NMR-Based Rational Drug Design of G:G Mismatch DNA Binding Ligand Trapping Transient Complex via Disruption of a Key Allosteric Interaction

Abstract

Small molecules that bind to mismatched DNA have been applied in various fields, including nanotechnology, bioimaging, and therapeutics. However, the intrinsic dynamic nature of mismatched DNA complicates the prediction of structural changes upon ligand binding, hindering rational ligand design. In this study, NMR was used for structure-based drug design, with a focus on the G:G mismatch binder ND and the structural dynamics of the DNA-ND complex. Through comprehensive NMR analysis with isotope labeling, two complex structures, the transient and stable complexes, were successfully determined. The nucleobase flip-outs and the distortion of the phosphate backbone of the complex structures were characterized by residual dipolar coupling (RDC) and ³¹P NMR, respectively. The RDC-refined stable complex structure suggested that the ligand linker–nucleobase interaction allosterically regulates a structural transition. This interaction was experimentally validated by ¹H–¹⁵N HSQC spectra using a ¹⁵N-labeled ligand. Disruption of this key allosteric interaction facilitated the design of a new ligand, sND, that traps the transient complex structure. In conclusion, comprehensive NMR analysis using a weak binder aids in designing nucleic acid-binding ligands based on transient complex structures.

Introduction

Mismatched base pairs are typical noncanonical Watson–Crick (WC) base pairs consisting of G:G, G:A, A:A, G:T, A:C, C:C, T:T, and C:T.¹ These mismatched base pairs are produced by various mechanisms, such as errors in DNA replication,² mutagenic chemicals,³ chemical modifications,⁴ and ionizing radiation.⁵ Such genomic errors are usually quickly repaired by a DNA repair system, but its dysfunction can lead to the accumulation of genomic errors and the development of cancer.⁶⁻¹¹ DNA mismatches also occur in trinucleotide repeats and minisatellite sequences (i.e., CGG, CTG, CAG) present in the genome and are associated with repeat expansion diseases, including fragile X syndrome, myotonic dystrophy, and Huntington’s disease.¹²⁻¹⁷ Such elongated repeat DNAs exhibit an increased propensity for slipped DNA (S-DNA), producing a metastable hairpin structure with X:X mismatches in the CXG/CXG motif.¹⁸⁻²³

To date, several synthetic molecules targeting mismatched base pairs have been reported.²⁴⁻³⁴ Nakatani and his collaborators have reported a series of mismatch-binding molecules, such as naphthyridine-azaquinolone (NA) and naphthyridine carbamate dimer (NCD), which target CAG repeats (Huntington’s disease) and CGG repeats (fragile X syndrome), respectively.^{26,33,35−38} The binding of NA to the CAG/CAG motif in the slipped CAG DNA structure induced the contraction of CAG repeats in the striatum of the R6/2 mouse model of Huntington’s disease.³⁹ These findings highlight the potential of mismatch-binding molecules as therapeutic agents for repeat expansion diseases and cancer.

While the demand for the development of mismatch-binding molecules is increasing, rationally designing such small molecules remains challenging. Mismatched base pairs have lower thermodynamic stability than WC base pairs,^40,41 leading to increased conformational fluctuations and facilitating ligand insertion. Most of the reported mismatch-binding molecules utilize this conformational flexibility for binding. Although structural information on mismatched DNA-ligand complexes has accumulated,²⁴⁻³³ the intrinsic flexibility of mismatched sites makes it difficult to predict structural changes in nucleic acids upon ligand binding, hindering structure-based drug design targeting nucleic acids.

For the rational design of mismatch-binding molecules, it is necessary to understand the recognition mechanism of ligands, including their structural dynamics. Here, we revisited a small molecule naphthyridine dimer (ND)^42,43 to design new mismatch-binding molecules. ND is the molecule that served as the basis for designing NA and NCD, and its chemical structure is similar to those of NA and NCD. Moreover, ND binds more weakly to the d(CGG/CGG) motif than does NCD, making it suitable for detecting transient complexes. A comprehensive NMR study was performed to understand the dynamic nature of the DNA structure upon ND binding by focusing on the differences between ND, NA, and NCD. We succeeded in obtaining precise structural information about the DNA-ND complex through advanced NMR studies, including stable isotope (SI) labeling, natural abundance ¹H–¹³C residual dipolar coupling (RDC) measurements, and ³¹P NMR analyses. The chemical synthesis of the isotope-labeled ligands and the use of site-specifically isotope-labeled DNA were critical for assigning NMR signals of the transient complex and provided detailed information on the DNA–ligand interactions. Based on the obtained structural insights, we achieved the rational design of a novel small molecule trapping the transient structure.

Results

ND Stabilizes the DNA Duplex Containing the d(CGG/CGG) Motif

We used a model DNA named CGG-dsDNA (Figure 1A), in which the d(CGG/CGG) motif is embedded within a double-stranded DNA (dsDNA), to investigate the recognition mechanism of the d(CGG/CGG) motif by the G:G mismatch binder ND (Figures 1B and S1A). This sequence is the same as that previously used for the structural study of the DNA-ligand complexes possessing a d(CXG/CXG) motif (NA (X = A) and NCD (X = G)).^26,33 The UV absorbance changes at 260 nm of free dsDNA (5 μM) showed a typical sigmoidal curve with a melting temperature (T_m) of 26.8 ± 0.1 °C (Figure S2A). In the presence of 20 μM of ND, the T_m value of CGG-dsDNA increased to 44.1 ± 0.3 °C, corresponding to a melting increase of 17 °C. Isothermal titration calorimetry (ITC) was then performed to obtain the thermodynamic parameters (Figure S2B). The significant heat generation during ND injection indicated that the binding is predominantly driven by an enthalpic factor. The best fit for the binding isotherm using a single-site model yielded an apparent K_D of 391 ± 94 nM with the binding stoichiometry (n) of 2.1 ± 0.1, which is consistent with two ND molecules binding to CGG-dsDNA. To investigate the kinetic properties, we performed surface plasmon resonance (SPR) assays with a sensor chip of surface-immobilized hairpin DNA (hpDNA) containing the d(CGG/CGG) motif with ND as an analyte. Fitting with a single-site binding model yielded the following apparent kinetic parameters: k_on of 1.4 × 10⁴ M^–1 s^–1, k_off of 5.1 × 10^–3 s^–1, and K_D of 350 nM (Figure S2C).

Figure 1.

Naphthyridine dimer (ND) binds to the d(CGG/CGG) motif in two ways: a 1:1 complex and a 1:2 complex. (A) Double-stranded DNA sequence containing a d(CGG/CGG) motif named CGG-dsDNA. (B) Chemical structure of the naphthyridine dimer (ND). The atom numbering of ND is described in Figure S1. (C) Imino proton region of the 1D ¹H NMR titration spectra (500 MHz, 10 °C) of CGG-dsDNA with various concentrations of ND. The ratios of [DNA]:[ND] are shown on the left. NMR signals that appear or disappear upon complex formation are indicated by dotted lines, solid lines, or triangles. The broadened peaks marked by stars indicate terminal guanine imino protons. These protons are not assigned due to the lack of NOEs. NPn (n = 1–4) indicates the amide proton in 2-amino-1,8-naphthyridine (NP). Conditions: 0.1 mM dsDNA, indicated molar ratio of ND. (D) Synthetic scheme of ¹⁵N₄–ND. The N01 and N03 atoms (colored red) of 2-amino-1,8-naphthyridine are labeled with ¹⁵N. (E) 1D ¹⁵N spectrum (500 MHz spectrometer, 10 °C) of free ¹⁵N₄–ND and 2D ¹H–¹⁵N HSQC spectra (500 MHz, 10 °C) of CGG-dsDNA with ¹⁵N₄–ND. The ratio of [DNA]:[¹⁵N₄–ND] is shown in the spectra. The green and orange signals correspond to the 1:2 and 1:1 dsDNA-ND complexes, respectively. Conditions: 1.5 mM dsDNA, indicated molar ratio of [ND].

