supplementary Fig. 7a, c). Both the m⁶₂A-C2 and U-N3 chemical shifts of the m⁶₂A-modified dsGGACU (${{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}_{2}^{6}{{{{{\mathrm{A}}}}}}}$) were in very good agreement with those measured for the ES in ${{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ using RD (Fig. 4g). In addition, we observed an upfield shifted imino proton resonance (at ~10 p.p.m.), which could unambiguously be assigned via site labeling to the m⁶A partner U13-H3 (Fig. 4f and Supplementary Fig. 7a). This along with nuclear Overhauser effect-based distance connectivity (Supplementary Fig. 7a) indicate that the m⁶(syn)A···U ES likely retains a weaker (m⁶A6)N1···H-N3(U13) Watson–Crick H-bond, although we cannot rule out that the H-bond is mediated by water (see Supplementary Fig. 7e). Similar chemical-shift agreement including for Δω_H3 was obtained for m⁶₂A in dsA6RNA (Supplementary Fig. 7b).

Taken together, these data provide strong support for a singly H-bonded m⁶(syn)A···U bp (Fig. 4e), which is distinct from the bp open state (Supplementary Fig. 8 and Supplementary Note 2). To our knowledge, this alternative m⁶A-specific conformational state has not been documented previously.

m⁶(syn)A···U behaves like a mismatch

Although we initially dismissed hybridization pathways in which the major syn isomer hybridizes to form a dsRNA intermediate, our data indicate that this is indeed possible because m⁶A can pair with uridine to form the m⁶(syn)A···U conformation. Several lines of evidence indicate that m⁶(syn)A···U behaves like a mismatch when it comes to hybridization kinetics.

Like many mismatches⁴⁰, m⁶(syn)A···U loses a H-bond and is destabilized relative to the Watson–Crick m⁶(anti)A–U bp by ~3 kcal/mol. In addition, based on the three-state fit of the RD data measured for ${{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ at T = 55 °C (Fig. 3b), the m⁶(syn)A···U containing duplex intermediate anneals at a ~20-fold slower rate compared to the unmethylated control, whereas it melts with an ~80-fold faster rate. These changes in hybridization kinetics relative to the unmethylated control are also in line with those previously reported when introducing single mismatches to dsRNA^22,23,24.

We were able to verify the mismatch-like hybridization kinetics of m⁶(syn)A···U containing duplex by using NMR RD to measure the hybridization kinetics of the ${{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}_{2}^{6}{{{{{\mathrm{A}}}}}}}$ ES mimic (Supplementary Fig. 7d). For ${{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}_{2}^{6}{{{{{\mathrm{A}}}}}}}$, k_on was ~16-fold slower, whereas k_off was ~100-fold faster relative to the unmethylated RNA. Therefore, depending on the isomer, m⁶A can behave either like a Watson–Crick (anti) or mismatch (syn) when paired to the same partner uridine.

Kinetic model for m⁶A hybridization via conformation selection and IF

The RD data measured for ${{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ at T = 55 °C provided direct evidence for hybridization via an IF pathway. However, the standalone IF pathway fails to account for the data measured at both 65 °C (Supplementary Fig. 3c) and 55 °C (Supplementary Fig. 4e) based on constrained fits. Since the RD data measured at T = 65 °C is consistent with hybridization via CS, with no evidence for flux along IF, we tested a general model that includes both pathways (CS + IF) (Fig. 5a).

**Fig. 5: Testing a four-state CS + IF kinetic model.**

We used the four-state CS + IF model along with the exchange parameters (Δω, R₁, and R₂ values) determined independently (“Methods”) to simulate the RD data measured for ${{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ at T = 55 °C. The exchange parameters associated with isomerization in ssRNA were again fixed to the values obtained from temperature-dependent RD measurements on ${{{{{\mathrm{ssGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ (Supplementary Fig. 2c). ${k}_{{{{{{\mathrm{off}}}}}},{anti}}$ was again assumed equal to ${k}_{{{{{{\mathrm{off}}}}}}}$ and ${k}_{{{{{{\mathrm{on}}}}}},{anti}}$ was deduced by using the melting free energy obtained from RD measurements (see “Methods”) (Fig. 2a). ${k}_{{{{{{\mathrm{on}}}}}},{syn}}$ and ${k}_{{{{{{\mathrm{off}}}}}},{syn}}$ describing the hybridization of ssRNA^syn and methyl isomerization in dsRNA were fixed to the values obtained from the three-state fit of the RD data for ${{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ (Fig. 3b).

Indeed, the RD profiles simulated for m⁶A-C2 using the four-state model were in much better agreement (${\chi }_{{{{{{\mathrm{red}}}}}}}^{2}$ = 10.7) (Supplementary Fig. 9a) with the experimental data relative to simulations using the CS model (${\chi }_{{{{{{\mathrm{red}}}}}}}^{2}$ = 563.7) (Supplementary Fig. 4c) or constrained three-state fits to the CS model (${\chi }_{{{{{{\mathrm{red}}}}}}}^{2}$ = 43.3) (Fig. 3c). A constrained fit of the RD data to the four-state model (see “Methods”) improved the agreement further (${\chi }_{{{{{{\mathrm{red}}}}}}}^{2}$ = 9.6) (Supplementary Fig. 9a) to a level comparable to the three-state fit (Fig. 3a). The ${\chi }_{{{{{{\mathrm{red}}}}}}}^{2}$ values from globally fitting both m⁶A-C2 and m⁶A-C8 show similar trends (Fig. 5b).

These results provide a plausible explanation for how m⁶A selectively slows ${{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ annealing at T = 55 °C via both the CS and IF pathways. Based on optimized kinetic rate constants obtained from the constrained four-state fit of the RD data, the flux (see “Methods”) was ~50:50 through the CS and IF pathways at T = 55 °C (Fig. 5c). Along the CS pathway, m⁶A reduces the apparent rate of annealing due to the ~20-fold lower population of the ssRNA^anti intermediate. However, as described for the data measured at T = 65 °C, m⁶A does not affect melting because the dominant isomer in the duplex is anti, which behaves similarly to unmethylated adenine. Along the IF pathway, m⁶A reduces the apparent rate of annealing by 20-fold because m⁶(syn)A···U behaves as a mismatch, reducing hybridization rate to form the dsRNA^syn intermediate by 20-fold. Like a mismatch-containing duplex, this intermediate melts at a rate ~100-fold faster relative to the unmethylated duplex. However, the intermediate does not accelerate the apparent melting rate of the methylated duplex along the IF pathway relative to the unmethylated control because its equilibrium population is only ~1%.

We also reanalyzed the RD data measured at T = 65 °C and obtained good agreement with the constrained four-state fit (${\chi }_{{{{{{\mathrm{red}}}}}}}^{2}$ = 3.0) (Fig. 5b). The level of agreement is similar to that obtained using the constrained three-state fit to the CS model (Fig. 2f), which is expected considering that majority (90%) of the flux is through the CS pathway (Fig. 5c). The smaller flux along the IF pathway at 65 versus 55 °C can be attributed to a slower annealing rate along the IF pathway at 65 °C due to a 2-fold reduction in the population of the ssRNA^syn relative to ssRNA^anti and comparatively 2.5-fold slower annealing rate constant of ssRNA^syn along the IF pathway relative to ssRNA^anti along the CS pathway.

A quantitative model predicts how m⁶A reshapes the hybridization kinetics of DNA and RNA duplexes

To test the generality and robustness of our proposed mechanism, we developed and tested a quantitative CS + IF model that predicts how methylating a central adenine residue impacts the hybridization kinetics for any duplex. The model assumes that the temperature-dependent isomerization kinetics in ssRNA and dsRNA does not vary, consistent with the small deviations (<2-fold) seen with sequence, as supported by our data (Supplementary Table 1). The model assumes that ${k}_{{{{{{\mathrm{off}}}}}},{anti}}$ = ${k}_{{{{{{\mathrm{off}}}}}}}$ and ${k}_{{{{{{\mathrm{on}}}}}},{anti}}$ is deduced based on the known energetics of annealing the m⁶A-containing duplex. The value of ${k}_{{{{{{\mathrm{on}}}}}},{syn}}$ was assumed to be 20-fold slower than the unmethylated RNA and ${k}_{{{{{{\mathrm{off}}}}}},{syn}}$ was then deduced by closing the thermodynamic cycle (see “Methods”). Using these rate constants and the CS + IF model, kinetic simulations (see “Methods”) were used to predict ${k}_{{{{{{\mathrm{on}}}}}},{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}^{{{{{{\mathrm{app}}}}}}}$ and ${k}_{{{{{{\mathrm{off}}}}}},{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}^{{{{{{\mathrm{app}}}}}}}$.

We used the model to predict the ${k}_{{{{{{\mathrm{on}}}}}},{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}^{{{{{{\mathrm{app}}}}}}}$ and ${k}_{{{{{{\mathrm{off}}}}}},{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}^{{{{{{\mathrm{app}}}}}}}$ values recently reported²¹ for two duplexes (${{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ and ${{{{{\mathrm{dsHCV}}}}}}{{{{{\mathrm{m}}}}}}^{6}{{{{{\mathrm{A}}}}}}$) under a range of different salt (Mg²⁺ and Na⁺) concentrations and temperatures and for an additional dataset involving ${{{{{\mathrm{dsHCV}}}}}}{{{{{\mathrm{m}}}}}}^{6}{{{{{\mathrm{A}}}}}}$ at T = 55 °C in 3 mM Mg²⁺ (Supplementary Fig. 5). Across these duplexes and conditions, m⁶A slowed the apparent annealing by ~5-fold to ~20-fold while minimally impacting the melting rate (<2-fold). As shown in Fig. 6a, a good correlation (R² = 0.8–0.9) was observed between the measured and predicted ${k}_{{{{{{\mathrm{on}}}}}},{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}^{{{{{{\mathrm{app}}}}}}}$ and ${k}_{{{{{{\mathrm{off}}}}}},{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}^{{{{{{\mathrm{app}}}}}}}$, as well as the overall impact on the apparent annealing and melting rates induced by methylation, with all deviations being <1.5-fold.

**Fig. 6: Testing the predictive power of the CS + IF model.**

In all the above examples, the equilibrium flux was primarily (~50–95%) via the CS pathway. The differences in the m⁶A-induced slowdown (~5–20-fold) of annealing across different duplexes are primarily driven by differences in the annealing rate of ssRNA^anti along the CS pathway relative to that of unmethylated RNA, with the slowdown being more substantial the more stable the unmethylated duplex (Supplementary Fig. 9c). It should be noted that the slowdown is predicted to be even more substantial when hybridization is fast and isomerization of methylamino group becomes rate-limiting, as observed for an RNA conformational transition, as described below.

As an additional test, we used the model to predict the impact of m⁶A on the apparent hybridization kinetics of an A-rich duplex DNA (dsA6DNA, Supplementary Fig. 5). Based on the unmethylated duplex hybridization kinetics measured previously²¹, the model predicts that m⁶A should reduce the apparent ${k}_{{{{{{\mathrm{on}}}}}},{{{{{\mathrm{m}}}}}}^{6}{{{{{\mathrm{A}}}}}}}^{{{{{\mathrm{app}}}}}}$ by ~6-fold while having a little effect (<2-fold) on ${k}_{{{{{{\mathrm{off}}}}}},{{{{{\mathrm{m}}}}}}^{6}{{{{{\mathrm{A}}}}}}}^{{{{{\mathrm{app}}}}}}$. We used NMR RD measurements (Supplementary Fig. 5) on methylated dsA6DNA to test these predictions and the results show that m⁶A reduces ${k}_{{{{{{\mathrm{on}}}}}},{{{{{\mathrm{m}}}}}}^{6}{{{{{\mathrm{A}}}}}}}^{{{{{\mathrm{app}}}}}}$ by ~8-fold while having a little effect (<2-fold) on ${k}_{{{{{{\mathrm{off}}}}}},{{{{{\mathrm{m}}}}}}^{6}{{{{{\mathrm{A}}}}}}}^{{{{{\mathrm{app}}}}}}$, in good agreement with the predictions (Fig. 6a).

Finally, we extended our model to also predict NMR CEST data by imposing additional constraints on NMR exchange parameters (Δω, R₁, and R₂) needed to simulate the RD data (see “Methods”). In addition to providing a rationale for the kinetic basis of the m⁶A-induced hybridization slow down, such a model would also validate the existence of the IF and CS intermediates in diverse sequence contexts under a variety of experimental conditions. Thus, we subjected all of the above RD data to a constrained four-state fit to the CS + IF model. A reasonable fit (${\chi }_{{{{{\mathrm{red}}}}}}^{2}$ ~ 3.5–14) could be obtained in all cases (Supplementary Fig. 9d). This suggests that m⁶A-induced hybridization slowdown in DNA is likely mediated by similar IF and CS intermediates as RNA.

Testing kinetic model on RNA conformational transitions

Beyond duplex hybridization, our kinetic model predicts that m⁶A should also slow intramolecular conformational dynamics in which m⁶A transitions between an unpaired conformation, in which the methylamino group is predominantly syn, and a paired conformation, in which the methylamino group is predominantly anti. In addition, the model predicts that the slowdown can be much more substantial for conformational transitions that are much faster than the hybridization kinetics measured under our experimental conditions.

To test these predictions, we methylated A35 in the apical loop of transactivation response element (TAR) (Fig. 6b) from human immunodeficiency virus type-1 (HIV-1)⁴¹ and examined whether m⁶A reduces the rate constant of a previously described intramolecular conformational transition in which unpaired A35 in the GS forms a wobble A35⁺-C30 mismatch in the ES⁴². As in the Watson–Crick A–U bp, the methylamino group needs to be anti to form one of the H-bonds in the m⁶A⁺-C wobble (Fig. 6b). TAR therefore also allowed us to test the generality of the model to non-Watson–Crick bps.

We prepared a TAR NMR sample containing m⁶A35 and ¹³C8-labeled G34 as an RD probe⁴². Based on the chemical-shift perturbations, m⁶A destabilized the TAR ES relative to the GS by ~2 kcal/mol, in a manner analog to duplex destabilization¹² (Supplementary Fig. 10a and “Methods”). The CS + IF kinetic model predicts that m⁶A will reduce k_ex, k_forward, and k_backward for the TAR conformational transition by ~17-, ~400-, and ~14-fold, respectively. The much greater m⁶A induced reduction in the forward rate constant relative to hybridization arises because the TAR conformational transition is intrinsically faster, and this pushes the isomerization step in the dominant CS pathway away from equilibrium, leading to a slowdown much greater than that due to the equilibrium population (~10%) of the ssRNA^anti CS intermediate when hybridization is limiting. Here, the IF pathway is highly disfavored (flux < 1%) because the ES with m⁶A in the syn conformation is predicted to be highly energetically disfavored.

Based on NMR RD measurements (Supplementary Fig. 10b), m⁶A reduced k_ex, k_forward, and k_backward by ~15-, ~300-, and ~12-fold in very good agreement with predictions from our model (Fig. 6c). The TAR experimental RD data could be satisfactorily fit to a constrained three-state fit to the CS model with ${\chi }_{{{{{{\mathrm{red}}}}}}}^{2}$ = 0.2 (Supplementary Fig. 10c) comparable to that obtained from an unconstrained two-state fit. These results indicate that m⁶A can also slow down RNA conformational transitions and potentially to a much greater degree than observed in our duplex hybridization experiments.

As a negative control, m⁶A minimally (<2-fold) affects the exchange rate of conformational transition in the HIV-1 Rev response element stem IIB (RREIIB; Fig. 6d)⁴³ in which the m⁶A remains unpaired in the two conformations (Fig. 6e, Supplementary Fig. 10d, and Supplementary Note 3).

