Structural basis of DNA duplex distortion induced by thiazole-containing hairpin polyamides.

This manuscript reports the molecular basis for double-stranded DNA (dsDNA) binding of hairpin polyamides incorporating a 5-alkyl thiazole (Nt) unit. Hairpin polyamides containing an N-terminal Nt unit induce higher melting stabilisation of target dsDNA sequences relative to an archetypical hairpin polyamide incorporating an N-terminal imidazole (Im) unit. However, modification of the N-terminus from Im to Nt-building blocks results in an increase in dsDNA binding affinity but lower G-selectivity. A general G-selectivity trend is observed for Nt-containing polyamide analogues. G-selectivity increases as the steric bulk in the Nt 5-position increases. Solution-based NMR structural studies reveal differences in the modulation of the target DNA duplex of Nt-containing hairpin polyamides relative to the Im-containing archetype. A structural hallmark of an Nt polyamide•dsDNA complex is a more significant degree of major groove compression of the target dsDNA sequence relative to the Im-containing hairpin polyamide.

INTRODUCTION

Sequence-selective recognition of double-stranded, B-type DNA (dsDNA) by small to medium size molecules holds considerable promise as a general strategy for the exogenous modulation of gene expression (1–12). Hairpin n class="Chemical">pyrrole-imidazole polyamides (PIPs) are particularly powerful exemplars of this approach as their sequence-selective dsDNA binding is inherently programmable (13–17). Using a generalized set of pairing rules (Figure 1A), (13) the primary sequence of pyrrole (Py) and imidazole (Im) amino acid heterocycles defines the sequence selectivity of the target DNA duplex up to 24 base-pairs in length (18–33). One of the most important recognition profiles in hairpin PIP designs is the G-selective hydrogen-bonding recognition signature of the endocyclic N3 of the Im building block (ImN3) with the exocyclic N2 amine of G (34–36). Repulsive interactions between ImN3 and the electron-rich C, T and A nucleotides provide an auxiliary level of sequence discrimination (11). One distinct limitation of current hairpin PIPs for targeting G-rich sequences is their cell permeability, which generally decreases with increasing Im content (37,38). Although a variety of strategies exist to mitigate the enhanced polarity of Im-rich PIPs—such as modification of the C-terminus and the turn motifs (39–41)—these approaches tend to have an unpredictable impact on their overall physicochemical properties and toxicity (39,42,43).

Figure 1.

(A) General binding mode of hairpin polyamides. (B) Structures of polyamides used in this study.

(A) General binding mode of hairpin pan class="Chemical">polyamides. (B) Structures of pan class="Chemical">polyamides used in this study.

Expanding the heterocyclic repertoire of current hairpin n class="Chemical">polyamide designs offers a powerful alternative for broadening the dsDNA binding lexicon currently available and provides a means to optimize cell uptake by tuning their physicochemical parameters (44–51). Thiazole-4-carboxylic acids (Nt) (46) are attractive analogues of Im (52) building blocks in this regard as they contain endocyclic heteroatoms (S1 and N3), which are potential hydrogen-bond acceptors for the exocyclic N2 amine of G (50,51,53–55). Moreover, facile synthetic routes are available to prepare Nt analogues containing alkyl substituents in the 5-position, thereby allowing the hydrophobicity of these building blocks to be tuned accordingly (56–58). Recent high-resolution X-ray crystal structures of cyclic 8-ring PIP•dsDNA complexes have revealed significant widening of the minor groove, and concomitantly, major groove compression, which results in an overall bending of the DNA duplex (17,59) relative to that observed when a hairpin polyamide binds to nucleosomal dsDNA (18,60). To date, solution-based structural characterizations of hairpin PIPs have been confined to six-ring hairpin and cyclic PIP•dsDNA complexes (14,15,61,62).

In this manuscript, we expand the heterocyclic repertoire of hairpin n class="Chemical">PIPs by replacing the N-terminal Im unit of PA1 with Nt-building blocks (PA2–4, Figure 1B) (63). We show that Nt-building blocks incorporated in PA2–4 differ in their relative binding preferences for a target G nucleotide in the minor groove with G-selectivity increasing with increasing steric bulk in the 5-position. Finally, NMR-derived models of PA3•dsDNA and PA4•dsDNA complexes show that Nt-containing polyamides (PA3–4) bind to dsDNA in a hairpin conformation and induce a more pronounced degree of major groove compression relative to an Im-containing PA1•dsDNA complex.