NMR Study of the DNA-ND Complex Revealed the Presence of Stable 1:2 and Transient 1:1 Complexes

1D ¹H NMR titration experiments were subsequently performed to further investigate the binding of ND to CGG-dsDNA. The assignment of the free dsDNA is presented in the bottom panel of Figure 1C (detailed signal assignments are shown in Figure S3). Upon the addition of ND, new peaks appeared at 10.0–12.5 ppm with a concomitant decrease in the free dsDNA signals without changes in the chemical shifts or peak shapes, indicating the slow exchange within the NMR time scale. These new peaks appearing at 10.0–12.5 ppm are derived from the ND amide and guanine imino protons, which form WC-like naphthyridine-guanine base pairs. The addition of 3 or 4 equiv of ND did not induce significant signal changes (Figure S4), indicating that ND binding was saturated when 2 equiv of ligand were added. These results suggest that the 1:2 dsDNA-ND complex is formed at the saturation point and that dsDNA-ND complexes with a stoichiometry higher than 1:2 are not formed even with excess ligands, analogous to NCD and NA. On close inspection of the spectra, another set of peaks (marked by orange triangles in Figure 1), which belong to neither the free dsDNA nor the 1:2 complex, was also observed. With 1 equiv of ND, the signal intensities of these transiently appearing peaks reached a maximum. Similar behavior was also observed for the methyl proton signals (Figure S4B). In summary, ND binds to CGG-dsDNA via two types of binding modes: one is assigned to the transient peaks (marked by orange triangles in Figure 1C), and the other is the 1:2 complex (marked by green lines in Figure 1C). The observation of this transient complex was a breakthrough in the rational design of new mismatch-binding ligands, as discussed below.

To identify ligand-derived peaks and binding stoichiometries unambiguously, we synthesized ¹⁵N₄–ND, in which the N01 and N03 of 2-amino-1,8-naphthyridine of ND are labeled with ¹⁵N (Figure 1D), using the inexpensive and readily handled solid reagent ¹⁵NH₄Cl as the ¹⁵N source. 1D ¹⁵N spectrum of the synthesized ¹⁵N₄–ND gave two signals corresponding to the symmetric N01 and N03 atoms (Figures 1E top, S5, S6, and S7).

When 2 or more equiv of ¹⁵N₄–ND were added to dsDNA, 1D ¹⁵N and 2D ¹H–¹⁵N HSQC spectra gave four signals derived from hydrogen bond (H-bond) donors or methyl groups of ¹⁵N₄–ND (Figures 1E middle marked in green, S4C, S7, S8A top, and S8B top). This indicates that four naphthyridine rings are located in different chemical environments, which are consistent with the 1:2 dsDNA-ND complex as discussed above (Figure 1C). With 1 equiv of ¹⁵N₄–ND, two additional peaks were observed (Figures 1E bottom marked in orange, S7, S8A bottom, and S8B bottom), indicating that the two naphthyridine rings are located in different chemical environments, which is consistent with the 1:1 dsDNA-ND complex. In summary, ND forms both 1:1 and 1:2 complexes with CGG-dsDNA, unlike NA and NCD, which form 1:2 complexes with d(CXG/CXG) motifs without forming 1:1 complexes.^26,33 The transient 1:1 complex observed for ND is expected to provide new insights into the molecular recognition mechanism of the ligand to mismatched DNA.

RDC-Refined NMR Structure of Stable 1:2 Complex Reveals Linker–Nucleobase Interactions

For a detailed analysis of the mode of recognition of the d(CGG/CGG) motif by ND, we performed 3D structural analysis of the 1:1 and 1:2 complexes. First, we tackled the NMR signal assignment and structural determination of the 1:2 complex to confirm whether ND adopts the same binding mode as NA and NCD and to obtain a clue for the subsequent NMR analysis of the 1:1 complex.

The 1:2 complex structure was determined via the conventional method,^44,45 acquiring a series of 2D NMR spectra (¹H–¹H NOESY, ¹H–¹H DQF-COSY, ¹H–¹H TOCSY, ¹H–¹³C HSQC, ¹H–¹⁵N HSQC, ¹H–³¹P CPMG-HSQC, Figures 2A,B and S9–16). The structure was then refined with RDC giving long-range distance constraints.^46,47 The superimposition of the 30 accepted structures is shown in Figure 2C (a comparison before and after RDC refinement is shown in Figure S17). The structures converged well with an all-atom RMSD of 1.18 ± 0.38 Å (Table S1). Many intermolecular NOEs (ND1-ND2: 20, DNA-ND1/ND2: 56) support the validity of the obtained NMR structures (Table S1, and NOE maps are shown in Figure S18). The calculated 1:2 complex structure showed almost the same binding mode as those of NA and NCD (Figure S19). That is, the ligand binds to the d(CGG/CGG) motif with a 1:2 dsDNA-ligand stoichiometry and recognizes nucleobases through H-bonds with interstrand binding, resulting in the flip-out of cytosines (C5 and C16) in the d(CXG/CXG) triad. These flip-outs are also supported by the characteristic ¹H chemical shifts of H5 and H6 of these cytosines (Figure S14) and their distinct C5–H5 RDC values compared with those of the stem regions (Figure 2D).

Figure 2.

NMR assignment and solution structure of the 1:2 dsDNA-ND complex. (A) Expansion of the 2D ¹H–¹H NOESY spectrum (800 MHz, 10 °C) of the 1:2 dsDNA-ND complex. Imino proton region (10.0–14.5 ppm) is expanded. Conditions: 2.0 mM dsDNA, 4.0 mM ND. (B) Expansion of 2D natural abundance ¹H–¹⁵N HSQC spectra (950 MHz, 10 °C) of the 1:2 complex. Conditions: 2.0 mM dsDNA, 4.0 mM ND. (C) Superimposed representations of the 30 lowest-energy NMR structures of the 1:2 complex (PDB ID: 8ZD2). Two ND molecules (ND1 and ND2) are colored blue and purple, respectively. The DNA strands are colored in light blue and light green. The flipped-out cytosine bases are colored in orange. The structures are seen from the side (left) and minor groove (right). The structures are aligned at the ligand-bound site (G6, G7, G17, G18, ND1, and ND2 residues). (D) RDC values (¹D_CH) of the 1:2 complex. The C2–H2 (blue), C5–H5 (orange), C6–H6 (gray), and C8–H8 (yellow) RDCs are shown as a function of the secondary structure. The spectral resolution value was used as the experimental uncertainty (error bars). (E) 1D ³¹P spectrum (600 MHz spectrometer, 40 °C) of the 1:2 complex. Trimethyl phosphate (TMP) was used as an external standard for the phosphorus chemical shift (δ = 0 ppm). The intense peak marked by Pi corresponds to the phosphate buffer. Conditions: 2.0 mM dsDNA, 4.0 mM ND.

Gorenstein reported that the ³¹P chemical shifts of various DNA-drug complexes correlate with the observed degree of unwinding of the duplex DNA upon drug binding.⁴⁸ We then investigated the ³¹P chemical shifts of the 1:2 complex by 1D ³¹P (Figure 2E) and 2D ¹H–³¹P CPMG-HSQC spectra (Figure S15). The ³¹P chemical shifts of all the residues observed between −4.8 and −3.2 ppm were assigned to a right-handed conformation within the range of the B_I and B_II conformations defined by Gorenstein.⁴⁸ Notably, slight downfield shifts were observed for the flipped-out cytidines and their adjacent residues on the 3′ side (C5, G6, C16, and G17). The dihedral angle parameters of the calculated structure were consistent with the chemical shifts of ³¹P with perturbations around the flipped-out cytosines (Figure S20). These results indicate that the flip-outs of cytosine do not induce significant structural distortions on the phosphate backbone and, conceivably, minimize structural distortions in the phosphate backbone associated with ligand binding.

