Characterizing Watson-Crick versus Hoogsteen base-pairing in a DNA-protein complex using NMR and site-specifically ¹³C/¹⁵N labeled DNA

Abstract

A(syn)-T and G(syn)-C⁺ Hoogsteen base pairs in protein-bound DNA duplexes can be difficult to resolve by X-ray crystallography due to ambiguous electron density and by NMR spectroscopy due to poor chemical shift dispersion and size limitations with solution-state NMR spectroscopy. Here we describe an NMR strategy for characterizing Hoogsteen base pairs in protein-DNA complexes, which relies on site-specifically incorporating 13C/¹⁵N labeled nucleotides into DNA duplexes for unambiguous resonance assignment and to improve spectral resolution. The approach was used to resolve the conformation of an A-T base pair in a crystal structure of a ~43 kD complex between a 34-bp duplex DNA and the integration host factor (IHF) protein. In the crystal structure (PDB: 1IHF), this base pair adopts an unusual Hoogsteen conformation with a distorted sugar-backbone that is accommodated by a nearby nick used to aid in crystallization. The NMR chemical shifts and inter-proton NOEs indicate that this base pair predominantly adopts a Watson-Crick conformation in the intact DNA-IHF complex under solution conditions. Consistent with these NMR findings, substitution of 7-deazaadenine at this base pair resulted in only a small (~2-fold) decrease in the IHF-DNA binding affinity. The NMR strategy provides a new approach for resolving crystallographic ambiguity and more generally for studying the structure and dynamics of protein-DNA complexes in solution.

Graphical Abstract

INTRODUCTION

Following the discovery of the DNA double helix¹, much effort was directed towards solving single crystal X-ray structures of isolated purine-pyrimidine dimers to test the fine details of Watson-Crick base pairing proposed by Watson and Crick^2–8. The first successful attempt was reported in 1959, when Karst Hoogsteen solved the X-ray co-crystal structure of base pair derivatives containing 9-methyladenine and 1-methylthymine². Rather than observing an A–T Watson-Crick base pair (bp), an alternative pairing mode called ‘Hoogsteen’ was observed (Figure 1A). Relative to a Watson-Crick bp, the adenine base flips 180° to adopt a syn rather than anti conformation (Figure 1A). The base forms a unique set of hydrogen bonds (H-bonds) that require that the two bases come into closer proximity by ~2.0–2.5 Å^{3, 9} (Figure 1A), which in turn induces a kink in the DNA double helix^{10, 11}. Analogous G–C⁺ Hoogsteen bps¹² form by flipping the guanine base but also require protonation of cytosine N3 to form a second H-bond (Figure 1A).

Figure 1.

Structures of Watson-Crick and Hoogsteen base pairs and crystal structure of IHF-DNA complex. (A) Chemical structures of Watson-Crick and Hoogsteen base pairs. Differences in hydrogen bonding, cytosine protonation, purine base orientation (anti versus syn) and C1′–C1′ distance are highlighted. Close atomic distances involving imino protons that give rise to characteristic NOEs in A-T Watson-Crick (AH2-TH3) and Hoogsteen (AH8-TH3) bps are indicated using black arrows. (B) X-ray structure of the IHF-DNA complex (PDBID: 1IHF¹³). The consensus sequence for specific protein recognition is shown in blue. The A21–T48 Hoogsteen-like (HG-like) bp, the A56-T13 Watson-Crick (WC) bp, and the control A9-T60 Watson-Crick bp are shown in red, orange and grey rectangles, respectively. The nick in one DNA strand is indicated with an arrow. Two pseudosymmetry-related large kink sites are marked with dashed lines in the DNA sequence, corresponding to positions of intercalating proline residues.

Since their discovery, Hoogsteen bps have been observed in X-ray structures of AT-rich sequences that form duplexes entirely composed of Hoogsteen bps^14–17; in structures of DNA bound to small molecule bis-intercalators^{12, 18–22}; in damaged DNA bases incapable of forming Watson-Crick bps where they are thought to play roles in damage accommodation^23–25, recognition²⁶, and repair²⁷; in Y-family ‘low fidelity polymerases’ which replicate DNA using Hoogsteen pairing as a means of bypassing lesions on the Watson-Crick face during replication^28–31. More recently, Nuclear Magnetic Resonance (NMR) relaxation dispersion (RD) methods^32–34 revealed that in canonical duplex DNA, A–T and G–C Watson-Crick bps exist in dynamic equilibrium with short-lived (lifetimes of 0.2–2.5 ms) and sparsely populated (0.1–5%) Hoogsteen bps^{11, 35–39}.