MATERIALS AND METHODS

Synthesis of polyamides PA1-4

n class="Gene">PA1–4 were prepared by manual solid phase synthesis using Boc-β-Ala-PAM resin (0.5 mmol/eq, Sigma-Aldrich). HATU, HOAt and TFA were purchased from Fluorochem. DCC and Dp were purchased from Alfa Aesar.

UV melting analysis

Melting temperature analyses were performed on a Shimadzu UV spectrometer UV 1800 equipped with a thermo-controlled cell holder possessing a cell path length of 1 cm. An n class="Chemical">aqueous solution consisting of 10 mM monobasic phosphate, 10 mM dibasic phosphate, 100 mM NaCl, at pH 7.0 was used as the analysis buffer. The corresponding duplex DNA and the PA ligands were mixed at a stoichiometry of 1:1.75 providing a final dsDNA concentration of 1 μM for each experiment. Denaturating profiles were recorded at λ = 260 nm from 20 to 90°C, with a heating rate of 1°C/min. The reported melting temperatures were defined as averages of the maximum of first derivative plots of three independent denaturing profiles.

DNase I Footprinting

DNA fragments for footprinting were prepared by ligating n class="Species">synthetic oligonucleotides, which had been designed to contain six heptanucleotide potential ligand binding sites, each separated by GCGC, into the BamHI site of pUC19 (64,65). Successful clones were sequenced by Eurofins MWG Operon, Ebersberg, Germany. The sequence of the inserted fragment is shown in Supplementary Figure S1.

Radiolabelled DNA fragments were obtained by cutting the plasmid with HindIII and SacI and labelling the 3′-end of the HindII site with α-32P[dATP] using exo-Klenow fragment of DNA polymerase I (New England Biolabs). The labelled DNA was separated from the remainder of the plasmid on a 6% pan class="Chemical">polyacrylamide gel, eluted and dissolved in 10 mM Tris–HCl pH 7.5, containing 0.1 mM pan class="Chemical">EDTA at a concentration of at least 10 cps/μl as determined on a hand-held Geiger counter (∼10 nM DNA).

DNase I footprinting was performed by mixing 1.5 μl radiolabelled DNA with 1.5 μl ligand (diluted in 10 mM Tris–HCl containing 10 mM n class="Chemical">NaCl, from a 10 mM stock solution in DMSO). Samples were equilibrated for at least 30 min before adding 2 μl DNase I (typically 0.01 units/ml) which had been diluted in 20 mM NaCl, 2 mM MgCl2, 2 mM MnCl2. The reaction was stopped after 1 min by adding 4.5 μl 80% formamide containing 10 mM EDTA, 1 mM NaOH and bromophenol blue. Samples were denatured by heating at 100°C for 3 min before cooling on ice and loading onto a 40 cm, 8% denaturing polyacrylamide gel containing 8 M urea, which was run at 1500 V for ∼90 min. The gel was then fixed in 10% (v/v) acetic acid, dried onto 3MM filter paper and exposed overnight to a phosphor screen. The screen was scanned using a Typhoon phosphorimager.

NMR sample preparation

Two separated batches of HPLC purified self-complementary d(CGATGTACATCG)2 (ODN1) were purchased from Eurogentec. Each sample was dissolved in 1.2 ml of n class="Chemical">H2O buffered with 100 mM phosphate at pH 7.4 and containing 0.1 mM deuterated trimethylsilylpropionic acid sodium salt-d4, divided in two aliquots (600 μl each) and freeze-dried to dryness. The samples were re-dissolved in 600 μl of 90% H2O with 10% D2O and in 99% D2O. The solutions in two separate 5 mm high-precision NMR tubes were used for ODN1 data acquisition. After acquisition, the two 99% D2O samples were freeze-dried and re-dissolved in 10% D2O/90% H2O. A concentrated solution of the PA1–3-4 (16 mM in up to 10% d6-DMSO in water) was added in small aliquots. After each addition of polyamide to the buffered solution containing ODN1, 1D 1H NMR spectra were acquired until the end point of the titration was reached as determined by inspection of the imino proton resonance region of the data. After acquisition of the full set of data in 10% D2O/90% H2O, the samples were freeze-dried and re-dissolved in D2O to complete the acquisition of all data sets.

NMR data acquisition

NMR data were typically acquired on a Bruker AVANCE-II+ 600 NMR spectrometer operating at a magnetic field strength of 14.1 T (600.13 MHz for proton resonance) under TopSpin (version 3.5 patch level 5, Bruker, Reinstetten, Germany) running on a Hewlett n class="Chemical">Packard Z420 workstation under Windows Professional version 7 (Microsoft Inc.). A triple resonance probehead [TBI-z] equipped with actively shielded z-gradient coil for delivery of pulsed field gradients was manually tuned for a [1H, 13C, 31P] frequency configuration on each sample studied to allow uninterrupted data acquisition to progress using all channels in their turn. The probe temperature was maintained at 298 K in all instances.

