Kinetics and thermodynamics of BI-BII interconversion altered by T:G mismatches in DNA

Abstract

T:G mismatches in DNA result in humans primarily from deamination of methylated CpG sites. They are repaired by redundant systems, such as thymine DNA glycosylase (TDG) and methyl-binding domain enzyme (MBD4), and maintenance of these sites has been implicated in epigenetic processes. The process by which these enzymes identify a canonical DNA base in the incorrect basepairing context remains a mystery. However, the conserved contacts of the repair enzymes with the DNA backbone suggests a role for protein-phosphate interaction in the recognition and repair processes. We have used ³¹P NMR to investigate the energetics of DNA backbone BI-BII interconversion, and for this work have focused on alterations to the activation barriers to interconversion and the effect of a mismatch compared with canonical DNA. We have found that alterations to the ΔG of interconversion for T:G basepairs are remarkably similar to U:G basepairs in the form of stepwise differences in ΔG of 1–2 kcal/mol greater than equivalent steps in unmodified DNA, suggesting a universality of this result for TDG substrates. Likewise, we see perturbations to the free energy (∼1 kcal/mol) and enthalpy (2–5 kcal/mol) of activation for the BI-BII interconversion localized to the phosphates flanking the mismatch. Overall our results strongly suggest that the perturbed backbone energetics in T:G basepairs play a significant role in the recognition process of DNA repair enzymes.

Graphical abstract

Significance

DNA is under constant attack causing damage to its chemical structure. An elegant repair process has evolved to help prevent the negative effects of these chemical modifications. The details of the initial steps of damage detection remain a mystery. Our results use ³¹P NMR to demonstrate changes to the backbone conformational equilibrium due to the presence of a T:G mismatch, and are localized to the positions directly adjacent to the mismatched basepair, and may play a role in the initial steps of the repair process. We propose that our system represents an example of how a small model system can provide evidence for the role of conformational dynamics in protein-DNA recognition.

Introduction

Improperly basepaired nucleotides in double-stranded DNA can occur due to a variety of sources. T:G mismatches in mammals primarily arise by deamination of 5-methyl-cytosine in CpG steps (1, 2, 3) and are associated with carcinogenesis and other deleterious effects (4,5). Low-population high-energy states of T:G basepairs involving base tautomerization may be involved in copying errors during basepair replication (6,7). T:G mismatches are repaired by redundant systems of DNA glycosylases to remove the thymidine, leaving an orphaned deoxyguanosine. The enzymes in the repair mechanism include those in the SMUG family, such as thymine DNA glycosylase (TDG) and methyl-binding domain (MBD4) proteins (3). TDG has also been implicated for a major role in mammalian epigenetics (8, 9, 10). As with other glycosylases, TDG and MBD4 employ a nucleotide flipping mechanism during the later stages of recognition to allow access to the target base, which is rotated out of the DNA helix and stabilized in the binding pocket of the glycosylase (11, 12, 13, 14, 15).

Specific recognition and removal of the mismatched nucleotide present an interesting biophysical problem for the repair enzymes. They must be able to search for and select a normal DNA base (thymine) in the incorrect pairing context (T:G) from the vast excess of proper basepairs (T:A), and overcome the energetic cost to disrupt hydrogen bonds and base stacking. The exact process of mismatch and base lesion recognition by DNA glycosylases is not fully understood, whether structural, dynamics or energetic aspects contribute. X-ray (16) and solution NMR studies (17, 18, 19, 20, 21) observe structural changes involving T:G mismatches primarily in the form of wobble basepairs, whereas the backbone torsion angles have only minor differences based on static or average structures. Solution NMR has, however, demonstrated unusual ³¹P chemical shifts in DNA sequences containing mismatches (22,23). There is evidence for an energetic aspect to recognition, where local “flexibility” in damaged DNA has long been suggested to play a role in recognition (24,25), and local distortions in DNA mismatches “pre-pay” some of the energetic cost of recognition and removal (26). Isaacs and Spielmann state (27) “Equilibria between different conformations in the phosphodiester backbone may provide flexibility … that can be exploited by DNA-binding proteins. Dinucleotide steps that have a large fraction of BII conformation may be preorganized to bend, which may facilitate protein binding.” In previous work, we observed dramatic, local changes to the DNA backbone conformational equilibrium on the introduction of uracil and dihydrouracil (in U:G basepairs), 1,N⁶-ethenoadenine, and T:G mismatches (28). These disruptions were expressed as large stepwise shifts of the %BII conformation and ΔG of the equilibrium relative to unmodified, control DNA sequences. There remained an open question about the effect of DNA damage on the activation barrier and kinetics of the backbone phosphate dynamics, which we explore presently.

The DNA phosphodiester backbone conformation is described by a series of dihedral angles (29,30). Specifically, the torsion angles (Fig. 1: ε, C4′ — C3′ — O3′ — P; and ζ, C3′ — O3′ — P — O5′) define two primary DNA backbone conformations known as BI and BII (29, 30, 31). The values of these torsion angles for the two conformations have been traditionally defined as trans/g- in BI (ε – ζ ∼ −90°) and g-/trans in BII (ε – ζ ∼ +90°). A large body of evidence has concluded that the DNA backbone is in a dynamic conformational equilibrium between these two primary states (29,30). The BI-BII dynamic interconversion is primarily modeled as a two-site intramolecular exchange with unequal forward and reverse rates (32):

$$B_I\underset{k_r}{\overset{k_f}{\rightleftharpoons }}{B}_{II},$$ (1)

where a particular phosphate is described with %BII, designating the population of the traditionally higher energy state (Fig. 1).

Figure 1.

