Dynamic Basis for dG•dT misincorporation via tautomerization and ionization

Abstract

Tautomeric and anionic Watson-Crick-like mismatches play important roles in replication and translation errors through mechanisms that are not fully understood. Using NMR relaxation dispersion, we resolved a sequence-dependent kinetic network connecting G•T/U wobbles with three distinct Watson-Crick mismatches consisting of two rapidly exchanging tautomeric species (G^enol•T/U⇌G•T^enol/U^enol; population <0.4%) and one anionic species (G•T⁻/U⁻; population ≈0.001% at neutral pH). Inserting the sequence-dependent tautomerization/ionization step into a minimal kinetic mechanism for correct incorporation during replication following initial nucleotide binding leads to accurate predictions of dG•dT misincorporation probability across different polymerases, pH conditions, and for a chemically modified nucleotide, and provides mechanisms for sequence-dependent misincorporation. Our results indicate that the energetic penalty for tautomerization/ionization accounts for ≈10⁻²−10⁻³–fold discrimination against misincorporation, which proceeds primarily via tautomeric dG^enol•dT and dG•dT^enol with contributions from anionic dG•dT⁻ dominating at pH ≥8.4 or for some mutagenic nucleotides.

In their paper describing the structure of the DNA double helix¹, Watson and Crick proposed that if nucleotide bases adopted their energetically unfavorable tautomeric forms, mismatches (Fig. 1a) could pair up in a Watson-Crick-like (WC-like) geometry (Fig. 1b) and potentially give rise to spontaneous mutations. Decades later, it is well established that the replicative and translational machineries form a tight grip around the WC geometry to discriminate against mismatches^2–5. There is also evidence that both tautomeric^6–13 (Fig. 1b) and anionic^7–9,14,15 (Fig. 1c) WC-like mismatches can evade such fidelity checkpoints and give rise to replication^6,7 and translation errors¹⁶. Despite their centrality to the fidelity of information transfer in the central dogma of molecular biology, and growing evidence showing the involvement of spontaneous mutations in generating cancer causing mutations¹⁷, the very existence of these species and their contributions to replication and translation errors remain to be definitively established.

Figure 1. Tilting rapid tautomeric equilibria in excited state WC-like mismatches.

Chemical structures of ground state wobble (a) and excited state tautomeric (b) and anionic (c) WC-like G•T/U (X = H or CH₃ for uridine and thymidine, respectively) mismatches. d, Tilting rapid tautomeric equilibrium using sequence (cyan), structure (red), and chemical modifications (green). X-X′ and Y-Y′ denote WC base pairs adjacent to the G•T/U mismatches. e, Perturbations that differentially tilt the tautomeric equilibrium are expected to give rise to anti-correlated linear changes in the <∆ω_G-N1> and <∆ω_T/U-N3> values (color-coded).

Tautomeric and anionic mismatches come in a variety of chemical forms (Extended Data Fig. 1). For example, WC-like G•T/U mismatches can form when either the guanine (G^enol•T/U and G⁻•T/U) or thymidine/uridine (G•T^enol/U^enol and G•T⁻/U⁻) base assumes a rare enolic (Fig. 1b) or anionic (Fig. 1c) form. While it remains unclear which WC-like mismatch contributes to replication and translation errors, factors (e.g. changes in pH^7,8,14,18 and chemical modifications¹⁹) that stabilize different forms have been shown to increase misincorporation probabilities^14,20. Misincorporation probabilities can also vary significantly with sequence context through mechanisms that remain poorly understood^21,22. Resolving these different WC-like mismatches and their chemical dynamics is key for elucidating their potential roles in replication, transcription, and translation errors. However, this presents a formidable challenge to current biophysical methods because these mismatches differ by the placement of a single proton and π-bond (Fig. 1b,c and Extended Data Fig. 1). Protons are generally invisible to X-ray crystallography and cryo-EM¹², and consequently it has not been possible to unambiguously resolve the identity of WC-like mismatches captured within active sites of polymerases^6,7,15,23 and the ribosome decoding site^9,24. Moreover, WC-like mismatches are predicted to exist in rapid tautomeric (G^enol•T/U⇌G•T^enol/U^enol)^25,26 equilibria (Fig. 1b,c and Extended Data Fig. 1) making them exceptionally difficult to capture experimentally.