All of the methylene linkers of ND in the 1:2 complex gave sharp NMR signals and were unambiguously assigned (Figure S11). Their sharp signals imply that ND linkers adopt a single conformation or are in fast exchange between two or more conformations. To understand the chemical structure of the ND molecule, crystal structure analysis of ND was attempted. While powder X-ray diffraction confirmed the presence of crystals, single crystals suitable for X-ray crystal structure analysis could not be obtained. Therefore, the microcrystal electron diffraction (MicroED) technique was applied to microcrystals, successfully yielding the crystal structure of ND (Figure 3A). In the crystal, the secondary amine in the ND linker exists in a protonated NH₂⁺ state, forming salt bridges with the surrounding chloride ions.

Figure 3.

Experimental validation of the linker–nucleobase interactions observed in the 1:2 dsDNA-ND complex. (A) Crystal structure of ND (HCl salt) analyzed by MicroED. The Coulomb potential map is shown as a green mesh. (B) Expansion of the NMR structure of the 1:2 complex seen from the major groove. Hydrogen bonds are shown as yellow dotted lines. For the drawing of H-bonds, the default criterion (2.8 Å of maximum distance) of Maestro (Schrödinger) was used. (C) Schematic illustration of the expanded region of Figure 3B. (D) 1D ¹⁵N NMR spectra (500 MHz spectrometer, 10 °C) of the ¹⁵N₁–ND. Top: Free ¹⁵N₁–ND in buffer. Conditions: 0.2 mM ¹⁵N₁–ND. Bottom: 1:2 dsDNA-¹⁵N₁–ND complex. Conditions: 1 mM dsDNA, 2 mM ¹⁵N₁–ND. (E) 2D ¹H–¹⁵N HSQC spectrum (500 MHz, 10 °C) of the 1:2 dsDNA-¹⁵N₁–ND complex. For coherence transfer by INEPT, 2.5 ms of delay (1/4 J) was used. 1D ¹H spectrum of the 1:2 complex measured with unlabeled ND is attached on top of the 2D spectra for reference.

In RDC-refined solution structures of the 1:2 complex, the interactions between the ND linker amine and nucleobases of G6/G17 on the major groove side were found (Figure 3B,C). The boxplot in Figure S21 illustrates that 58 of the 60 interatomic distances between the ND linker nitrogen and the G6/G17 O6 atoms fall within a range of H-bond distances. To experimentally validate the formation of the linker–nucleobase H-bonds, we synthesized ¹⁵N-labeled ND (¹⁵N₁–ND), in which the nitrogen of the ND linker is labeled with ¹⁵N, to observe the NMR signals of the amino groups in the ND linker (Scheme 1 and Figure S22). 1D ¹⁵N spectrum of the synthesized ¹⁵N₁–ND exhibited one resonance at 41.6 ppm (Figure 3D top), confirming the labeling of the linker amine. Using ¹⁵N₁–ND, 1D ¹⁵N and 2D ¹H–¹⁵N HSQC spectra of the 1:2 complex were measured (Figures 3D bottom,E and S7). 1D ¹⁵N spectra of the complex exhibited two resonances corresponding to the two bound ND molecules, ND1 and ND2. The ¹H–¹⁵N HSQC spectra exhibited two sets of NH₂ signals, demonstrating that in the 1:2 complex the ND linker exists in the protonated NH₂⁺ state as found in the MicroED crystal structure of ND. The shoulder peaks at the upper portion likely arise from the ¹⁵NHD species (Figure S23), further supporting the NH₂⁺ state of the linker amine. Examining the ¹H chemical shift, one of the two protons of the NH₂ signal showed a canonical chemical shift with amino groups (7.65 and 8.02 ppm) such as lysine (typical ¹H chemical shifts of lysine in protein: 7–8 ppm.⁴⁹ In contrast, the other showed a significant downfield shift to ∼11 ppm (10.82 and 11.21 ppm), indicating a reduced electron density of the protons due to the H-bond formation. H-bond formation was further confirmed by the chemical shift of N04 in ¹⁵N₁–ND, where the formation of the 1:2 complex induced a downfield shift of approximately 6 ppm, changing from 41.6 ppm for free ¹⁵N₁–ND to 48.2 and 47.8 ppm for ND1 and ND2, respectively (Figures 3D,E and S7). The difference in the chemical shift between ^ND1N04 and ^ND2N04 (Δδ ∼ 0.4 ppm) is likely due to differences in the residues adjacent to the d(CGG/CGG) motif. These findings support the presence of the H-bonds between the ND linker amine and the G6/G17 O6 atoms in the major groove (Figure S16), and this ligand linker–nucleobase H-bonds allosterically stabilize the 1:2 complex.

Scheme 1. Synthetic Scheme of ¹⁵N₁–ND.

The N04 atom (colored in red) in the amino group of the ND linker is labeled with ¹⁵N.

NMR Signal Assignment and Solution Structure of the Transient 1:1 Complex

Next, we worked on the NMR signal assignment and structural analysis of the transient 1:1 complex under the condition where 1 equiv of ND was added to CGG-dsDNA (Figure 1C). A series of NMR measurements (¹H–¹H NOESY, ¹H–¹H TOCSY, ¹H–³¹P CPMG-HSQC, Figures 4A,B and S24–S30) was performed at 20 °C, and the signals derived from the 1:1 complex were assigned. Due to the high symmetry of the ligand-binding site, two possible binding topologies were allowed without any inconsistency in signal assignments (Figure S24). To unambiguously determine the binding topology, we introduced 10% of the residue-specific isotope labels (¹³C and ¹⁵N) into the DNA,⁵⁰ enabling unambiguous assignments of all guanine imino protons in the ligand-binding site (Figures 4A and S25) with a unique binding topology. Most of the 1:1 complex signals were assigned, but some residues remained unassigned in the NOESY spectrum, because of the signal broadening.

Figure 4.

Assignment and solution structure of the 1:1 dsDNA-ND complex. (A) 1D projection of the 2D ¹H–¹⁵N HSQC spectra (600 MHz, 20 °C) of 10% ¹³C, ¹⁵N-guanine site-specifically (G6, G7, G17, or G18) labeled CGG-dsDNA in the presence of 1 equiv of ND. Conditions: 0.1 mM dsDNA (10% ¹³C, ¹⁵N-labeled), 0.1 mM ND. (B) 2D ¹H–¹H NOESY spectrum (950 MHz, 20 °C) of the mixture of the free dsDNA, the 1:1 and 1:2 complexes. The imino proton region (11.2–12.8 ppm) is expanded. NOEs derived from the 1:1 complex are colored in orange. Conditions: 2.7 mM dsDNA, 2.7 mM ND. (C) Superimposed representations of the 10 lowest-energy NMR structures of the 1:1 complex (PDB ID: 8ZD8). The ND molecule is colored in blue. The DNA strands are colored in light blue and light green. The structures are seen from the side (left) and minor groove (right). The structures are aligned at the ligand-bound site (C5, G6, G7, C16, G17, G18, ND residues). (D) 1D ³¹P spectra (500 and 600 MHz spectrometer, 20 °C) of the free dsDNA (lower), the mixture of the free dsDNA, the 1:1, and 1:2 complexes (middle), and the 1:2 complex (upper). Conditions: 2.0–2.7 mM dsDNA, the molar ratio of the [DNA]:[ND] is shown on the left of the spectra. The NMR signals of the 1:1 complex are marked by orange triangles. Trimethyl phosphate (TMP) was used as an external standard for the phosphorus chemical shift (δ = 0 ppm). The intense peak marked by Pi corresponds to the phosphate buffer.

Strong NOEs were observed between the G6 and G17 imino protons and the naphthyridine amide protons (Figure 4B), indicating H-bond formation between these guanines and the 2-amino-1,8-naphthyridine (NPs). Strong NOEs were also observed between guanine (G7, G18) imino protons and cytosine (C5, C16) amino protons (Figure S24), indicating that these cytosines (C5 and C16), which were flipped out in the 1:2 complex, formed WC base pairs with guanines in the 1:1 complex as in the free dsDNA. The downfield shift of C5 and C16 observed in the 1:2 complex was not observed in the 1:1 complex, suggesting that these cytosines are stacked within the double helix (Figure S26).