(A). Definition of the backbone dihedral angles most associated with the BI-BII equilibrium (ε and ζ). (B). Illustration of the putative BI-BII interconversion equilibrium energy diagram, the backbone conformations, and the definition of the Gibbs free energies associated with the equilibrium with respect to the forward process. To see this figure in color, go online.

The dynamic process of BI-BII interconversion has been assumed to be in the fast exchange limit (30,32), occurring on the nanosecond timescale based on MD computer simulations (27,29, 30, 31,33). To quantify the populations of equilibrium states, the Hartmann group originally proposed an empirical relationship between the observed ³¹P chemical shifts (δP) and the %BII (31), and Tian et al. further extended the work by deriving an empirical relationship between δP and %BII as a function of temperature (32). Our work further enhances these investigations by evaluating %BII for DNA containing lesions and mismatches, and generated a new empirical relationship between δP and %BII that spans the full 1.6 ppm range originally predicted (28). The relationship between the thermodynamics and kinetics of chemical exchange from NMR has been extensively developed elsewhere (32,34, 35, 36, 37, 38, 39, 40, 41). The ³¹P linewidths (Δν_½) can be used to determine the temperature-dependent forward rate constant k_f and therefore the free energy of activation ΔG^‡. Determination of the free energy of activation (ΔG^‡) provides a direct link to the exchange kinetics and the combination of ΔG and ΔG^‡ allows for determination of the overall exchange rate constant, k_ex. Temperature studies are used to determine the enthalpy of activation (ΔH^‡) using an Eyring plot, where the slope of the graph of natural log of the rate constant (ln k_f or ln k_ex) vs. 1/T is directly related to ΔH^‡ (40). Recent studies have used solution NMR methodologies to study the thermodynamics of slow motions in DNA containing 5-formylcytosine by chemical exchange saturation transfer (42) and by ¹⁹F NMR in T:G and U:G mismatches (43). For the purposes of this work, we are using the ΔG obtained from δP and the activation energy ΔG^‡ from the ³¹P linewidth to develop a more complete understanding of the effect of T:G mismatches on the DNA backbone conformational equilibrium (Fig. 1).

We have analyzed three DNA oligonucleotides (Table 1) and determined their ³¹P isotropic chemical shifts and linewidths as a function of temperature. The control sequence remains the Dickerson-Drew Dodecamer (DDD), a sequence (44) that is exceptionally well studied by numerous techniques. The test sequences contain T:G mismatches at two different positions (replacing the C3 and C9 with Ts and with sequence IDs T3 and T9, respectively; Table 1). The ¹H and ³¹P isotropic chemical shifts for T3 were previously assigned (18,22). The T9 sequence has been studied regarding its basepair opening kinetics by NMR (45) and by x-ray crystallography (16) but its ³¹P peaks remained unassigned until our previous paper (28). These sequences offer an interesting companion to the DDD, meC3, and meC9 sequences analyzed previously (32), as deamination of 5-methylcytosine produces a thymine, so C:G → meC:G → T:G.

Table 1.

List of all DNA sequences analyzed for this work

Sample ID	Sequence
DDD
T9
T3

All are double-stranded, palindromic sequences.

The goal of this work is twofold. First, we compare the %BII and ΔG values for the T3 sequence to our previous results and evaluate trends within the data for all T:G and U:G mismatches that might indicate a universal phenomenon among substrates of TDG. Second, we determine the Gibbs free energies and enthalpies of activation, ΔG^‡ and ΔH^‡ respectively, to evaluate the effect of T:G mismatches on the energy barriers and kinetics of backbone interconversion. These results will hopefully provide significant insight into the energetic properties of the DNA backbone in the presence of T:G mismatches and perhaps demonstrate a role in mismatch recognition, binding, interrogation, or nucleotide flipping.

Materials and methods

Sample preparation and NMR spectroscopy

Custom DNA oligonucleotides were synthesized by IDTDNA (Coralville, IA). Deuterium oxide was purchased from Cambridge Isotope Labs (Tewksbury, MA). All other chemicals were purchased from Sigma-Aldrich (St. Louis, MO).

DNA samples were annealed in ∼0.75–1.0 mL of 25 mM phosphate buffer (sodium salt), adjusted to pH 7.4 with a double-stranded DNA concentration between 0.5 and 1.0 mM at a final volume of 1.0 mL. Our 1D experiments for the T9 sequence were performed on a more concentrated sample, with the DNA concentration around 5 mM. Samples were heated in a water bath at 90°C for 15 min before being allowed to slowly cool for at least 3 h. The pH was readjusted with dilute phosphoric acid and the samples were lyophilized. The dried samples were dissolved in D₂O, pH adjusted as necessary, and lyophilized. This procedure was repeated two to three times in high-purity D₂O to remove exchangeable protons.

NMR experiments were performed using a Varian 400 MHz INOVA NMR spectrometer with VnmrJ 4.2 software, a 5 mm inverse detection double resonance gradient solution probe, temperature control, and wet1D solvent suppression, following standard protocols (46, 47, 48). For chemical shift assignments, spectra were collected at 10 and 25°C. The spectral width for the proton channel was set to 4000 Hz. Unless otherwise noted, NOESY spectra (see Fig. S1 in the supporting material for a representative spectrum) used for assignments were obtained using 64 or 128 t1 increments with 256 scans per increment, a mixing time of 200 ms, and a relaxation time of 1 s.

1D ³¹P temperature study experiments (see Fig. S3 for representative data) were performed with a spectral window of 6000 Hz, a relaxation delay of 1–2 s, and 1024 scans, and were recorded between 10 and 35°C for temperature studies with 5°C increments. The experiments at 10 and 25°C were repeated with a coaxial insert containing fresh 85% phosphoric acid used as an external reference set to 0.00 ppm for all ³¹P peaks. ¹H-³¹P HSQC experiments were performed with 2048 complex points for the t2 dimension, 256 t1 increments, and a relaxation delay of 1–2 s. The spectral width was 4000 Hz in f2 and 1950 Hz in f1.