Hoogsteen bps are also observed in some X-ray structures of DNA-protein complexes where they are thought to play roles in DNA shape recognition^{13, 40–42}. For example, two consecutive G–C⁺ Hoogsteen bps are observed in the complex between DNA and the TATA box-binding protein⁴¹. In one of these Hoogsteen bps, the syn guanine base appears to alleviate a steric clash between the guanine-amino group and a nearby leucine side chain⁴¹. In addition, one of the syn guanines partially stacks with the phenylalanine residue that inserts between the two G-C Hoogsteen bps²². In the structure of DNA in complex with the DNA binding domain of p53 tumor suppressor protein, two consecutive A–T Hoogsteen bps in the consensus sequence contribute to a narrowed and more negatively charged minor groove in the flanking regions, which may favor electrostatic interactions with a positively charged arginine residue⁴². Thus, proteins seem to employ a variety of mechanisms to interact with Hoogsteen bps.

Considering that it is now well established that Hoogsteen bps form robustly in naked duplex DNA^{23, 35, 40} with a preference for sites of major groove kinking^{10, 11}; it comes as something of a surprise that Hoogsteen bps have not been more widely observed in structures of protein-DNA complexes, where structural distortions could destabilize Watson-Crick bps. Indeed, recent studies on DNA-drug complexes show that recognition of DNA can result in a variety of behaviors; with some bps predominantly (>90% populations) adopting Watson-Crick or Hoogsteen conformations, while others having high (~10%) fractional Hoogsteen populations relative to the naked DNA (<1%)^{22, 40, 43}. Depending on the quality of the electron density, it can be difficult to resolve Hoogsteen bps from Watson-Crick bps when solving X-ray structures of DNA^44–46.

For example, Aggarwal and coworkers showed using X-ray crystallography and biochemical experiments that human DNA polymerase ɩ (Polɩ), a member of the Y-family polymerases, employs Hoogsteen base-paring to replicate damaged and undamaged DNA^{28, 31, 47}. The X-ray structure of Polɩ showed a template syn adenine in the active site forming a Hoogsteen bp with an incoming dTTP²⁸. This structure was met with some skepticism when Wang⁴⁴ argued that the weak electron density of the active site A–T bp makes it is difficult to unambiguously assign a Watson-Crick versus Hoogsteen geometry. Subsequent structural and biochemical studies, including studies examining the impact of single-atom substitutions that destabilize Hoogsteen pairing in the DNA template, confirmed that this polymerase does indeed employ Hoogsteen mechanism to replicate certain base pairs^{30, 48}. Other studies have noted the difficulty in resolving Hoogsteen versus Watson-Crick bps in X-ray structures of protein-DNA complexes including the DNA-homeodomain⁴⁵ and DNA-p73 complexes⁴⁶. Thus, it is conceivable that other Hoogsteen bps in crystal structures of DNA have been potentially mis-modeled as pure Watson-Crick bps. Sites that experience significant Hoogsteen breathing can also be difficult to resolve by crystallography, as the elevated local dynamics could contribute to the lack of features in the local electron density⁴⁹ or be suppressed by cryocooling of the crystals that is routinely performed to improve resolution of crystal structure⁵⁰. Conversely, there is also a need for methods that can independently verify Hoogsteen bps observed in crystal structures of DNA. For example, AT-rich duplexes composed entirely of Hoogsteen bps based on X-ray crystallography have been shown to form regular Watson-Crick bps when examined by solution NMR¹⁵. Here, another potential example is given by the structure of the integration host factor (IHF) protein in complex with DNA¹³ (Figure 1B).