Discussion

Our results help explain how m⁶A selectively and robustly slows annealing while minimally impacting the rate of duplex melting under our experimental conditions. The minor ssRNA^anti isomer hybridizes with kinetic rate constants similar to unmethylated adenine. m⁶A slows the apparent annealing rate along the CS pathway relative to the unmethylated control due to the low equilibrium population of the ssRNA^anti isomer. Once in a duplex, anti is the dominant isomer and m⁶A does not substantially impact the apparent rate of duplex melting along the CS pathway. The major ssRNA^syn isomer can also hybridize via an IF pathway to form a singly H-bonded bp and with kinetic rate constants similar to that of a mismatch-containing duplex. This intermediate forms slowly, explaining why m⁶A also slows the apparent annealing rate along the IF pathway. However, because its equilibrium population is only ~1%, the intermediate does not accelerate the apparent melting rate along the IF pathway. While we have focused on relatively short duplexes with m⁶A located in the middle, the impact of the modification on the hybridization kinetics will likely vary and be diminished when placed near the terminal ends, as observed for mismatches²².

By treating the two m⁶A isomers as two modular elements that have Watson–Crick or mismatch-like kinetic properties independent of sequence context⁴⁴, we were able to build a model that can predict the impact of m⁶A on the overall hybridization kinetics and RNA conformational dynamics from component reactions. The power of such a quantitative and predictive kinetic model is that it obviates the need to carry out time-consuming kinetics experiments to measure the universe of kinetics data that is of biological interest. For example, when combined with an existing computational model that can predict the hybridization kinetics of unmethylated DNA duplexes from sequence⁴⁵, our model could be used to predict how a central m⁶A impacts the hybridization kinetics of any arbitrary DNA duplex. This allowed us to predict the impact of m⁶A on hybridization kinetics for all ~6000 m⁶A sites reported in the mouse genome⁹ (Fig. 6f). Our model may also aid the design and implementation of studies that harness the kinetic effects of m⁶A as a chemical tool that can bring conformational transitions within detection or aid kinetic studies of RNA and DNA biochemical mechanisms.

Our model also makes a number of interesting biological predictions. The model predicts that m⁶A should slow any process in which the unpaired m⁶A in the predominantly syn isomer has to transition into a conformation in which m⁶A is predominantly anti. This should include all templated processes that create canonical A–U Watson–Crick bps and many mismatches (A⁺ (anti)–C (anti), A (anti)–G(anti), and A⁺ (anti)–G (syn)), in which the methylamino group adopts the anti conformation. m⁶A is found in a variety of RNAs involved in processes that require base pairing, including R-loop formation⁴⁶, microRNA RNA target recognition⁴⁷, snoRNA–pre-rRNA base pairing⁴⁸, snRNA–pre-mRNA base pairing⁴⁹, and the assembly of the spliceosome⁵⁰ and ribosome⁵¹. The model also predicts that the m⁶A-induced slowdown could exceed 1000-fold for fast conformational transitions such as the folding of short hairpins and this could have important consequences on RNA folding, conformational switches, RNA protein recognition, and processes that occur co-transcriptionally. Further studies are needed to examine whether m⁶A does indeed slow these processes and whether this has any biological consequences.

Our NMR measurements had to be performed under high-temperature conditions so that hybridization falls within the detection limits of RD. However, we were able to observe isomerization of the methylamino group in both ssRNA and dsRNA at T = 37 °C (Supplementary Figs. 2b and 6a). Based on the temperature dependence of the hybridization steps in the CS and IF pathways, our model predicts (see “Methods”) that m6A will slow down annealing by ~5-fold while minimally impacting the melting rate consistent with our measurements at higher temperatures. A comparable level of the slowdown in annealing is also obtained when predicting the m⁶A-induced slowdown at T = 37 °C using rate constants for hybridization of unmethylated RNA reported previously²² at T = 37 °C and assuming that m6A destabilizes dsRNA by 1 kcal/mol⁵² (see “Methods”).

The mismatch-like m⁶(syn)A···U bp is interesting not only because of its role in hybridization kinetics but also because it could potentially prime the methylamino group for recognition by reader proteins, which recognize the methylamino group in a syn conformation⁵³. Upon surveying ~50,000 unmethylated A–U bps in Protein Data Bank (PDB), we found 428 bps that share the conformational signatures of the singly H-bonded m⁶A···U bp (see “Methods”). More than 60% of these bps are found in noncanonical regions, such as junctions, terminal ends, tertiary structural elements, and protein-bound RNA (Supplementary Fig. 7f). It will be interesting to examine whether the mismatch-like m⁶(syn)A···U forms as the dominant conformation in certain structural contexts where it may facilitate recognition by reader proteins both by locally destabilizing the bp, so that m⁶A is more accessible and by adopting a preformed syn conformation.

Methods

Sample preparation

AMP and m⁶AMP

Unlabeled adenosine and N⁶-methyladenosine 5ʹ-monophosphate monohydrate (AMP and m⁶AMP) were purchased from Sigma-Aldrich (A2252 and M2780). Powders were directly dissolved in NMR buffer (25 mM sodium chloride, 15 mM sodium phosphate, 0.1 mM EDTA, and 10% D₂O at pH 6.8 with or without 3 mM Mg²⁺). The final concentrations of AMP and m⁶AMP were 50 mM.

Oligonucleotides

Unmethylated, methylated (N⁶-methylated adenosine, N⁶,N⁶-dimethyl adenosine), and ¹³C- or ¹⁵N-site-labeled (¹⁵N3-labeled U, ¹³C8,¹³C2-labeled A/m⁶A, and ¹³C10-labeled m⁶A) RNA oligonucleotides were synthesized using a MerMade 6 Oligo Synthesizer employing 2ʹ-tBDSilyl-protected phosphoramidites and 1 μmol standard synthesis columns (1000 Å) (BioAutomation). Unlabeled m⁶A, m⁶₂A, rU, and n-acetyl-protected rC, rA, and rG phosphoramidites were purchased from Chemgenes. ¹⁵N3-labeled U and ¹³C8,¹³C2-labeled rA/m⁶A phosphoramidites were synthesized in-house according to published procedures^21,36. ¹³C10-labeled m⁶A phosphoramidite was synthesized as described in Supplementary Note 4. RNA oligonucleotides were synthesized with the option to retain the final 5ʹ-protecting group, 4,4ʹ-dimethoxytrityl (DMT). Synthesized oligonucleotides were cleaved from columns using 1 ml AMA (1:1 ratio of 30% ammonium hydroxide and 30% methylamine), followed by 2-h incubation at room temperature. The solution was then air-dried and dissolved in 115 μl dimethyl sulfoxide, 60 μl triethylamine (TEA), and 75 μl TEA.3HF, followed by 2.5-h incubation at T = 65 °C for 2ʹ-O deprotection. The solutions were then quenched using Glen-Pak RNA quenching buffer and loaded onto Glen-Pak RNA cartridges (Glen Research Corporation) for purification and subsequently ethanol precipitated. Following ethanol precipitation, RNA oligonucleotides were dissolved in water (200–500 μM for duplex samples, 50 μM for hairpin samples) and annealed by heating an equimolar amount of complementary single strands or hairpins at T = 95 °C for 10 min, followed by cooling at room temperature for 2 h for duplex samples or 30 min on ice for hairpin samples. Extinction coefficients for concentration calculation were obtained from the atdbio online calculator (https://www.atdbio.com/tools/oligo-calculator). The extinction coefficients for modified single strands were assumed to be equal to that of their unmodified counterparts. All samples were buffer exchanged using centrifugal concentrators (Amicon Ultra-15 3-kDa cut-off EMD Millipore) into NMR buffer (25 mM sodium chloride, 15 mM sodium phosphate, 0.1 mM EDTA, and 10% D₂O at pH 6.8 with or without 3 mM Mg²⁺).

The ¹³C8,¹³C2-labeled m⁶dA ssA6DNA oligonucleotide was synthesized in-house using a MerMade 6 oligo synthesizer. The ¹³C8,¹³C2-labeled m⁶dA phosphoramidite was synthesized as described in Supplementary Note 5. Standard DNA phosphoramidites (n-ibu-dG, bz-dA, ac-dC, and dT) were purchased from Chemgenes. DNA oligonucleotides were synthesized with the option to retain the final 5ʹ-DMT group. Synthesized oligonucleotides were cleaved from columns using 1 ml AMA, followed by 2-h incubation at room temperature. The DNA sample was then purified using Glen-Pak DNA cartridges and ethanol precipitated. The complementary ssDNA of the m⁶A containing ssDNA is uniformly ¹³C/¹⁵N labeled and was synthesized and purified by in vitro primer (see Supplementary Table 8) extension⁵⁴ using ¹³C/¹⁵N isotopically labeled dNTPs (Silantes), and purified using 20% 29:1 polyacrylamide denaturing gel with 8 M urea, 20 mM Tris borate, and 1 mM EDTA, and electroelution (Whatmann, GE Healthcare) in 40 mM Tris acetate and 1 mM EDTA. DNA duplexes were prepared and buffer exchanged in a manner analogous to that described above for RNA duplexes.

Definition of rate constants

1.
${k}_{1}$ and ${k}_{-1}$ are the forward and backward rate constants for methylamino isomerization in ssRNA, respectively.
2.
${k}_{2}$ and ${k}_{-2}$ are the forward and backward rate constants for methylamino isomerization in dsRNA, respectively.
3.
${k}_{{{{{\mathrm{on}}}}}}$ and ${k}_{{{{{\mathrm{off}}}}}}$ are the annealing and melting rate constants, respectively, for unmethylated RNA.
4.
${k}_{{{{{{\mathrm{on}}}}}},{{{{{\mathrm{anti}}}}}}}$ and ${k}_{{{{{{\mathrm{off}}}}}},{{{{{\mathrm{anti}}}}}}}$ are the annealing and melting rate constants, respectively, when m⁶A adopts anti conformation in both ssRNA and dsRNA.
5.
${k}_{{{{{{\mathrm{on}}}}}},{{{{{\mathrm{syn}}}}}}}$ and ${k}_{{{{{{\mathrm{off}}}}}},{{{{{\mathrm{syn}}}}}}}$ are the annealing and melting rate constants, respectively, when m⁶A adopts syn conformation in both ssRNA and dsRNA.
6.
${k}_{{{{{{\mathrm{on}}}}}},{{{{{\mathrm{m6A}}}}}}}^{{{{{\mathrm{app}}}}}}$ and ${k}_{{{{{{\mathrm{off}}}}}},{{{{{\mathrm{m6A}}}}}}}^{{{{{\mathrm{app}}}}}}$ are the apparent annealing and melting rate constants, respectively, for m⁶A-methylated RNA.
7.
${k}_{{{{{\mathrm{forward}}}}}}$ and ${k}_{{{{{\mathrm{backward}}}}}}$ are the forward and backward rate constants, respectively, for conformational transitions measured using RD.

NMR experiments

Resonance assignments

All NMR experiments (except for the imino proton exchange experiment) were performed on a Bruker Avance III 600 MHz spectrometer equipped with a 5 mm triple-resonance HCPN cryogenic probe. Resonance assignments for ${{{{{\mathrm{hpGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ have been reported previously³⁶. Resonance assignments for m⁶₂A-modified dsGGACU and dsA6 were obtained using 2D [¹H,¹H] nuclear magnetic resonance spectroscopy experiments with 150 ms mixing time along with 2D [¹³C, ¹H] and [¹⁵N, ¹H] Heteronuclear single quantum coherence spectroscopy (HSQC) experiments. The assignments for ${{{{{\mathrm{ssGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$, ${{{{{\mathrm{ssA6RNA}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$, dsGGACU A/m⁶A, ${{{{{\mathrm{dsA6DNA}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$, and dsHCV A/m⁶A could be readily obtained since the samples were site-specifically labeled. The assignments for AMP and m⁶AMP were obtained from a prior study²⁵ (Supplementary Fig. 1). Data were collected using TopSpin 3.2 (Bruker), processed using NMRpipe software package⁵⁵, and analyzed using SPARKY (T.D. Goddard and D.G. Kneller, SPARKY 3, University of California, San Francisco).

¹³C and ¹⁵N R _1ρ relaxation dispersion

¹³C and ¹⁵N R_1ρ experiments were performed using 1D R_1ρ schemes as described previously^56,57,58. The spin-lock powers (ω/2π Hz) and offsets (Ω_eff/2π Hz, where Ω_eff = ω_obs − ω_rf, where ω_obs is the Larmor frequency of the spin and ω_rf is the carrier frequency of the applied spin-lock) are listed in Supplementary Table 5. The spin-lock was applied for a maximal duration (<120 ms for ¹⁵N and <60 ms for ¹³C) to achieve ~70% loss of peak intensity at the end of the relaxation delay.

Analysis of R _1ρ data

1D peak intensities were measured using NMRpipe⁵⁵. R_1ρ values for a given spin-lock power and offset combination were calculated by fitting the intensities as a function of delay time to a monoexponential decay³⁴. A Monte Carlo approach was used to calculate R_1ρ uncertainties⁵⁹. Alignment of initial magnetization during the B–M fitting was performed based on the k_ex/|Δω_major| ratio (k_ex/|Δω_major| ≥ 1 and k_ex/|Δω_major | > 1 corresponding to GS alignment and AVG alignments, respectively)¹⁸. Chemical exchange parameters were obtained by fitting experimental R_1ρ values to numerical solutions of the B–M) equations³² that describe N-site chemical exchange³⁴. Errors in exchange parameters were determined using a Monte Carlo approach³⁴. When available, R_1ρ data measured for the same exchange process under the same condition were globally fitted, sharing ES population and exchange rate constants. Reduced χ² (${\chi }_{red}^{2}=\mathop{\sum }\nolimits_{i=1}^{N}{\left( \frac{{R}_{1\rho (i)}^{Calc}-{R}_{1\rho (i)}^{exp}}{{\sigma }_{\exp (i)}}\right)}^{2}$, ${R}_{1\rho (i)}^{Calc}$, and ${R}_{1\rho (i)}^{exp}$ are experimentally measured and calculated R_1ρ data using the B–M equations, ${\sigma }_{\exp (i)}$ is the experimental uncertainty in R_1ρ determined using a Monte Carlo approach) was calculated to assess the goodness of fitting¹⁸. In general, similar exchange parameters were obtained from individual fitting and global fitting. All exchange parameters are summarized in Supplementary Table 1.

Estimate p _ES of methylated TAR from chemical shifts

The RD signal of methylated TAR is weak probably due to small p_ES and fast k_ex (Supplementary Fig. 10b). We used chemical-shift perturbation-based method⁶⁰ as an alternative approach to estimate the population of ES⁴² (${p}_{{ES},{{m}^{6}{A}}}$) of methylated TAR. Specifically, in methylated TAR, ${{{{{\mathrm{\omega}}}}}}_{obs}={{{{{\mathrm{\omega}}}}}}_{GS}\times (1-{p}_{{ES},{{m}^{6}{A}}})+{{{{{\mathrm{\omega}}}}}}_{ES}\times {p}_{{p}_{{ES},{{m}^{6}{A}}}}$. ${{{{{\mathrm{\omega}}}}}}_{GS}$ and ${{{{{\mathrm{\omega}}}}}}_{ES}$ are chemical shifts of GS and ES of unmethylated TAR and were determined previously⁶⁰. Based on 2D [¹³C, ¹H] HSQC spectra, G34-C8 peak shifts toward GS (Supplementary Fig. 10a), and the calculated ${p}_{{ES},{{m}^{6}{A}}}$ is ~1%.

¹³C and ¹⁵N CEST

¹³C and ¹⁵N CEST experiments were performed using 1D schemes without equilibration of GS and ES magnetization prior to the relaxation delay²¹. The RF field strengths (ω/2π Hz) and offset combinations (${{{\Omega}}/2{\pi}}$ Hz, where Ω = ω_rf − ω_obs) used in CEST measurements are listed in Supplementary Table 6. The relaxation delay for all CEST experiments was 200 ms.