Techniques based on NMR relaxation dispersion (RD)^27–29 make it possible to characterize low-abundance short-lived conformational states, or ‘excited states’ (ESs), in biomolecules³⁰. Using NMR RD, we recently provided evidence that wobble G•T/U mismatches exist in dynamic equilibrium with tautomeric (ES1) and anionic (ES2) WC-like mismatches within DNA and RNA duplexes^8,31. The guanine N1 (G-N1) and thymidine/uridine N3 (T/U-N3) chemical shifts measured for tautomeric ES1 were consistent with G^enol•T/U, but were partially skewed toward G•T^enol/U^enol. This was interpreted as evidence for a rapid (on the chemical shift timescale) equilibrium between a major G^enol•T/U and minor G•T^enol/U^enol species⁸. The anionic ES2 was only detectable at high pH (≥7.8) and was heavily skewed in favor of G•T⁻/U⁻ with no evidence for G⁻•T/U. The roles of these various WC-like mismatches in replication and translation errors remains unknown. Here, by combining NMR RD and measurements of misincorporation rates, we resolved a kinetic network connecting two distinct tautomeric and one anionic WC-like mismatches, and established their relative contributions to dG•dTTP misincorporation.

Tilting the tautomeric equilibrium

If ES1 does indeed represent two tautomeric species in rapid equilibrium (Fig. 1b), it should be feasible to tilt the equilibrium (K_t = p_Genol/p_Tenol/Uenol) by changing the local sequence or structural context around the mismatch, or by using base modifications (Fig. 1d). This in turn should lead to very specific changes in the ES1 G-N1 and T/U-N3 chemical shifts, which are population weighted averages over the two species (Fig. 1e; left). Tilting the equilibrium in favor of G^enol•T/U should induce a downfield shift in the ES1 G-N1, because it increases the population of deprotonated G^enol, and an upfield shift in ES1 T/U-N3, because it decreases the population of deprotonated T^enol/U^enol, and vice versa (Fig. 1e; left). ∆ω_T/U-N3 versus ∆ω_G-N1 is predicted to be linear (Fig. 1e; right) with negative slope and intercept determined by the fundamental chemical shifts of the tautomeric species (Equation 1).

We measured ¹⁵N RD for five dG•dT mismatches within distinct sequence contexts and for thirteen rG•rU mismatches in nine structurally unique non-coding RNAs (Fig. 2a and Extended Data Fig. 2a). Experiments were carried out at near-neutral pH (6.4–6.9) so as to lower the anionic ES2 below detection⁸ (Extended Data Fig. 2b). The RD experiments measure spin-relaxation rates in the rotating frame (R_1ρ) during a relaxation period in which a radiofrequency field is applied with variable offset (Ω 2π⁻¹, in Hz) and power (ω 2π⁻¹, in Hz) to suppress the chemical exchange contribution (R_ex) to the transverse spin relaxation rate (R₂) arising due to chemical exchange between the energetically more stable ground state (GS) and ES^27,28.

Figure 2. Resolving rapidly interconverting tautomers.

a, Representative hairpin (“hp”) DNA, RNA, and chemically modified constructs. Name denotes mismatch and sequence context (5′-3′). b, Representative G-N1 and T/U-N3 RD profiles (pH 6.8-6.9 and 25 °C). Best fits to the B-M equations are shown. c, <∆ω_T/U-N3> versus <∆ω_G-N1> ES1 chemical shift differences measured by NMR RD. Blue line indicates fit to the ∆ω’s using the fundamental tautomer chemical shifts as variables (Methods). Red line indicates DFT-predictions. d, Lower bounds for the rate of tautomeric G^enol•T/U⇌G•T^enol/U^enol exchange. Contour plots showing scaled ${\overline{\chi }}^2$ weights for combinations of k_Tenol/Uenol versus k_t; red indicates better fit. e, A plot of K_t versus tautomeric ES1 population for DNA (blue, n=5) and RNA (cyan, n=6) constructs determined at pH 6.9 and 25 °C. f, Change in stacking overlap (∆Ų=Å^2(WC)–Å^2(WB)) between wobble and Watson-Crick-like mismatches versus ES1 population (pH 6.9 and 25 °C) for five DNA sequence contexts (Methods). Error bars in b, c, e reflect experimental uncertainty (one s.d., Methods).