IHF is a bacterial DNA-binding protein that aids in the compaction of prokaryotic genomes by inducing significant DNA bending⁵¹. It plays important roles in site-specific recombination of λ phage into the bacteria host genes⁵², enhancing bacterial DNA replication⁵³, regulation of gene expression^{54, 55}, and in Cas1-Cas2-mediated spacer integration⁵⁶. IHF induces a global conformational change in the DNA duplex (i.e. ~160° bending) upon binding, resulting in two sharp major groove directed kinks separated by ~7 Watson-Crick bps in pseudo-symmetry related sites (Figure 1B). In the crystal structure (PDB: 1IHF), a nick between T48 and G49 (Figure 1B) was introduced at a kinked site to aid crystallization¹³. Next to this site is an unusual Hoogsteen-like A21–T48 bp (Figure 1B). The adenine base adopts an anti, rather than syn conformation that is uncharacteristic of Hoogsteen bps observed so far in duplexes. Additionally, the entire T48 nucleotide is flipped about the sugar-phosphate backbone to adopt a left-handed Z-DNA like conformation. This large sugar-backbone distortion at T48 is likely made possible by the nick between T48 and G49¹³ (Figure 1B). Indeed, the corresponding A56–T13 bp at the pseudo-symmetry related site adopts a Watson-Crick conformation, though the bp is distorted¹³ (Figure 1B). Whether or not the A21–T48 bp forms a Hoogsteen bp remains to be established experimentally in the native IHF-DNA complex without nicks.

Thus, there is a need for complementary approaches that can help resolve Watson-Crick versus Hoogsteen conformations and their dynamics in DNA under solution conditions. Watson-Crick and Hoogsteen bps can in principle readily be resolved using solution NMR spectroscopy and this could provide a means to resolve crystallographic ambiguity^{11, 23, 35}. However, application of NMR-based approaches can be difficult for large DNA-protein complexes owing to severe spectral overlap and unfavorable relaxation properties, which leads to significant line-broadening and losses in sensitivity. Here, we propose and employ an approach that relies on using commercially available ¹³C/¹⁵N labeled nucleotides to site-specifically incorporate ¹³C/¹⁵N nucleotides at specific positions in the DNA duplex. Results from the NMR analysis are then verified biochemically by examining the impact of site-specifically incorporating 7-deazapurine (7dA), which destabilizes Hoogsteen bps^{30, 36}, on DNA-protein binding affinity. We apply this approach to resolve the conformation of the A21–T48 bp in the intact IHF-DNA complex with full length DNA under solution conditions. The results indicate that the A21–T48 bp predominantly forms a Watson-Crick bp.

METHODS

Sample Preparation

Preparation of DNA and IHF protein samples for NMR and FP studies:

Single-stranded DNA oligonucleotides with one (A21 or A9) or two (A56 and T48) uniformly ¹³C/¹⁵N-labeled deoxynucleotides were purchased from the Yale Keck Oligo Synthesis Resource with DNA cartridge (Glen-Pak) purification. The only exception was the A21-labeled DNA strand, which was purchased with standard desalting and subjected to in-house purification via 20% denaturing polyacrylamide gel electrophoresis (PAGE). Final yields of purchased site-specifically labeled DNA oligonucleotides were 150 – 200 nmol and ~90 nmol for cartridge-purified and PAGE-purified oligonucleotides, respectively.

5′-Fluorescein-dT labeled oligonucleotides for FP studies were purchased with reverse-phase HPLC purification from IDT (Integrated DNA Technologies, Inc.). DNA oligonucleotides with 7dA21 and 7dA56 modifications were purchased from the Midland Certified Reagent Company and the Yale Keck Oligo Synthesis Resource, respectively, with reverse-phase HPLC purification. Unmodified DNA oligonucleotides were purchased with standard desalting from IDT (Integrated DNA Technologies, Inc.). Complementary DNA strands were annealed at a final duplex concentration of 1 mM. Annealing was performed by heating the mixed single strands at 95 °C for 5 min and slow cooling at room temperature for 20–30 minutes.