One-dimensional (1D) n class="Chemical">1H NMR data were acquired using either presaturation (pulse program zggppr) or double pulsed field gradient spin-echo (pulse program zgesgp) to eliminate the residual solvent resonance. Data were typically acquired for 90% H2O samples with between 64 and 256 transients into 32K data points (acquisition time: 1.09 s) over a frequency width equivalent to 20.0276 ppm centred at δ1H = 4.692 ppm. Hard proton 90° pulses were typically calibrated at 9.6 W with an average value of 9.7 μs across the samples studied. Data acquired for D2O-dissolved samples were acquired over a frequency width equivalent to 12.0166 ppm (acquisition time: 2.27 s).

One-dimensional (1D) n class="Chemical">31P–{1H} (66) NMR spectra were acquired over a frequency width of 3 ppm into 512 data points (acquisition time: 0.35 s) centred at δ31P = –1.185 ppm with the decoupler offset set to δ1H = 4.00 ppm. Between 512 and 4096 transients were acquired with a relaxation delay of 0.5 s.

Proton inversion recovery profiles were measured on one sample to gauge the requirement for relaxation delays during the acquisition of 2D n class="Chemical">NOESY data sets used in structure calculations. The largest T1 values were measured as 5.65 s (for adenine H2 protons). Relaxation delays were typically set to 2.0 s for most data acquisitions with the exception of NOESY data sets whose relaxation delays were typically set to 7 s to allow at least one T1 period of magnetization recovery without excessively extending 2D NOESY data acquisitions.

Two-dimensional (2D) NMR data sets were acquired as follows for samples dissolved in n class="Chemical">D2O. Proton-only data were acquired over frequency widths equivalent to 12 ppm (7194 Hz) in both F2 and F1 centred at an offset of δ1H = 4.702 ppm. 2D DQF-COSY data (pulse program: cosydfphpr) were acquired with 16 transients for each of 1024 States-TPPI t1 increments for acquisition times (aq) of 285 ms (ω2) and aqmax = 71 ms (ω1) and a total data accumulation time of 10 hrs. 2D TOCSY data (pulse program: dipsi2phpr) were acquired into 2048 data points (ω2 acquisition time = 142 ms) with eight transients for each of 512 States-TPPI t1 increments (aqmax ω1 = 35.4 ms) using a spin-lock time of 70 ms. The same conditions were used to acquire 2D NOESY data (pulse program: noesyphpr) with mixing times ranging from 100 ms to 250 ms. 2D [1H, 31P] correlation data were acquired for assignment purposes only according to the constant-time procedure (pulse program: nahpctco) (67). Data were acquired over frequency widths equivalent to 7 ppm (ω2) and 2 ppm (ω1) centred at δ1H = 4.701 ppm and δ31P = –1.185 ppm respectively with 128 transients for each of 128 State-TPPI t1 increments and a relaxation delay of 1.5 s to give a total data accumulation period of 2.75 h. 2D [1H, 13C] correlation data were acquired using echo/anti-echo-TPPI data acquisition schemes with sensitivity improvement, with (pulse program: hsqcedetgpsisp.2) and without (pulse program: hsqcetgpsisp.2) DEPT-editing and without/with non-uniform sampling (NUS) respectively. NUS data were acquired with 25% sampling of 512 t1 increments into 2048 data points for frequency widths equivalent to 10 ppm (F2) and 170 ppm (F1) with 128 transients per t1 increment for a total data accumulation time of 10 h. DEPT-edited HSQC data were acquired with a traditional acquisition mode using the same conditions and 64 transients per t1 increment for a total data accumulation time of 20 hrs. For samples dissolved in 90% H2O, 2D NOESY data were acquired with excitation sculpting for solvent suppression (pulse program: noesyesgpph) using State-TPPI for 1024 t1 increments and 4096 data points over F2 and F1 frequency widths equivalent to 25 ppm centred at the solvent frequency resonance. Data were acquired with 16 transients per t1 increment with a relaxation delay of 7 s between transients for a total data accumulation time of 33 h. Mixing times were defined as 180 and 250 ms. All data were processed according to established NMR data processing protocols and transferred into SPARKY for data assignment and reduction in preparation for molecular structure calculations.