2D CEST for ¹³C methyl probes

The pulse sequence for the ¹³C methyl CEST was derived by modifying the 2D CEST experiment for ¹³C from Zhang and co-workers²⁹ in accordance with considerations described in Bouvignies and Kay³¹ outlining a 2D CEST experiment for ¹³C methyl groups. The following changes were made to the CEST experiment from Zhang and co-workers²⁹:

Given that the samples for methyl CEST in this study were site specifically ¹³C labeled at the methyl group, we removed shaped pulse c that was used to refocus carbon–carbon scalar couplings.
The delay τ between ¹³C pulses of phases ϕ2 and ϕ3 and ϕ3 and ϕ5 was set to be as close as possible to the optimal value of τ $=\,\;\scriptstyle\frac{\arccos(\sqrt{2/3})}{2\pi{J}_{HC}}$, where J_HC is the scalar coupling between the methyl carbon and protons, and for optimal transfer of in-phase methyl carbon magnetization to antiphase, as described by Bouvignies and Kay, while ensuring that the delays between the pulses in the sequence were positive. J_HC was measured using an F1-coupled 2D [¹³C, ¹H] HSQC experiment.
The τ delay flanking shaped pulse b was set to be equal to $\scriptstyle\frac{\arccos(\sqrt{1/3})}{2\pi{J}_{HC}}$. The duration of shaped pulse b was shortened as needed so as to ensure that the delays between the pulses in the sequence were positive.
A gradient pulse was inserted between the ¹³C and ¹H π/2 pulses after T1 evolution, as described by Bouvignies and Kay³¹, to purge transverse magnetization.

Analysis of the CEST data

1D or 2D peak intensities were calculated using NMRpipe⁵⁵. The intensity error for all offsets for a given spin-lock power was set to be equal to the standard deviation of three measurements of peak intensity with zero relaxation delay under the same spin-lock power. The intensities were normalized to the average intensity of the three zero delay measurements. Exchange parameters were then obtained by fitting experimental intensity values to numerical solutions of the B–M equations and RF field inhomogeneity was taken into account during CEST fitting⁶¹. No equilibration of GS magnetization was assumed when integrating the B–M equations for non-methyl probes⁶¹, while equilibration was assumed for the methyl CEST given that the sequence employs nonselective hard pulses. Fits of CEST data were carried out assuming unequal R₂ or assuming equal R₂ for duplex melting²¹ and other ES measurements, respectively. Alignment of the initial magnetization during CEST fitting was chosen based on the k_ex/|Δω_major| ratio as described in the previous section⁶¹. Errors in exchange parameters were determined using a Monte Carlo approach with 500 iterations⁶². Global fitting of CEST data was carried out for the same exchange process under identical conditions. ${\chi }_{{{{{\mathrm{red}}}}}}^{2}$ was calculated to assess the goodness of fitting as described in the previous section¹⁸. Note that the different ${\chi }_{{{{{\mathrm{red}}}}}}^{2}$ values for different fits are most likely due to differences in the quality of the NMR data and poor estimation of the real experimental uncertainty (Supplementary Table 1). Model selection (three state with triangular, linear, or starlike topology; Supplementary Fig. 4d) was carried out by calculating Watanabe–Akaike information criterion and Watanabe–Bayesian information criterion weights for each model and selecting the model with the highest relative probability³⁴.

¹H CEST experiment

A transverse relaxation optimized spectroscopy-based spin-state selective ¹H CEST experiment⁶³ was carried as described previously⁶⁴. The power of the B₁ field was set to be 60 or 120 Hz and the offset of the B₁ field ranged from 8.5 p.m. to 15.5 p.p.m. with a step of 30 Hz. The relaxation delay was 400 ms. The ¹H CEST data were collected in a pseudo-3D mode and were analyzed using NMRPipe⁵⁵. The intensities in the N^α and N^β CEST profiles were normalized to a reference intensity with B₁ frequency = −20 p.p.m. The N^β CEST profile was then subtracted from the N^α CEST profile to result in a difference CEST profile, from which the Δω of the ES was fitted with predetermined fitting parameters such as p_ES, k_ex, and ¹⁵N R₁ from the ¹³C/¹⁵N R_1ρ experiments. Errors in the CEST intensity profiles were estimated based on the scatter in regions of 1D profiles that did not contain any intensity dips. The Python package ChemEx (https://github.com/gbouvignies/chemex) is used to carry out the fitting.

Imino proton exchange experiment

Experiments were carried out on a 700 MHz Bruker NMR spectrometer equipped with hydrogen cyanide room-temperature probe to measure the proton exchange between imino proton and water⁶⁵, following the same pulse programs and protocols as described in a prior study⁶⁶. Briefly, the water proton longitudinal relaxation rate constant R₁ was first measured using a standard saturation-recovery method⁶⁶. A pre-saturation pulse was used for solvent suppression. The relaxation delay time for measuring water proton R₁ was set to be 0.0, 0.4, 0.8, 1.2, 1.6, 2.0, 2.4, 2.8, 3.2, 3.6, 4.0, 4.4, 4.8, 5.2, 6.0, 7.0, 8.0, 9.0, 10.0, 12.0 and 15.0 s. The apparent solvent exchange rate constant of the imino protons was then measured using an inversion-recovery scheme by initially selectively inverting the bulk water magnetization, followed by detecting transfer of the water magnetization to the imino proton during the solvent exchange. A sinc-shaped π-pulse was optimized and used to invert the water magnetization. A binominal water-suppression scheme was used to suppress water. The delay times used to measure water and imino proton exchange rate constants are listed in Supplementary Table 7.

The apparent exchange rate (${k}_{ex}$) of imino and water proton was obtained by fitting the imino magnetization as a function of exchange time upon solvent exchange according to Eq. (1),

$$W(t)={W}^{0}-E\times {W}^{0}\times \frac{{k}_{ex}}{{R}_{1{w}}-{R}_{1{n}}}\times ({e}^{-{R}_{1{n}}\times t}-{e}^{-{R}_{1{w}}\times t})$$

(1)

where $W(t)$ is the imino peak volume as a function of exchange time $t$, ${W}^{0}$ is the initial peak volume (at t = 0 s), $E$ is the efficiency of the inversion pulse, ${k}_{{{{{\mathrm{ex}}}}}}$ is the apparent solvent exchange rate constant between imino and water proton, ${R}_{1{{{{{\mathrm{w}}}}}}}$ is water proton R₁, ${R}_{1{{{{{\mathrm{n}}}}}}}$ is the summation of imino proton R₁, and exchange rate constant ${k}_{{{{{\mathrm{ex}}}}}}$. In the equation, ${R}_{1{{{{{\mathrm{w}}}}}}}$ and $E$ values are fixed parameters that are predetermined, while ${k}_{{{{{\mathrm{ex}}}}}}$ and ${R}_{1{{{{{\mathrm{n}}}}}}}$ are fitted parameters. The error of the fitted parameters is the standard fitting error, which is the square root of the diagonal elements of the covariance matrix. The efficiency of the selective shaped pulse used for water inversion (E) was calculated by Eq. (2):

$$E=1-\frac{{W}_{inv}}{{W}_{eq}}$$

(2)

where ${W}_{{{{{\mathrm{inv}}}}}}$ and ${W}_{{{{{\mathrm{eq}}}}}}$ represent the peak volumes of the water proton with and without the shaped pulse for inversion, respectively (at zero delay time and without binominal water suppression).

Determining the methylamino isomerization rate constants from temperature-dependent RD measurements for methylated ssRNA and dsRNA. The observed temperature dependence of k₁, k₋₁ in m⁶AMP and ssRNA (Supplementary Fig. 2c) and k₂, k₋₂ in dsRNA (Supplementary Fig. 6d) determined using RD were fit to a modified van’t Hoff equation that accounts for statistical compensation effects and assumes a smooth energy surface⁵⁷:

$${ln}\left (\frac{{k}_{i}(T)}{T}\right)=\,{ln}\left(\frac{{k}_{B}\kappa }{h}\right)-\frac{{\Delta }{G}_{i}^{^\circ \,T}({T}_{hm})}{R{T}_{hm}}-\frac{{\Delta} {H}_{i}^{^\circ T}}{R}\left(\frac{1}{T}-\frac{1}{{T}_{hm}}\right)$$

(3)

where k_i (i = 1, −1 or 2, −2) is the rate constant, ${{\Delta}} {G}_{i}^{^\circ T}$ and ${{\Delta}} {H}_{i}^{^\circ T}$ are the free energy and enthalpy of activation (i = 1, 2) or deactivation (i = −1, −2), respectively, R is the universal gas constant (kcal/mol/K), T is temperature (K), and T_hm is the harmonic mean of the experimental temperatures (${T}_{i}$ in K) computed as ${T}_{{{{{\mathrm{hm}}}}}}=n/\mathop{\sum }\limits_{i=1}^{n}(1/{T}_{i})$, k_B is the Boltzmann’s constant, $\kappa$ is the transmission coefficient (assumed to be 1). The goodness-of-fit indicator R² between the measured and fitted rate constants was calculated as follows: ${R}^{2}=1-\frac{S{S}_{res}}{S{S}_{total}},\; S{S}_{res}=\sum {({k}_{i,{fit}}-{k}_{i,{exp}})}^{2}, \; S{S}_{total}=\sum {({k}_{i,{exp}}-\overline{{k}_{i,{exp}}})}^{2}\cdot {k}_{i,{fit}}$, and ${k}_{i,{exp}}$ (i = 1, −1 or 2, −2) are fitted and experimentally measured rate constants. $\overline{{k}_{i,{exp}}}$ is the mean of all ${k}_{i,{exp}}$. Errors of fitting for ${{\Delta}} {G}_{i}^{^\circ \,T}$ and ${{\Delta}} {H}_{i}^{T}$ were calculated as the square root of the diagonal elements of the covariance matrix. Given these fitted ${{\Delta}} {G}_{i}^{^\circ \,T}$ and ${{\Delta}} {H}_{i}^{T}$ values, k_i at T = 55 and 65 °C used for kinetic modeling was computed using Eq. (3).

Measuring the kinetics of duplex hybridization from CEST data

k_off (s⁻¹) and k_on (M⁻¹ s⁻¹) for duplex hybridization were determined based on the forward rate (k_forward) and backward (k_backward) rate constants obtained from a two-state fit of the dsHCV/${{{{{\mathrm{dsHCV}}}}}}{{{{{\mathrm{m}}}}}}^{6}{{{{{\mathrm{A}}}}}}$ A11-C8 and dsA6DNA m⁶A16-C2 RD data (two-state fit of other constructs were reported previously²¹) and a three-state fit of m⁶A-C2 ${{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ at T = 55 °C:

$${k}_{forward}={k}_{off}$$

(4)

$${k}_{backward}={k}_{on}\times [{ss}2]$$

(5)

where $[{{{{{\mathrm{ss}}}}}}2]$ is the free concentration of the complementary single strand.

$${k}_{backward}={k}_{ex}(1-{p}_{ss})$$

(6)

where ${p}_{ss}$ is the single-strand population. The annealing rate constant k_on is given by:

$${k}_{on}=\frac{{k}_{ex}(1-{p}_{ss})}{[{ss}2]}$$

(7)

The uncertainty in $[{{{{{\mathrm{ss}}}}}}2]$, and p_ss and k_ex from CEST measurements were propagated to determine the uncertainty in of k_on. From two-state CEST fit, $[{ss}2]={C}_{{{{{\mathrm{t}}}}}}\times {p}_{{{{{\mathrm{ss}}}}}}$, where ${C}_{{{{{\mathrm{t}}}}}}$ is the total concentration of the duplex, which was obtained using the extinction coefficient as described in the “Sample preparation” section. The uncertainty of ${C}_{{{{{\mathrm{t}}}}}}$ was assumed to be 20%²¹. $[{{{{{\mathrm{ss}}}}}}2]$ from a three-state fit were calculated as described in the energetic decomposition section below.

Validation of NMR RD measurements on m⁶A RNA hybridization

We have previously²¹ shown that hybridization kinetics measured from NMR RD on unmodified DNA and RNA duplexes are consistent with those measured using other techniques employing fluorescence spectroscopy. As an additional test, we performed temperature-dependent RD measurements for ${{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ (Supplementary Fig. 11a). The annealing rate constant k_on did not have a strong temperature dependence, consistent with prior studies reporting non-Arrhenius behavior for k_on in unmodified duplexes^67,68. On the other hand, the melting rate constant k_off showed a strong temperature dependence, which was also consistent with prior studies^67,68. The extrapolated annealing thermodynamic parameters including ${{\Delta}} {{G}}_{{{{{{\mathrm{anneal}}}}}}}^{^\circ }$, ${{\Delta}} {{H}}_{{{{{{\mathrm{anneal}}}}}}}^{^\circ }$, and ${{\Delta}} {{S}}_{{{{{{\mathrm{anneal}}}}}}}^{^\circ }$ measured from NMR experiments are in good agreement with those measured from ultraviolet (UV) melting experiments³⁶ (Supplementary Fig. 11b, c). We also observed a good agreement between the annealing free energy $({{\Delta}} {{G}}_{{{{{{\mathrm{anneal}}}}}}}^{^\circ })$ measured using CEST and UV melting experiments for nine additional DNA/RNA duplexes at temperatures ranging from 45 to 65 °C²¹ (Supplementary Fig. 11d).

UV melting experiments

UV melting experiments were conducted on a PerkinElmer Lambda 25 UV/VIS spectrometer with an RTP 6 Peltier Temperature Programmer and a PCB 1500 Water Peltier System. At least three measurements were carried out for each sample (3 µM in NMR buffer without D₂O) with a volume of 400 µl in a Teflon-stoppered 1 cm path length quartz cell. The absorbance at 260 nm (A₂₆₀) was monitored at temperatures ranging from 15 to 95 °C, and at a ramp rate of 1.0 °C/min. The melting temperature (T_m) and standard enthalpy change (ΔH°) of hybridization reaction for duplexes were obtained by fitting the absorbance of the optical melting experiment to Eqs. (8) and (9)⁶⁹,

$${A}_{260}=(({m}_{{{{{\mathrm{ss}}}}}}\times T)+{b}_{{{{{{\mathrm{ss}}}}}}})\times {p}_{{{{{{\mathrm{ss}}}}}}}+(({m}_{{{{{{\mathrm{ds}}}}}}}\times T)+{b}_{{{{{{\mathrm{ds}}}}}}})\times (1-{p}_{{{{{{\mathrm{ss}}}}}}})$$

(8)

$${p}_{{ss}}=1-\frac{1+4{{e}}^{{\left(\frac{1}{{{T}}_{{m}}}-\frac{1}{{T}}\right)}\frac{{{\Delta} {H}}^{^\circ }}{{R}}}-\sqrt{1+8{{e}}^{{\left(\frac{1}{{{T}}_{{m}}}-\frac{1}{{T}}\right)}\frac{{{\Delta} {H}}^{^\circ }}{{R}}}}}{4{{e}}^{{\left(\frac{1}{{{T}}_{{m}}}-\frac{1}{{T}}\right)}\frac{{{\Delta} {H}}^{^\circ }}{{R}}}}$$

(9)

where ${m}_{{{{{{\rm{ss}}}}}}}$, ${b}_{{{{{{\rm{ss}}}}}}}$, ${m}_{{{{{{\rm{ds}}}}}}}$, and ${b}_{{{{{{\rm{ds}}}}}}}$ are coefficients describing the temperature dependence of the molar extinction coefficient of single strands and double strands, respectively, T is the temperature (K), R is the gas constant (kcal/mol/K), and p_ss is the population of the single strand. Standard entropy change (ΔS°) and ΔG° of double-strand hybridization were therefore computed from Eqs. (10) and (11):

$${{\Delta}} {{S}}^{^\circ }=\frac{{{{\Delta}} {H}}^{^\circ }}{{{T}}_{{{{{{\rm{m}}}}}}}}-R\,{{{{{\mathrm{ln}}}}}}(\frac{{{C}}_{{{{{\mathrm{t}}}}}}}{2})$$

(10)

$${{\Delta}} {{G}}^{^\circ }={{\Delta}} {{H}}^{^\circ }-T{{\Delta}} {{S}}^{^\circ }$$

(11)

where C_t is the total concentration of duplex. The uncertainty in T_m and ΔH° were obtained based on standard deviation in triplicate measurements that were propagated to the uncertainty of ΔS° and ΔG°.