We observed G-N1 and T/U-N3 RD consistent with WC-like ES1 exchange for all five dG•dT and eight rG•rU mismatches located within helical environments (Fig. 2b and Extended Data Fig. 3a), thus establishing their robust occurrence in DNA and RNA. No RD was observed (Extended Data Fig. 3b) for rG•rU mismatches adjacent to apical loops, three-way junctions, or bulges (Extended Data Fig. 2a). This could be due to the lower abundance of WC-like mismatches when outside the grip of the helical environment, though we cannot rule out that the exchange is orders of magnitude faster and beyond detection.

As predicted based on variably tilting the G^enol•T/U⇌G•T^enol/U^enol equilibrium (Fig. 1e; right), the fitted <∆ω_G-N1> and <∆ω_T/U-N3> values obtained from two-state analysis (GS⇌ES1) of the RD profiles (Fig. 2b; Extended Data Figure 3a; and Supplementary Table 1) fell along a line with negative slope (Fig. 2c). As a negative control, no correlation was observed between the corresponding GS G-N1 and T/U-N3 chemical shifts (Extended Data Fig. 3c). We confirmed these linear trends using chemical modifications that tilt the tautomeric equilibrium toward enolic dT (dG•dU and dG•^5BrdU) or enolic dG (^8BrdG•dT) (Fig. 2c; Extended Data Fig. 3a; and Supplementary Discussion 1).

Sequence-dependent G^enol•T/U⇌G•T^enol/U^enol

A linear fit to the <∆ω_T/U-N3> versus <∆ω_G-N1> values, assuming physically reasonable ranges, yielded fundamental chemical shifts for the tautomeric species that are in excellent agreement with values predicted by DFT calculations (Fig. 2c and Supplementary Tables 2,3)⁸. The tautomeric equilibria (Supplementary Table 2) obtained from this analysis and from re-fitting the RD data using a 3-state model with linear topology²⁵ (wobble⇌G•T^enol/U^enol⇌G^enol•T/U) are slightly tilted in favor of dG^enol•dT in DNA (K_t = 2.1-4.6) whereas the populations of rG^enol•rU and rG•rU^enol are more comparable in RNA (K_t = 0.5-1.1). These differences may be attributed to the electron-donating methyl group in dT which destabilizes dT^enol relative to rU^enol(32). The RD data also allowed us to estimate a lower bound for the fast tautomeric exchange rate $k_t=k_{G^{\mathit{enol}}\to T^{\mathit{enol}}}+k_{T^{\mathit{enol}}\to G^{\mathit{enol}}}$ >≈500,000-1,000,000 s⁻¹ (Fig. 2d and Extended Data Fig. 4)³³ and a G•T^enol/U^enol→G^enol•T/U transition state barrier <9-10 kcal mol⁻¹ (pre-exponential factor = k_BT h⁻¹⁽³⁴⁾ and κ = 1) that is in good agreement with values (≈11.5 kcal mol⁻¹) reported using computational methods²⁵. These results establish the existence of G•T^enol/U^enol and G^enol•T/U in an ultra-fast equilibrium, each of which can potentially contribute to replication and translation errors.

Interestingly, the exchange parameters vary significantly with sequence context (Supplementary Table 1). The ES1 population (p_ES1 = p_Genol + p_Tenol/Uenol) varies 3-fold in DNA and 8-fold in RNA while the forward (k_{GS →ES1}) and reverse (k_ES1→GS) rate constants vary by 4- and 5-fold, respectively, for DNA, and by 38- and 6-fold, respectively, for RNA (Supplementary Table 1). A linear correlation is observed between p_ES1 and K_t (Fig. 2e and Supplementary Table 2), indicating that the G^enol•T/U population dominates these variations with sequence and structural context. In DNA, these variations can potentially be explained by sequence-specific changes in stacking with the immediate neighbors that accompany the transition from the wobble to Watson-Crick geometry (Fig. 2f). For example, GGG, has the highest p_ES1 and is predicted to gain stacking overlap whereas CGG has the lowest p_ES1 and is predicted to lose stacking overlap. Similar sequence-dependent effects have been reported in lesion repair by methyltransferases³⁵. Interestingly, dG dominates the changes in stacking, potentially explaining the stronger sequence dependence of the G^enol•T/U population compared to G•T^enol/U^enol.