Preliminary molecular dynamics

Preliminary molecular dynamics for distance restraint calculations of free ODN1 and the complexes were run in explicit solvent using the SANDER module of AmberTools16 (68). The starting model of ODN1 was generated using the UCSF CHIMERA program (69). The structures of each n class="Chemical">polyamide•ODN1 complex were generated using an X-ray-derived structure of a cyclic polyamide•dsDNA complex as the starting structure (PDB ID 3OMJ) (59). Input files (prmtop and inpcrd) were generated using the nucleic acid force field (LEAP module) ff99bsc1 and GAFF for ODN1 and ligands respectively. 22 and 21 Na+ ions for the free ODN1 and the complexes were added to neutralize the negative charge of the resultant structures. An octahedral box with outer edges of ∼10.0 Å of TIP3P water was placed around the structures. A 10 Å cut-off for non-bonded interaction were applied with the particle mesh Ewald (PME) method to account for long range electrostatic interactions. Initial minimizations of 1000 steps (500 steps of steepest descent and 500 steps of conjugate gradient minimization) were performed on each complex keeping the solute fixed with a 500 kcal/mol.Å2 force constant. The entire system was then minimized over 2500 steps. A first MD run of 10 000 steps was performed (20 ps) using the SHAKE algorithm. The initial temperature of 0 K was brought gradually up to 300 K and kept constant using Langevin dynamics and keeping the solute fixed with weak restraints (10 kcal/mol.Å2). The RST file was then used for a second MD run of 50 000 steps (100 ps) removing the restraints. A snapshot was recorded every 100 steps to generate the trajectory file. RST files were checked for general consistency with the NOE data and used for distance calculations.

NMR distance restraints

In all cases, the distance restraints were obtained from 2D [n class="Chemical">1H–1H] NOESY (250 ms). NMR-FAM SPARKY was used for the assignment process (70). Non-overlapping peaks were integrated using the Gaussian fit to generate the intensity file for MARDIGRAS (71). The RST files obtained from preliminary molecular dynamics were converted to pdb file format to be used in MARDIGRAS through CORMA.IN using an isotropic model and relative error of 10%. The correlation times used were roughly established in each case checking MARDIGRAS calculation outputs from 1 to 8 ns. Absolute un-normalized noise was defined as a fraction of the smallest peak that could be integrated in the spectrum. The methyl jump 3 model was used for all models. MARDIGRAS was run for 50 cycles. The resulting .dst files were associated with RAND-RESTR to generate an average distance file. These coordinates were converted to restraints to be used in AMBER using M2AHOMO or MARDI2AMBER and filtering out selected out-of-range distances. The force constants applied were 10 kcal/mol.Å for lower and upper bounds. The width of the parabola was set to be 2 Å in all the cases. Calculations were repeated at the beginning of each molecular dynamics cycle as reported below.

NMR-restrained molecular dynamics

Initial molecular dynamics were run using the generalized Born implicit-solvent model. After initial minimization (500 total steps, 250 steepest descent and 250 conjugated gradient), 100 ps MD were calculated keeping the temperature constant. NMR restraints were gradually increased over the first 20 ps and kept constant during the remaining 80 ps. After error checking, the RST file was used for a second cycle of MARDIGRAS distance calculation. The second round of molecular dynamics was carried out in explicit solvent applying the same condition reported for preliminary molecular dynamics. Newly generated NMR restraints were gradually applied over the first 20 ps of MD simulation and kept constant over the remaining 80 ps. The RST file was then used for the last cycle of MARDIGRAS calculation and the obtained restraints used for the third MD simulation. Production runs of 500 ps were obtained. The production runs were analyzed using the cpptraj module of AMBERTools16. Clustering of the last 400 ps was carried out using UCSF Chimera to produce ten representative structures, which were minimized and deposited in the ProteinDataBank (PDB accession codes 5OCZ, 5OE1, 5ODF and 5ODM). An average minimized structure obtained from this ensemble was obtained using the CPPTRAJ module of AMBERTools16 and used to calculate the DNA parameters taking a complementary approach using Curves+ and 3DNA software (72,73).

RESULTS AND DISCUSSION

The heterocyclic core of the anti-n class="Disease">tumour hairpin PIP (25,63) PA1 was chosen as our benchmark design to explore the impact of replacing the N-terminal Im unit (Im8) with Nt analogues. PA1 has an established high affinity binding profile for 5΄-WGWWCW (where W = A/T), which is recognized by the Androgen Receptor (59,63).

Synthesis of hairpin polyamides PA1-4

n class="Gene">PA1 was prepared via Boc-based solid phase synthesis on a β-Ala PAM resin (0.5 mmol/eq) (74–78). For the synthesis of PA2–4, the resin-bound hexamer 5 was used (Scheme 1). Amide coupling of dimer carboxylic acids 6–8 (Supplementary Scheme S1) (79) with 5 using HATU as the condensing agent, followed by aminolytic cleavage with (N,N-dimethylamino)propylamine (Dp) afforded PA2–4 after purification by RP-HPLC.