The 1:1 complex structure was determined via the conventional method (Figure 4C). Although overall convergence was limited because of the lower percentage of assigned signals, the ligand-binding sites converged well, which was supported by 16 intermolecular NOEs between DNA and ND (Figure S31 and Table S2). Refinement by RDC could not be achieved because of the lack of enough sensitivity in natural abundance ¹H–¹³C HSQC spectra and severe signal overlap. Notably, the 1:1 complex exhibited no cross-peak in the ¹H–¹⁵N HSQC spectrum for the linker amine in ¹⁵N₁–ND, suggesting that there was no H-bond between the linker amine and the nucleobases (Figure S32). In the 1:1 complex, ND forms WC-like hydrogen bonds with the central G:G mismatch, maintaining the adjacent G:C base pairs, unlike the 1:2 complex. The helical orientation of the linker of ND is right-handed and aligned with the DNA helix in the 1:1 complex, whereas it is left-handed in the 1:2 complex.

Next, we analyzed the phosphate backbone of the 1:1 complex by ³¹P NMR experiments (1D ³¹P (Figures 4D and S29) and 2D ¹H–³¹P CPMG-HSQC (Figure S30)). By comparison with the free dsDNA and 1:2 complex spectra, the 1:1 complex signals were assigned (Figure 4D marked in orange and Supplementary Discussion). In the 1:1 complex, G7 and G18 residues, in which the phosphate backbone structure was elongated and unwound by the insertion of the aromatic ring of ND, and adjacent G6 and G17 residues presented significantly large downfield shifts in ³¹P signals. These shifts are consistent with the observation that the ligand-binding region of the 1:1 complex is structurally distorted and deviates from the canonical B-form DNA (Figure 4C). In the 1:2 complex, such shifts were not observed (Figure 4D), which is consistent with the calculated structure of the 1:2 complex maintaining the B-form phosphate backbone structure. Notably, the significantly large downfield shift of ³¹P signals is reported for the complex of DNA with intercalators such as propidium iodide,⁵¹ actinomycin D,^52,53 echinomycin,⁵⁴ and daunorubicin.⁵⁵ The phosphate backbone structures of these DNA-intercalator complexes are unwound, and therefore, it is likely that the significantly large downfield shift of the ³¹P signals corresponds to the unwinding of the phosphate backbone structures, supporting the unwinding of the 1:1 dsDNA-ND complex.

Thermal Stability and Binding Kinetics of the DNA-ND Complexes

During the thermal melting of the dsDNA-ND complex monitored by UV, a simple sigmoid curve was observed, indicating a two-state transition. In contrast, NMR experiments enabled the detection of three states, the free dsDNA, 1:1, and 1:2 complexes, allowing investigation of the thermal stability of each state. When 1 equiv of ND was added to dsDNA at 10 °C, an almost equal amounts of the free dsDNA, 1:1 complex, and 1:2 complex states coexisted (Figure 1C). The imino proton signals were subsequently monitored by changing the temperature in 10 °C increments, from 10 to 40 °C (Figure 5A). The free dsDNA and the 1:1 complex peak disappeared between 30 and 40 °C, indicating that they melted within this temperature range. In contrast, the 1:2 complex was still observed at 40 °C, indicating that it did not melt even at 40 °C. These results demonstrate that the thermal stability is in the following order: 1:2 dsDNA-ND complex >1:1 dsDNA-ND complex ∼ free dsDNA.

Figure 5.

Comparison of the thermodynamic and kinetic properties of the 1:1 and 1:2 complexes. (A) 1D ¹H spectra (950 MHz, 10–40 °C) of the dsDNA-ND complexes measured at various temperatures. The free dsDNA, 1:1, and 1:2 complexes are indicated by blue, orange, and green triangles, respectively. Conditions: 0.1 mM dsDNA, 0.1 mM ND. (B) 2D ³¹P–³¹P EXSY spectra (500 MHz spectrometer, 20 °C) of the free dsDNA and the dsDNA-ND complexes. EXSY cross-peaks corresponding to exchange between the free dsDNA and the 1:1 complex were observed. The peaks marked by an asterisk (*) are likely to be from an alternative conformation of free dsDNA induced by the ligand. Conditions: 2.7 mM dsDNA, 2.7 mM ND. The mixing time was 300 ms. 1D ³¹P spectrum is attached on top of the 2D spectra for reference. (C) 2D ¹H–¹H NOESY spectra (800 MHz, 20 °C) of the dsDNA-ND complex. EXSY cross-peaks of G7 and G18 between the free dsDNA and the 1:1 complex are colored in red. Conditions: 2.7 mM dsDNA, 2.7 mM ND. The mixing time was 200 ms. (D) Determination of the exchange rate between the free dsDNA and the 1:1 complex. The noise level of the spectra was used as an experimental uncertainty. (E) Schematic illustration of the structural transitions of the d(CGG/CGG) triad induced by ND binding.

To investigate why three states were detected by NMR but not by UV melting, the effect of DNA concentration was investigated by 1D ¹H NMR measurements with different DNA concentrations (100, 50, 5, and 1 μM DNA, Figure S33A). All NMR spectra were similar, indicating that the difference between NMR (100 μM DNA) and UV melting (5 μM DNA) was unlikely to be due to variation in the DNA concentration. Next, the effect of the relative concentration of ND on the dsDNA was investigated by T_m measurement by varying the [dsDNA]:[ND] ratio (1:4.2, 1:3.6, 1:3.0, 1:2.4, 1:1.8, 1:1.2, 1:0.6, and 1:0, Figure S33B). Two melting points were detected corresponding to free dsDNA (26.5 °C) and the 1:2 dsDNA-ND complex (45.4 °C). These results support the results described above that the T_m of the 1:1 dsDNA-ND complex is similar to that of free dsDNA, which may explain why three states were detected by NMR but not by UV. However, the T_m of the 1:1 complex could be higher than that of free dsDNA, if the relatively faster exchange averages these T_m values.

We then investigated the conformational exchange between each complex state using ³¹P–³¹P EXchange SpectroscopY (EXSY) experiments, which employ the same pulse sequence as NOESY and can track chemical exchanges on millisecond to second time scales. In a ³¹P–³¹P EXSY experiment, NOE is usually not observed between ³¹P atoms because of long ³¹P–³¹P distances (the closest distance in B-form DNA: ∼6.5 Å), allowing the conformational exchange to be easily traced based on the presence of the EXSY cross-peaks. The ³¹P–³¹P EXSY spectrum, measured in the presence of three states (free dsDNA, 1:1, and 1:2 complex), showed cross-peaks between the free dsDNA and the 1:1 complex, but none for the 1:2 complex (Figure 5B). These EXSY peaks were also found in the ¹H–¹H NOESY spectrum (Figures 5C and S34). The exchange rate between the free dsDNA and the 1:1 complex was determined to be k_ex = 1.4 s^–1 by measuring the ¹H–¹H NOESY spectra with different mixing times, where nonlinear fitting based on the McConnell equation was applied to well-resolved diagonal and cross-peaks of the G7 and G18 imino protons (Figure 5D). Similarly, the fitting with the T15 methyl yielded a comparable exchange rate (k_ex = 1.4 s^–1, Figure S35). These results demonstrate that the formation of the 1:1 complex is relatively fast, whereas the formation of the 1:2 complex is much slower than EXSY-detectable time scales. That is, formation of the 1:1 complex has a lower activation barrier and is kinetically more favorable than the 1:2 complex. Considering these results, it is likely that a pre-equilibrium between the free dsDNA and the 1:1 complex exists before the thermally stable 1:2 complex is formed (Figure 5E).