NMR experiments for assignment purposes were analyzed using ACD/Spectrus 2017.1.1 software (ACD/Labs, Toronto, Canada). All proton peaks were referenced to the temperature-dependent residual HDO peak (49). NOESY data were interpreted by analyzing several distinct regions: the aromatic-H2′/H2″ region, the aromatic-H1′ region, the H1′-H2′/H2″ region, the H2′/H2″-H3′ region, the H1′-H4′ region, and the H2′-H2″ region. Spectra were compared with corresponding regions in the TOCSY, as is the established protocol for distinguishing through-space and through-bond correlations (46, 47, 48).

³¹P isotropic chemical shifts (δP) were assigned (±0.01 ppm) by identifying ¹H-³¹P couplings in HSQC experiments as a function of the same temperature range as the 1D spectra (i.e., 10–35°C) (28,32). The primary correlations were between the phosphate peak for a dinucleotide pair (5′-XpY-3′) and the H4′ from the 3′ nucleotide in the pair (32). Additional peaks in an HSQC were used as possible, including crosspeaks between the H3′ for the 5′ nucleotide in the same dinucleotide pair and the phosphate peak. Fig. S2 shows a representative spectrum. Once the individual phosphates were assigned from the 2D experiments, their chemical shift values were monitored and extrapolated for all temperatures from the 1D temperature study (see Table S3 for the T3 DNA) and used to determine the temperature dependence of the %BII for each peak in each sequence. As previously (28), from our estimated error of ±0.01 ppm for δP, using standard error propagation, this provides an error in %BII of ±1% and an error in ΔG of ±0.01 kcal/mol.

To determine the ³¹P linewidths (Δν_½) for each phosphate as a function of temperature (Tables S1, S5, and S6), the following procedure was used. First, the decoupling parameters for the ³¹P channel were calibrated and optimized, so no residual ¹H-³¹P coupling would contribute to the linewidth. Second, ¹H-³¹P-HSQC experiments were obtained in 5° steps from 10 to 35°C as discussed above. From there, 1D projections of each phosphate peak in the ³¹P dimension were generated without apodization and exported as ascii files. Each peak present in the 1D projections were fit with Python code as discussed below (see Fig. 3 for examples of the fit quality), and the relevant linewidths used for determination of the activation barrier. The variation of the linewidth was determined to be less than ±0.35 Hz based on repeated ³¹P experiments.

Figure 3.

Example linefits and ΔG^‡ fits. (A) 1D slice from DDD HSQC at 298 K for the G4pA5 phosphate (largest peak). (B) 1D slice from T9 HSQC at 298 K for the G4pA5 phosphate. The data are the red dashed lines, the best fits are the solid lines, and the residuals are the dashed green lines. (C) Comparison of ΔG^‡ curve fits for the G4pA5 phosphate in DDD (blue) vs. T9 (orange). (D) Comparison of ΔG^‡ curve fits for the A5pA6 phosphate in DDD (blue) vs. T9 (orange). Error bars arise from the estimated variation in linewidth due to replicate measurements (0.35 Hz). To see this figure in color, go online.

In addition to using 1D projections from 2D experiments, we have also tested fitting the 1D ³¹P NMR experiments (Bloch decay) to obtain linewidths. ³¹P spectra of DNA have notoriously low dispersion among their chemical shifts, leading to many overlapping peaks where individual linewidths cannot be resolved. However, the relatively high dispersion of the T9 sequence (Fig. S20 in the supporting material of Westwood et al. (28) and Fig. S3) offers an opportunity to test 1D spectra for analysis. This would be generally beneficial as a single 1D spectrum with reasonable signal-to-noise can be obtained in 1 h, whereas an HSQC can be as much as 10–20 h. For the peaks in the T9 1D spectra that have measurable resolution (Table S2), we have demonstrated that the values for ΔG^‡ determined are very similar to the 2D analysis (Table S9). The error in linewidth deconvolution was always smaller than the experimental reproducibility error of 0.35 Hz; therefore all further analysis propagated the larger experimental error. All conclusions are drawn from the 1D projections of 2D experiments and do not rely on the 1D experiments, but it is nevertheless satisfying that the two experiments are equivalent.

Exchange kinetics and thermodynamics

All NMR studies of BI-BII interconversion have followed under the assumption of the fast exchange time regime, defined as the rate constant for the exchange between the two conformers (k_ex) from the sum of the unequal forward and reverse rate constants, assumed to be much larger than the difference in chemical shift: $k_f+k_r=k_{ex}\gg \left|{\text{Ω}}_{BII}-{\text{Ω}}_{BI}\right|=\text{ΔΩ}$ (24, 25, 26, 27, 28, 29, 30) or equivalently $\frac{k_{ex}}{\text{ΔΩ}}\gg 1$. The observed ³¹P isotropic chemical shifts (δP) of the backbone phosphates are therefore the population-weighted average of the chemical shifts from the two pure states (i.e., Ω_BI and Ω_BII), $\overline{\text{Ω}}=p_{BII}{\text{Ω}}_{BII}+p_{BI}{\text{Ω}}_{BI}=\delta P$ (34, 35, 36, 37, 38, 39, 40). The range in chemical shifts (in ppm) between a pure BI and a pure BII state was determined to be approximately 1.6 ppm (23,30). As stated above, there has been previous important work developing empirical relationships between δP and %BII (31,32). Our work expands on this to include mismatch and damaged DNA and utilizes δP values spanning the full 1.6 ppm range originally predicted for ΔΩ (28). From the %BII, one can determine the equilibrium constant and Gibbs free energy of equilibrium.