Scheme 1.

Synthesis of PA1–4.

Synthesis of pan class="Gene">PA1–4.

G-Selectivity of PA2-4 binding to dsDNA sequences increases with Nt steric bulk

The dsDNA binding affinity and sequence selectivity of n class="Gene">PA1–4 were determined using a combination of DNase I footprinting and UV–vis melting experiments (65). Footprinting of PA1 [3–0.01 μM] on the radiolabelled DNA fragment containing the 75 base-pair insert with variants of the putative binding sites revealed a bias towards binding to 5΄-ATTACT (Figure 2A, Supplementary Figure S1). For PA2, a more promiscuous DNA binding profile was observed, characterized by footprints at 5΄-ATTACT, 5΄-ATTACT and 5΄-ATTACT. In contrast, a bias towards G-selective binding was observed for PA4, with PA4 exhibiting a clear preference for 5΄-ATTACT; i.e. a similar trend observed for PA1. To gain further insight into the relative sequence preferences of PA1–4, UV–vis melting experiments were conducted using dsDNA sequences covering the four possible nucleotides (5΄-ATTACT, where X = G, C, A or T; Figure 2B; Supplementary Tables S1 and S2). A 19.8°C stabilization was observed for PA1 in the presence of the target sequence 5′-ATTACT (Figure 2B), with a clear binding preference for G at this position (Supplementary Tables S1 and S2). Enhanced thermal stabilization for all three Nt-containing polyamides (PA2–4) relative to PA1 was observed with PA2 exhibiting the highest degree of thermal stabilization of the polyamide series. However, enhanced dsDNA binding affinity results in sequence selectivity for G nucleotides being somewhat compromised. In agreement with the sequence selectivity observed in the DNase I footprinting experiments, sequence discrimination of PA4 was higher than PA2–3, with a ΔΔTm of 1.5°C stabilization for 5′-ATTACT relative to 5′-ATTACT.

Figure 2.

(A) DNase I footprint titration using an engineered DNA 75 bp fragment in the presence of PA1–4 at concentrations [3–0.01 μM]. (B) UV melt stabilization of PA1–4.

(A) DNase I footprint titration using an engineered DNA 75 bp fragment in the presence of pan class="Gene">PA1–4 at concentrations [3–0.01 μM]. (B) UV melt stabilization of pan class="Gene">PA1–4.

NMR structural characterization of dsDNA•polyamide complexes

Data arising from DNase I footprinting and UV–vis melts revealed a G-selective binding profile for n class="Chemical">Nt-containing polyamides as steric bulk is increased in the 5-position. NMR studies were undertaken to gain structural insight into the binding profile of PA3–4 compared to PA1 when in complex with the dodecamer oligodeoxyribonucleotide (ODN1) sequence d(CGATTCG)2 (Figure 3A and B). A 1:1 stoichiometric complex for each polyamide was observed by 1D 1H NMR and characterized by a doubling of the number of DNA imino proton resonances and a single set of polyamide resonances (Figure 3C–F). Assignment of selected NMR data for ODN1 and the polyamide complexes PA1•ODN1, PA3•ODN1 and PA4•ODN1 was achieved by a combination of 2D [1H,1H] NOESY (150 ms and 250 ms mixing times), DQF-COSY, TOCSY, 2D [1H, 13C] HSQC and 2D [1H, 31P] COSY for samples in 99% D2O and 2D [1H, 1H] NOESY for samples in 10% D2O/90% H2O. For PA3•ODN1 and PA4•ODN1, the NtH2 proton was diagnostic (8.97 ppm for PA3; 8.87 ppm for PA4) and used as a starting point for the structural assignment of these polyamides. A strong NOE cross-correlation from NtH2 to the adjacent Py7NH and Py7H3 was observed in conjunction with TOCSY correlations to identify Py7H5 (Supplementary Figure S2A). NOE correlations between Py7H5/Py7H3 and (Me)Py7 completed the assignment of Py7 resonances. This general assignment pattern was found for all three polyamide complexes. Finally, 2D [1H, 1H] TOCSY and COSY NMR data were used to assign the alkyl protons of γ-turn, β-alanine and Dp moieties.

Figure 3.

(A) Schematic representations and nomenclature of PA1, PA3 and PA4 in complex with ODN1. (B) Schematic depiction of a hairpin polyamide in complex with the target DNA duplex. (C) Selected region of 1D 1H NMR of free ODN1, (D) PA1•ODN1, (E) PA3•ODN1, (F) PA4•ODN1.

