delta-Melt: Nucleic acid conformational penalties from melting experiments

Thermodynamic propensities of biomolecules to adopt non-native conformations are crucial for understanding how they function, but prove difficult to measure experimentally. Combining optical melting experiments with chemical modifications and mutations, we developed delta-Melt for measuring the energetic penalties associated with nucleic acid conformational rearrangements and how they vary with sequence and physiological conditions. delta-Melt is fast, simple, cost effective, and can characterize conformational penalties inaccessible to conventional biophysical methods.


Abstract
Thermodynamic propensities of biomolecules to adopt non-native conformations are crucial for understanding how they function, but prove difficult to measure experimentally.
Combining optical melting experiments with chemical modifications and mutations, we developed delta-Melt for measuring the energetic penalties associated with nucleic acid conformational rearrangements and how they vary with sequence and physiological conditions. delta-Melt is fast, simple, cost effective, and can characterize conformational penalties inaccessible to conventional biophysical methods.

Main text
The biological functions of nucleic acids often require changes in structure that occur in response to binding of proteins, ligands, and other nucleic acids, as well as changes in physiological conditions 1 . Changing the conformation of a biomolecule upon binding to a partner molecule comes with an energetic cost or penalty which has to be paid by favorable intermolecular interactions 2 . Although widespread in biology, and central to molecular recognition, we know very little about the magnitude of these conformational penalties and to what extent they determine binding affinity and specificity 2,3 . This is in part due to challenges in experimentally measuring these conformational penalties, which requires accurately measuring the population of a minor (<10%) conformation amongst thousands of conformations in an ensemble.
Recent developments in NMR relaxation dispersion (RD) techniques 4 have made it possible to accurately quantify the population of minor conformations and to deduce the conformational penalties accompanying their formation (Fig. 1a). Despite their success, these NMR-based approaches are technically demanding, laborious, expensive, and require the preparation of large quantities of isotopically enriched samples. Therefore, they do not lend themselves to high throughput investigations to comprehensively explore how conformational penalties vary with sequence, post-transcriptional modifications, and other physiological conditions. They are also limited to minor conformations with populations >0.01% and transitions on the micro-to-milli second timescales.
To address these limitations, we developed a new approach we call "delta-Melt" to measure conformational penalties in nucleic acids. delta-Melt combines two widely used and simple techniques in molecular biology, melting experiments 5 and perturbations in the form of mutations or chemical modifications. In delta-Melt, a mutation or chemical modification is introduced so as to substantially bias (> 90%) the conformational ensemble to the desired i th minor conformation for which we wish to determine a conformational penalty G conf  (i) (Fig. 1b). The unmodified wild-type (WT) and In theory, unlike NMR and other techniques, delta-Melt has no limit on the population and lifetime of the minor conformation and the size of biomolecules that can be examined.
However, delta-Melt does not provide any information regarding the kinetics of interconversion, and the analysis might prove difficult for certain molecules that exhibit multiple melting transitions. As an initial test, we used delta-Melt to measure the conformational penalty associated with opening Watson-Crick A-T base pairs (bps) in DNA. Base opening is the elementary step that drives melting and annealing of nucleic acids, and their unwinding by helicases 6 . Measurements of imino proton exchange rates 7 have shown that Watson-Crick A-T bps transiently open to form exceptionally low-populated (population~0.001%) and short-lived (lifetime~0.1 s) conformations (Fig. 2a).
We hypothesized that substitution of T by N 3 -methylated thymine (m 3 T), in which the imino proton of thymine is replaced by a methyl group to disrupt the A(N1)-T(H3) hydrogen bond, could be used to substantially bias the conformational ensemble towards the minor base-open conformation 8 (Fig. 2a). Indeed, NMR spectra (Extended Data Fig.   1) revealed that the modification disrupts the targeted bp with the m 3 T residue being intrahelical.