$$\text{Δ}G=-RTln{K}_{eq}=-RTln\left(\frac{\%BII}{\%BI}\right)=-RTln\left(\frac{k_f}{k_r}\right)$$ (2)

We determined the free energy of activation ΔG^‡ and subsequently the forward rate constants k_f as a function of temperature using the Eyring equation:

$$k_f=\kappa \frac{k_{B}T}{h}{e}^{-\text{Δ}{G}^{‡}/RT}$$ (3)

The transmission coefficient κ is generally assumed to have a value of 1. Temperature studies of the rate constants are used to determine the enthalpy of activation (ΔH^‡) using an Eyring plot, where the slope of the graph of natural log of the rate constant (ln k_f or ln k_ex) vs. 1/T is directly related to ΔH^‡ (40). Determination of the free energy of activation (ΔG^‡) provides a direct link to the exchange kinetics and the combination of ΔG and ΔG^‡ allows for determination of the overall exchange rate constant, k_ex.

Data fitting and analysis

We rederived the relationship between linewidth and rate constant (and thus ΔG^‡) from the original work by Jeener et al. (50). Our results expressed in Eq. 4 are different from those obtained by Tian et al. for an equivalent expression (32) and we have included a full derivation in the supporting material.

$$\text{Δ}{\nu }_{\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$2$}\right.}=\frac{1}{\pi T_2}+\frac{K_{eq}{\left({\omega }_{BII}-{\omega }_{BI}\right)}^2}{4\pi \left(1+K_{eq}\right)}\left(\frac{h}{k_{B}T}\right)e^{\text{Δ}{G}^{‡}/RT}$$ (4)

A plot of linewidth (Δν_½) versus temperature can be fit using Eq. 4 with ΔG^‡ and T₂ as the fitting parameters. Best fits provide values for ΔG^‡ from which values for k_f can be determined using the Eyring equation (Eq. 3).

Determination of ³¹P linewidth parameters was performed by fitting the appropriate peaks in 1D projections of HSQC spectra to single Lorentzian peak lineshapes using an in-house Python tkinter-based GUI developed to simultaneously fit multiple, possibly overlapping peaks. Each Lorentzian was parameterized by a center frequency, half-width at half max (subsequently doubled for full-width at half max), and height. The multiparameter fitting was performed using the curve_fit() function in the standard Python Scipy.optimize library (https://docs.scipy.org/doc/scipy/reference/generated/scipy.optimize.curve_fit.html). Fit parameter error bars were determined from the covariance matrix returned by the curve_fit() function. Similarly, the ΔG^‡ and T₂ values were determined by fitting Eq. 4 as a function of temperature in a two-parameter fit (i.e., ΔG^ǂ and T₂) once the ΔΩ and K_eq parameters had been determined as described above. The largest error determined for ΔG^‡ from the fit was determined to be ±0.19 kcal/mol (although this was often considerably lower) and the largest error in T₂ was 0.025 s, and for those phosphates with better resolution the error in ΔG^‡ was often less than 0.1 kcal/mol (Table S7).

Results

Following the complete ³¹P assignment of the phosphates in mismatch-containing DNA duplexes, the isotropic chemical shifts (δP) and linewidths (Δν_½) were measured as a function of temperature, and differences noted in mismatched sequences relative to the DDD duplex. The %BII analysis followed the protocol we developed recently (28), which is a modification of previous work (31,32). The temperature-dependent δP values for DDD and T9 were determined previously (28), whereas the values for T3 are new for this work and are listed in the supporting material (Table S4) and match well with the previous results (23). Determining the %BII from δP allows calculation of the equilibrium constant K_eq for the backbone interconversion and therefore the ΔG (Eq. 2). We measured ³¹P linewidths for all three sequences as a function of temperature for all resolved phosphate peaks. Some were not sufficiently resolved to provide unambiguous values and are noted where relevant. These linewidths were used to calculate the free energies and enthalpies of activation (ΔG^‡ and ΔH^‡, respectively) as a function of position as well as the associated rate constants k_f and k_ex. The typical errors for kinetic and thermodynamic quantities were found to be less than 1% of the determined value, but see Tables 2, 3, and 4, and Table S7 for more specifics.

Table 2.

Kinetic and thermodynamic parameters for DDD at 298 K as a function of phosphate position

Position	ΔGa (kcal/mol)	Keq	ΔHvHa (kcal/mol)	ΔGǂ (kcal/mol)	T2 (s)	ΔHǂ (kcal/mol)	kf (kHz)	kex (kHz)
C1pG2	−0.03	1.05	n.d.	10.59	0.076	10.76	106	207
G2pC3	0.14	0.78	−1.47	10.49	0.056	11.31	126	288
C3pG4	−0.17	1.32	−2.26	10.59	0.070	11.49	105	185
G4pA5	0.14	0.78	1.00	10.59	0.061	10.17	91.8	209
A5pA6	0.34	0.56	−0.51	11.01	0.066	11.22	52.5	146
A6pT7	0.58	0.38	−1.24	11.19	0.058	12.40	38.5	141
T7pT8	0.58	0.38	−2.61	10.77	0.049	13.29	78.3	286
T8pC9	0.34	0.56	−0.60	10.98	0.062	11.20	54.8	152
C9pG10	−0.29	1.63	−1.14	10.68	0.074	11.04	90.9	147
G10pC11	−0.03	1.05	−1.03	10.17	0.055	11.01	215	421
C11pG12	−0.29	1.63	n.d.	10.38	0.074	10.52	151	243

n.d., value not determined. Errors were determined from standard error propagation and were on average less than 1% of the determined value for all kinetic and thermodynamic properties listed, except the rate constants for which the error was determined to be approximately 16% on average. Specific errors for ΔG^‡ are listed in Table S7.

^aProvided from reference (28).

Table 3.