The 1:1 complex was observed under special conditions: low ligand concentration at low temperature. EXSY results indicate that the exchange rate of the 1:1 complex with free dsDNA is much faster than that of the 1:2 complex with free dsDNA. These results suggest that when ND is added to DNA, a relatively faster pre-equilibrium between free DNA and the 1:1 complex is established prior to the formation of the 1:2 complex. Short-lived states are referred to as “transient” states in recent NMR studies,^56,57 and the 1:1 complex is in a short-lived state due to the relatively faster pre-equilibrium as mentioned above. Therefore, we refer to the 1:1 complex as a “transient complex” in this paper.

Design of a New Molecule Stabilizing 1:1 Complex and Destabilizing 1:2 Complex

Previous titration experiments of NA and NCD, which recognize the d(CAG/CAG) and d(CGG/CGG) motifs, respectively, revealed that the 1:2 dsDNA-ligand binding mode was found without the formation of a 1:1 complex.^26,33 In our study of ND, we discovered that CGG-dsDNA can accommodate both 1:1 and 1:2 binding modes. The 1:2 binding mode recognizes the CXG:CXG motif as reported for NA and NCD, whereas the 1:1 binding mode recognizes the G:G mismatch as demonstrated in this report. Therefore, the 1:1 binding mode provides a promising platform for designing a new ligand recognizing X:X mismatches rather than the CXG:CXG motif. Moreover, developing a ligand mimicking the 1:1 dsDNA-ND complex allows for a more detailed study of this complex. To achieve selective formation of the 1:1 complex, either stabilization of the 1:1 complex or destabilization of the 1:2 complex is necessary. The 1:1 and 1:2 complexes form the same number of base pairs (1:1 complex = 10 WC bps +2 NP:G bps, 1:2 complex = 8 WC bps +4 NP:G bps; see Figure 5E), but the interaction between the linker amine and the base is unique to the 1:2 complex (Figures 3E and S32). Therefore, we designed and synthesized a new molecule, sND, with a shorter linker structure, to expose the linker amine to the solvent as observed in the 1:1 dsDNA-ND complex (Figure 6A, the atom numbering is shown in Figure S1). This modification of the linker structure will disrupt the linker–nucleobase interaction, which is a key interaction stabilizing the 1:2 complex. Consequently, sND is expected to specifically destabilize the 1:2 complex structure without affecting 1:1 complex formation.

Figure 6.

Design and analysis of a ligand that selectively forms a 1:1 complex with the d(CGG/CGG) triad. (A) The concept of the molecular design of sND, which aims to stabilize the 1:1 complex by destabilizing the linker–base interaction in the 1:2 complex (top). Synthetic scheme of sND (bottom). The atom numbering of sND is described in Figure S1. (B) Imino proton region of the 1D ¹H NMR titration spectra (500 MHz, 10 °C) of the dsDNA with various concentrations of sND. The ratios of [DNA]:[sND] are shown on the left. NMR signals that appear or disappear upon complex formation are indicated by dotted or solid lines. The broadened peaks marked by stars indicate terminal guanine imino protons. These protons are not assigned due to the lack of NOEs. As shown by the red dotted square, the peaks corresponding to NP2 and NP3, which were observed at 10.0–11.0 ppm in the 1:2 dsDNA-ND complex, are not observed here. Conditions: 0.1 mM dsDNA, indicated molar ratio of sND. (C) Superimposed representations of the 30 lowest-energy NMR structures of the dsDNA-sND complex (PDB ID: 8ZD7). The sND molecule is colored in blue. The DNA strands are colored in light blue and light green. The structures are seen from the side (left) and minor groove (right). The structures are aligned at the ligand-bound site (C5, G6, G7, C16, G17, G18, and sND residues). (D) RDC values (¹D_CH) of the dsDNA-sND complex. The C2–H2 (blue), C5–H5 (orange), C6–H6 (gray), and C8–H8 (yellow) RDCs are shown as a function of the secondary structure. The spectral resolution value was used as the experimental uncertainty (error bars). (E) 1D ³¹P spectrum (600 MHz spectrometer, 30 °C) of the dsDNA-sND complex. Conditions: 2.0 mM dsDNA, 2.0 mM sND. Trimethyl phosphate (TMP) was used as an external standard for the phosphorus chemical shift (δ = 0 ppm). The intense peak marked by Pi corresponds to the phosphate buffer.

The binding of sND to CGG-dsDNA was investigated via NMR titration experiments (Figure 6B). The signal change saturated upon the addition of 1 equiv of sND, and no further changes were observed with additional sND. This demonstrates that sND selectively forms a 1:1 complex with CGG-dsDNA, as designed. To determine the 3D structure of the complex, a series of 2D NMR spectra (¹H–¹H NOESY, ¹H–¹H DQF-COSY, ¹H–¹H TOCSY, ¹H–¹³C HSQC, ¹H–¹⁵N HSQC, and ¹H–³¹P CPMG-HSQC) was measured (Figures S36–S40). The superimposition of the 30 RDC-refined structures is shown in Figures 6C and S41. The structures converged well with an all-atom RMSD of 0.70 ± 0.28 Å. 40 intermolecular NOEs between DNA and sND were observed, supporting structural validity (Table S3 and Figure S42). As expected, the dsDNA-sND complex structure is similar to the 1:1 dsDNA-ND complex; that is, the WC G:C base pairs adjacent to the G:G mismatch are maintained, and the G:G mismatched bases form an H-bond with naphthyridine moiety. All of the RDC values of the base ¹H–¹³C pairs (¹D_CH) for the dsDNA-sND complex were comparable throughout the strands, suggesting that all of the bases were well stacked without flipping out (Figure 6D). The structural similarity between the dsDNA-sND complex and the 1:1 dsDNA:ND complex is also supported by the similarity of the chemical shift values of H8/6 and H1′ protons (Figure S43). It should be noted that the small and simple modification of the chemical structure of the ligand has induced drastic changes in the binding mode and three-dimensional structure of the complex.

In the dsDNA-sND complex, the ³¹P signals of G7 and G18, where the aromatic ring of the ligand is inserted, exhibited remarkable downfield shifts, similar to those observed in the 1:1 dsDNA-ND complex (Figures 6E and S40A). The ³¹P signals of G6 and G17 showed a slight (∼−0.5 ppm) downfield shift and the cross-peaks with H2′/2″ signals were observed in 2D ¹H–³¹P CPMG-HSQC spectrum (Figure S03B). According to the previous studies by Bloomers, the cross-peak of ³¹P with H2′/2″ indicates that the dihedral angle of ε (C4′–C3′–O3′–P) adopts a gauche(−) conformation.⁵⁸ Thus, these dihedral angle constraints were used in the structure calculations of the dsDNA-sND complex. The downfield shift of the ³¹P signals around the ligand-bound site including G6, G7, G17, and G18 is consistent with the unwinding structure of the phosphate backbone (Figure S44). The significant upfield shifts of the H2′/2″ signals of C5 and C16 are likely due to the ring current effect from the guanine on the 3′ side, which is induced by changes in the helical twist accompanying the unwinding of the phosphate backbone (Figure S45). These upfield shifts were also observed in the 1:1 dsDNA-ND complex, where the unwinding structure of the phosphate backbone was observed.

T_m measurements of CGG-dsDNA were performed to estimate the thermal stability of the dsDNA-sND complex (Figure S46A). sND increased the T_m of CGG-dsDNA by 9 °C. In the 1D ¹H spectrum of the dsDNA-sND complex, the complex is maintained even at 40 °C, similar to the 1:2 dsDNA-ND complex, unlike the 1:1 dsDNA-ND complex and free dsDNA (Figure S46B). ND increased the T_m by 17 °C forming the 1:2 dsDNA-ND complex. These results indicate that the thermal stability is in the following order: 1:2 dsDNA-ND complex > dsDNA-sND complex >1:1 dsDNA-ND complex ∼ free dsDNA. Our data demonstrated that relying solely on T_m measurements in binding assays may overlook ligands with novel binding modes. The importance of detailed NMR studies for the discovery of nucleic acid-targeting molecules is emphasized here.