MD simulations

To generate an ensemble of RNA duplexes with different m⁶A geometries, we performed MD simulations on dsGGACU with the m⁶A–U bp in either syn or anti conformations, or an m⁶₂A···U bp. All MD simulations were performed using the ff99 AMBER force field with bsc0 and χ_OL3 corrections for RNA, using periodic boundary conditions as implemented in the AMBER MD package. Starting structures for MD of unmethylated dsGGACU were generated by building an idealized A-RNA duplex using the fiber module of the 3DNA suite of programs⁷⁰. The starting structures for ${{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ with an m⁶A–U bp in either the anti or syn conformation were generated by replacing the anti and syn adenine amino hydrogen atoms in the idealized unmethylated dsGGACU structure with a methyl group. The starting structure for the dsGGACU duplex with the m⁶₂A–U bp was generated by replacing both of the amino hydrogen atoms of the adenine in the idealized unmethylated dsGGACU structure with methyl groups. All starting structures were solvated with an octahedral box of SPC/E water molecules with box size chosen such that the boundary was at least 10 Å away from any of the DNA atoms. Na⁺ ions treated using the Joung–Cheatham parameters were then added to neutralize the charge of the system. The system was then energy minimized in two stages with the solute heavy atoms (except for the atoms comprising the m⁶₂A···U bp and the m⁶(syn)A···U bp) being fixed (with a restraint of 500 kcal/mol/Å²) during the first stage. Heating, equilibration, and production runs (500 ns) were performed as described previously⁷¹. To maintain the methyl group in the syn conformation during the MD simulation of the dsGGACU duplex with the m⁶(syn)A···U bp, a torsion angle restraint was applied on the angle spanning the methyl carbon-N6-C6-C5 atoms of m⁶A. The restraint was chosen to be square welled between 160° and 200°, parabolic between 159–160° and 200–201°, and linear beyond 201° and <159°, with a force constant of 32 kcal/mol/Å². Force field parameters for m⁶A were derived from those by Aduri et al.⁷². In particular, the atom types and charges for the methyl group were taken from those by Aduri et al., while retaining atom types and charges (apart from N6, see below) for the remaining atoms from those of adenine in the AMBER ff99bsc0χOL3 force field. Charges on the amino N6 atom of m⁶A were adjusted to maintain a net charge for the m⁶A nucleoside of −1. An analogous procedure was followed to generate the parameters for the m⁶₂A nucleoside. Missing force field parameters were generated using the antechamber and parmchk utilities of the AMBER suite (16.0). All the structure visualization was performed in PyMOL (https://pymol.org/).

Automated fragmentation quantum mechanics/molecular mechanics chemical-shift calculations

We generated mono-nucleoside models of m⁶A with the N1-C6-N6-methyl carbon dihedral angle ranging from 0° to 360° in steps of 20° (syn conformation is 0°, whereas anti conformation is 180°). Coordinates of the m⁶A residue were derived from Aduri et al.⁷². We subjected the various mono-nucleoside models and all the RNA duplex MD ensembles (each with N = 100) to QM/MM chemical-shift calculations using a fragmentation procedure as described previously⁷³. The parameters of geometric minimization for RNA structures were described in a prior study⁷⁴. For all the RNA duplex ensembles, the chemical-shift calculations were solely focused on A6 and U13 residues in dsGGACU; therefore, each conformer in the RNA duplex ensembles was broken into only two quantum fragments centered on A6 or U13, respectively, whereas for all the mono-nucleoside models, each quantum fragment was the single mono-nucleoside. We then used a distribution of point charges on the fragment surface to represent the effects of RNA that is outside the quantum fragment and solvent⁷⁵. The local dielectric ε value was set to be 1, 4, and 80 for RNA inside the quantum fragment, RNA outside the quantum fragment, and RNA outside the solvent, respectively. We then performed the GIAO chemical-shift calculations for each quantum fragment with the OLYP functional and the pcSseg-0 basis set, using demon-2k program (http://www.demon-software.com/public_html/download.html). Reference shieldings were computed for TMS and nitromethane at the same level of theory.

Free energy decomposition along the CS pathway

The free energy of annealing the methylated duplex can be decomposed into two steps (CS pathway):

$$Step\;1:\;{{{{{\mathrm{ssRNA}}}}}}^{syn}\rightleftharpoons {{{{{\mathrm{ssRNA}}}}}}^{anti}$$

(12)

$$Step\;2:\;{{{{{\mathrm{ssRNA}}}}}}^{anti}+{{{{{\mathrm{ss2}}}}}}\rightleftharpoons {{{{{\mathrm{dsRNA}}}}}}^{anti}$$

(13)

${k}_{1}$ and ${k}_{-1}$ were determined from two-state fits or temperature dependence of the RD data (see ‘Determining the methylamino isomerization rate constants from temperature-dependent RD measurements for methylated ssRNA and dsRNA’ section above):

$${{\Delta}} {G}_{{{{{{\mathrm{iso}}}}}},{{{{{\mathrm{ss}}}}}}}^{^\circ }=-RT\,{{{{{\mathrm{ln}}}}}}{\left(\frac{{k}_{1}}{{k}_{-1}}\right)}$$

(14)

The apparent free energy of annealing methylated dsRNA was determined using:

$${{\Delta}} {G}_{{{{{{\mathrm{anneal}}}}}},{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}^{^\circ {{{{{\mathrm{app}}}}}}}=-RT\,{{{{{\mathrm{ln}}}}}}{\left(\frac{[{{{{{\mathrm{ssRNA}}}}}}^{syn}][{{{{{\mathrm{ss2}}}}}}]}{[{{{{{\mathrm{dsRNA}}}}}}^{anti}]}\right)}$$

(15)

in which the concentrations of the relevant species were measured based on two-state fits of the RD data²¹:

$$[{ssRNA}^{anti}]=\frac{[{ssRNA}^{{{{{\mathrm{total}}}}}}]\times {k}_{1}}{{k}_{1}+{k}_{-1}}$$

(16)

$$[{ssRNA}^{syn}]=\frac{[{ssRNA}^{{{{{\mathrm{total}}}}}}]\times {k}_{-1}}{{k}_{1}+{k}_{-1}}$$

(17)

$$[{ssRNA}^{{{{{\mathrm{total}}}}}}]={C}_{{{{{\mathrm{t}}}}}}\times {p}_{{{{{\mathrm{ss}}}}}}$$

(18)

$$[{dsRNA}^{anti}]={C}_{{{{{\mathrm{t}}}}}}\times {p}_{{{{{\mathrm{GS}}}}}}$$

(19)

$$[{ss}2]=[ss2]_{{{{{\mathrm{total}}}}}}-[{dsRNA}^{anti}]-[{dsRNA}^{syn}]$$

(20)

$$[{dsRNA}^{syn}]={C}_{{{{{\mathrm{t}}}}}}\times {p}_{{{{{\mathrm{ES}}}}}}$$

(21)

${p}_{{{{{\mathrm{ss}}}}}}$ and ${p}_{{{{{\mathrm{GS}}}}}}$ are the populations of the ssRNA^total (ssRNA^syn + ssRNA^anti) and dsRNA^anti species obtained from the RD data. ${[{{{{{\mathrm{ss}}}}}}2]}_{{{{{\mathrm{total}}}}}}$ is the total complementary strand concentration. Note that at T = 65 °C, dsRNA^syn has a negligible contribution to RD profiles ($[{{{{{\mathrm{dsRNA}}}}}}^{anti}]=0$), while at T = 55 °C, dsRNA^syn population (${p}_{{{{{\mathrm{ES}}}}}}$) was obtained from a three-state fit of the m⁶A-C2 CEST data for ${{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$. Also note that the ${{\Delta}} {G}_{{{{{{\mathrm{anneal}}}}}},{{{{{\mathrm{m}}}}}}^{6}{{{{{\mathrm{A}}}}}}}^{^\circ \,{{{{{\mathrm{app}}}}}}}$ here differs slightly (by ~0.1 kcal/mol) from the prior study²¹, where ssRNA^syn and ssRNA^anti were not distinguished.

The free energy of annealing ssRNA^anti is given by:

$$\begin{array}{c}{{\Delta}} {G}_{{anneal},anti}^{^\circ } = {{\Delta}} {G}_{{anneal},{m}^{6}{A}}^{^\circ {app}}\,-\, {{\Delta}} {G}_{{{{{{\mathrm{iso}}}}}},{{{{{\mathrm{ss}}}}}}}^{^\circ }\end{array}$$

(22)

$${k}_{{off},anti}=\frac{{k}_{{{{{{\mathrm{on}}}}}},anti}}{{e}^{\frac{{{\Delta}} {G}_{{{{{{\mathrm{anneal}}}}}},anti}^{^\circ }}{-RT}}}$$

(23)

$$\begin{array}{c}{{\Delta}} {{\Delta}} {G}_{{anneal},anti}^{^\circ }={{\Delta}} {G}_{{anneal},anti}^{^\circ }-{{\Delta}} {G}_{{anneal},A}^{^\circ }\end{array}$$

(24)

At T = 55 °C, ${{\Delta}} {{\Delta}} {G}_{{anneal},anti}^{^\circ }$ = 0.5 ± 0.2 kcal/mol, and the m⁶A methyl group in anti conformation slightly destabilizes the duplex, whereas it stabilized it by a comparable amount at T = 65 °C (${{\Delta}} {{\Delta}} {G}_{{anneal},anti}^{^\circ }$ = −0.5 ± 0.2 kcal/mol).

B–M simulations and constrained fits

When dealing with three- or four-state exchange, there is always a danger of overfitting the RD data. For this reason, we initially performed simulations in which all of the relevant kinetic rate constants, populations, Δω, R₁, and R₂ of the different species were approximated to values measured experimentally using the appropriate RNA constructs (Supplementary Figs. 2b and 6a and Supplementary Table 1). These values were then used in a three- or four-state simulation to simulate CEST profiles without any adjustable parameters. We then performed constrained fits in which the parameters (population, rate constants, Δω, R₁, and R₂) were allowed to float by an amount determined by the experimentally measured uncertainty (1 s.d.).

The simulations and constrained fits were performed by numerically integrating the appropriate B–M equations³⁴. Briefly, the simulations were performed by directly predicting RD profiles for a given set of exchange parameters that are defined below. In the constrained fitting, the same numerical integration was used to fit exchange parameters applying specific constraints as detailed below.

Three-state CS simulations and constrained fits for the ${{{{{\mathrm{dsA6DNA}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$RD data measured at T = 65 °C

These analyses used the following input exchange parameters:

1.
${k}_{1}$ and $\,{k}_{-1}$ were obtained from the temperature-dependent RD measurements on ${{{{{\mathrm{ssGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ (Supplementary Fig. 2c).
2.
${k}_{{off},{anti}}$ was assumed equal to ${k}_{off}$ measured for the unmethylated dsGGACU and ${k}_{{{{{{\mathrm{on}}}}}},{{{{{\mathrm{anti}}}}}}}$ was obtained from the energetic decomposition described above (Eq. 23).
3.
The longitudinal (R₁) and transverse (R₂) relaxation rate constants for all three species (ssRNA^syn, ssRNA^anti, and dsRNA^anti) were obtained from two-state fits of the CEST RD data probing duplex melting at T = 65 °C²¹. R₁(ssRNA^anti) = R₁(ssRNA^syn) = R₁(dsRNA^anti) = R_1,GS = R_1,ES. R₂(ssRNA^anti) = R₂(ssRNA^syn) = R_2,ES. R₂(dsRNA^anti) = R_2,GS.
4.
The equilibrium populations ${p}_{({{{{{\mathrm{ssRNA}}}}}}^{syn})},\;{p}_{({{{{{\mathrm{ssRNA}}}}}}^{anti})},\;{{{{{\mathrm{and}}}}}}\;{p}_{({{{{{\mathrm{dsRNA}}}}}}^{anti})}$ were obtained from kinetic simulations (see differential equations below) that were sufficiently long to ensure equilibration. The same equilibrium populations were obtained from analytical expressions outlined in ref. ⁷⁶:
$$\frac{{{{{{\rm{d}}}}}}[{ssRNA}^{syn}]}{{{{{{\rm{d}}}}}}t}=-{k}_{1}[{ssRNA}^{syn}]+{k}_{-1}\big[{ssRNA}^{anti}\big]$$
(25)
$$\frac{{{{{{\rm{d}}}}}}[{{{{{\mathrm{ssRNA}}}}}}^{anti}]}{{{{{{\rm{d}}}}}}t} = \,{k}_{1}[{ssRNA}^{syn}]-{k}_{-1}[{ssRNA}^{anti}]-{k}_{{{{{{\mathrm{on}}}}}},anti}[{ssRNA}^{anti}][{ss2}]\\ \,+{k}_{{{{{{\mathrm{off}}}}}},anti}[{dsRNA}^{anti}]$$
(26)
$$\frac{{{{{{\rm{d}}}}}}[{{{{{\mathrm{dsRNA}}}}}}^{anti}]}{{{{{{\rm{d}}}}}}t}={k}_{{{{{{\mathrm{on}}}}}},anti}[{ssRNA}^{anti}][{ss2}]-{k}_{{{{{{\mathrm{off}}}}}},anti}[{{{{{\mathrm{dsRNA}}}}}}^{anti}]$$
(27)
$$\frac{{{{{{\rm{d}}}}}}[{{{{{\mathrm{ss2}}}}}}]}{{{{{{\rm{d}}}}}}t}=-{k}_{{{{{{\mathrm{on}}}}}},anti}[{ssRNA}^{anti}][{ss2}]+{k}_{{{{{{\mathrm{off}}}}}},anti}[{dsRNA}^{anti}]$$
(28)
5.
${{\Delta}} {\omega}$ of ${{{{{\mathrm{ssRNA}}}}}}^{syn}$ and ${{{{{\mathrm{ssRNA}}}}}}^{anti}$ for C2: ${{\Delta}} {{\omega }}_{{{{{{\rm{ss}}}}}},syn}={{\omega }}_{{{{{{\rm{ss}}}}}},syn}-{{\omega }}_{{{{{{\rm{ds}}}}}},anti},$ in which ${{\omega }}_{{{{{{\rm{ss}}}}}},syn}={{\omega }}_{{{{{{\rm{ss}}}}}}}-\frac{{p}_{({{{{{\mathrm{ssRNA}}}}}}^{syn})}}{{p}_{({{{{{\mathrm{ssRNA}}}}}}^{syn})+}{p}_{({{{{{\mathrm{ssRNA}}}}}}^{anti})}}\times {{\Delta}} {{\omega }}_{{{{{{\rm{ss}}}}}},anti-syn}$. ${{\Delta}} {{\omega }}_{{{{{{\rm{ss}}}}}},anti}={{\omega }}_{{{{{{\rm{ss}}}}}},anti}-{{\omega }}_{{{{{{\rm{ds}}}}}},anti},\,$ in which ${{\omega }}_{{{{{{\rm{ss}}}}}},anti}={{\omega }}_{{{{{{\rm{ss}}}}}}}+\frac{{p}_{({{{{{\mathrm{ssRNA}}}}}}^{anti})}}{{p}_{({{{{{\mathrm{ssRNA}}}}}}^{syn})+}{p}_{({{{{{\mathrm{ssRNA}}}}}}^{anti})}}\times {{\Delta}} {{\omega }}_{{{{{{\rm{ss}}}}}},anti-syn}$. ${{\omega }}_{{{{{{\rm{ss}}}}}}}\,$ and ${{\omega }}_{{{{{{\rm{ds}}}}}},anti}$ were obtained from 2D HSQC spectra (Supplementary Fig. 1) and ${{\Delta}} {{\omega }}_{{{{{{\rm{ss}}}}}},anti-syn}$ was obtained from ${{{{{\mathrm{ssGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ RD measurements at T = 25 °C and was assumed to be temperature independent, as supported by the data collected in this study (Supplementary Fig. 2, Supplementary Table 1). Since C8 is not sensitive to methylamino isomerization (Supplementary Fig. 2), $\Delta {{{{{{\rm{\omega }}}}}}}_{{{{{{\rm{ss}}}}}},anti}=0$, while ${{\Delta}} {{{{{{\rm{\omega }}}}}}}_{{{{{{\rm{ss}}}}}},syn}$ is obtained from the two-state fit of the CEST RD data probing duplex melting at T = 65 °C²¹.