Kinetic and thermodynamic parameters for T9 at 298 K as a function of phosphate position

Position	ΔGa (kcal/mol)	Keq	ΔHvHa (kcal/mol)	ΔGǂ (kcal/mol)	T2 (s)	ΔHǂ (kcal/mol)	kf (kHz)	kex (kHz)
C1pG2	−0.03	1.05	n.d.	10.26	0.059	10.40	186	364
G2pC3	0.02	0.96	−1.15	n.d.	n.d.	n.d.	n.d.	n.d.
C3pG4	−0.11	1.21	−1.79	10.20	0.056	11.04	206	376
G4pA5	1.25	0.12	1.22	11.79	0.082	10.74	13.9	130
A5pA6	0.41	0.50	−1.04	11.03	0.076	11.84	50.1	151
A6pT7	0.87	0.23	−1.77	11.25b	0.046b	12.77	34.8b	186b
T7pT8	0.41	0.50	−2.57	11.08	0.059	12.82	46.7	141
T8pT9	0.76	0.28	−3.61	10.99	0.052	13.72	54.2	251
T9pG10	−0.88	4.41	2.42	10.44	0.076	10.01	136	167
G10pC11	0.13	0.81	−0.97	10.63	0.058	11.27	98.9	221
C11pG12	−0.20	1.40	n.d.	10.37	0.064	10.70	154	264

n.d., value not determined. Errors were determined from standard error propagation and were on average less than 1% of the determined value for all kinetic and thermodynamic properties listed, except the rate constants for which the error was determined to be approximately 16% on average. Specific errors for ΔG^‡ are listed in Table S7.

^aProvided from reference (28).

^bDetermined from 1D NMR data (see main text and Table S2).

Table 4.

Kinetic and thermodynamic parameters for T3 at 298 K as a function of phosphate position

Position	ΔGa (kcal/mol)	Keq	ΔHvHa (kcal/mol)	ΔGǂ (kcal/mol)	T2 (s)	ΔHǂ (kcal/mol)	kf (kHz)	kex (kHz)
C1pG2	−0.17	1.32	n.d.	10.51	0.065	11.37	121	213
G2pT3	1.17	0.14	−5.36	11.38	0.060	17.24	27.8	228
T3pG4	−0.58	2.68	−0.68	10.55	0.068	11.22	114	157
G4pA5	0.30	0.60	0.05	11.12	0.074	11.13	43.6	116
A5pA6	0.34	0.56	−0.21	11.09	0.067	11.53	46.0	128
A6pT7	0.62	0.35	−1.36	n.d.	n.d.	n.d.	n.d.	n.d.
T7pT8	0.54	0.40	−2.32	11.41	0.074	13.74	26.6	92.3
T8pC9	0.11	0.83	−0.93	10.77	0.072	11.74	78.7	173
C9pG10	−0.25	1.54	−1.00	10.57	0.065	11.54	109	181
G10pC11	0.32	0.58	−0.50	10.85	0.054	11.55	68.2	186
C11pG12	−0.06	1.11	n.d.	10.38	0.059	11.04	152	289,

n.d., value not determined. Errors were determined from standard error propagation and were on average less than 1% of the determined value for all kinetic and thermodynamic properties listed, except the rate constants for which the error was determined to be approximately 16% on average. Specific errors for ΔG^‡ are listed in Table S7.

^aProvided from reference (28).

ΔG as a function of temperature

Under the assumption that the DNA backbone conformations interconvert between the two primary states BI and BII in the fast exchange limit (29,30), we determined the %BII from the weighted average of the ³¹P chemical shift (δP). ΔG is defined based on the equilibrium between the two states BI and BII (Fig. 1). As one might expect, a –ΔG indicates that the BII conformation is preferred, a +ΔG indicates that the BI form is preferred, and a ΔG near zero indicates near equal populations of both conformations. Temperature-dependent ΔG values (within a typical error of ±0.01 kcal/mol) as a function of sequence position for the DDD, T9, and T3 sequences are shown in Fig. 2. Both mismatched sequences also have a large +ΔG for the 5′ neighboring phosphate, which becomes more positive with increasing temperature, and a large –ΔG for the phosphate position immediately 3′ to the mismatch site. The results for the T9 were demonstrated in our previous work (28), but the T3 results are new and offer a second T:G mismatch position for comparison.

Figure 2.

ΔG values for the DNA sequences as a function of temperature and sequence position (within a typical error of ±0.01 kcal/mol based on standard error propagation). The temperature values range from 283 K (lightest) to 313 K (darkest) in 5 K steps. The U3 and U9 data are adopted from Westwood et al. (28). The numbers on the x axes refer to the phosphates in order, starting with 1 = C1pG2. To see this figure in color, go online.

ΔG^ǂ, rate constants as a function of temperature

Measuring NMR linewidths (Δν_½) as a function of temperature allows determination of the activation energy (ΔG^‡) (34, 35, 36, 37, 38, 39, 40, 41) and therefore the kinetic rate constants (k_f and k_ex) as well as the enthalpy of activation (ΔH^‡) using Eyring plots. The linewidths were determined by a Python fitting program for each phosphate peak, taken from a 1D projection of the temperature-dependent HSQC. Linewidths as a function of temperature are plotted and the data fit with Eq. 4 to calculate ΔG^‡. Example fits are shown in Fig. 3.