Sequence-dependent anion equilibria

Next, we examined whether anionic WC-like G•T⁻/U⁻ (Fig. 1c) also form robustly in DNA and RNA and whether anionic G⁻•T/U remains undetectable under these different environments. We measured RD at high pH (≥7.8) for G•T/U mismatches in a subset of our RNA (Fig. 3a and Extended Data Fig. 5a) and DNA constructs (Fig. 3b and Extended Data Fig. 5b). In all cases, we observed RD consistent with wobble⇌anion exchange, thus establishing the robustness of this process across different sequence contexts (Fig. 3a,b and Extended Data Fig. 5). 3-state fitting of the RD data assuming a starlike topology (Extended Data Fig. 6a) yielded large ∆ω_T/U-N3(ES2) ≈55 p.p.m. and much smaller ∆ω_G-N1(ES2) ≈5 p.p.m., consistent with a dominant G•T⁻/U⁻ species and with no evidence for G⁻•T/U. Again, we observe strong sequence-specific variations in ES2 population (p_ES2) and the k_GS→ES2 and k_ES2→GS rates robustly across different temperatures and pH conditions (Supplementary Tables 4,5).

Figure 3. 3-State exchange with triangular topology and minor exchange between tautomeric and anionic WC-like excited states.

Comparison of 3-state B-M fit with triangular (left) and starlike (right) topology to the RD profiles measured in (a) hpUG-CGC RNA and (b) hpTG-GGC DNA. Statistical AIC (w_AIC) and BIC (w_BIC) weights comparing starlike and triangular topologies are shown. Error bars reflect experimental uncertainty (one s.d., Methods).

In a previous study⁸, the emergence of anionic ES2 at high pH was accompanied by unexpected changes in the ES1 tautomeric chemical shifts. Similar deviations are observed here for both RNA and DNA (Supplementary Table 4). We postulated that ‘minor’ exchange³⁶ between ES1 and ES2 could ‘mix’ their chemical shifts and give rise to such deviations (Extended Data Fig. 6a,b and Supplementary Table 4). Indeed, all five RD profiles with peculiar ES1 chemical shifts showed a statistically significant improvement when fitting data to a 3-state model with minor exchange in a triangular rather than starlike topology (Fig. 3a,b; Extended Data Fig. 6c,d; and Supplementary Tables 4–6). The resulting ES1 ¹⁵N rG-N1 and rU-N3 chemical shifts vary less significantly with pH (Extended Data Fig. 6e and Supplementary Table 4) and the rate constants (k_ES1→ES2 and k_ES2→ES1) exhibit the expected temperature dependence (Extended Data Fig. 6f), neither of which is expected if the data were being spuriously over fitted. Additionally, limited or poor quality RD data can make it difficult to resolve different topologies³⁷ (Supplementary Table 6).

Tautomerization/ionization during misincorporation

dG•dT misincorporation is the most frequent base substitution error committed by high fidelity DNA polymerases with misincorporation frequency F_pol= (k_pol/K_d)_incorrect/(k_pol/K_d)_correct ≈10⁻⁴−10⁻⁵ for most studied polymerases^38,39, i.e. an error is committed with frequency of 1 in 10⁴−10⁵ nucleotide incorporations, in which k_pol is the maximum rate of nucleotide incorporation and K_d is the apparent nucleotide equilibrium dissociation constant. Differences in nucleotide binding affinities (K_d^{−1(incorrect)}/K_d^{−1(correct)}) only account for a factor of ≈10⁻¹−10⁻² in discrimination⁴⁰, whereas differences in the polymerization rates (k_pol^incorrect/k_pol^correct) account for ≈10⁻³.