Preparation of IHF protein:

IHF protein (UniprotKB: P0A6X7 and P0A6Y1) was expressed in E. coli Rosetta(DE3)plysS cells from a pET21a vector (available from Addgene). Cells from a frozen glycerol stock were streaked onto an agar plate containing LB (Luria-Bertani broth), ampicillin (50 μg/mL) and chloramphenicol (30 μg/mL) and grown overnight at 37 °C. Several single colonies were then resuspended into 20 mL liquid LB with 100 μg/mL ampicillin, which was added to 500 mL of the same medium, grown at 37 °C until A₆₀₀ reached ~0.7 and kept at 4°C overnight. About 80 mL of that culture was used to inoculate 6 1-liter cultures, which were also grown at 37 °C. Once reached A₆₀₀=0.7–0.8, IPTG was added to a final concentration of 0.5 mM to induce protein expression. Cells were harvested 2–3 hours later by centrifugation at 8,000 rpm for 7 minutes (Sorvall F10S rotor), and the pellet was stored at −20 °C. Cell pellets were thawed on ice, then resuspended in 200–300 mL 100 mM Tris (pH 8.0), 1 mM EDTA, 10% Surcrose (w/v), 10% (v/v) Glycerol, 1 M NaCl, with 6 tablets of Complete Mini Protease Inhibitor cocktail (Roche). After lysozyme was added to a final concentration of 200 μg/mL, cells were incubated on ice for 10–20 minutes, then sonicated. Cell debris was pelleted by centrifugation for 1 hour at 18,000 rpm in an SS-34 rotor.

The supernatant was collected and subjected to (NH₄)₂SO₄ fractionation. Briefly, IHF remained in the supernatant at 50% saturation and was in the pellet at 80% saturation. Solid (NH₄)₂SO₄ was first added to a final saturation of 50% by gradual addition of solid (NH₄)₂SO₄ with gentle stirring over ice, with an additional for 30 minutes of gentle stirring after all the solid had dissolved. The supernatant was then collected after centrifugation for 30 minutes at 18,000 rpm in an SS-34 rotor. (NH₄)₂SO₄ was gently stirred into the supernatant as above to a final saturation of 80%, followed by a similar centrifugation, and the pellet was stored at 4 °C. The sample was further purified by chromatography on heparin and monoS columns. The (NH₄)₂SO₄ pellet was dissolved in 50 mL or more buffer A (20 mM MES pH 5.5, 0.1 mM EDTA, 5% glycerol) and loaded onto a 20 mL heparin column (Amersham). After washing with 20% buffer B (buffer A with 2 M NaCl), a gradient of 20–80% for buffer B was run in 90 minutes, with IHF eluting at ~1 M NaCl. IHF-containing fractions were pooled, dialyzed into 20% buffer B (80% buffer A), and rechromatographed on the same column. Fractions were then pooled and dialyzed overnight into 2.5% buffer B and loaded onto a 1 ml monoS column (Amersham). The column was washed with 2.5% buffer B, and eluted with a gradient of 2.5% to 47.5% buffer B in 45 minutes. IHF eluted at ~0.35 M NaCl. Pooled fractions were concentrated to about 10 mg/mL, then were dialyzed into 25 mM HEPES pH 7.5, 1 mM EDTA, 20% Glycerol, 0.2 M NaCl, and flash frozen in small aliquots, 100μl each. Samples were tested for endonuclease contamination by incubating 10 μg/μL IHF with 1 μg/μL pUC19 plasmid in the presence of 10 mM MgCl₂. Samples were examined by agarose gel electrophoresis after a 2-hour incubation at 37 °C, and no degradation of the plasmid was seen. The concentration of IHF was determined using a calculated absorption coefficient (IHF at 2.95 mg/mL has OD₂₈₀ of 1).

Preparation of NMR sample of the DNA-IHF complex:

NMR samples of IHF-DNA complex were prepared by slowly titrating and mixing 50 μM duplex DNA into 50 μM IHF protein. The resulting 25 μM DNA-IHF complex was then buffer exchanged by centrifugal concentrators (EMD Millipore with 3 kDa cut-off) to obtain the NMR sample (~250 μL) and then supplied with 10% D₂O. Final concentrations of free DNA and DNA-protein complex NMR samples were ~0.3 mM.

NMR buffer:

DNA or DNA-protein complex samples were buffer exchanged into the NMR buffer consisting of 25mM HEPES, 100 mM NaCl, 0.1 mM EDTA with pH 7.0 and 10% D₂O three times using a centrifugal concentrator at 4 °C (EMD Millipore with 3 kDa cut-off) until containing >99.9% of the desired buffer.