Discussion

Most of the reported 3D structural analyses of nucleic acid-small molecule complexes have focused on base pairs and their stacking, paying little attention to the phosphate backbone. However, for the dsDNA-ND complex, the 1:1 and 1:2 complexes have the same number of base pairs, indicating that base pair H-bonding and stacking cannot explain their difference in thermal stability. ³¹P NMR is known to be sensitive to the structural changes of nucleic acids, particularly in the phosphate backbone, but there are a limited number of examples in which the ³¹P signal has been assigned in the NMR analysis of nucleic acids. Here, we applied ³¹P NMR to evaluate the distortion in the phosphate backbone structure as a potential driving force for the structural transition from the 1:1 to the 1:2 complex. The flip-out of cytosine in the 1:2 complex was thought to be entropically unfavorable from the viewpoint of hydrophobic interactions, but our ³¹P NMR analysis suggested that these flip-outs contributed to releasing the distortion of the phosphate backbone structure caused by ligand binding. As a result, the 1:2 complex settles in a thermally stable canonical B-form with support of the linker–base interactions. A comprehensive study of the ³¹P NMR will provide further structural constraints on the phosphate backbone, and ³¹P chemical shifts will be a new benchmark to evaluate the structural and thermodynamic properties of the complexes of nucleic acids.

In the 3D structural analysis of complexes of nucleic acids and ligands by NMR, stable isotope (SI) labeling has been limited to nucleic acids and the SI labeling of ligands has rarely been performed. Most nucleic acid-binding molecules, with the exception of intercalators, recognize targets via hydrogen bonding. SI-label and signal detection of the hydrogen bonding donors of biomolecules or ligands are effective strategies for detecting hydrogen bonds by NMR.^59,60 In the studies of the formation of hydrogen bonds between ligands and nucleic acids, there have been few examples of the use of SI labels on the ligand side, and there have also been no successful cases of the detection of NMR signals from the amino groups of ligands. Based on the obtained 3D structure of the complex, we hypothesized that the amino group of the ligand is important for complex formation. We then prepared a ¹⁵N-labeled ligand amino group, which was predicted to act as a hydrogen bond donor based on NMR structure determination, and succeeded in detecting the signal from the ligand amino group. This amino group is indeed important for complex formation, leading to the design of new ligands with a novel binding motif. As in our case, SI-labeled ligands prepared by organic synthesis are effective for the experimental verification of hydrogen bond formation between ligands and nucleic acids. In addition, it allows for the detection of interactions of ligands with the phosphate backbone, which has been difficult to verify by standard NMR techniques. SI-label of ligands is a powerful technique for visualizing nucleic acid–ligand interactions.

The importance of minor conformations in nucleic acids has long been recognized, and Al-Hashimi et al. experimentally demonstrated that minor conformations, such as Hoogsteen base pairs^61,62 and HIV-1 TAR RNA,^63,64 significantly influence their functions. Minor conformations have also been confirmed in the DNA-ligand systems.⁶⁵ These conformations have been successfully controlled by base mutations to modulate their population. We focused on the chemical structure of the ligand rather than the nucleic acids and demonstrated that the population of the minor conformations can be controlled by simple organic synthetic modification of the ligand. The conventional principle of molecular design has been to promote the stabilization of complexes, but we succeeded in developing a new molecule with unexplored binding modes by intentionally removing the stabilizing factors of the ligand. In our case, allosteric stabilization by the linker–nucleobase interaction was key to destabilizing the most stable complex. By removing the allosteric stabilization factor of the ligand based on NMR analysis, the conformational population can be altered, leading to the development of novel molecules with a different binding mode to the target. We propose this strategy as a new method to design new nucleic acid-targeting molecules.

Conclusions

In this report, sND, the new molecule with an unexplored binding mode, is rationally designed through NMR-based drug design focusing on minor conformations, that is, the transient complex structure. By designing sND, we successfully converted the recognition sequence from the CGG:CGG to the G:G mismatch. In principle, the utilization of both NCD (CGG:CGG binding) and sND (G:G binding) enables the specific recognition of any type of G:G mismatch DNA. Moreover, the use of NCD and sND with other mismatch-binding ligands, such as NA (A:G and CAG:CAG binding),^26,35AmND (C:C binding),^66,67 and ANP77 (CC:X binding),⁶⁸ expands the recognition sequence of mismatched DNA. Our strategy for designing mismatch-binding ligands offers a new avenue for recognizing any type of mismatch DNA by small molecules, leading to the development of groundbreaking therapeutic drugs capable of improving the symptoms of repeat expansion diseases and preventing their onset.

Materials and Methods

Chemical Synthesis of the Compounds

Reagents and solvents were purchased from standard suppliers without further purification. Non-isotope labeled ND was synthesized according to the previously reported protocol.⁶⁹ For the synthesis of ¹⁵N-labeled ND, ¹⁵NH₄Cl and ¹⁵N-phthalimide were purchased from Sigma-Aldrich, and Shoko Science Co., Ltd. (Yokohama, Japan), respectively. The detailed synthetic protocols are described in the Supporting Information.

Melting Temperature (T_m) Measurements

T_m profiles of the sodium phosphate buffer (10 mM, pH 7.0) solutions of CGG-dsDNA (5′-(CTAA CGG AATG)-3′/5′-(CATT CGG TTAG)-3′: 5 μM) containing NaCl (100 mM), in the absence or presence of ND (20 μM) or sND (10 μM), were recorded by a UV–Vis spectrometer (UV-2550, SHIMADZU) equipped with a temperature controller (TMSPC-8, SHIMADZU). At least triplicate experiments were performed under the same conditions to obtain the mean value and standard deviation. For concentration-dependent measurements, the T_m measurements were conducted using CGG-dsDNA (5 μM) with varying ND concentrations (21, 18, 15, 12, 9, 6, 3, and 0 μM).

Isothermal Titration Calorimetry (ITC)

ITC measurements were performed on a calorimeter (MicroCal iTC200, Malvern Analytical), and data analysis was carried out with Origin 7.0 software. A 2000 μL solution of CGG-dsDNA (2.5 μM) in sodium phosphate buffer (pH 7.0, 10 mM) containing NaCl (100 mM) was prepared, and then a part of the solution was put in the cell of the calorimeter. The temperature inside the cell was kept at 25 °C during the experiment. Next, a 200 μL sodium phosphate buffer (pH 7, 100 mM) solution of ND (50 μM) containing NaCl (100 mM) was prepared as a titrant, and then a portion of the solution was put in the 40 μL syringe of the calorimeter. The concentration of ND used for titrant was calibrated by the qNMR technique with a DSS-d6 standard solution (FUJIFILM Wako). The titration experiment was performed by the 19 injections of the titrant (0.4 μL for the first injection and 2 μL for the other injections) from the syringe into the sample cell at 25 °C (stirred at 750 rpm). The duration for each injection was 4 s, and the interval between the two nearest injections was 150 s. In addition, the titrant was injected into the cell containing a buffer solution without dsDNA in the same manner to obtain the data for the heat of dilution of the titrant. Prior to fitting analysis, the data point corresponding to the first injection was removed, and the heat data were corrected by subtracting the heat of dilution. Binding isotherms were fitted to a one-site binding model by least-squares analysis. At least triplicate experiments were performed under the same conditions to obtain the mean value and standard deviation.