The mechanisms that lower the values of k_pol^incorrect relative to k_pol^correct remain poorly understood. Many decades ago, Topal and Fresco postulated that the frequency of tautomerization may be an important determinant of misincorporation probability¹¹. Interestingly, the population of the tautomeric species ≈10⁻³ is comparable to the k_pol^incorrect/k_pol^correct values. In addition, the rate at which the wobble dG•dT forms either the WC-like tautomeric (k_GS→ES1 = 0.3-10 s⁻¹; Supplementary Tables 1,5) or anionic (k_GS→ES2 = 1.1-124 s⁻¹; Supplementary Tables 1,5) mismatches (Fig. 4a) is comparable to k_pol^incorrect (0.16-1.16 s⁻¹) measured for incorrect dG•dTTP or dGTP•dT misincorporation^22,39 whereas it is up to ≈1000-fold slower than k_pol^correct (25-275 s⁻¹) measured for correct dG•dCTP or dGTP•dC^22,39. If formation of WC-like dG•dT mismatches (Fig. 4a) is required for misincorporation following initial dNTP binding in a wobble conformation, it could provide a mechanism for lowering k_pol^incorrect relative to k_pol^correct. Indeed, prior studies have shown that DNA polymerases cannot undergo the necessary conformational changes needed for catalysis when dG•dT is in a wobble conformation⁷ and all available structures of catalytically active polymerases with bound mismatches within the active site feature WC-like dG•dT or dA•dC geometries^6,7. Similarly, WC-like rG•rU mismatches have been shown to form in the first and second codon positions of catalytically active ribosomes⁹, in which wobbles are typically rejected⁵, potentially helping to explain translational error hotspots⁴¹.

Figure 4. Kinetic mechanism for dG•dT misincorporation.

a, Triangular exchange between wobble G•T/U mispair, rapidly interconverting WC-like tautomers (G^enol•T/U⇌G•T^enol/U^enol), and anionic WC-like G•T⁻/U⁻. Exchange between anionic G•T⁻/U⁻ and a low-abundance short-lived anionic G⁻•T/U or other non-WC species that fall outside RD detection cannot be ruled out. WC-like G•T/U populations and ranges at pH 6.4-8.9 and 10-25 °C (Supplementary Tables 1, 5). b, Minimal kinetic mechanism for polymerization⁴². Incorporation of an incorrect dNTP includes an additional tautomerization/ionization step allowing for the formation of a Watson-Crick-like dG•dT mismatch. Discriminatory steps are in red.

To examine this possibility, we built a kinetic model for dG•dTTP misincorporation by inserting a tautomerization/ionization step (Fig. 4a) following initial nucleotide binding in a wobble conformation and prior to the pre-chemistry conformational change in the existing minimal kinetic model for correct incorporation⁴² (Fig. 4b). All other steps, including the pre-chemistry conformational change and phosphodiester bond formation, are assumed to have identical kinetic parameters as measured for correct nucleotide incorporation^39,43–46 (Supplemental Table 7). The model assumes that misincorporation directly from the wobble conformation is negligible and that the tautomerization and ionization rates measured in duplex DNA by NMR approximate the rates in the polymerase active site. We tested models (Extended Data Fig. 7) in which either the tautomeric (M_ES1), anionic (M_ES2), or both (M_ES1+ES2) species can be misincorporated as well as models that excluded the triangular network all together (M_Kd).

Strikingly, the most general M_ES1+ES2 model robustly predicts the measured F_pol values for three polymerases (T7, polymerase ε, and polymerase β) that have varying rate limiting steps and microscopic rate constants (Fig. 5a; Extended Data Fig. 8; and Supplementary Table 7). Similar results are obtained with M_ES1 under these neutral conditions in which the ES2 population is insignificant (<10⁻⁶ at pH 6.9) (Fig. 5a and Extended Data Fig. 8). In contrast, M_ES2 consistently underestimates F_pol by one to two orders of magnitude whereas M_Kd overestimates F_pol by one to two orders of magnitude (Fig. 5a)⁴⁷. Variants of the M_Kd model in which only pre-formed tautomeric dNTP with populations of 10⁻⁴−10⁻⁵ bind in a productive WC-like geometry overestimates k_pol and K_d by several orders of magnitude (data not shown). These data indicate that formation of tautomeric WC-like dG^enol•dT and dG•dT^enol with population of ≈0.1% can account for the ≈10²−10³–fold lower value of k_pol^incorrect relative to k_pol^correct and that at neutral pH, >99% of misincorporation proceeds via the tautomeric species, which form predominantly via direct exchange from the wobble (Fig. 5b).