We compared our ΔG^‡ results for the DDD with those generated by Tian et al. (32) and our results are different by about 1–2 kcal/mol, although the sequence trends are equivalent. It should be noted that there are three significant differences between the analyses: Δω² (our range is 1.6 ppm compared with theirs of 0.86 ppm), the empirical equation to determine %BII equation leading to different values in K_eq, and the actual equation to determine ΔG^‡ (our Eq. 4 compared with their Eq. 11). We discuss the differences in the form of the ΔG^‡ equation in more detail in the supporting material. We next analyzed the sequences containing T:G mismatches. Tables 2, 3, 4, S7, and Figs. 4 and 6 summarize our results for the three sequences. For the T9 (Fig. 4 A), the largest relative difference in ΔG^‡ from the equivalent position DDD $\left(\text{Δ}{G}_{DDD}^{‡}-\text{Δ}{G}_{T9}^{‡},\text{or}\text{ΔΔ}{G}^{‡}\right)$ is 1.2 kcal/mol at the G4pA5 position, which is significant as G4 is the basepairing partner of the T in the mismatched pair. A comparable result was observed for T3, in that the difference in ΔG^‡ from DDD is 0.90 kcal/mol at the G2pT3 step, which is the phosphate immediately 5′ to the mismatched T. The next largest differences in both sequences involve a phosphate neighboring the basepairing partner as well. The difference at the T6pT7 position originates from the very poor resolution of this phosphate peak with the A6pT7 peak in both sequences (see Fig. S2).

Figure 4.

Comparison of relative values (free energy of the DDD control minus the free energy of the mismatched sequence, or ΔΔG) of ΔG (blue bars) and ΔG^‡ (red bars) for T9 (A) and T3 (B) compared with DDD as a function of phosphate position at 298 K. Arrows indicate the positions of the mismatched T and the basepairing G. Typical errors were less than ±0.01 kcal/mol for ΔG and ±0.19 kcal/mol for ΔG^‡. To see this figure in color, go online.

Figure 6.

Comparison of relative values of ΔH (enthalpy of DDD minus enthalpy of mismatched sequence or ΔΔH) from van ’t Hoff analysis of K_eq (ΔH_vH) (blue bars) and from Eyring analysis of k_ex (ΔH^‡) (red bars) for T9 (A) and T3 (B) compared with DDD as a function of phosphate position. Arrows indicate the positions of the mismatched T and the basepairing G. Typical errors in the values for the enthalpies were less than ±0.006 kcal/mol for ΔH_vH and ±0.095 kcal/mol for ΔH^‡. To see this figure in color, go online.

Using the Eyring equation (Eq. 3), the forward rate constant (k_f) can be calculated from the activation energy ΔG^‡. Subsequently, an Eyring plot (ln k/T vs. 1/T) can be used to determine the enthalpy of activation, ΔH^‡. As discussed above, after calculating k_f we use those results and K_eq to determine the overall rate constant for the exchange process, k_f + k_r = k_ex. Tables 2, 3, 4, S8, and Fig. S6 summarize the sequence dependence of k_f and k_ex for our three DNA duplexes. Whereas absolute values of these kinetic and thermodynamic properties offer interesting biophysical insights for canonical DNA duplexes, the effect of T:G mismatches is also relevant for the process of mismatch detection by the DNA glycosylases. Tables 2, 3, and 4 contain all relevant thermodynamic and kinetic parameters obtained from the NMR data analysis.

Effect on conformational enthalpies

Previous variable temperature studies of K_eq (obtained from ΔG) (28) with van’t Hoff plot analysis (plotting ln(K_eq) vs. 1/T) provide ΔH_vH. Calculating the van’t Hoff enthalpy (Tables 2, 3, and 4) demonstrates that the results for T3 are comparable to those of a similar U3 sequence (Table S26 in the supporting material of Westwood et al. (28)). This again indicates strong similarity in the energetic properties of different TDG substrates in identical sequence contexts. Variable temperature studies of k_f and k_ex and the use of Eyring plots also provide ΔH^‡ (Tables 2, 3, and 4) from the slope of ln(k/T) vs. 1/T graphs. Representative Eyring plots for ΔH^‡ (using k_ex) are shown in Fig. 5.

Figure 5.

Comparison of ln(k_ex/T) vs. 1/T Eyring plots comparing certain phosphates from DDD to T9 (A) and T3 (B). The data presented in (A) are the DDD sequence A5pA6 (red) and the T8pC9 (blue) positions, as well as the T9 sequence A5pA6 (green) and T8pT9 (orange) positions. The data presented in (B) are the DDD sequence A5pA6 (red) and the G2pC3 (blue) positions, as well as the T3 sequence A5pA6 (purple) and G2pT3 (pink) positions. Values for ΔH^‡ are obtained from the slopes of these curves. Note that those curves with the most different slope (orange and pink) correspond to the phosphates with the largest difference in ΔH^‡ (T8pT9 and G2pT3, respectively). To see this figure in color, go online.

The results showing changes to the enthalpies for both T9 and T3 are striking (Fig. 6), and perhaps the most significant result reported here. Tables 2, 3, and 4, and Fig. 5 show that the ΔH^‡ most dramatically increases for the position 5′ to the mismatch; this would be the T8pT9 for the T9 sequence and the G2pT3 for the T3 sequence and correlates with the ΔH_vH. The values for ΔH^‡ for T9 and T3 increase by 2.57 and 5.51 kcal/mol, respectively, relative to the control DDD.

Discussion

The dynamic and energetic landscape of damaged or mismatched DNA and the role (if any) in recognition by repair enzymes remain a mystery. We previously quantified aspects of the backbone energies (28), specifically ΔG and ΔH_vH, in the context of several lesion types. Herein, we take a closer look at T:G mismatches, and attempt to quantify the activation energies and kinetic parameters for the BI-BII backbone interconversion relative to a control DNA. Our work includes the temperature-dependent ³¹P NMR chemical shifts (δP) and linewidths (Δν_½) for three DNA oligonucleotides, of which two contain a T:G mismatch and the third is a control. The NMR properties allow us to calculate %BII, ΔG, K_eq, ΔH_vH, ΔG^‡, ΔH^‡, k_f, and k_ex.