The above parameters were fixed to simulate the CEST profiles using a three-state B–M equation³⁴. For the constrained three-state fit, the ratio (but not absolute magnitude) of ${k}_{{{{{{\mathrm{on}}}}}},{{{{{\mathrm{anti}}}}}}}$ to ${k}_{{{{{{\mathrm{off}}}}}},{{{{{\mathrm{anti}}}}}}}$ was constrained to preserve the free energy of the hybridization step. All other parameters (population, $\,{k}_{1},\,{k}_{-1}$, Δω, R₁, and R₂ for all species) were allowed to float by an amount determined by the uncertainty (1 s.d.). When possible, global constrained three-state B–M fits were carried out on both m⁶A C8 and C2 CEST data (Fig. 2f). ${\chi }_{{{{{\mathrm{red}}}}}}^{2}$ was calculated to assess the goodness of fitting¹⁸.

Three-state IF simulations and constrained fits for the ${{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ RD data were measured at T = 65 and 55 °C

These analyses used the following input exchange parameters:

1.
${k}_{2}$, ${k}_{-2}$, $\,{k}_{{{{{{\mathrm{on}}}}}},syn}$, ${k}_{{{{{{\mathrm{off}}}}}},syn},$ C2 Δω_ds,syn, C2 R₁(dsRNA^syn) = R₁(dsRNA^anti) = R_1,GS, and R₂(dsRNA^syn) = R₂(dsRNA^anti) = R_2,GS were obtained from a three-state fit to the ${{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ m⁶A-C2 RD data (Fig. 3a and Supplementary Table 3) using the triangular topology at T = 55 °C or from RD measurements done on the hairpin constructs at T = 65 °C (Supplementary Fig. 6a). C8 Δω_ds,syn = 0 because C8 is not sensitive to methylamino isomerization (Supplementary Fig. 6a). C8 R₁(dsRNA^syn) = R₁(dsRNA^anti) = R_1,GS, and R₂(dsRNA^syn) = R₂(dsRNA^anti) = R_2,GS were obtained from a two-state fit to the ${{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ m⁶A-C8 RD data (Supplementary Table 1), ${{\Delta}} {{\omega }}_{{{{{{\rm{ss}}}}}},syn}$ is obtained from the two-state fit of the CEST RD data probing duplex melting at T = 65 and 55 °C²¹.
2.
The equilibrium populations ${p}_{({{{{{\mathrm{ssRNA}}}}}}^{syn})},\;{p}_{({{{{{\mathrm{dsRNA}}}}}}^{syn})},\;{p}_{({{{{{\mathrm{dsRNA}}}}}}^{anti})}$ were obtained from kinetic simulations (see differential equations below) that were sufficiently long to ensure equilibration:
$$\frac{{{{{{\rm{d}}}}}}[{{{{{\mathrm{ssRNA}}}}}}^{syn}]}{{{{{{\rm{d}}}}}}t}={k}_{{{{{{\mathrm{off}}}}}},syn}[{{{{{\mathrm{dsRNA}}}}}}^{syn}]-{k}_{{{{{{\mathrm{on}}}}}},syn}[{{{{{\mathrm{ssRNA}}}}}}^{syn}][{{{{{\mathrm{ss2}}}}}}]$$
(29)
$$\frac{{{{{{\rm{d}}}}}}[{{{{{\mathrm{dsRNA}}}}}}^{syn}]}{{{{{{\rm{d}}}}}}t} = \,-{k}_{{{{{{\mathrm{off}}}}}},syn}[{{{{{\mathrm{dsRNA}}}}}}^{syn}]+{k}_{{{{{{\mathrm{on}}}}}},syn}[{{{{{\mathrm{ssRNA}}}}}}^{syn}][{{{{{\mathrm{ss2}}}}}}]\\ +{k}_{2}[{{{{{\mathrm{dsRNA}}}}}}^{anti}]-{k}_{-2}[{{{{{\mathrm{dsRNA}}}}}}^{syn}]$$
(30)
$$\frac{{{{{{\rm{d}}}}}}[{{{{{\mathrm{dsRNA}}}}}}^{anti}]}{{{{{{\rm{d}}}}}}t}={k}_{-2}[{{{{{\mathrm{dsRNA}}}}}}^{syn}]-{k}_{2}[{{{{{\mathrm{dsRNA}}}}}}^{anti}]$$
(31)
$$\frac{{{{{{\rm{d}}}}}}[{{{{{\mathrm{ss2}}}}}}]}{{{{{{\rm{d}}}}}}t} = \,{k}_{{{{{{\mathrm{off}}}}}},syn}[{{{{{\mathrm{dsRNA}}}}}}^{syn}]-{k}_{{{{{{\mathrm{on}}}}}},syn}[{{{{{\mathrm{ssRNA}}}}}}^{syn}][{{{{{\mathrm{ss2}}}}}}]\\ +{k}_{{{{{{\mathrm{off}}}}}},anti}[{{{{{\mathrm{dsRNA}}}}}}^{anti}]$$
(32)

The same approach was used to simulate/fit CEST profiles for the IF pathway as described in the previous section.

Four-state CS + IF simulations and constrained fits for the ${{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ RD data at T = 55 °C

These analyses used the following input exchange parameters:

1.
All of the exchange parameters related to the CS pathway (${k}_{1},{k}_{-1},{k}_{{{{{{\mathrm{on}}}}}},anti},{k}_{{{{{{\mathrm{off}}}}}},anti},{{\Delta}} {{\omega }}_{{{{{{\rm{ss}}}}}},syn},{{\Delta}} {{\omega }}_{{{{{{\rm{ss}}}}}},anti},\;$R₁(ssRNA^anti), R₁(ssRNA^syn), R₁(dsRNA^anti R₂(ssRNA^anti), R₂(ssRNA^syn), and R₂(dsRNA^anti)) and the IF pathway (${k}_{-2},{k}_{2},{k}_{{{{{{\mathrm{on}}}}}},syn},{k}_{{{{{{\mathrm{off}}}}}},syn},$ R₁(dsRNA^syn), and ${{\Delta}} {{\omega }}_{{{{{{\rm{ds}}}}}},syn}$) were obtained as described in the previous sections for the three-state CS and IF analysis, respectively.
2.
The population of all four species was obtained from four-state kinetic simulations using the eight rate constants (${k}_{1},{k}_{-1},{k}_{{{{{{\mathrm{on}}}}}},anti},{k}_{{{{{{\mathrm{off}}}}}},anti},{k}_{-2},{k}_{2},{k}_{{{{{{\mathrm{on}}}}}},syn},{k}_{{{{{{\mathrm{off}}}}}},syn}$) based on the CS + IF model (see differential equations below). The same equilibrium populations were obtained from analytical expressions outlined in ref. ⁷⁶:
$$\frac{{{{{{\rm{d}}}}}}[{{{{{\mathrm{ssRNA}}}}}}^{syn}]}{{{{{{\rm{d}}}}}}t}= -{k}_{1}[{{{{{\mathrm{ssRNA}}}}}}^{syn}]+{k}_{-1}[{{{{{\mathrm{ssRNA}}}}}}^{anti}]\\ +{k}_{{{{{{\mathrm{off}}}}}},syn}[{{{{{\mathrm{dsRNA}}}}}}^{syn}]-{k}_{{{{{{\mathrm{on}}}}}},syn}[{{{{{\mathrm{ssRNA}}}}}}^{syn}][{{{{{\mathrm{ss2}}}}}}]$$
(33)
$$\frac{{{{{{\rm{d}}}}}}[{{{{{\mathrm{ssRNA}}}}}}^{anti}]}{{{{{{\rm{d}}}}}}t} = \,{k}_{1}[{{{{{\mathrm{ssRNA}}}}}}^{syn}]-{k}_{-1}[{{{{{\mathrm{ssRNA}}}}}}^{anti}]\\ -{k}_{{{{{{\mathrm{on}}}}}},anti}[{{{{{\mathrm{ssRNA}}}}}}^{anti}][{{{{{\mathrm{ss2}}}}}}]+{k}_{{{{{{\mathrm{off}}}}}},anti}[{{{{{\mathrm{dsRNA}}}}}}^{anti}]$$
(34)
$$\frac{{{{{{\rm{d}}}}}}[{{{{{\mathrm{dsRNA}}}}}}^{syn}]}{{{{{{\rm{d}}}}}}t}= \,-{k}_{{{{{{\mathrm{off}}}}}},syn}[{{{{{\mathrm{dsRNA}}}}}}^{syn}]+{k}_{{{{{{\mathrm{on}}}}}},syn}[{{{{{\mathrm{ssRNA}}}}}}^{syn}][{{{{{\mathrm{ss2}}}}}}]\\ +{k}_{2}[{{{{{\mathrm{dsRNA}}}}}}^{anti}]-{k}_{-2}[{{{{{\mathrm{dsRNA}}}}}}^{syn}]$$
(35)
$$\frac{{{{{{\mathrm{d}}}}}}[{{{{{\mathrm{dsRNA}}}}}}^{anti}]}{{{{{{\rm{d}}}}}}t} = \,{k}_{{{{{{\mathrm{on}}}}}},anti}[{{{{{\mathrm{ssRNA}}}}}}^{anti}][{{{{{\mathrm{ss2}}}}}}]-{k}_{{{{{{\mathrm{off}}}}}},anti}[{{{{{\mathrm{dsRNA}}}}}}^{anti}]\\ +{k}_{-2}[{{{{{\mathrm{dsRNA}}}}}}^{syn}]-{k}_{2}[{{{{{\mathrm{dsRNA}}}}}}^{anti}]$$
(36)
$$\frac{{{{{{\rm{d}}}}}}[{{{{{\mathrm{ss2}}}}}}]}{{{{{{\rm{d}}}}}}t} = \,{k}_{{{{{{\mathrm{off}}}}}},syn}[{{{{{\mathrm{dsRNA}}}}}}^{syn}]-{k}_{{{{{{\mathrm{on}}}}}},syn}[{{{{{\mathrm{ssRNA}}}}}}^{syn}][{{{{{\mathrm{ss2}}}}}}]\\ -{k}_{{{{{{\mathrm{on}}}}}},anti}[{{{{{\mathrm{ssRNA}}}}}}^{anti}][{{{{{\mathrm{ss2}}}}}}]+{k}_{{{{{{\mathrm{off}}}}}},anti}[{{{{{\mathrm{dsRNA}}}}}}^{anti}]$$
(37)

The exchange parameters described above were then used to simulate the CEST profile using the four-state B–M equation (see below)⁶²:

{GS/${{{{{\mathrm{ES}}}}}}_i$}{x/y/z} (i = 1, 2, 3) denotes the magnetization of the GS or ESs in the specified direction. R_2,GS, R_2,ES1, R_2,ES2, and R_2,ES3 are the transverse relaxation rate constants for the GS (dsRNA^anti), ES1 (dsRNA^syn), ES2 (ssRNA^syn), and ES3 (ssRNA^anti), respectively. R_1,GS, R_1,ES1, R_1,ES2, and R_1,ES3 are corresponding longitudinal relaxation rate constants. ⍵ is the RF field power; k_{ij} and k_{ji} are the forward and backward rate constants of reactions shown in Fig. 5a. Specifically, k₁₂ = k₂, and k₂₁ = k₋₂ are the forward and backward rate constants of methylamino isomerization in dsRNA. k₂₃ = ${k}_{{{{{{\mathrm{off}}}}}},syn}$, k₃₂ = ${k}_{{{{{{\mathrm{on}}}}}},syn}[{{{{{\mathrm{ss2}}}}}}]$. k₃₄ = k₁ and k₄₃ = k₋₁ are the forward and backward rate constants of methylamino isomerization in ssRNA. k₄₅ = ${k}_{{{{{{\mathrm{on}}}}}},anti}[{{{{{\mathrm{ss2}}}}}}]$ and k₅₄ = ${k}_{{{{{{\mathrm{off}}}}}},anti}$. $I_{\{{{{{{{\mathrm{GS}}}}}}/{{{{{\mathrm{ES}}}}}}_i}\}z,{{{{{\mathrm{eq}}}}}}}$ (i = 1, 2, 3) denotes the longitudinal magnetization of the GS or ESs at the start of the experiment. ω_i (i = 1, 2, 3, 4) are the offset frequencies of the GS, or ES resonances in the rotating frame of the RF field⁶¹.

We carried out two independent constrained four-state fits at T = 55 °C that differ with regards to how ${k}_{{{{{{\mathrm{on}}}}}},syn}$ and ${k}_{{{{{{\mathrm{off}}}}}},syn}$ were defined. In one case, ${k}_{{{{{{\mathrm{on}}}}}},syn}$ was assumed to be equal to the ${k}_{{{{{{\mathrm{ss}}}}}}\rightarrow {{{{{\mathrm{ES}}}}}}}$ rate constant obtained from a three-state fit to the CEST data measured for ${{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ at T = 55 °C (Fig. 3b) using the triangular topology. Note that this is an approximation since the ssRNA represents the major ssRNA^syn and minor ssRNA^anti species in fast exchange. ${k}_{{{{{{\mathrm{off}}}}}},syn}$ was then calculated by closing the thermodynamic cycle:

$${{\Delta}} {G}_{{anneal},syn}^{^\circ }={{\Delta}} {G}_{{anneal},{m}^{6}{{{{{\mathrm{A}}}}}}}^{^\circ \,{{{{{\mathrm{app}}}}}}}\,-\,{{\Delta}} {G}_{{iso},{ds}}^{^\circ }$$

(39)

$${{\Delta}} {G}_{{{{{{\mathrm{iso}}}}}},{{{{{\mathrm{ds}}}}}}}^{^\circ }=-RT\,{{{{{\mathrm{ln}}}}}}{\left(\frac{{k}_{-2}}{{k}_{2}}\right)}$$

(40)

$${k}_{{off},syn}=\frac{{k}_{{on},syn}}{{e}^{\frac{{\Delta} {G}_{{anneal},syn}^{^\circ }}{-RT}}}$$

(41)

All other exchange parameters were then allowed to float by an amount determined by the experimental uncertainty (one standard deviation). In the second case, only the ratio (but not absolute magnitude) of ${k}_{{{{{{\mathrm{on}}}}}},syn}$ to ${k}_{{{{{{\mathrm{off}}}}}},syn}$ was constrained to preserve the free energy of the hybridization step. The fitted ${k}_{{{{{{\mathrm{on}}}}}},syn}$ and ${k}_{{{{{{\mathrm{off}}}}}},syn}$ values were similar using these two independent methods. The results from the second method were reported in Fig. 5a and Supplementary Table 2. When possible, global constrained four-state B–M fits were carried out on both m⁶A C8 and C2 CEST data. ${\chi }_{{{{{\mathrm{red}}}}}}^{2}$ was calculated to assess the goodness of fitting¹⁸.