Figure 5. Measured versus predicted misincorporation probability and rates.

a, F_pol measured experimentally for dTTP•dG misincorporation for human DNA polymerase ε, rat DNA polymerase β, and T7 DNA polymerase with values simulated using M_ES1, M_ES2, M_ES1+ES2 and M_Kd (error bars represent one s.d., Methods). b, Flux pathways for dT•dG(GGC) and ^5BrdU•dG(CGC). c, Left. Measured and simulated F_pol values for dTTP/^5BrdUTP in AMV RT¹⁴. Right, measured and simulated k_obs values for dTTP/^5BrdUTP misincorporation for human DNA polymerase β. Error is s.d. of n=3 biological replicates for kinetic assays, or previously published error for AMV RT¹⁴. Error for kinetic simulations is described in Methods. d, Measured and simulated k_obs for dTTP misincorporation for human DNA polymerase β in different sequence contexts. (*) indicates that ES2 exchange rates were extrapolated (Methods). Error is s.d. of n=3 biological replicates for kinetic assays. Error for kinetic simulations is described in Methods.

Impact of pH, modifications, and sequence

We also examined whether the M_ES1+ES2 model can reproduce the dependence of misincorporation probability on pH, base modification, and sequence. M_ES1+ES2 accurately predicts the ≈3-fold increase in misincorporation probability observed with increasing pH (Fig. 5c; left). This can be attributed to an increase in the population of dG•dT⁻, which accounts for >70% of the net misincorporation at pH 8.4 (Fig. 5b,c). In contrast, M_ES1 fails to predict this increase in misincorporation probability (Fig. 5c, left; M_Kd not shown due to absence of pH-dependent K_d values). Under high pH, the tautomeric and anionic species have comparable populations, and there is significant flux (>20%) toward both tautomeric and anionic species through the indirect minor exchange pathway (Fig. 5b). In this manner, the tautomeric and anionic contributions to misincorporation are coupled.

M_ES1+ES2 and NMR RD measurements also accurately predict F_pol and k_obs for 5-bromo-2′-deoxyuridine triphosphate (^5BrdUTP) (Fig. 5c). This includes a steeper ≈6-fold increase in F_pol^dG•5BrdUTP measured for AMV RT when increasing the pH from 7.0 to 8.4 (Fig. 5c; left). This can be attributed to the lower pK_a of dG•^5BrdU⁻ (pK_a ≈9)¹⁴ relative to dG•dT⁻ (pK_a ≈11.8)⁸. We further verified the robustness of these predictions by measuring k_obs^dG•5BrdUTP and k_obs^dTTP for human DNA polymerase β at high pH (8.4). The model accurately predicts the ≈4-fold enhancement in k_obs^dG•5BrdUTP relative to k_obs^dTTP (Fig. 5c; right). Again, M_ES1 fails to predict these variations (Fig. 5c). Indeed, at both neutral and high pH, ^5BrdUTP is predicted to be predominantly misincorporated via the more populated dG•dT⁻ (Fig. 5b,c). These data indicate that misincorporation due to dG•dT⁻ can dominate at pH≥8.4 or for chemically modified nucleotides at neutral pH.

Importantly, due to the sequence-dependence of tautomerization and ionization, M_ES1+ES2 also predicts ≈8-fold sequence-specific variations in F_pol at pH 8.4 (Fig. 5d). Comparable (5-fold) sequence-specific variations have been reported previously²¹. We tested these predictions using human DNA polymerase β at pH 8.4 for nine different sequence contexts (Supplementary Table 8). While k_obs^dG•dCTP varied weakly (<1.2-fold) with sequence, k_obs^dG•dTTP varied ≈45-fold (Extended Data Fig. 9), with larger changes observed when varying the base pair at the n-1 position, which stacks with dG•dTTP in the polymerase active site (Fig. 5d). While the M_ES1+ES2 predictions slightly underestimate the sequence-specific variations in k_obs^dG•dTTP, this is not too surprising considering that other microscopic steps could also vary with sequence. The predictions do recapitulate the lower k_obs^dG•dTTP for CGA and comparable values for GGC and CGC (Fig. 5d). Interestingly, the two major outliers (TGA and GGG) arise primarily due to a large ES2 population. It is likely that the polymerase environment, including absence of base pairs at the n+1 position (Fig. 5d), can influence the sequence-specific dependence of tautomerization/ionization and consequently misincorporation.