Optical melting experiments were used to measure the energetics of duplex melting with or without m 3 T substitution at ten sites corresponding to six different trinucleotide sequence contexts (Supplementary Table 1  (Methods) obtained from weak R1 RD profiles are denoted using "*". Data point 6 with a flat R1 RD profile is denoted using "x", indicating that the minor conformation in this case falls outside the detection limit.  Tables 1 and 9). Errors in Coopmelt were determined by propagating the uncertainties from the UV melts as described in Methods. Black dashed curves in panels a, c and e denote steric clashes. For panels b, d, f and g, error bars for NMR and delta-Melt measurements were obtained using a Monte-Carlo scheme as described in Methods, and by propagating the uncertainties from UV melts (and c(i) for panel g), respectively, as described in Methods.
Pearson's correlation coefficient (r) and root mean square error (RMSE), were computed as described in Methods. Blue shaded region denotes estimate of error of linear regression obtained using Monte-Carlo simulations, while open symbols denote data derived from weak RD profiles (Methods).
As a second test, we used delta-Melt to measure the conformational penalty associated with the Watson-Crick to Hoogsteen transition in duplex DNA (Fig. 2c, e) 9 . In this conformational transition, the purine base flips 180 about the glycosidic bond to adopt a syn conformation, and this is accompanied by constriction of the DNA backbone to allow hydrogen bonding between the bases (Fig. 2c, e) 10 . The conformational penalty associated with forming Hoogsteen bps is proposed to play important roles in DNAprotein recognition 2, 11 and in DNA damage induction 12 . Using NMR RD 4 , Hoogsteen bps have been shown to form transiently in naked DNA duplexes with populations of ~0.5% (A-T) and ~0.1% (G-C + ), and with lifetimes of ~1 millisecond 9 .
We previously showed that N 1 -methylated adenine (m 1 A) and N 1 -methylated guanine (m 1 G) can be used to substantially bias the conformational ensemble towards the minor A-T and G-C + Hoogsteen bps, respectively 9, 13 (Fig. 2c Next, we tested the throughput of delta-Melt by applying it to measure the G-C + Hoogsteen conformational penalties for all sixteen trinucleotide sequence contexts in a DNA duplex that was not used for determination of calibration curves (Fig. 2g Comparison of a total of six conformational penalties measured using delta-Melt Two adjacent Hoogsteen bps are often observed in X-ray structures of DNA bound to proteins and drugs 16 . We therefore also used delta-Melt to estimate the conformational penalty of forming a Hoogsteen bp when the neighboring bp is a preformed Hoogsteen. This allowed us to obtain insights into the cooperativity of forming tandem Hoogsteen bps (Methods). Indeed, we find that the conformational penalty associated with forming tandem Hoogsteen bps is smaller by 1-3 kcal/mol than the sum of penalties for forming two individual Hoogsteen bps for five different bp steps (Fig. 2h  Error bars for NMR and delta-Melt measurements were obtained using a Monte-Carlo scheme and by propagating the uncertainties from UV melts respectively, as described in Methods. Pearson's correlation coefficient (r) and root mean square error (RMSE) were obtained as described in Methods. Blue shaded region denotes estimate of error of linear regression obtained using Monte-Carlo simulations as described in Methods.
Finally, we applied delta-Melt to measure the conformational penalties accompanying formation of minor conformations in RNAs 19 . N 6 -methyl adenine (m 6 A) is an abundant epitranscriptomic RNA modification 20 . When paired with uridine, the methylamino group in m 6 A has recently been shown to isomerize between anti (major) and syn (minor) conformations which results in the loss of a hydrogen-bond 21 (Fig. 3a).
This conformational change has been proposed to play roles slowing a variety of biochemical processes that involve duplex melting and base pairing 22 . It has been shown that N 6 ,N 6 -dimethyl adenosine (m 6 2A), when paired with uridine, substantially biases the ensemble toward the minor syn m 6 A-U bp conformation 21 . As expected, we observed a good correlation between the differences in the energetics of melting of m 6 2A relative to Many unmodified RNAs undergo functionally important transitions to form minor conformations by reshuffling bps in and around non-canonical motifs 19 . Prior studies have shown that point substitution mutations can render these minor conformations the major state 19 . By using such a mutant to substantially bias the conformational ensemble toward a minor conformation in HIV-1 TAR 18 , we used delta-Melt to examine how the conformational penalty to adopt the minor conformation varies with and without 1 mM Mg 2+ . The difference in delta-Melt derived melting energetics with and without Mg 2+  In summary, a broad range of applications indicate that delta-Melt can be used to rapidly and accurately quantify conformational penalties in nucleic acids and to examine how they vary with sequence and physiological conditions. delta-Melt can also be extended to proteins, its throughput can be increased using advanced melting experiments 23 , and it may help guide the discovery of new minor conformations in biomolecules (Supplementary Note 2).