The first general observation is also the most straightforward, and that is the presence of a T:G mismatch perturbs the populations of the BI and BII primary conformers (as exhibited in K_eq and ΔG) relative to equivalent positions in an unmodified DNA duplex. Indeed, the δP values (and subsequently the %BII) in standard DNA are generally observed (32) within a much smaller range of about 0.8 ppm. This is approximately half the range proposed by Gorenstein and co-workers (23,30). Some of the first observations of δP outside this range for canonical DNA were due to the presence of mismatches (22,23). Our current work demonstrates that the presence of a mismatch produces stepwise differences (i.e., p₁ to p₂ in a two phosphate sequence, such as 5′-X-p₁-Y-p₂-Z-3′) in populations of the BII conformer relative to those expected for canonical DNA.

There are several additional results observed in our data that demonstrate that the backbone energetics near the lesion are different than other positions. First, as with our previous work (28), the ΔG values in our mismatch-containing sequences differ from the unmodified DNA duplexes and the positions of greatest difference correlate with important enzyme-phosphate contacts. We found that the presence of a T:G or U:G mismatch alters the ΔG often by more than 1.0–1.5 kcal/mol relative to the same position in the DDD DNA, and that these differences are localized to the phosphates flanking the damaged or mismatched nucleotide or its basepaired partner. The T3 and T9 sequences both have large stepwise differences in %BII and ΔG near the mismatched nucleotide (Figs. 2, S4, and S5). The phosphate step immediately 5′ to the mismatched nucleotide has a large +ΔG value (BI conformation is preferred) as does the basepairing partner. The phosphate step immediately 3′ to the mismatched nucleotide has a large –ΔG value (BII is preferred).

Temperature studies also provide some insights into the energetics of DNA backbone interconversion. The linewidth as a function of temperature allowed us to determine the free energy of activation. The values of ΔG^‡ for the three sequences are provided in Tables 2, 3, and 4. In Fig. 4, we plot the difference in ΔG^‡ for equivalent positions in the T9 and T3 relative to DDD, and the results mirror those for ΔG. In the T3 sequence, the largest energy variation (1.3 kcal/mol) relative to the control is the position immediately 5′ to the mismatch at G2pT3. In the T9 sequence, the largest energy variation (1.1 kcal/mol) relative to the control is the position immediately 3′ to the mismatched basepairing partner (G4pA5). A common result is the increase in activation energy near the mismatched pair, and the fact that the free energies of both strands are impacted is significant because glycosylases, such as TDG, have important phosphate contacts with both strands of the DNA duplex (11, 12, 13, 14, 15). While 1 kcal/mol may not appear as a change of significant magnitude, it should be noted that the relative difference in Boltzmann populations for two states of 1 kcal/mol difference in energy is ∼85%. To describe the model of differences in the associated free energies, we have included an illustration in Fig. 7.

Figure 7.

For a Figure360 author presentation of Fig. 7, see https://doi.org/10.1016/j.bpj.2022.03.031.

Graphic demonstrating the observed effects on ΔG and ΔG^‡ in DNA containing a T:G mismatch, using the T9 sequence as an example. Following the color scheme of Fig. 4, the blue (ΔG) and red bars (ΔG^‡) represent the differences relative to the DDD (or ΔΔG/ΔΔG^‡, respectively) as a function of position, from 5′ to 3′ for both palindromic strands. The largest changes to both are associated with the mismatched basepair. The associated energy diagrams are illustrated in the graphics on the right-hand side. To see this figure in color, go online. For a Figure360 author presentation of Fig. 7, see https://doi.org/10.1016/j.bpj.2022.03.031.

From these values of the ΔG^‡, we determined the forward rate constant k_f as a function of temperature using the Eyring equation (Eq. 3). As the rate constants are directly related to ΔG^‡, the trends for their variation are equivalent. The values for k_r are then determined from our values of k_f and K_eq using Eq. 2 and Eq. 3, respectively. The sum of k_f and k_r specifies k_ex and provides an overall idea of the timescale of the BI-BII interconversion. The relative differences from the control for the rate constants are provided in Table S8 and Fig. S6.

Our current results for the two mismatched sequences demonstrate a stepwise difference (i.e., p₁ to p₂ in a two phosphate sequence, such as 5′-X-p₁-Y-p₂-Z-3′) in backbone energies flanking the mismatched basepair. In the T3 and T9 sequences the position 5′ to the mismatch exhibits (Figs. 5 and 6) the largest decrease in van ’t Hoff enthalpy (ΔH_vH) and the largest increase in the enthalpy of activation (ΔH^‡). In both sequences the positions 3′ to the mismatch also exhibit large increases in van ’t Hoff enthalpy. In terms of free energies, there are also large stepwise differences in both ΔG and ΔG^‡ (Fig. 4).

TDG removes several types of DNA mismatches and lesions, including thymine in T:G basepairs, uracil in U:G basepairs, 3,N⁶-ethenocytosine, halogenated pyrimidines, and others (51, 52, 53, 54, 55). In addition, TDG also has significant sequence-dependent activity, where T:G mismatches in a CpG:T context (which also happens to be the case for our two test sequences) have the highest activity and those in ApG:T have the lowest activity (53, 54, 55), and there is modest activity differences based on the second nearest neighbor as well (43). Extensive biochemical research has been done to determine the minimal kinetic mechanism of TDG (43,51, 52, 53, 54, 55), by breaking up the enzymatic cycle into initial binding, interrogation, and nucleotide flipping steps followed by later catalytic and base removal steps. We propose that our results indicate an energetic contribution to those earliest steps.