NMR Experiments

NMR data were collected on a 600 MHz Bruker NMR spectrometer equipped with an HCN cryogenic probe, a 600 MHz Bruker Avance III spectrometer equipped with an HCPN cryogenic probe, a 700 MHz Bruker Avance III spectrometer equipped with a triple-resonance HCN cryogenic probe, and an 800 MHz Varian DirectDrive2 spectrometer equipped with a triple-resonance HCN cryogenic probe and. Data were processed and analyzed using NMRpipe⁵⁷ and SPARKY (T. D. Goddard and D. G. Kneller, SPARKY 3, University of California, San Francisco), respectively. Chemical shifts data were obtained using both TROSY and conventional 2D HSQC experiments; resonance assignments were analyzed using ¹⁵N-edited 2D [¹H, ¹H] NOESY (mixing time 180 ms), 2D H1′-C1′-AN9/TN1 HCN and conventional HSQC experiments with broadband ¹⁵N decoupling and selective decoupling on purine-N9 or pyrimidine-N1. Spectra were recorded at 25 °C unless otherwise stated.

The resonances in the IHF-DNA complex could be unambiguously assigned for single-site labeled samples (H′-DNA^A21 and H′-DNA^A9). The aromatic resonances (A56-C8 and T48′-C6) in IHF bound H′-DNA^A56T48 complex were assigned by turning on and off the selective carbon decoupling on T-C5 (~110 ppm) in the conventional 2D HSQC experiment⁵⁸ (data not shown). A56-C1′ and T48-C1′ in the free H′-DNA^A56T48 were assigned by a ¹H-¹⁵N 2D version of HCN experiment⁵⁹ that correlates H1′ to adenine-N9 or thymine-N1 which has distinct chemical shifts (i.e. adenine-N9 ~ 170 ppm while thymine-N1~140 ppm) (Figure S2A). However, the same HCN experiment could not detect signals in the DNA-protein complex due to significant transverse relaxation before acquisition. The A56-C1′ and T48-C1′ resonances in the DNA-protein complex were assigned by selectively decoupling adenine-N9 or thymine-N1, in comparison with the broadband ¹⁵N decoupling in a regular 2D HSQC experiment (Figure S2B).

A weak NOE cross peak is observed which can be tentatively assigned to A21H1’ in the ¹⁵N-edited NOESY spectra of the complex (see Figure 4B). Such an NOE is not expected for a regular Watson-Crick bp in which the distance between the protons is >6 Å nor is it expected for the distorted Watson-Crick bp in the IHF-protein crystal structure (>6 Å). Such a weak NOE could arise if T48 and A21 came slightly into closer proximity in solution, perhaps even transiently, when bound to IHF protein. This is consistent with observation of ~0.22 ppm downfield shifted T48-H3 in the complex form relative to that in free DNA, which suggests stronger hydrogen bonding. It is also consistent with previously reported (Dhavan et al J Mol Biol 1999) reduced imino proton exchange rates measured in a short 19 bp DNA upon binding to IHF. While a Hoogsteen bp could also bring these protons into close proximity (~5.5 Å), we can rule out this possibility because we do not observe any evidence for the T48H3-A21H8 NOE (Figure S3).

Figure 4.

NMR analysis of IHF-DNA complex formation. (A) 1D ¹H spectra showing imino proton resonances in the IHF-H′-DNA^A56T48 complex overlaid with unbound DNA. (B) 2D [¹³C, ¹H] TROSY HSQC spectra for IHF-DNA complexes (colored) overlaid with unbound DNA (in grey). Chemical shift perturbations are indicated by the black arrows. (C) Overlay of 2D [¹⁵N,¹H] SOFAST HMQC spectra for IHF-H′-DNA^A56T48 complex (in green) and unbound H′-DNA^A56T48 (in gray) at 37 °C. Minor unbound DNA population is observed in the spectrum consistent with slow exchange. Complex-only spectra are shown in Figure S3. (D) 2D ¹⁵N-edited [¹H, ¹H] NOESY spectrum for IHF-H′-DNA^A56T48 complex shows T48H3-A21H2 NOE cross-peaks consistent with Watson-Crick pairing at A21-T48 bp. The T48H3-A21H1′ NOE assignment (labeled with *) is tentative and we cannot rule out that the resonance assigned to A21H1′ corresponds to other neighboring sugar or amino protons.