Buffer preparation
With the exception of 1 H proton exchange measurements (see below), the buffer were prepared by titrating ammonium hydroxide solution (14.8 M, Millipore Sigma) to the NMR buffer, followed by adjusting pH to 8.8 by adding hydrochloric acid (HCl) 24 . The effective concentration of ammonia [NH3] was computed using the buffer pH (8.8) and total ammonia concentration added to the buffer ([NH3]0) as follows: 10 −pK a NH3 + 10 −pH (1) where pK a NH3 is the pKa for ammonia. The buffer used in imino 1  Standard RNA phosphoramidites (n-ac-rA, n-ac-rG, n-ac-rC, rU, rm 6  concentrations, in particular ATP, and that of Mg 2+ during primer extension was seen to be necessary to avoid spurious NTP addition to form dangling ends in the resulting oligonucleotides. This was essential to obtain good agreement between the NMR and delta-Melt derived conformational penalties. The site-specifically labeled DNA strands comprising duplexes (A6-DNA(A16lb), scaf2_CGT GC (G6lb) and scaf2_TGC GC (G6lb)) were purchased from Yale Keck Oligonucleotide Synthesis Facility with cartridge purification.

NMR experiments
The imino 1 H exchange experiments were carried out on a Bruker Avance III 700 MHz spectrometer equipped with a HCN room temperature probe while the remaining NMR data was collected on Bruker Avance III 600 MHz or 700 MHz NMR spectrometers equipped with HCPN and HCN cryogenic probes, respectively.

C/ 15 N R 1ρ relaxation dispersion
Off-resonance 13 C/ 15 N R1ρ relaxation dispersion (RD) experiments were implemented using a 1D selective excitation scheme as described in prior studies 38,41,42 .
The spin-lock power (ω1/2π) used ranged from 100 to 1000 Hz, while the offsets used ranged from 10 and 3.5 times the spin-lock power for 13 C and 15 N experiments, respectively (Supplementary Table 6). For each resonance, six to ten delay times were selected during the relaxation period with maximum duration up to 60 ms and 150 ms for 13 C and 15 N, respectively. The experimental conditions for all the 13 C/ 15 N R1ρ RD experiments (temperature, magnetic field, solvent) are summarized in Supplementary   Table 5.

Analysis of R 1ρ data
The R1ρ data was analyzed as described previously 43 . Briefly, 1D peak intensities as a function of delay times extracted using NMRPipe 40 were fitted to a mono-exponential decay to obtain the R1ρ value for the different spin-lock power and offset combinations.
The error in R1ρ was estimated using a Monte Carlo procedure as described previously 19 .
Exchange parameters of interests, such as the population of the i th minor conformation  Table 7).
It should be noted that in addition to sensing conformational changes at the bp  Fig. 7). 13 C/ 15 N CEST experiments on the aromatic C6/C8/C2 and C1' carbons, and imino nitrogen U-N3 atoms were carried out using a pulse sequence employing a selective excitation scheme in a 1D manner, as described previously 22,47,48 . The 13 C CEST experiments on the methyl carbon in hpGGACU m6A6 (m 6 A6 C10 lb) were carried out in a 2D mode using a pulse sequence derived from prior 13 C CEST pulse sequences 21,47,49 .  Supplementary Table 5.

Analysis of CEST data
The analysis of CEST data was performed as described previously 22,48 . Briefly, 1D peak intensities were obtained in a manner similar to that for R1ρ. All intensities at a given radio-frequency (RF) power were normalized by the average of the intensities over the triplicate CEST measurements with zero relaxation delay using the same RF power, to obtain normalized CEST profiles. Normalized CEST profiles were plotted as a function of offset  = RF -OBS, in which OBS is the Larmor frequency of the observed resonance and RF is the angular frequency of the applied spin-lock (Extended Data Fig. 6). The Treatment of spin-lock inhomogeneity and alignment of the initial magnetization during CEST fitting was performed as described previously 48 . Only GS magnetization was considered to be present at the start of the relaxation delay during CEST fitting for the C2/C6/C8/C1' and imino nitrogen spins, owing to the absence of the equilibration delay in the pulse sequence. GS and ES magnetization was considered to be equilibrated while fitting the methyl CEST data on hpGGACU m6A6 (m 6 A6 C10 lb) as the pulse sequence employs hard pulses to excite spins 21 . The sensitivity of the fit to the CEST data to changes in pi was examined by computing the reduced χ 2 while fixing pi to a range of different values (Extended Data Fig. 7). All the fitting parameters from CEST experiments have been summarized in Supplementary Table 7. The errors of all the fitting parameters were estimated using 100 Monte Carlo iterations as described previously 22 .

Imino 1 H exchange
The kinetics of base opening were determined using a combination of experiments including a saturation recovery experiment to measure water proton R1 (R1w) and a magnetization transfer experiment to measure the exchange rate of the imino proton with water, as described previously 24 . The relaxation delay times for measuring water proton  Table 2.