(A) Schematic representations and nomenclature of n class="Gene">PA1, PA3 and PA4 in complex with ODN1. (B) Schematic depiction of a hairpin polyamide in complex with the target DNA duplex. (C) Selected region of 1D 1H NMR of free ODN1, (D) PA1•ODN1, (E) PA3•ODN1, (F) PA4•ODN1.

Nt polyamides bind to target dsDNA sequences in a hairpin conformation

Upfield perturbation of the H4΄ chemical shift was observed for nucleotides located in the central 5΄-ATTACT sequence, which is a diagnostic marker for minor groove binding (Figure 4A) (14–16,34). This was corroborated by strong n class="Chemical">NOE correlations between amide NH, PyH3 and H1΄ sugar protons of the adjacent base (Supplementary Figure S2B). NOE cross-peaks from NtH2 of both Nt building blocks to β-alanine and Py1 protons in PA3•ODN1 and PA4•ODN1 suggest that both Nt-containing polyamides bind to their target sequence in a hairpin conformation (Figure 4B). NOE data also provide insight into the nature of Nt-binding to G5 in the minor groove. NOE correlations from NtH2 of PA3 and PA4 to G5H1′ and the G5N2 exocyclic amine suggest both Nt building blocks have the endocyclic N3 directed towards the floor of the minor groove. The presence of a hydrogen bond between NtN3 and the exocyclic G5N2 amine is implied by slower exchange rate of the exocyclic N–H of G5N2 not involved in Watson–Crick base-pairing (G5H22) and NOE cross-correlations from G5H22 to G5H1 and T6H3, all of which are absent in the NMR data for the free ODN1 duplex (Figure 5A–D). Relevant differences in the chemical shift of G5H22 in PA1•ODN1 (7.87 ppm) and Nt polyamides (PA3•ODN1 6.76 ppm and PA4•ODN1 6.59 ppm) indicate that the presence of the endocyclic sulphur in Nt building blocks influences Nt•G5N2 pairing. Taken collectively, the NMR data of each polyamide•dsDNA complex correlates with the implied formation of a hydrogen bond between the exocyclic amine of G5N2 (i.e. G5H22) and the endocyclic N3 of Im8 and both Nt building blocks.

Figure 4.

(A) Chemical shift perturbations plots of H4′ protons of ODN1. (B) Schematic representations of strong inter and intramolecular NOE cross-correlations of NtH2 (PA3 and PA4) in complex with ODN1.

Figure 5.

(A–D) Strip plots from 2D [1H, 1H] NOESY NMR data of free ODN1 and PA1•ODN1, PA3•ODN1 and PA4•ODN1. (E) Chemical shift perturbations plots of imino proton resonances of ODN1. (F–I) 31P-{1H} NMR spectra of free ODN1, PA1•ODN1, PA3•ODN1 and PA4•ODN1 complexes. (J) Chemical shift perturbations plots of phosphorous atom NMR resonances of ODN1.

(A) Chemical shift perturbations plots of H4′ protons of ODN1. (B) Schematic representations of strong inter and intramolecular n class="Chemical">NOE cross-correlations of NtH2 (PA3 and PA4) in complex with ODN1.

(A–D) Strip plots from 2D [n class="Chemical">1H, 1H] NOESY NMR data of free ODN1 and PA1•ODN1, PA3•ODN1 and PA4•ODN1. (E) Chemical shift perturbations plots of imino proton resonances of ODN1. (F–I) 31P-{1H} NMR spectra of free ODN1, PA1•ODN1, PA3•ODN1 and PA4•ODN1 complexes. (J) Chemical shift perturbations plots of phosphorous atom NMR resonances of ODN1.

Differences in DNA duplex distortion imparted by Nt-containing polyamides

Analysis of chemical shift perturbations of n class="Chemical">PA3•ODN1 and PA4•ODN1 reveal subtle, but distinct differences in DNA duplex distortion relative to PA1•ODN1. For example, a significant upfield shift of H4΄ (T6 and A7) and imino (G5H1) protons in the PA1•ODN1 complex is observed relative to PA3•ODN1 and PA4•ODN1 (Figures 4A and 5E). In the case of T6H3 and T18H3 imino protons, a more significant upfield shift is evident for the Nt-containing polyamides PA3•ODN1 and PA4•ODN1 relative to PA1•ODN1 (Figure 5E). 31P NMR spectra were assigned by scalar cross-correlations from the sugar H3΄ to the O3′-P. 31P NMR spectra of the polyamide•dsDNA complexes revealed broadening of the 31P resonances with downfield shifts of G5P and G17P, and upfield shifts of A9P and A21P, standout features relative to free ODN1 (Figure 5F–J). Further distinct chemical shift changes for G5P, T6P and A7P were observed across all three complexes indicative of different levels of phosphodiester backbone distortion imparted by each polyamide (Figure 5J, Supplementary Table S22).