R_1ρ RD experiments on the free H′-DNA^A56T48 duplex were performed on the 700 MHz Bruker spectrometer as previously described^{35, 61, 62} at 37 °C, with spinlock powers (ω_SL 2π⁻¹ Hz) and offset frequencies (Ω 2π⁻¹ Hz) listed in Table S1. Because of the large size of the duplex and low sample concentration (0.3 mM), a large number of scans (ns=512) were used in the RD experiments targeting A56-C1′ and the T48-C1′ to achieve 10–20 signal-to-noise ratio in the absence of the spin-lock. Because of the low sensitivity, only two delay points (0 ms and 20 ms) were used to monitor mono-exponential decay. RD measurements for aromatic spins (adenine-C8/C2) were not performed due to the much lower sensitivity arising due to the larger R₂⁶¹ and potentially larger chemical exchange contributions.

AFQM/MM Calculations

Structures of the IHF-DNA complex for the automated fragmentation quantum mechanics/molecular mechanics (AFQM/MM) calculations were prepared based on the crystal structure (PDBID: 1IHF). Due to the presence of the nick near A21, we focused on computing chemical shifts based on the pseudo-symmetry related kink site without any nick. We directly took the crystal structure (PDB: 1IHF) as the model structure for 100% Watson-Crick conformation for the calculation. For modeling A56 at 100% Hoogsteen conformation, the glycosidic bond at A56 was manually flipped 180°. The structure with flipped A56 was hydrated using SPC/E water, and then neutralized by adding Na⁺ ions, followed by energy-minimization for 1000 steps and heating gradually from 0 K to 298 K in 20 ps. The system was subsequently equilibrated for 1 ns using Langevin dynamics with the collision frequency 3 ps⁻¹ and NPT ensemble at 298 K. We observed Hoogsteen H-bonds (AN6-H6---TO4 and AN7---TH3-N3) formed in the structure after the equilibration step and the resultant structure was used in the AFQM/MM calculation.

Chemical shift calculations followed the procedure described earlier^{39, 63}. For structures of IHF-DNA complex with either Hoogsteen or Watson-Crick bp modeled at A56-T13, quantum fragments were constructed centered on nucleotides A9 and A56. The effects of atoms in the IHF-DNA complex outside the quantum region (nucleotide A9 or A56), and of water and ions in the solvent, were represented by point charges uniformly distributed on the molecular surface of the quantum region, were resolved by fitting to Poisson–Boltzmann calculations using the “solinprot” program from the MEAD package⁶⁴. The quantum region (A9 or A56) was assigned a local dielectric of ε = 1 (vacuum), the remaining nucleic acid region had ε = 4, and the solvent region ε = 80. GIAO chemical shift calculations were carried out for each quantum fragment, using version 4.3.6 of the demon-2k program⁶⁵, using the OLYP functional⁶⁶ with the TZVP basis set and the GEN-A2* fitting set. Further details are available elsewhere^{39, 63, 67}.

Fluorescence Polarization Assay for Measurement of Binding Affinity

Fluorescence polarization (FP) experiments were performed using a Tecan F500 plate reader using 485/20 nm excitation and 535/25 nm emission polarization filters suitable for fluorescein and a Beacon2000 Fluorescence Polarization System. Experiments were conducted using a black polypropylene 384-well round-bottom plate (Corning Inc.). The concentration of fluorescein-labeled DNA was kept constant at 3 nM, and the IHF protein concentration was varied from 0.1 nM to 20 nM. The FP assay was carried out at 25 °C in the NMR buffer (25mM HEPES, 100 mM NaCl, 0.1 mM EDTA with pH 7.0) and the anisotropy values measured in triplicates or duplicates. Fluorescence anisotropy (A) was calculated based on the parallel and perpendicular emission light intensities:

$$A=\left(I_{\Vert }-I_{\perp }\right)/\left(I_{\Vert }+2I_{\perp }\right)$$

where A is the measured anisotropy at a given concentration. The normalized anisotropy, A_norm, was obtained using:

$$A_{\mathit{\text{norm}}}=\left(A-A_0\right)/\left(A_{\mathit{\text{max}}}-A_0\right)$$