Analysis of imino 1 H exchange data
The net exchange rate (kex) between imino and water proton can be measured by fitting the 1D imino 1 H peak volume at each delay time (t) in the magnetization transfer experiment to Equation 2 shown below, where W(t) is the imino 1 H peak volume as a function of relaxation delay time t, W 0 is the initial imino 1 H peak volume at zero delay, W(t)/W 0 is the normalized peak volume, E is the efficiency of the pulse for inverting water, R1n represents the summation of imino 1 H where 0 is the inverse of the base opening rate (lifetime of the closed state), is the base catalyst accessibility factor which is generally assumed to be 1 50 , [NH3] is the effective ammonia concentration, Kdiss is the bp dissociation constant defined by, and kB is the rate constant for exchange catalysis which is given by, B = coll 1 + 10 pK a Nu −pK a NH3 (7) where kcoll is the bi-molecular collision rate constant between the imino proton and  Table 4).
To benchmark our implementation of the imino 1 H exchange measurements in this study, we compared our measurements on T6 in G-DNA at 25 °C to a prior study 50 and obtained a consistent conformational penalty for opening (Extended Data Fig. 5). pi values for T3, T5, T16, T17, T18, T19 and T21 in TBP-DNA at 15 °C were obtained from a prior study 25 .

Experiments and sample conditions
Optical melting experiments were carried out on a PerkinElmer Lambda 25 UV/VIS spectrometer with a RTP 6 Peltier Temperature Programmer and a PCB 1500 Water Peltier System. At least three measurements were carried out for each DNA and RNA duplex using a sample volume of 400 µL in a Teflon-stoppered 1 cm path length quartz cell. The absorbance at 260 nm was monitored while the temperature was varied at a ramp rate of 1 °C/min.

Data analysis
The melting temperature (Tm) and standard enthalpy change of melting (H melt,WT ○ or H melt,Mut ○ ) (Fig. 1d), respectively, were obtained by fitting the absorbance at 260 nm where melt , melt , fold and fold are coefficients describing the temperature dependence of the extinction coefficients for the melted and folded species, respectively, and pmelt is the population of the melted duplex/hairpin species. pmelt for a WT duplex and hairpin are given by the following expressions (analogous expressions can also be written for melting of Mut by replacing WT by Mut),  (12) where CT is total concentration of the duplex/hairpin. Using the obtained enthalpies and entropies, the free energy of melting was then computed as follows, For UV melting profiles of TAR with 1 mM Mg 2+ , TAR G28U with and without 1mM Mg 2+ , scaf2_AGG m1GC and scaf2_TGG m1GC (Extended Data Fig. 2), minor deviations from two-state fits were observed in the lower baseline which could potentially arise from the existence of multiple folded species in equilibrium with each other 51 . Due to the lack of knowledge about the second folded species for the above samples, we have chosen to fit the UV curves for them assuming a two-state model. Melting data with similar minor deviations has also been fit assuming a two-state approximation to extract thermodynamics parameters in the literature 52 . Nevertheless, understanding the conformational dynamics that gives rise to such minor deviations and how that affects the interpretation of the melting data will be the subject of future studies.

Comparison of conformational penalties between NMR experiments and delta-Melt
For a given equilibrium between the major and i th minor conformation, Major ⇔ Minor, the conformational penalty ( ΔG conf ○ (i) ) measured by NMR experiments was computed as follows, where T is the temperature (K), R is the gas constant (kcal/mol) and pi is the population of the i th minor conformation, respectively. Errors in ΔG conf ○ (i) were obtained by propagating the error in pi obtained from the NMR measurements.
The enthalpy (ΔH conf ○ (i)) and entropy (ΔS conf ○ (i)) differences between the major and i th minor conformation were obtained by fitting ΔG conf ○ (i) as a function of temperature to the following equation, Errors in ΔH conf ○ (i) and ΔS conf ○ (i) were determined using a Monte-Carlo procedure.  Fig. 11.
Given conformational equilibria for the melting of WT and Mut nucleic acids, we will have the following equation according to Fig. 1d,  (i), obtained as described above.
Given that the enthalpy and entropy change are also state variables in a manner analogous to the free energy, equations analogous to Equation 20 above, can also be written for them.
We also define,  (29) The Coopmelt term was determined using delta-Melt by using duplexes containing N 1methylated purines at tandem bps, and their singly methylated counterparts (Fig. 2h, Extended Data Fig. 9). Errors in Coopmelt were obtained by propagating the errors in ΔΔG melt ○ (i) values obtained as described above. In a manner analogous to defining cooperativity of Hoogsteen bp formation at two adjacent bps, we can also define cooperativity of forming Hoogsteen bps at three adjacent bps as follows  (33)