Four-state constrained fits for the CS + IF model for ${{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ at T = 65 °C

Because the dsRNA^syn ES was not directly detected at T = 65 °C, the RD data were analyzed as described for T = 55 °C, with the exception that k₂ and k₋₂ were measured in ${{{{{\mathrm{hpGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ at T = 65 °C using R_1ρ RD (Supplementary Fig. 6a), ${k}_{{{{{{\mathrm{on}}}}}},syn}$ was assumed to be equal to ${k}_{{{{{\mathrm{on}}}}}}/20$. This 20-fold slowdown in annealing of ssRNA^syn relative to unmethylated ssRNA was observed for ${{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ at T = 55 °C. ${k}_{{{{{{\mathrm{off}}}}}},syn}$ was then calculated by closing the thermodynamic cycle (Eq. 37). Similar results were obtained when assuming ${k}_{{{{{{\mathrm{off}}}}}},syn}$ is equal to ${k}_{{{{{\mathrm{off}}}}}}\times 80$ as observed for ${{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ at T = 55 °C, and closing the cycle (Eq. 37) to calculate ${k}_{{{{{{\mathrm{on}}}}}},syn}$.

Four-state constrained fits for the CS + IF model for ${{{{{\mathrm{dsHCV}}}}}}{{{{{\mathrm{m}}}}}}^{6}{{{{{\mathrm{A}}}}}}$ and ${{{{{\mathrm{dsA6DNA}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$

RD data measured for ${{{{{\mathrm{dsHCV}}}}}}{{{{{\mathrm{m}}}}}}^{6}{{{{{\mathrm{A}}}}}}$ and ${{{{{\mathrm{dsA6DNA}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ were analyzed in a similar manner as described in the previous sections.

1.
k₁, k₋₁ and k₂, k₋₂ were assumed to be the same as those measured in GGACU^m6A constructs using temperature-dependent RD measurements (Supplementary Figs. 2c and 6d).
2.
R₁(ssRNA^anti) = R₁(ssRNA^syn) = R₁(dsRNA^anti) = R_1,GS = R_1,ES. R₂(ssRNA^anti) = R₂(ssRNA^syn) = R_2,ES. R₂(dsRNA^anti) = R_2,GS. R_1,ES and R_2,GS were obtained from a two-state fit to the RD data probing duplex melting (Supplementary Table 1).
3.
Δω_ss,anti = Δω_ds,syn = 0 for A11-C8 in ${{{{{\mathrm{dsHCV}}}}}}{{{{{\mathrm{m}}}}}}^{6}{{{{{\mathrm{A}}}}}}$ since A11 is not the m⁶A site. Δω_ss,syn was assumed to be equal to the Δω value for A11-C8 in ssRNA obtained from a two-state fit of the A11-C8 RD data²¹.
4.
Δω_ss,syn and Δω_ss,anti for m⁶A16-C2 in ${{{{{\mathrm{dsA6DNA}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ were determined as described in CS three-state simulation for ${{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ at T = 65 °C, assuming ${{\Delta}} {{\omega }}_{{{{{{\rm{ss}}}}}},anti-syn}$ of ${{{{{\mathrm{ssA6DNA}}}}}}{{{{{\mathrm{m}}}}}}^{6}{{{{{\mathrm{A}}}}}}$ is the same as that of ${{{{{\mathrm{ssGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$. Δω_ds,syn was assumed to be equal to that measured for ${{{{{\mathrm{hpGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ at T = 55 °C (Supplementary Table 1).

Flux calculations

Flux through the of CS (F_CS) and IF (F_IF) pathways was calculated as the harmonic mean of the forward rates along the CS and IF pathways²⁷:

$${F}_{{{{{\mathrm{CS}}}}}}={\left(\frac{1}{{k}_{1}[{ssRNA}^{syn}]}+\frac{1}{{k}_{{{{{{\mathrm{on}}}}}},anti}[{ssRNA}^{anti}][{{{{{\mathrm{ss2}}}}}}]}\right)}^{-1}$$

(42)

$${F}_{{{{{\mathrm{IF}}}}}}={\left(\frac{1}{{k}_{{{{{{\mathrm{on}}}}}},syn}[{ssRNA}^{syn}][{ss2}]}+\frac{1}{{k}_{-2}[{{{{{\mathrm{dsRNA}}}}}}^{syn}]}\right)}^{-1}$$

(43)

All concentrations are equilibrium concentrations obtained using constrained four-state fit of CEST data (Fig. 5c) or CS + IF kinetic modeling.

Model to predict apparent k _on and k _off for methylated RNA/DNA duplexes and TAR

The four-state CS + IF model was used to simulate time traces describing the evolution of all four species as a function of time starting from 100% ssRNA^syn at t = 0. Similar results were obtained when performing simulations starting with an equilibrium population of ssRNA^syn (${k}_{-1}/({k}_{1}+{k}_{-1})$) and ssRNA^anti (${k}_{1}/({k}_{1}+{k}_{-1})$). ${k}_{1},\;{k}_{-1},\;{k}_{-2},\;{{{{{\mathrm{and}}}}}}\;{k}_{2}$ were all assumed equal to the corresponding values measured for ${{{{{\mathrm{ssGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ and ${{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ at the appropriate temperature based on the temperature-dependent RD measurements (Supplementary Figs. 2c and 6d). ${k}_{{{{{{\mathrm{off}}}}}},anti}$ was assumed to be equal to ${k}_{{{{{\mathrm{off}}}}}}$, and ${k}_{{{{{{\mathrm{on}}}}}},anti}$ was deduced from closing the thermodynamic cycle (Eq. 23). ${k}_{{{{{{\mathrm{on}}}}}},syn}$ and ${k}_{{{{{{\mathrm{off}}}}}},syn}$ were obtained using two different approaches and yielded similar predictions for the apparent k_on and k_off for methylated RNA/DNA duplexes and TAR. In one case, $\,{k}_{{{{{{\mathrm{on}}}}}},syn}={k}_{{{{{\mathrm{on}}}}}}/20$ and ${k}_{{{{{{\mathrm{off}}}}}},syn}$ was deduced from closing the thermodynamic cycle (Eq. 37). Alternatively, ${k}_{{{{{{\mathrm{off}}}}}},syn}={k}_{{{{{\mathrm{off}}}}}}\times 80$ and ${k}_{{{{{{\mathrm{on}}}}}},syn}$ was deduced from closing the thermodynamic cycle (Eq. 37). The predictions shown in Fig. 6a were obtained using the former approach. ${k}_{{{{{{\mathrm{on}}}}}},{{{{{\mathrm{m}}}}}}^{6}{{{{{\mathrm{A}}}}}}}^{{{{{\mathrm{app}}}}}}$ and ${k}_{{{{{{\mathrm{off}}}}}},{{{{{\mathrm{m}}}}}}^{6}{{{{{\mathrm{A}}}}}}}^{{{{{\mathrm{app}}}}}}$ were obtained by fitting simulated time course of [dsRNA^syn] + [dsRNA^anti] at multiple time points to numerical solutions of Eqs. (40) and (41) for a two-state hybridization model ${{{{{{\mathrm{ss}}}}}}1}+{{{{{{\mathrm{ss}}}}}}2}\,{{\Delta}} {{{{{\mathrm{d}}}}}}s$, ${k}_{{{{{{\mathrm{on}}}}}},{{{{{\mathrm{m}}}}}}^{{{{{\mathrm{6}}}}}}{{{{{\mathrm{A}}}}}}}^{{{{{\mathrm{app}}}}}}$, and ${k}_{{{{{{\mathrm{off}}}}}},{{{{{{{\rm{m}}}}}}^{6}}}{{{{{\mathrm{A}}}}}}}^{{{{{\mathrm{app}}}}}}$ are the annealing and melting constants, respectively:

$$\frac{{d}[{d}s]}{{d}t}={k}_{{on},{{m}^{6}}{A}}^{app}[{ss1}][{ss2}]-{k}_{{off},{{m}^{6}}{A}}^{app}[{d}s]$$

(44)

$$\frac{{{{{{\mathrm{d}}}}}}[{{{{{\mathrm{ss}}}}}}1]}{{{{{{\mathrm{d}}}}}}t}=\frac{{{{{{\mathrm{d}}}}}}[{{{{{\mathrm{ss}}}}}}2]}{{{{{{\mathrm{d}}}}}}t}=-{k}_{{{{{{\mathrm{on}}}}}},{{{{{{\mathrm{m}}}}}}^{6}}{{{{{\mathrm{A}}}}}}}^{{{{{\mathrm{app}}}}}}[{{{{{\mathrm{ss}}}}}}1][{{{{{\mathrm{ss}}}}}}2]+{k}_{{{{{{\mathrm{off}}}}}},{{{{{{\mathrm{m}}}}}}^{6}}{{{{{\mathrm{A}}}}}}}^{{{{{\mathrm{app}}}}}}[{{{{{\mathrm{d}}}}}}s]$$

(45)

Similar results were obtained when fitting simulated time course of [dsRNA^anti] only. However, it should be noted that for certain kinetic regimes outside those examined here, particularly when ${k}_{{{{{{\mathrm{on}}}}}},syn}$ is ultra-fast, there can be a substantial accumulation of the dsRNA^syn. In this scenario, the system is poorly defined with the apparent two-state approximation and separate rate constants are needed to describe the evolution of all species. In addition, similar results were obtained from fitting the traces to the appropriate two-state second-order kinetic equation (see ref. ⁷⁷). Finally, similar results were obtained when simulating m⁶A-C8 RD profiles using the four-state CS + IF model together with exchange parameters (Δω, R₁, and R₂ values for all species) derived from the ${{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}$ 55 °C m⁶A-C8 CEST data, and then fitting the data to a two-state model. Note that C8 was used the probe instead of C2 because the two-state fit results vary depending on the three Δω values used in the C2 CEST simulation. On the other hand, varying the one Δω value used in the C8 CEST simulation does not affect the two-state fit results. As the choice of exchange parameters (R₁ and R₂ values) had a minor effect on the two-state fit results, we show results from the kinetic simulations in Fig. 6a and that from the two-state fitting to the simulated C8 RD data in Supplementary Fig. 9b.

A similar approach was used to compute the apparent ${k}_{{{{{\mathrm{forward}}}}}}$ and ${k}_{{{{{\mathrm{backward}}}}}}$ rate constants for methylated TAR, except that ${k}_{1},\,{k}_{-1}$ were assumed to be equal to the values measured for m⁶AMP, which is a better mimic of the environment of the flipped out and unstacked A35 in TAR than ssRNA. Apparent ${k}_{{{{{\mathrm{forward}}}}}}$ and ${k}_{{{{{\mathrm{backward}}}}}}$ rate constants were obtained by fitting simulated time course of $[{{{{{\mathrm{ES}}}}}}]$ at multiple time points to the equation $[{{{{{\mathrm{ES}}}}}}]=A(1-{e}^{-{k}_{{{{{\mathrm{ex}}}}}}t})$, where A is a pre-exponential factor. Note that for the energetics decomposition and kinetic simulations of TAR, the $[{{{{{\mathrm{ss}}}}}}2]$ term in all equations above was removed since the TAR conformational transition is a first-order reaction.

Predict m⁶A-induced slowdown of DNA hybridization in the mouse genome

We used our four-state CS + IF model to predict the hybridization kinetics for 12-mer DNA duplex representing 5951 m⁶A sites in the mouse genome⁹, in which m⁶A was positioned at the sixth nucleotide. ${k}_{{{{{\mathrm{on}}}}}}$ of unmethylated DNA was predicted as described previously⁴⁵ (http://nablab.rice.edu/nabtools/kinetics.html). The free energy (${{\Delta}} {G}_{{{{{{\mathrm{anneal}}}}}},A}^{\circ }$) of each sequence was predicted using the MELTING program (https://www.ebi.ac.uk/biomodels-static/tools/melting/). ${k}_{{{{{\mathrm{off}}}}}}=\frac{{k}_{{{{{\mathrm{on}}}}}}}{{e}^{\frac{{{\Delta}} {G}_{{{{{{\mathrm{anneal}}}}}},A}^{^\circ }}{-RT}}}$. In all cases, the thermodynamic destabilization of the duplex by m⁶A (${{\Delta}} {{\Delta}} {G}_{{{{{{\mathrm{anneal}}}}}},{{{{{{{\rm{m}}}}}}^{6}}}{{{{{\mathrm{A}}}}}}}^{\circ }$) was assumed to be 1 kcal/mol based on prior studies^12,78 and our measurements (Supplementary Table 4). ${{\Delta}} {G}_{{{{{{\mathrm{anneal}}}}}},{{{{{{\mathrm{m}}}}}}^{6}}{{{{{\mathrm{A}}}}}}}^{^\circ \,{{{{{\mathrm{app}}}}}}}$ was obtained from ${{\Delta}} {G}_{{{{{{\mathrm{anneal}}}}}},{{{{{{\mathrm{m}}}}}}^{6}}{{{{{\mathrm{A}}}}}}}^{\circ \,{{{{{\mathrm{app}}}}}}}$ = ${{\Delta}} {G}_{{{{{{\mathrm{anneal}}}}}},A}^{\circ }+{{{\Delta}} {{\Delta}} {G}_{{{{{{\mathrm{anneal}}}}}},{{{{{{\mathrm{m}}}}}}^{6}}{{{{{\mathrm{A}}}}}}}^{\circ \,{{{{{\mathrm{app}}}}}}}}$. ${k}_{{{{{\mathrm{on}}}}}}$, ${k}_{{{{{\mathrm{off}}}}}}$, and ${{\Delta}} {G}_{{{{{{\mathrm{anneal}}}}}},{{{{{{\mathrm{m}}}}}}^{6}}{{{{{\mathrm{A}}}}}}}^{\circ \,{{{{{\mathrm{app}}}}}}}$ were then used as inputs to predict ${k}_{{{{{{\mathrm{on}}}}}},{{{{{{\mathrm{m}}}}}}^6}{{{{{\mathrm{A}}}}}}}^{{{{{\mathrm{app}}}}}}$ and ${k}_{{{{{{\mathrm{off}}}}}},{{{{{{\mathrm{m}}}}}}^6}{{{{{\mathrm{A}}}}}}}^{{{{{\mathrm{app}}}}}}$ as described in the previous sections. The concentration of dsDNA was assumed to be 1 mM and T = 37 °C. We also used this approach to predict the impact of m⁶A on RNA hybridization kinetics at T = 37 °C using rate constants for hybridization of unmethylated RNA reported previously²² at T = 37 °C and assuming that m⁶A destabilizes dsRNA by 1 kcal/mol¹². m⁶A was predicted to slow ${k}_{{{{{\mathrm{on}}}}}}$ by ~5-fold while having a minor effect (<2-fold) on ${k}_{{{{{\mathrm{off}}}}}}$, consistent with our measurements at higher temperatures.