Surface Plasmon Resonance (SPR) Assay

A streptavidin-coated SA sensor chip (GE Healthcare) was washed with HBS-EP+ buffer (HEPES (10 mM, pH 7.4), NaCl (150 mM), EDTA (3 mM), and the surfactant P20 (0.005%, v/v)) for 6 min and then activated with three consecutive 1 min injections of an activation mixture (30 mL; NaOH (50 mM) and NaCl (1 M)). 5′-Biotinylated oligonucleotides (5′-biotin-CTAACGGAATG TTTT CATTCGGTTAG-3′) were adjusted to be (1 μM) in HBS-EP+ buffer and allowed to flow onto the SA chip (5 μL min^–1, 60 s) for immobilization. The binding of ND to the immobilized DNA was analyzed using a BIAcore T200 SPR system (GE Healthcare). ND (0.13, 0.25, 0.50, 1.0, or 2.0 μM in HBS-EP+ buffer) was injected over the flow cells on the sensor surface (60 μL min^–1, 120 s), followed by injection of running buffer (80 s) as a dissociation phase. Apparent binding kinetics and affinity were determined from the SPR curve applying a single-site binding model.

Sample Preparation for NMR Experiments

The chemically synthesized DNA oligomers 5′-d(CTAA CGG AATG)-3′ and 5′-d(CATT CGG TTAG)-3′ were purchased from Gene Design Inc. (Ibaraki, Japan). The 10%-enrichment site-specific ¹³C,¹⁵N-labeled oligonucleotides were synthesized by Gene Design using ¹³C,¹⁵N-labeled dG-phosphoramidite purchased from Cambridge Isotope Laboratories Inc. (Tewksbury, USA). The dsDNA solution was ultrafiltrated with VivaSpin 2 (MWCO 5000) three times against 20 mM sodium phosphate buffer (pH 6.8) containing 100 mM NaCl to remove ionic impurities. The ligand (ND and sND) was dissolved in H₂O to 10 mM. The conditions for NMR measurements are noted in the figure captions and summarized in Tables S4 and S5. For the sample used for structural determination, the solutions were further ultrafiltrated after the ligand was added to remove the excess amount of ligand and impurities. All NMR measurements were performed in the same buffer conditions (20 mM sodium phosphate buffer (pH 6.8), 100 mM NaCl, and 5% D₂O), except for measurements of the ligand alone (10 mM sodium phosphate buffer (pH 6.8), 50 mM NaCl, and 5% D₂O). ¹H chemical shift was directly referenced with DSS-d6 as an external standard, and ¹³C and ¹⁵N chemical shifts were indirectly referenced with DSS. ³¹P chemical shift was referenced with trimethyl phosphate (TMP) as an external standard (δ = 0) according to the work by Gorenstein.⁴⁸ For NMR measurements of DNA-ligand complex, all experiments were started more than 10 min after ligand addition at room temperature.

NMR Measurements of DNA-Ligand Complex

The NMR experiments were performed on AVANCE III HD 500 MHz (BBO CryoProbe), AVANCE III HD 600 MHz (QCI-P CryoProbe), AVANCE III HD 800 MHz (TXI CryoProbe), and AVANCE III 950 MHz (TCI CryoProbe). For water suppression, jump-and-return with 1–1 echo,^70,71 3–9–19,⁷² or excitation sculpting⁷³ pulses were used. The NMR spectra were processed with Topspin 3.6.4 (Bruker Biospin) and analyzed by NMRFAM-Sparky⁷⁴ (UCSF). NMR titration experiments were performed with 1D ¹H spectra with water suppression of jump-and-return with 1–1 echo pulses at 10 °C.

For the assignment of free CGG-dsDNA, ¹H–¹H NOESY (Nuclear Overhauser Enhancement/Effect SpectroscopY, mixing time: 200 ms), ¹H–¹H TOCSY (TOtal Correlation SpectroscopY, mixing time: 200 ms), natural abundance ¹H–¹³C HSQC (Heteronuclear Single Quantum Coherence), and ¹H–³¹P CPMG (Carr–Purcell–Meiboom–Gill)-HSQC⁷⁵ spectra were measured at 10 or 20 °C.

For the assignment of the 1:2 dsDNA-ND complex, ¹H–¹H NOESY (mixing time: 30 and 200 ms), ¹H–¹H TOCSY (mixing time: 200 ms), ¹H–¹H DQF-COSY, natural abundance ¹H–¹³C HSQC, natural abundance ¹H–¹⁵N HSQC, ¹H–¹⁵N HSQC (with ¹⁵N₁–ND or ¹⁵N₄–ND), and ¹H–³¹P CPMG-HSQC spectra were measured at between 10 and 40 °C. H2′ and H2″ protons were distinguished based on zero quantum-filtered ¹H–¹H NOESY⁷⁶ spectra with 30 ms of mixing and ¹H–¹H DQF-COSY (Double Quantum Filtered COrrelation SpectroscopY) spectra. Aromatic atoms of ND were assigned by using ¹H–¹H NOESY, ¹H–¹H TOCSY, and natural abundance ¹H–¹³C HSQC spectra. Aliphatic atoms of ND (linker moiety) were assigned using ¹H–¹H NOESY and natural abundance ¹H–¹³C HSQC spectra based on the correlation with the amide protons of ND. Distance constraints were produced from the ¹H–¹H NOESY spectra measured at 40 °C with 7.3 s of repetition delay. RDC was obtained from natural abundance ¹H–¹³C HSQC spectra without ¹³C decoupling during acquisition.

For the assignment of the 1:1 dsDNA-ND complex, ¹H–¹H NOESY (mixing time: 200 ms), ¹H–¹H TOCSY (mixing time: 200 ms), ¹H–¹⁵N HSQC (with site-specifically ¹³C,¹⁵N-labeled DNA or ¹⁵N₄–ND), ¹H–³¹P CPMG-HSQC, and ³¹P–³¹P EXSY (EXchange SpectroscopY) spectra were measured at 10 or 20 °C. Distance constraints were produced from the ¹H–¹H NOESY spectra measured at 40 °C with 2.3 s of repetition delay. Aromatic atoms of ND were assigned by using ¹H–¹H NOESY and ¹H–¹H TOCSY spectra.

For the assignment of the dsDNA-sND complex, ¹H–¹H NOESY (mixing time: 30 and 200 ms), ¹H–¹H TOCSY (mixing time: 200 ms), ¹H–¹H DQF-COSY, natural abundance ¹H–¹³C HSQC, natural abundance ¹H–¹⁵N HSQC, and ¹H–³¹P CPMG-HSQC spectra were measured at 30 °C. H2′ and H2″ protons were distinguished based on zero quantum-filtered ¹H–¹H NOESY spectra with 30 ms of mixing and ¹H–¹H DQF-COSY spectra. Aromatic atoms of sND were assigned by using ¹H–¹H NOESY, ¹H–¹H TOCSY, and natural abundance ¹H–¹³C HSQC spectra. Aliphatic atoms of sND (linker moiety) were assigned using ¹H–¹H NOESY and natural abundance ¹H–¹³C HSQC spectra based on the correlation with amide protons of sND. Distance constraints were produced from the ¹H–¹H NOESY spectra measured at 30 °C with 7.3 s of repetition delay. RDC was obtained from natural abundance ¹H–¹³C HSQC spectra without a ¹³C decoupling pulse during acquisition.

For the assignment of ³¹P signals in the 1:2 dsDNA-ND complex and dsDNA-sND complex, conventional methods, which use the correlation between H3′(n)–³¹P(n–1) and H4′(n)–³¹P(n) in ¹H–³¹P CPMG-HSQC spectra,⁴⁴ were employed (n: residue number). For free dsDNA and the 1:1 dsDNA-sND complex, since H3′ signals at ligand-binding sites were missing due to exchange broadening, ³¹P–³¹P EXSY was used for assignment (for further details, see the Supplementary Discussion and Figure S47).