NMR retrained molecular dynamics of polyamide•dsDNA complexes

NMR-derived structures of free ODN1 and n class="Gene">PA1–3-4 complexes were obtained by NMR restrained molecular dynamics (57). The ten most representative geometric conformational states of each model obtained from molecular dynamics simulations were then minimized and chosen as representative ensembles (Supplementary Figure S3). Although all three polyamides retain a ring-over-ring conformation in each complex, (14–17,59) the Nt-containing polyamides (PA3–4) adopt a distinct dsDNA binding geometry relative to PA1. In the case of PA1•ODN1, the smaller Im8 building block induces a more pronounced curvature within the minor groove relative to PA3•ODN1 (Figure 6A and B). In the PA4•ODN1 complex, the steric clash between isopropyl and (Me)Py1 prevents close contact between NtN3 and the exocyclic G5N2 amine (Figure 6C). In addition, more pronounced base inclination and axis bending is seen in the PA1•ODN1 complex compared with PA3•ODN1 and PA4•ODN1 (Supplementary Figure S4, Supplementary Tables S24, S28, S32 and S36).

Figure 6.

View of the curvature and anticipated hydrogen-bonding pattern between (A) PA1•ODN1, (B) PA3•ODN1 and (C) PA4•ODN1.

View of the curvature and anticipated n class="Chemical">hydrogen-bonding pattern between (A) PA1•ODN1, (B) PA3•ODN1 and (C) PA4•ODN1.

Analysis of the minor and major groove geometries in each complex reveals a number of distinct structural features. n class="Gene">PA1•ODN1 and PA3•ODN1 complexes are characterized by an enlarged minor groove relative to PA4•ODN1 (17,59). This is most likely due to deeper minor groove penetration of the less hindered Im8 and (Me)Nt8 (Figure 7A, Supplementary Table S23). An unexpected observation is the structural impact on the major groove width when PA1•ODN1 is compared with PA4•ODN1. A more pronounced narrowing of the major groove is observed at the site of (iPr)Nt binding to G (i.e. NtN3•G5N2), changing from 22.4 Å in the free duplex (i.e. ODN1) to 12.2 Å in PA4•ODN1 (Figure 7B, Supplementary Table S23). In contrast, only a slight narrowing (21.0 Å) of the major groove is observed for the PA1•ODN1 complex, which is consistent with crystal structural studies of a cyclic PIP•dsDNA complex (59). For the PA3•ODN1 complex, an intermediate change (16 Å) in the major groove width relative to PA4•ODN1 suggests that the increased steric bulk of the isopropyl substituent in PA4•ODN1 plays a significant role in inducing major groove compression of the DNA duplex.

Figure 7.

(A) Minor Groove and (B) Major Groove width of ODN1 and PA1•ODN1, PA3•ODN1, PA4•ODN1. Major groove view of (C) PA1•ODN1 (D) PA3•ODN1 (E) PA4•ODN1.

(A) Minor Groove and (B) Major Groove width of ODN1 and n class="Gene">PA1•ODN1, PA3•ODN1, PA4•ODN1. Major groove view of (C) PA1•ODN1 (D) PA3•ODN1 (E) PA4•ODN1.

Structural changes to the Watson-Crick pairing profile in each n class="Chemical">polyamide•DNA complex also differ. In PA1•ODN1, the larger propeller twist in T4•A21 and increased buckle of T6•A19 contribute in inducing a widening of the major groove, which is also typified by the larger shift changes of the imino and H4΄ protons (Supplementary Tables S25, S29, S33, S37). In the case of the PA4•ODN1 complex, the sterically more encumbered (iPr)Nt induces significant changes in the χ torsional angle (O4΄–C1΄–N9–C4) of the G5 (–87.9°) nucleosidic bond relative to PA3•ODN1 (–103.2°) and PA1•ODN1 (–117.1°; Supplementary Figure S3 and Tables S27, S31, S35, S39). Finally, the significant deviation in the 31P resonances of G5P, T6P and A7P in PA4•ODN1 relative to PA1•ODN1 is evidence of significant changes in the phosphodiester bond angles ζ (C3΄–O3΄–P-O5΄) and ε (C4΄–C3΄–O3΄-P) located in close proximity to the site of Nt-binding (Supplementary Figure S5 and Tables S27, S31, S35, S39).