Our data indicate that formation of WC-like anionic and tautomeric mismatches help determine the frequency of dG•dT misincorporation and its dependence on pH, chemical modifications, and possibly sequence. Our analysis indicates that F_pol is determined primarily by the ES1 population and that significant reductions in k_ex = k_GS→ES1 + k_ES1→GS, outside the range detected here, would be required to significantly reduce F_pol (Extended Data Fig. 10). While it is likely that differences in the polymerase active site environment will tune tautomerization/ionization dynamics, the robustness of the predictions across different polymerases, pH conditions, and modified nucleotides suggests that it will not cause significant perturbations relative to the broad kinetic range examined here. Indeed, very small differences in tautomerization/ionization dynamics are observed for DNA and RNA, which have different helical structures and stabilities. It is possible that tautomerization/ionization is dominated by the energetics of hydrogen-bonding and proton transfer and that the natural grip for the WC geometry in the double helix is similar to that achieved by the polymerase in the context of an isolated dNTP paired to the template. Other mechanisms may be applicable for purine-purine mismatches where alterations in the active site have been proposed rather than adoption of a WC-like base pair^43,48. The approach presented here can be applied to examine the roles of other tautomeric and anionic mismatches in replication, transcription, translation, and mismatch repair⁴⁹.

Extended Data

Extended Data Figure 1. Watson-Crick-like mismatches.

a, Watson-Crick-like mismatches stabilized by tautomeric and ionic base forms. Tautomeric purine•pyrimindine and purine•purine mismatches were first proposed by Topal and Fresco¹¹. For G•T/U mismatches, X = H or CH₃ for uridine and thymidine, respectively.

Extended Data Figure 2. DNA and RNA constructs used in this study.

a, Secondary structures of the various DNA and RNA constructs used in this study. G•T/U mismatches that show signs of chemical exchange directed toward tautomeric and/or anionic WC-like mismatches are highlighted in blue and green, respectively. G•T/U mismatches that show no evidence for WC-like RD are highlighted in brown. The value of K_t measured at near-neutral pH is shown next to each mismatch. The DickersonTG-CGA, hpTG-CGC and hpUG-CGC sequences contexts were studied in a prior publication⁸. b, 2D [¹⁵N, ¹H] HSQC spectra of DNA and RNA constructs used in this study showing the imino resonances of G-N1/H1 and T/U-N3/H3 targeted for RD measurements. Spectra shown for xptG was collected at pH 6.7 and 25 °C in potassium acetate buffer described elsewhere⁸.

Extended Data Figure 3. RD profiles measured in DNA and RNA at near-neutral pH.

a, ¹⁵N G-N1 and T/U-N3 RD measured for G•T/U mismatches at pH 6.4-6.9 and 10-25 °C showing wobble⇌tautomer exchange. Note that in addition to wobble⇌tautomer exchange, tp5abc undergoes an independent slower exchange process involving a change in secondary structure that is described in detail elsewhere⁵³. The trend lines represent B-M 2-/3-state fits. Constructs containing a chemically modified base are indicated with (▼).b, Absence of ¹⁵N RD for rG•rU mismatches near bulges, apical loops, or three-way junctions. Error bars reflect experimental uncertainty (one s.d., see Methods). c, No correlation is observed between ground state (GS) wobble G-N1 and T/U-N3 chemical shifts for DNA (n = 5) or RNA (n = 8). Error bars in a, b reflect experimental uncertainty (one s.d., see Methods).

Extended Data Figure 4. Establishing lower limits for rates of base pair tautomeric exchange.

Agreement between measured and predicted R_1ρ values (scaled ${\overline{\chi }}^2$ weight, Eq. 2) when varying the wobble⇌G•T^enol/U^enol ( $k_{GS\rightleftharpoons T^{\mathit{enol}}/U^{\mathit{enol}}}$) and G•T^enol/U^enol ⇌G^enol•T/U (k_t) rate of exchange. See Methods for additional details.