NOE-Based Structure Determination

NOE distance restraints were obtained from a ¹H–¹H NOESY spectrum with 200 ms of mixing time. Cross-peaks in the spectrum were integrated using NMRFAM-SPARKY. Interproton distance bounds were determined from the integrated peak intensities by the program MARDIGRAS.⁷⁷ Using the RANDMARDI (random error MARDIGRAS⁷⁸) procedure of the complete relaxation matrix analysis method, the effect of experimental noise was estimated. For some distance restraints containing exchangeable protons, the maximum limit is manually reduced. Based on ¹H–¹H DQF-COSY, ¹H–¹H NOESY, and phosphorus spectra, sugar puckers and backbone torsion angles were restrained to maintain an S-type sugar conformation and right-handed helix, respectively. In the ¹H–³¹P HSQC spectrum of the dsDNA-sND complex, a gauche (−) constraint was imposed on the dihedral angle ε for residues where coupling between H2′/H2″ and ³¹P was observed.⁵⁸ Based on the MicroED structure, amide moieties (H–N–C–O) of the ligand were constrained to trans conformation and the amino group was prepared as a protonated state. H-bonding restraints were imposed on the WC base pairs and the guanine-naphthyridine H-bonding pairs. Simulated annealing protocols were performed using Crystallography and NMR System (CNS) version 1.3.⁷⁹ The structures without a distance violation greater than 0.5 Å were selected as accepted structures.

For the 1:2 dsDNA-ND complex and the dsDNA-sND complex, the peak volume error was set to 100%. For the 1:1 dsDNA-ND complex, it was set at 200% due to chemical exchange. In the presence of chemical exchange, NOE peak intensities are underestimated, resulting in overestimated distances.⁸⁰ For the 1:1 complex, an underestimation of approximately 15% of the NOE peak intensity was expected from the EXSY analysis, and this underestimation was factored into the 200% peak volume error. To distinguish between the NOE and chemical exchange peaks in the NOESY spectrum, the free dsDNA and the 1:2 complex were analyzed in the absence of the 1:1 complex prior to analysis of the 1:1 complex. The cross-peaks between the free dsDNA and the 1:1 complex in the NOESY spectrum are assumed to be due to chemical exchange; therefore, the remaining peaks were assigned to NOE peaks.

Refinement of NOE-Based Structures with Residual Dipolar Couplings (RDC) Restraints

The refinement of RDC was applied to the 1:2 complex of ND and the dsDNA-sND complex. RDCs were recorded using liquid crystalline Pf1 phage as an alignment medium (ASLA Biotech, Latvia). 10 and 12 mg/mL Pf1 phages were added to the 1:2 dsDNA-ND complex and dsDNA-sND complex, respectively. The buffer of the DNA solution is exchanged by ultrafiltration with VivaSpin 2 (MWCO 5000). The alignment was confirmed by quadrupolar splits of ²H NMR signals (dsDNA-ND: 10.6 Hz at 950 MHz, dsDNA-sND: 10.2 Hz at 800 MHz). Using these samples, natural abundance ¹³C-coupled ¹H-¹³C HSQC spectra were measured. The spectra were analyzed by NMRFAM-Sparky, and aromatic ¹D_CH values were obtained. Rhombicity and axial values are estimated by the calcTensor program in Xplor-NIH 3.1.⁸¹ RDC refinements were performed by adding constraint conditions, including RDC, to the 30 accepted structures obtained in the previous structure calculation without RDC. Flip-out cytosines and terminal residues considered to have high mobility were excluded from the RDC restraints. Five structures were calculated for each initial structure, and those that did not violate NOEs (greater than 0.5 Å) or dihedral constraints and had a correlation factor between the experimental and calculated RDC values of 0.95 or higher were adopted. Maestro (Schrödinger, Inc.) and UCSF Chimera 1.16⁸² were used for structure visualization. The structure parameters were analyzed by w3DNA.⁸³ Beeswarm plots were created by using matplotlib and seaborn libraries on Python 3.0.

EXSY Spectra and Kinetic Analysis of the Complex

The ³¹P–³¹P EXSY spectra of the free dsDNA, dsDNA-ND, and dsDNA-sND were measured by using the same pulse sequence as the NOESY spectrum with ¹H-decoupling during acquisition. The spectra were measured at 20 °C with mixing times of 200 ms.

For the analysis of the rate constant of the binding of ND, a series of ¹H–¹H NOESY spectra with 1–1 echo water suppression with mixing times of 20, 40, 60, 80, 100, 125, 150, 175, 200, 250, and 300 ms was used. The intensities of the cross and diagonal EXSY peaks were estimated by NMRFAM-Sparky. A two-state model with a pseudo-first-order kinetic model was used to determine rate constants. Only the exchange between the free DNA and the 1:1 complex was considered and the 1:2 complex was ignored because the 1:2 complex did not show EXSY peaks. R₁ values of the 1:1 complex and free dsDNA were assumed to be equal, and the exchange rate was assumed to be common for each residue (G6 and G17). Upon these hypotheses, the intensities of cross and diagonal peaks are governed by the below equations:

where p_A and p_B are the relative populations for state A (free dsDNA) and state B (1:1 complex), k_ex is the sum of the forward, k_on[ND], and reverse, k_off, kinetic rate constants (k_ex = k_on[ND] + k_off) for the interconversion between the two conformations, and R₁ is the longitudinal relaxation rate of the imino protons. The population of each complex (p_A and p_B) was obtained from the deconvolution of the 1D spectra by using TopSpin 3.6.3. In the fitting, the sum of the intensity of the EXSY cross peak (I_BA+I_AB) and each diagonal peak intensity (I_AA, and I_BB) were used, and k_ex and R₁ were used as variants. Experimental data were fitted with SOLVER program in Microsoft Excel to minimize the factor below:

where the D_exp and D_fit are experimental and fitted values, respectively.

MicroED Data Collection

Crystals suitable for MicroED analysis were obtained by vapor diffusion of Et₂O into solutions of HCl salt of ND in MeOH. Nanocrystals of ND were loaded on a Quantifoil grid (Cu R 1.2/1.3) and measured on Talos Arctica (Thermo Fisher Scientific) operating at 200 kV. The virtual camera distance was nominally 670 mm, calibrated to be ∼618 mm. Automatic data collection was performed with SerialEM⁸⁴ and Verlox in a way similar to those reported.⁸⁵⁻⁸⁸ Diffraction patterns were recorded on a Ceta camera at 1 frame/s, while the sample was continuously rotated at 0.953°/s with an electron flux of ∼0.05 electron/Ų/s. More than 750 crystals were measured. The crystals were thin plates (Figure S48A).

MicroED Data Processing

Diffraction patterns were indexed and integrated with DIALS.^89,90 Data processing was parallelized with GNU Parallel.⁹¹ Out of 616 successfully indexed crystals (Figure S48B), 245 good crystals were selected, scaled, and merged by xia2.multiplex⁹² and dials.scale,⁹³ in which the indexing ambiguity (real space operator: x,-y,-z) was resolved by dials.cosym.⁹⁴ Although some crystals diffracted up to 0.7 Å (Figure S48C), we had to include many weakly diffracting crystals to improve the completeness due to the preferred orientation of the crystals on the grid and the anisotropy in the diffracting power. Merging statistics are listed in Table S6. The data set was phased by SHELXT⁹⁵ and kinematically refined by SHELXL⁹⁶ with the help of the Olex2 GUI.⁹⁷ Refinement statistics are summarized in Table S7.

Data Availability Statement

The coordinates for structures of the 1:2 dsDNA-ND complex (accession code: 8ZD2), 1:1 dsDNA-ND complex (accession code: 8ZD8), and dsDNA-sND complex (accession code: 8ZD7) have been deposited in the Protein Data Bank. The chemical shifts of the 1:2 dsDNA-ND complex (accession code: 36662), 1:1 dsDNA-ND complex (accession code: 36664), and dsDNA-sND complex (accession code: 36663) have been deposited in the Biological Magnetic Resonance Bank. The refined crystal structure of ND was deposited to CCDC (accession code: 2350096) and COD (accession code: 3000497). The raw MicroED images of ND were deposited to XRDa (accession code: XRD-335).

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/jacs.4c17538.

• Figures S1–S48; Tables S1–S7; supplementary discussion; chemical synthesis of the compounds; NMR spectra of the compounds (PDF)

The authors declare no competing financial interest.