Rationalisation of the differences in DNA selectivity and duplex distortion of Nt-containing polyamides

These experiments were designed to probe how structural changes to the N-terminal heterocycle of a hairpin n class="Chemical">PA affects overall dsDNA binding affinity and sequence selectivity. Our data reveal that the replacement of an N-terminal Im unit for an Nt heterocycle enhances overall dsDNA binding, but at the expense of sequence selectivity for a target G nucleotide. We hypothesise that the overall differences in hydrophobicity, hydrogen bond strength and steric bulk between Nt and Im heterocycles are contributing factors. Underpinning all of these is the replacement of the nitrogen N1 of Im with a sulfur (S1) atom, which in Nt is also oriented differently, occupying the same position of the smaller carbon (C5) atom in the Im unit. This key structural change appears to contribute to a general loss of G-selectivity for PA2–3 when there is minimal steric bulk in the 5-position relative to the archetypical hairpin PA1. This has implications in, for example, weaker hydrogen bonding between the exocyclic N-H of G5N2 (i.e. G5H22) and the N3 of Nt compared to the more basic ImN3, which is implied by a downfield shift of G5H22 for the Im-containing polyamide (PA1•ODN1, 7.87 ppm) compared to Nt polyamide (PA3•ODN1 6.76 ppm, and PA4•ODN1 6.59 ppm). The higher dsDNA melting stabilisation observed for Nt polyamides (PA2–4) compared to the archetypical Im polyamide (PA1) is attributed to a stronger overall hydrophobic interaction for the Nt-containing PA2–4, which appears to compensate for the weaker hydrogen bonding interaction with G5H22, and is most likely responsible for an increase in non-specific dsDNA binding for PA2 and PA3. However, the installation of an isopropyl group (i.e. PA4) increases G selectivity with respect to. Whilst these changes in G selectivity cannot be rationalised by this study alone, we hypothesise that the steric effect of the isopropyl group in PA4 biases the binding towards wider minor groove G•C regions.

A structural rationalization for the observed dsDNA binding differences of n class="Chemical">PA3–4 relative to PA1 would likely involve an interplay between the geometrical differences of the Nt and Im heterocycles, and their inherent differences in their respective non-covalent interactions with dsDNA in the minor groove. For example, the stronger hydrogen-bonding as well as the smaller overall dimension of the Im8 unit in PA1 results in deeper penetration into the dsDNA minor groove. From a topological perspective, this induces an overall increase in dsDNA bending in the PA1•ODN1 complex (Supplementary Figure S4), which concomitantly compresses the major groove around the central T6•A7•C8 region of the DNA duplex. In the case of PA3•ODN1, the weaker hydrogen-bonding interaction as well as the steric effect of the sulfur in (Me)Nt reduces the overall bending (Supplementary Figure S4), which in turn compresses the major groove over a larger DNA region (T4•G5•T6•A7•C8). The 5-position substituent of the (iPr)Nt building block of PA4•ODN1 only partially affects the overall DNA bending (Supplementary Figure S4). A unique characteristic is the larger steric bulk of the (iPr)Nt unit induces a more extensive compression of the major groove (in particular over G5), which is characterised by an increased hydrogen bond distance between the N3 of (iPr)Nt and G5H22 arising from reduced minor groove penetration. Taken collectively, the steric effect of the isopropyl Nt substituent in PA4 enhances the selective recognition of Nt-containing hairpin polyamides by compensating for the weaker hydrogen bonding interaction with the exocyclic N2 of G with an increase in the overall hydrophobicity of this N-terminal building block.

CONCLUSION

In conclusion, our data shows that hairpin n class="Chemical">polyamides which incorporate an N-terminal Nt unit (PA3–4) induce more pronounced compression of the major groove of target DNA duplexes compared to archetypical hairpin PIPs and untethered minor groove binders (e.g. PA1) (51). The enhanced level of major groove compression in PA3•ODN1 and PA4•ODN1 complexes is akin to Nt-containing polyamides ‘pinching’ the target dsDNA sequence at the polyamide extremities. These differences in dsDNA distortion could offer new opportunities for modulating gene expression either via disruption of transcription-dsDNA binding or by blocking transcriptional elongation (80–82). We envisage that further tuning of dsDNA binding affinity and sequence selectivity is indeed possible at other sites within the polyamide scaffold, which could be used as a design proxy for their continuing development as gene modulatory agents across a range of cell types.

AVAILABILITY

Atomic coordinates for the reported structures have been deposited with the Protein Data bank under accession numbers 5OCZ, 5OE1, 5ODF and 5ODM.

Click here for additional data file.