Survey of single H-bonded A–U bps in PDB structures

To identify singly H-bonded A–U bp conformations that mimic the m⁶(syn)A···U ES, we conducted a structural survey of the RCSB PDB⁷⁹. All X-ray (with resolution ≤ 3.0 Å) and NMR biological assemblies containing RNA molecules (including naked RNA, RNA protein complex, etc.) were downloaded from RCSB PDB on Aug 2017 and processed by X3DNA-DSSR⁸⁰ to generate a searchable database containing RNA structural information. Potential candidates of single H-bonded A–U bp were identified by applying the following filters in the database: (1) A–U bps are unmethylated; (2) the Leontis–Westhof (LW) classification⁸¹ is “cWW”; (3) both A and U are not in syn conformation at glycosidic bond; (4) A–U bps contain A(N1)–U(N3) H-bond (distance between A(N1) and U(N3) is <3.5 Å) but do not contain A(N6)–U(O4) H-bond (distance between A(N6) and U(O4) is >3.5 Å). We then manually inspected all the single H-bonded A–U bps, removed misregistered bps, and classified the structure context of all the resulting bps into the following categories (Supplementary Fig. 7e):

1.
Junction: A–U bp that is next to an internal bulge, a mismatch or an apical loop.
2.
Junction-1/2/3: 1/2/3 bp away from the junction.
3.
Tertiary: involved in tertiary interactions.
4.
Terminal: at terminal ends.
5.
Terminal-1/2/3: 1/2/3 bp away from the terminal end.
6.
Duplex: A–U bp at the canonical duplex context

Reporting summary

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

Data availability

The data supporting the findings of this study are available from the corresponding authors upon reasonable request. The NMR R_1ρ (Fig. 2c and Supplementary Figs. 2a, b, 5, 6a, and 10b, c, d), CEST (Figs. 2c, f, 3a, c, 4b–d, and 5b and Supplementary Figs. 2b, 3a–c, 4a, c–e, 5, 6a, and 9a, d) and imino exchange (Supplementary Fig. 8c, d) data as well as kinetic simulation and prediction results (Fig. 6a and Supplementary Fig. 9b) generated in this study are provided at https://github.com/alhashimilab/m6A_hybridization_kinetics (https://doi.org/10.5281/zenodo.5106694)⁸². The force field parameters for m⁶A and m⁶₂A used in MD simulations and PDB files of these structures that were submitted to the DFT calculations (Fig. 4g and Supplementary Fig. 7c) are provided at https://github.com/alhashimilab/m6A_ES (https://doi.org/10.5281/zenodo.5099581)⁸³. The results of PDB (RCSB Protein Data Bank) survey for singly H-bond A–U bps (Supplementary Fig. 7f) are provided at https://github.com/alhashimilab/Singly_HB_AU (https://doi.org/10.5281/zenodo.5099558)⁸⁴. The DNA m⁶A sites (Fig. 6f) used in this study were reported in a prior study⁹. See Supplementary Table 5 of the cited paper (https://www.nature.com/articles/nature17640#Sec26). Source data for Figs. 2d, 4g, and 6a, c, e and Supplementary Figs. 2c, 4b, 6d, 7c, d, f, 8c, d, 9b, c, 10b, and 11a–d are provided with this paper.

Code availability

In-house Python scripts used to perform CEST fitting, kinetic simulations, and predictions (duplex hybridization, TAR conformational transitions, and genome-wide DNA hybridization prediction) are provided at https://github.com/alhashimilab/m6A_hybridization_kinetics (https://doi.org/10.5281/zenodo.5099562)⁸².

References

Meyer, K. D. et al. Comprehensive analysis of mRNA methylation reveals enrichment in 3’ UTRs and near stop codons. Cell 149, 1635–1646 (2012).
Article CAS PubMed PubMed Central Google Scholar
Dominissini, D. et al. Topology of the human and mouse m6A RNA methylomes revealed by m6A-seq. Nature 485, 201–206 (2012).
Article ADS CAS PubMed Google Scholar
Fu, Y., Dominissini, D., Rechavi, G. & He, C. Gene expression regulation mediated through reversible m(6)A RNA methylation. Nat. Rev. Genet. 15, 293–306, (2014).
Article CAS PubMed Google Scholar
Roundtree, I. A., Evans, M. E., Pan, T. & He, C. Dynamic RNA modifications in gene expression regulation. Cell 169, 1187–1200 (2017).
Article CAS PubMed PubMed Central Google Scholar
Zaccara, S., Ries, R. J. & Jaffrey, S. R. Reading, writing and erasing mRNA methylation. Nat. Rev. Mol. Cell Biol. 20, 608–624 (2019).
Article CAS PubMed Google Scholar
Vanyushin, B. F., Belozersky, A. N., Kokurina, N. A. & Kadirova, D. X. 5-methylcytosine and 6-methylamino-purine in bacterial DNA. Nature 218, 1066–1067, (1968).
Article ADS CAS PubMed Google Scholar
Douvlataniotis, K., Bensberg, M., Lentini, A., Gylemo, B. & Nestor, C. E. No evidence for DNA N (6)-methyladenine in mammals. Sci. Adv. 6, eaay3335 (2020).
Article ADS CAS PubMed PubMed Central Google Scholar
Li, Z. et al. N(6)-methyladenine in DNA antagonizes SATB1 in early development. Nature 583, 625–630 (2020).
Article ADS CAS PubMed PubMed Central Google Scholar
Wu, T. P. et al. DNA methylation on N(6)-adenine in mammalian embryonic stem cells. Nature 532, 329–333 (2016).
Article ADS CAS PubMed PubMed Central Google Scholar
Liu, N. et al. N(6)-methyladenosine-dependent RNA structural switches regulate RNA-protein interactions. Nature 518, 560–564 (2015).
Article ADS CAS PubMed PubMed Central Google Scholar
Huang, L., Ashraf, S., Wang, J. & Lilley, D. M. Control of box C/D snoRNP assembly by N6-methylation of adenine. EMBO Rep. 18, 1631–1645 (2017).
Article CAS PubMed PubMed Central Google Scholar
Roost, C. et al. Structure and thermodynamics of N6-methyladenosine in RNA: a spring-loaded base modification. J. Am. Chem. Soc. 137, 2107–2115 (2015).
Article CAS PubMed PubMed Central Google Scholar
Choi, J. et al. N(6)-methyladenosine in mRNA disrupts tRNA selection and translation-elongation dynamics. Nat. Struct. Mol. Biol. 23, 110–115 (2016).
Article CAS PubMed PubMed Central Google Scholar
Slobodin, B. et al. Transcription Impacts the Efficiency of mRNA Translation via Co-transcriptional N6-adenosine Methylation. Cell 169, 326–337.e312 (2017).
Article CAS PubMed PubMed Central Google Scholar
Louloupi, A., Ntini, E., Conrad, T. & Orom, U. A. V. Transient N-6-methyladenosine transcriptome sequencing reveals a regulatory role of m6A in splicing efficiency. Cell Rep. 23, 3429–3437 (2018).
Article CAS PubMed Google Scholar
Du, K. et al. Epigenetically modified N(6)-methyladenine inhibits DNA replication by human DNA polymerase eta. DNA Repair 78, 81–90 (2019).
Article CAS PubMed Google Scholar
Aschenbrenner, J. et al. Engineering of a DNA polymerase for direct m(6) A sequencing. Angew. Chem. Int. Ed. Engl. 57, 417–421 (2018).
Article CAS PubMed Google Scholar
Rangadurai, A., Szymaski, E. S., Kimsey, I. J., Shi, H. & Al-Hashimi, H. M. Characterizing micro-to-millisecond chemical exchange in nucleic acids using off-resonance R1rho relaxation dispersion. Prog. Nucl. Magn. Reson Spectrosc. 112–113, 55–102 (2019).
Article PubMed PubMed Central CAS Google Scholar
Palmer, A. G. 3rd & Massi, F. Characterization of the dynamics of biomacromolecules using rotating-frame spin relaxation NMR spectroscopy. Chem. Rev. 106, 1700–1719 (2006).
Article CAS PubMed Google Scholar
Palmer, A. G. 3rd Chemical exchange in biomacromolecules: past, present, and future. J. Magn. Reson. 241, 3–1 (2014).
Article ADS CAS PubMed PubMed Central Google Scholar
Shi, H. et al. NMR chemical exchange measurements reveal that N(6)-methyladenosine slows RNA annealing. J. Am. Chem. Soc. 141, 19988–19993 (2019).
Article CAS PubMed PubMed Central Google Scholar
Cisse, I. I., Kim, H. & Ha, T. A rule of seven in Watson-Crick base-pairing of mismatched sequences. Nat. Struct. Mol. Biol. 19, 623–627 (2012).
Article CAS PubMed PubMed Central Google Scholar
Xu, S. C. et al. Real-time reliable determination of binding kinetics of DNA hybridization using a multi-channel graphene biosensor. Nat. Commun. 8, 14902 (2017).
Article ADS CAS PubMed PubMed Central Google Scholar
Tawa, K. & Knoll, W. Mismatching base-pair dependence of the kinetics of DNA-DNA hybridization studied by surface plasmon fluorescence spectroscopy. Nucleic Acids Res. 32, 2372–2377 (2004).
Article CAS PubMed PubMed Central Google Scholar
Engel, J. D. & von Hippel, P. H. Effects of methylation on the stability of nucleic acid conformations: studies at the monomer level. Biochemistry 13, 4143–4158 (1974).
Article CAS PubMed Google Scholar
Engel, J. D. & von Hippel, P. H. Effects of methylation on the stability of nucleic acid conformations. Studies at the polymer level. J. Biol. Chem. 253, 927–934 (1978).
Article CAS PubMed Google Scholar
Hammes, G. G., Chang, Y. C. & Oas, T. G. Conformational selection or induced fit: a flux description of reaction mechanism. Proc. Natl Acad. Sci. USA 106, 13737–13741 (2009).
Article ADS CAS PubMed PubMed Central Google Scholar
Sekhar, A. et al. Conserved conformational selection mechanism of Hsp70 chaperone-substrate interactions. Elife 7, https://doi.org/10.7554/eLife.32764 (2018).
Zhao, B., Hansen, A. L. & Zhang, Q. Characterizing slow chemical exchange in nucleic acids by carbon CEST and low spin-lock field R(1rho) NMR spectroscopy. J. Am. Chem. Soc. 136, 20–23 (2014).
Article CAS PubMed Google Scholar
Vallurupalli, P., Bouvignies, G. & Kay, L. E. Studying “invisible” excited protein states in slow exchange with a major state conformation. J. Am. Chem. Soc. 134, 8148–8161 (2012).
Article CAS PubMed Google Scholar
Bouvignies, G. & Kay, L. E. A 2D (1)(3)C-CEST experiment for studying slowly exchanging protein systems using methyl probes: an application to protein folding. J. Biomol. NMR 53, 303–310 (2012).
Article CAS PubMed Google Scholar
Mcconnell, H. M. Reaction rates by nuclear magnetic resonance. J. Chem. Phys. 28, 430–431 (1958).
Article ADS CAS Google Scholar
Mulder, F. A., Mittermaier, A., Hon, B., Dahlquist, F. W. & Kay, L. E. Studying excited states of proteins by NMR spectroscopy. Nat. Struct. Biol. 8, 932–935 (2001).
Article CAS PubMed Google Scholar
Kimsey, I. J., Petzold, K., Sathyamoorthy, B., Stein, Z. W. & Al-Hashimi, H. M. Visualizing transient Watson-Crick-like mispairs in DNA and RNA duplexes. Nature 519, 315–320 (2015).
Article ADS CAS PubMed PubMed Central Google Scholar
Abramov, G., Velyvis, A., Rennella, E., Wong, L. E. & Kay, L. E. A methyl-TROSY approach for NMR studies of high-molecular-weight DNA with application to the nucleosome core particle. Proc. Natl Acad. Sci. USA 117, 12836–12846 (2020).
Article CAS PubMed PubMed Central Google Scholar
Liu, B. et al. A potentially abundant junctional RNA motif stabilized by m(6)A and Mg(2). Nat. Commun. 9, 2761 (2018).
Article ADS PubMed PubMed Central CAS Google Scholar
Koss, H., Rance, M. & Palmer, A. G. 3rd General expressions for R1rho relaxation for N-site chemical exchange and the special case of linear chains. J. Magn. Reson. 274, 36–45 (2017).
Article ADS CAS PubMed Google Scholar
Bhaswati Goswami, B. L. G. & Jones, R. A. Nitrogen-15-labeled oligodeoxynucleotides. 5. Use of 15N NMR to probe H-bonding in an 06MeG-T base pair. J. Am. Chem. Soc. 115, 3832–3833 (1993).
Article Google Scholar
Van Charldorp, R., Heus, H. A. & Van Knippenberg, P. H. Adenosine dimethylation of 16S ribosomal RNA: effect of the methylgroups on local conformational stability as deduced from electrophoretic mobility of RNA fragments in denaturing polyacrylamide gels. Nucleic Acids Res. 9, 267–275 (1981).
Article PubMed PubMed Central Google Scholar
Aboul-ela, F., Koh, D., Tinoco, I. Jr. & Martin, F. H. Base-base mismatches. Thermodynamics of double helix formation for dCA3XA3G + dCT3YT3G (X, Y = A,C,G,T). Nucleic Acids Res. 13, 4811–4824 (1985).
Article CAS PubMed PubMed Central Google Scholar
Bannwarth, S. & Gatignol, A. HIV-1 TAR RNA: the target of molecular interactions between the virus and its host. Curr. HIV Res. 3, 61–71 (2005).
Article CAS PubMed Google Scholar
Dethoff, E. A., Petzold, K., Chugh, J., Casiano-Negroni, A. & Al-Hashimi, H. M. Visualizing transient low-populated structures of RNA. Nature 491, 724–728 (2012).
Article ADS CAS PubMed PubMed Central Google Scholar
Chu, C. C., Plangger, R., Kreutz, C. & Al-Hashimi, H. M. Dynamic ensemble of HIV-1 RRE stem IIB reveals non-native conformations that disrupt the Rev-binding site. Nucleic Acids Res 47, 7105–7117 (2019).
Article CAS PubMed PubMed Central Google Scholar
Bisaria, N., Greenfeld, M., Limouse, C., Mabuchi, H. & Herschlag, D. Quantitative tests of a reconstitution model for RNA folding thermodynamics and kinetics. Proc. Natl Acad. Sci. USA 114, E7688–E7696 (2017).
Article CAS PubMed PubMed Central Google Scholar
Zhang, J. X. et al. Predicting DNA hybridization kinetics from sequence. Nat. Chem. 10, 91–98 (2018).
Article PubMed CAS Google Scholar
Abakir, A. et al. N(6)-methyladenosine regulates the stability of RNA:DNA hybrids in human cells. Nat. Genet 52, 48–55 (2020).
Article CAS PubMed Google Scholar
Konno, M. et al. Distinct methylation levels of mature microRNAs in gastrointestinal cancers. Nat. Commun. 10, 3888 (2019).
Article ADS PubMed PubMed Central CAS Google Scholar
Decatur, W. A. & Fournier, M. J. RNA-guided nucleotide modification of ribosomal and other RNAs. J. Biol. Chem. 278, 695–698 (2003).
Article CAS PubMed Google Scholar
Seraphin, B., Kretzner, L. & Rosbash, M. A U1 snRNA:pre-mRNA base pairing interaction is required early in yeast spliceosome assembly but does not uniquely define the 5’ cleavage site. EMBO J. 7, 2533–2538 (1988).
Article CAS PubMed PubMed Central Google Scholar
Will, C. L. & Luhrmann, R. Spliceosome structure and function. Cold Spring Harb. Perspect Biol. 3, https://doi.org/10.1101/cshperspect.a003707 (2011).
Klinge, S. & Woolford, J. L. Jr. Ribosome assembly coming into focus. Nat. Rev. Mol. Cell Biol. 20, 116–131 (2019).
Article CAS PubMed PubMed Central Google Scholar
Kierzek, E. & Kierzek, R. The thermodynamic stability of RNA duplexes and hairpins containing N6-alkyladenosines and 2-methylthio-N6-alkyladenosines. Nucleic Acids Res. 31, 4472–4480 (2003).
Article CAS PubMed PubMed Central Google Scholar
Xu, C. et al. Structural basis for selective binding of m6A RNA by the YTHDC1 YTH domain. Nat. Chem. Biol. 10, 927–929 (2014).
Article CAS PubMed Google Scholar
Zimmer, D. P. & Crothers, D. M. NMR of enzymatically synthesized uniformly 13C15N-labeled DNA oligonucleotides. Proc. Natl Acad. Sci. USA 92, 3091–3095 (1995).
Article ADS CAS PubMed PubMed Central Google Scholar
Delaglio, F. et al. NMRPipe: a multidimensional spectral processing system based on UNIX pipes. J. Biomol. NMR 6, 277–293 (1995).
Article CAS PubMed Google Scholar
Nikolova, E. N., Gottardo, F. L. & Al-Hashimi, H. M. Probing transient Hoogsteen hydrogen bonds in canonical duplex DNA using NMR relaxation dispersion and single-atom substitution. J. Am. Chem. Soc. 134, 3667–3670 (2012).
Article CAS PubMed PubMed Central Google Scholar
Nikolova, E. N. et al. Transient Hoogsteen base pairs in canonical duplex DNA. Nature 470, 498–502 (2011).
Article ADS CAS PubMed PubMed Central Google Scholar
Hansen, A. L., Nikolova, E. N., Casiano-Negroni, A. & Al-Hashimi, H. M. Extending the range of microsecond-to-millisecond chemical exchange detected in labeled and unlabeled nucleic acids by selective carbon R(1rho) NMR spectroscopy. J. Am. Chem. Soc. 131, 3818–3819 (2009).
Article CAS PubMed Google Scholar
Bothe, J. R., Stein, Z. W. & Al-Hashimi, H. M. Evaluating the uncertainty in exchange parameters determined from off-resonance R1rho relaxation dispersion for systems in fast exchange. J. Magn. Reson. 244, 18–29 (2014).
Article ADS CAS PubMed PubMed Central Google Scholar
Abou Assi, H. et al. 2’-O-Methylation can increase the abundance and lifetime of alternative RNA conformational states. Nucleic Acids Res. https://doi.org/10.1093/nar/gkaa928 (2020).
Rangadurai, A., Shi, H. & Al-Hashimi, H. M. Extending the sensitivity of CEST NMR spectroscopy to micro-to-millisecond dynamics in nucleic acids using high-power radio-frequency fields. Angew. Chem. Int. Ed. Engl. 59, 11262–11266 (2020).
Article CAS PubMed PubMed Central Google Scholar
Vallurupalli, P., Sekhar, A., Yuwen, T. & Kay, L. E. Probing conformational dynamics in biomolecules via chemical exchange saturation transfer: a primer. J. Biomol. NMR 67, 243–271 (2017).
Article PubMed CAS Google Scholar
Yuwen, T. & Kay, L. E. Longitudinal relaxation optimized amide (1)H-CEST experiments for studying slow chemical exchange processes in fully protonated proteins. J. Biomol. NMR 67, 295–307 (2017).
Article CAS PubMed Google Scholar
Yanjiao Wang, G. H., Jiang, X., Yuwen, T. & Xue, Y. Chemical shift prediction of RNA imino groups: application toward characterizing RNA excited states. Nat. Commun. 12, 1595 (2021).
Gueron, M., Kochoyan, M. & Leroy, J. L. A single mode of DNA base-pair opening drives imino proton exchange. Nature 328, 89–92 (1987).
Article ADS CAS PubMed Google Scholar
Szulik, M. W., Voehler, M. & Stone, M. P. NMR analysis of base-pair opening kinetics in DNA. Curr. Protoc. Nucleic Acid Chem. 59, 7 20 21–18 (2014).
Article Google Scholar
Sorgenfrei, S. et al. Label-free single-molecule detection of DNA-hybridization kinetics with a carbon nanotube field-effect transistor. Nat. Nanotechnol. 6, 126–132 (2011).
Article ADS CAS PubMed PubMed Central Google Scholar
Wallace, M. I., Ying, L., Balasubramanian, S. & Klenerman, D. Non-Arrhenius kinetics for the loop closure of a DNA hairpin. Proc. Natl Acad. Sci. USA 98, 5584–5589 (2001).
Article ADS CAS PubMed PubMed Central Google Scholar
Bloomfield, V. A. et al. Nucleic Acids: Structure, Properties, and Functions (University Science Books, 2000).
Lu, X. J. & Olson, W. K. 3DNA: a software package for the analysis, rebuilding and visualization of three-dimensional nucleic acid structures. Nucleic Acids Res. 31, 5108–5121 (2003).
Article CAS PubMed PubMed Central Google Scholar
Rangadurai, A. et al. Why are Hoogsteen base pairs energetically disfavored in A-RNA compared to B-DNA? Nucleic Acids Res. 46, 11099–11114 (2018).
CAS PubMed PubMed Central Google Scholar
Aduri, R. et al. AMBER force field parameters for the naturally occurring modified nucleosides in RNA. J. Chem. Theory Comput. 3, 1464–1475 (2007).
Article CAS PubMed Google Scholar
Swails, J., Zhu, T., He, X. & Case, D. A. AFNMR: automated fragmentation quantum mechanical calculation of NMR chemical shifts for biomolecules. J. Biomol. NMR 63, 125–139 (2015).
Article CAS PubMed PubMed Central Google Scholar
Shi, H. et al. Rapid and accurate determination of atomistic RNA dynamic ensemble models using NMR and structure prediction. Nat. Commun. 11, 5531 (2020).
Article ADS CAS PubMed PubMed Central Google Scholar
Richardson, W. H., Peng, C., Bashford, D., Noodleman, L. & Case, D. A. Incorporating solvation effects into density functional theory: calculation of absolute acidities. Int. J. Quantum Chem. 61, 207–217 (1997).
Article CAS Google Scholar
Orlovsky, N. I., Al-Hashimi, H. M. & Oas, T. G. Exposing hidden high-affinity RNA conformational states. J. Am. Chem. Soc. 142, 907–921 (2020).
Article CAS PubMed Google Scholar
Meagher, N. E. & Rorabacher, D. B. Mathematical treatment for very rapid 2nd-order reversible kinetics as measured by stopped-flow spectrophotometry with corrections for the cell concentration gradient. J. Phys. Chem. 98, 12590–12593 (1994).
Article CAS Google Scholar
Guo, Q., Lu, M. & Kallenbach, N. R. Effect of hemimethylation and methylation of adenine on the structure and stability of model DNA duplexes. Biochemistry 34, 16359–16364 (1995).
Article CAS PubMed Google Scholar
Berman, H. M. et al. The Protein Data Bank. Nucleic Acids Res. 28, 235–242 (2000).
Article ADS CAS PubMed PubMed Central Google Scholar
Lu, X. J., Bussemaker, H. J. & Olson, W. K. DSSR: an integrated software tool for dissecting the spatial structure of RNA. Nucleic Acids Res. 43, e142 (2015).
Article PubMed PubMed Central CAS Google Scholar
Leontis, N. B. & Westhof, E. Geometric nomenclature and classification of RNA base pairs. RNA 7, 499–512 (2001).
Article CAS PubMed PubMed Central Google Scholar
Liu, B. et al. A quantitative model predicts how m6A reshapes the kinetic landscape of nucleic acid hybridization and conformational transitions. Bei-1205/m6A_hybridization_kinetics https://doi.org/10.5281/zenodo.5106694 (2021).
Liu, B. et al. A quantitative model predicts how m6A reshapes the kinetic landscape of nucleic acid hybridization and conformational transitions. alhashimilab/m6A_ES https://doi.org/10.5281/zenodo.5099581 (2021).
Liu, B. et al. A quantitative model predicts how m6A reshapes the kinetic landscape of nucleic acid hybridization and conformational transitions. Bei-1205/Singly_HB_AU https://doi.org/10.5281/zenodo.5099558 (2021).