While the T3 and T9 sequences have broad similarity to each other and their U:G counterparts, there are some differences. In the T9 sequence, the position of the basepairing partner to the mismatch (G4) has backbone conformational equilibrium essentially independent of temperature. The partner to the T3 thymine (G10) has only a small but measurable temperature dependence in ΔG. The phosphate 3′ to the mismatched T in the T9 sequence has a much more negative ΔG than T3 and opposite temperature dependence. These modest differences between T9 and T3 could be due to the fact that they are not in exactly identical sequence contexts (T3 is TpGpA and T9 is TpGpC), but this correlates with TDG activity as well, which has a small dependence on the nucleotide two steps away (43). In addition, the “steepness” of temperature dependence between T3 and U3 at the position immediately 5′ to the mismatched nucleotide is significantly different, with the ΔG in U3 increasing much more rapidly as temperature is increased. Nevertheless, the general trends are very similar.

It is important to note that, while T:G and U:G mismatches are both substrates for TDG, U:G mismatches are removed much more readily (52,53,55). How does this relate to our results? A close look at the data suggests at least a qualitative correlation, although further investigation is warranted. In Fig. 2 herein, as well as in Fig. 4 in Westwood et al. (28), it can be seen that the stepwise differences in ΔG in the phosphates flanking the mismatches are the largest and exhibit the most variation from the control. However, the stepwise differences for ΔG increase more dramatically for U:G compared with T:G, particularly as the temperature approaches physiological conditions. This indicates that the thermal energy necessary to favor the BII state more significantly impacts U:G mismatches. These temperature studies correlate well with the higher glycosylase activity for the U:G substrate compared with T:G.

The exchange constants k_ex are mostly on the order of 150–250 kHz (Tables 2, 3, and 4), which is in the fast exchange limit for the NMR timescale. These results match reasonably well with the timescale of enzyme motion along the DNA backbone. The model of facilitated diffusion by DNA glycosylases (56) is considered to have two primary modes: sliding and hopping. The sliding mode, which occurs over approximately a 10–12 basepair range, has a timescale on the microsecond range with a diffusion constant of D₁ = 6 × 10³ bp²/s for uracil DNA glycosylase (57), which matches well with our BI-BII interconversion rates. Assuming that the rates are 10³–10⁴ interconversions/s and the fact that the glycosylase samples at 10³–10⁴ bp²/s, the rates match reasonable well.

There has been important recent work with MD of mismatched DNA bound to TDG, which offers interesting insights into predicting structural properties by comparing calculated trajectories with crystal structures (58, 59, 60, 61). This is in addition to extensive efforts to computationally analyze the conformational properties of canonical DNA in general (62,63) as well as mismatched DNA (64). The value of the free energy of activation for DNA backbone interconversion is a source of discrepancy, as previous NMR approaches have determined values in the 10–14 kcal/mol range (32), as have some molecular modeling studies (65). However, other computational work has proposed much lower activation barriers for BI-BII interconversion, on the order of 1–2.5 kcal/mol (66,67). In addition, the work by Trieb et al. (66) calculated forward rate constants around 5 × 10⁶ Hz from their values of ΔG^‡ at 2.5 kcal/mol. Our results, those of Tian et al. (32), and other examples in the dynamic NMR literature consistently show activation energies for conformational interconversion on the order of 10–15 kcal/mol (39, 40, 41), even for small molecules, when using equivalent analysis.

Any role in lesion recognition due to changes in the backbone equilibrium may be as simple as the repair enzyme interactions being more favorable when one backbone population is dramatically different than canonical DNA, expressed as stepwise differences in ΔG (and therefore the relative populations of the two primary conformations). The common trends in %BII and ΔG differences for all U:G and T:G sequences relative to the control DNA duplex suggests a universality for substrates of a common glycosylase. In other words, different substrates, i.e., T:G, U:G, and even dhU:G all impact the DNA backbone equilibrium in similar ways and may contribute equivalently to the glycosylase recognition process. These differences in energies (such as ΔG, ΔG^‡, ΔH_vH, or ΔH^‡) may be a cue or trigger for the enzyme during its search for a proper substrate. Alternately, energy differences may assist the energetic process of nucleotide flipping. Maximally effective repair must have evolved mechanisms for efficient recognition as well as later chemical steps. Our ongoing work is pursuing correlations to enzyme activity based on sequence dependence and experimental conditions (temperature, ionic strength).

Conclusion

Our results offer important insight into the energetics of DNA backbone interconversion and a possible role in the mechanism of damage detection in base excision repair. We have demonstrated that different glycosylase substrate mismatches (U:G, T:G) have nearly identical effects in equivalent sequence contexts, suggesting a universality to the results. We have quantified the ΔG, ΔG^‡, the rate constants k_f and k_ex, and enthalpies for the backbone interconversion (ΔH_vH and ΔH^‡), and the impact of T:G mismatches on these quantities. The values observed in the mismatches have been compared with a control DNA, and the sites of greatest impact are those most closely associated with the mismatch in a stepwise fashion from the 5ʹ to 3ʹ phosphates flanking the mismatched basepair. The values of the free energies are perturbed up to 1.3 kcal/mol in the two basepair region containing the mismatche. Temperature studies allowed us to determine enthalpies associated with the backbone interconversion (ΔH_vH and ΔH^‡), again demonstrating that the values within the two basepair region neighboring the mismatch are those most significantly modified, and their values perturbed by up to ∼5 kcal/mol. These combined results suggest that perturbation to the DNA backbone thermodynamics and kinetics strongly correlate with the T:G mismatch. Recognition, interrogation, or nucleotide flipping by DNA repair enzymes are potentially initiated and more energetically favorable due to these energetic cues.

Author contributions

M.N.W., C.C.J., and G.A.M. performed the research. N.A.O. contributed analytical tools. M.N.W., C.C.J., and G.A.M. analyzed the data. G.A.M. wrote the manuscript.

Editor: Susan J Schroeder.