Extended Data Figure 5. RD profiles measured in DNA and RNA at high pH.

B-M 3-state fits of a, RNA and b, DNA ¹⁵N RD data for starlike and triangular topologies (indicated within the plots). The relative statistical Akaike’s Information Criterion (w_AIC) and Bayesian Information Criterion (w_BIC) weights^50,60 for each fit were used to select the model (representative starlike versus triangular, comparisons with linear models shown in Supplementary Table 6). Error bars reflect experimental uncertainty (one s.d., see Methods).

Extended Data Figure 6. Discerning minor exchange between WC-like tautomeric and anionic G•T/U mismatches.

a, Topologies used to model chemical exchange. Individual rate constants are shown for each leg of the different topologies. b, Left: B-M simulations showing that when R_2(GS) ≠ R_2(ES1) and R_2(GS) ≠ R_2(ES2) no apparent peak asymmetry is observed. Right: B-M simulations showing that minor exchange between two ESs in a triangular topology induces asymmetry in the RD profiles and opposite changes in the apparent chemical shift for the two ESs. c, B-M simulations (solid lines) showing the fitted exchange parameters for hpUG-CGC at pH 8.4 and 10 °C (Supplementary Table 5) when including minor exchange in a triangular topology. For comparison, simulations using the same parameters without minor exchange (k₃ = 0 and k₋₃ = 0) are also shown (dashed lines). d, Dashed lines denote 3-state B-M best fit to starlike topology (k_ES1→ES2 = 0 and k_ES2→ES1 = 0) to data simulated with triangular topology with minor exchange (solid lines). Shown to the right is the over/under estimation of the true (green) versus fitted (red) ES chemical shifts when fitting RD profiles with triangular topology with minor exchange to a starlike model that has no minor exchange. e, The ES1 <∆ω_rG-N1> and <∆ω_rU-N3> values as a function of pH derived from the 3-state B-M fit with triangular and starlike topology. f, Forward (k_ES1→ES2) and reverse (k_ES2→ES1) minor exchange rate constants for hpTG-GGC (pH 8 and 8.4) and hpUG-CGC (pH 7.9 and 8.4) as a function of temperature. Error bars in e, f reflect experimental uncertainty (one s.d., Methods).

Extended Data Figure 7. Kinetic mechanisms used to model misincorporation.

Rate constants for each step are listed in Supplementary Table 7.

Extended Data Figure 8. Benchmarking kinetic simulations of misincorporation.

a, Comparison of k_pol and K_d values for correct incorporation measured experimentally for human DNA polymerase ε with values computed based on pre-steady state simulations using the microscopic rate constants provided in ref. 39. Error bars reflect fitting uncertainty as previously published³⁹. b, Robustness of calculated k_pol values for human DNA polymerase ε when varying rate constants (forward rate constant: blue; and reverse rate constant: orange) for steps other than tautomerization/ionization by 2-fold (n = 200 independent simulations in which rate constants were randomly varied within 2-fold). As expected, the only rate constant with a substantial effect on the reported k_pol values was the rate limiting k₂ conformational change step (middle).

Extended Data Figure 9. k_obs values measured for human DNA polymerase β dCTP•dG (left) or dTTP•dG (right) insertion at pH 8.4, 25 °C, and 100 μM dNTP.

DNA template sequence (5´ to 3´) is read from bottom (n+1 position) to top (n-1 position). Individual replicates (n = 3 independent experiment) are indicated by grey circles. Bar height reflect average of replicates and error bars reflect one s.d.

Extended Data Figure 10. F_pol is primarily governed by ES1 populations.

Simulated F_pol values as a function of scaling up or down the kinetic exchange rate for ES1 formation (k_ex = k_GS→ES1 + k_ES1→GS) without altering the ES1 population. Increasing k_ex beyond values measured in this study experimentally (green dotted line) minimally affects F_pol. Decreasing the rate of exchange within the range measured experimentally in this study (purple dotted line) also minimally affects the value of F_pol. Much larger decreases in k_ex are required to significantly reduce the value of F_pol.