Download references

Acknowledgements

We thank members of the Al-Hashimi laboratory for their assistance and critical comments on the manuscript. We would like to thank Yanjiao Wang and Prof. Xue Yi (Tsinghua University) for the help with the ¹H CEST experiment, Prof. Terrence Oas (Duke University) for advice about kinetic simulations and calculations, and Prof. Qi Zhang for providing the 2D [¹³C, ¹H] CEST pulse sequence based on which the methyl CEST sequence was derived. This work was supported by the US National Institute for General Medical Sciences (1R01GM132899) and US National Institute of Health (R01GM089846) to H.M.A., the Austrian Science Fund (FWF, project P30370 and P32773) and the Austrian Research Promotion Agency FFG (West Austrian BioNMR, 858017) to C.K. and the National Institute for Allergy and Infectious Diseases (U54 AI150470) to D.A.C.

Author information

Chia-Chieh Chu
Present address: Department of Chemistry, University of Chicago, Chicago, IL, USA

Authors and Affiliations

Department of Biochemistry, Duke University School of Medicine, Durham, NC, USA
Bei Liu, Atul Rangadurai, Chia-Chieh Chu & Hashim M. Al-Hashimi
Department of Chemistry, Duke University, Durham, NC, USA
Honglue Shi & Hashim M. Al-Hashimi
Institute of Organic Chemistry and Center for Molecular Biosciences Innsbruck (CMBI), University of Innsbruck, Innsbruck, Austria
Felix Nussbaumer, Kevin Andreas Erharter & Christoph Kreutz
Department of Chemistry and Chemical Biology, Rutgers University, Piscataway, NJ, USA
David A. Case

Authors

Bei Liu
View author publications
You can also search for this author in PubMed Google Scholar
Honglue Shi
View author publications
You can also search for this author in PubMed Google Scholar
Atul Rangadurai
View author publications
You can also search for this author in PubMed Google Scholar
Felix Nussbaumer
View author publications
You can also search for this author in PubMed Google Scholar
Chia-Chieh Chu
View author publications
You can also search for this author in PubMed Google Scholar
Kevin Andreas Erharter
View author publications
You can also search for this author in PubMed Google Scholar
David A. Case
View author publications
You can also search for this author in PubMed Google Scholar
Christoph Kreutz
View author publications
You can also search for this author in PubMed Google Scholar
Hashim M. Al-Hashimi
View author publications
You can also search for this author in PubMed Google Scholar

Contributions

B.L. and H.M.A. conceived the project and experimental design. B.L. prepared NMR samples, performed NMR experiments, and analyzed NMR data with the help from H.S., A.R. and C.-C.C., F.N., K.A.E. and C.K. prepared (¹³CH₃)-m⁶A RNA phosphoramidite and ¹³C8,¹³C2-labeled m⁶dA phosphoramidite. B.L. performed kinetic simulations and predictions. H.S. performed proton CEST and imino proton exchange experiments. A.R. performed MD simulations. H.S. and D.A.C. performed Automated fragmentation quantum mechanics/molecular mechanics (AF-QM/MM) chemical-shift calculations. B.L. and H.S. performed the PDB survey. H.M.A. and B.L. wrote the manuscript with critical input from H.S. and A.R.

Corresponding author

Correspondence to Hashim M. Al-Hashimi.

Ethics declarations

Competing interests

H.M.A. is an advisor to and holds an ownership interest in Nymirum, an RNA-based drug discovery company. C.K. is an advisor to and holds an ownership interest in INNotope, a company providing RNA stable isotope labeling products. The remaining authors declare no competing interests.

Additional information

Peer review information Nature Communications thanks Joseph Puglisi and other, anonymous, reviewers for their contributions to the peer review of this work. Peer review reports are available.

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

Peer Review File

Reporting Summary

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions

About this article

Cite this article

Liu, B., Shi, H., Rangadurai, A. et al. A quantitative model predicts how m⁶A reshapes the kinetic landscape of nucleic acid hybridization and conformational transitions. Nat Commun 12, 5201 (2021). https://doi.org/10.1038/s41467-021-25253-8

Download citation

Received: 06 January 2021
Accepted: 21 July 2021
Published: 31 August 2021
DOI: https://doi.org/10.1038/s41467-021-25253-8

This article is cited by

Shedding light on the base-pair opening dynamics of nucleic acids in living human cells
- Yudai Yamaoki
- Takashi Nagata
- Masato Katahira
Nature Communications (2022)

Comments

By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.

Subjects

ABSTRACT

Similar content being viewed by others

DNA mismatches reveal conformational penalties in protein–DNA recognition

Interaction preferences between protein side chains and key epigenetic modifications 5-methylcytosine, 5-hydroxymethycytosine and N6-methyladenine

The regulation and functions of DNA and RNA G-quadruplexes

Introduction

Results

Kinetics of m6A methylamino isomerization in ssRNA

m6A(anti)–U and A–U have similar thermodynamic stabilities in dsRNA

Testing the CS kinetic model

An additional hybridization intermediate at T = 55 °C

The dsRNA hybridization intermediate features an m6(syn)A···U stabilized by a single H-bond

Chemical-shift fingerprinting the m6(syn)A···U ES using m6 2A

m6(syn)A···U behaves like a mismatch

Kinetic model for m6A hybridization via conformation selection and IF

A quantitative model predicts how m6A reshapes the hybridization kinetics of DNA and RNA duplexes

Testing kinetic model on RNA conformational transitions

Discussion

Methods

Sample preparation

AMP and m6AMP

Oligonucleotides

Definition of rate constants

NMR experiments

Resonance assignments

13C and 15N R 1ρ relaxation dispersion

Analysis of R 1ρ data

Estimate p ES of methylated TAR from chemical shifts

13C and 15N CEST

2D CEST for 13C methyl probes

Analysis of the CEST data

1H CEST experiment

Imino proton exchange experiment

Measuring the kinetics of duplex hybridization from CEST data

Validation of NMR RD measurements on m6A RNA hybridization

UV melting experiments

MD simulations

Automated fragmentation quantum mechanics/molecular mechanics chemical-shift calculations

Free energy decomposition along the CS pathway

B–M simulations and constrained fits

Three-state CS simulations and constrained fits for the \({{{{{\mathrm{dsA6DNA}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}\)RD data measured at T = 65 °C

Three-state IF simulations and constrained fits for the \({{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}\) RD data were measured at T = 65 and 55 °C

Four-state CS + IF simulations and constrained fits for the \({{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}\) RD data at T = 55 °C

Four-state constrained fits for the CS + IF model for \({{{{{\mathrm{dsGGACU}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}\) at T = 65 °C

Four-state constrained fits for the CS + IF model for \({{{{{\mathrm{dsHCV}}}}}}{{{{{\mathrm{m}}}}}}^{6}{{{{{\mathrm{A}}}}}}\) and \({{{{{\mathrm{dsA6DNA}}}}}}^{{{{{{\mathrm{m}}}}}}^6{{{{{\mathrm{A}}}}}}}\)

Flux calculations

Model to predict apparent k on and k off for methylated RNA/DNA duplexes and TAR

Predict m6A-induced slowdown of DNA hybridization in the mouse genome

Survey of single H-bonded A–U bps in PDB structures

Reporting summary

Data availability

Code availability

References

Acknowledgements

Author information

Authors and Affiliations

Contributions

Corresponding author

Ethics declarations

Competing interests

Additional information

Supplementary information

Supplementary information

Peer Review File

Reporting Summary

Rights and permissions

About this article

Cite this article

Share this article

This article is cited by

Shedding light on the base-pair opening dynamics of nucleic acids in living human cells

Comments

Search

Quick links

Kinetics of m⁶A methylamino isomerization in ssRNA

m⁶A(anti)–U and A–U have similar thermodynamic stabilities in dsRNA

The dsRNA hybridization intermediate features an m⁶(syn)A···U stabilized by a single H-bond

Chemical-shift fingerprinting the m⁶(syn)A···U ES using m⁶ ₂A

m⁶(syn)A···U behaves like a mismatch

Kinetic model for m⁶A hybridization via conformation selection and IF

A quantitative model predicts how m⁶A reshapes the hybridization kinetics of DNA and RNA duplexes

AMP and m⁶AMP

¹³C and ¹⁵N R _1ρ relaxation dispersion

Analysis of R _1ρ data

Estimate p _ES of methylated TAR from chemical shifts

¹³C and ¹⁵N CEST

2D CEST for ¹³C methyl probes

¹H CEST experiment

Validation of NMR RD measurements on m⁶A RNA hybridization

Model to predict apparent k _on and k _off for methylated RNA/DNA duplexes and TAR

Predict m⁶A-induced slowdown of DNA hybridization in the mouse genome