in which A₀ refers is the anisotropy in absence of protein, A_max is the maximum value of measured anisotropy. A_norm was measured at varying protein concentration (x) and then fitted to an equation describing single-site binding using the Levenberg-Marquardt non-linear curve fitting algorithm implemented in the OriginPro 2016 software (OriginLab Corporation):

$$A_{\mathit{\text{norm}}}=B+C\left(D+K_d+x-\sqrt{{\left(D+K_d+x\right)}^2-4Dx}\right)$$

where D is the constant fluorescein-labeled DNA concentration, B and C determine the anisotropies for free DNA and bound DNA, respectively, and K_d is the apparent dissociation constant.

Since 7dA56 and 7dA21 are located in different strands, we kept the fluorescein-dT at the 5′-end of the opposing strand to the one containing the 7dA substitution (Figure 6A). Therefore, two control DNAs without 7-deaza substitution were made with the fluorescein label at either of the two strands (Figure 6). Based on the FP assay, the two control DNAs yielded similar K_d values (K_d1 = 1.0 ± 0.3 nM and K_d2 = 1.4 ± 0.4 nM) for binding of unmodified H-DNA duplexes to IHF (Figure S4).

Figure 6.

Impact of 7-deazapurine substitutions on IHF-DNA binding affinity. (A) Sequences of DNA duplexes containing 7-deaza-dA modification (red circles) and the fluorescein-dT label (green star) at the 5′-end of the unmodified strand. Also shown is the chemical structure of 7-deaza-dA and its destabilization impact on Hoogsteen H-bonding. (B) Comparison of FP binding curves for 7-deaza-dA modified DNA with its unmodified counterpart (control DNA). Control DNA refers to the unmodified DNA duplex with the fluorescein-dT label at the same position as the modified duplex. Data is fit to single-site binding equation (see methods). Error represents standard deviation from triplicate (left) or duplicate (right) measurements. Fitted K_d and uncertainty values shown on the right.

RESULTS

NMR characterization of site-specifically labeled DNA duplex

Three site-specifically labeled 34 bp DNA duplexes were prepared (see methods) to study the IHF-DNA complex (Figure 1B and Figure S2A). In each case, specific nucleotide positions were uniformly ¹³C/¹⁵N labeled (Figure 1B and Figure S2A). In one sample (H′-DNA^A21), the A21 that forms a Hoogsteen bp in the crystal structure was labeled (Figure 1B and Figure S2A). A second sample (H′-DNA^A56T48) was prepared by labeling the A21 bp partner T48 as well as A56 at the pseudo-symmetry related site which forms a Watson-Crick bp in the crystal structure (Figure 1B and Figure S2A). A21, A56 and T48 are located at sites in which the DNA kinks sharply toward the major groove (Figure 1B and Figure S1). As a negative control, a third sample (H′-DNA^A9) was prepared labeling A9 which is located in a similar CAA sequence context but falls outside the consensus sequence in a region of the DNA structure that has a smaller degree of major groove kinking (Figure 1B and Figure S1). The buffer conditions used for the NMR studies (25 mM HEPES, 100 mM NaCl, 0.1 mM EDTA with pH 7.0) were similar to those used in the crystallization buffer (10 mM HEPES, 100 mM NaCl, 0.1 mM EDTA, 8% glycerol with pH 7.0)⁶⁸.

We first analyzed the NMR spectra of the unbound DNA duplexes (Figure 2A). As expected, the 1D ¹H spectra for all three DNA duplexes in the absence of IHF were essentially identical with well-resolved imino proton resonances that are consistent with a Watson-Crick B-form DNA duplex (Figure 2B). Excellent quality 2D NH and CH HSQC TROSY-based NMR spectra were obtained for all three duplexes clearly showing only a subset of resonances belonging to the labeled nucleotides (Figure 2C). The ¹H, ¹³C, and ¹⁵N chemical shifts in the absence of IHF were all consistent with Watson-Crick pairing in B-form DNA. Resonances could readily be assigned in these site-labeled duplexes (see Methods and Figure S2). Assigning large nucleic acids (>20 kDa), such as the 34-bp DNA duplex, with conventional 2D/3D NOESY experiments would be costly (~1 mM for ¹³C/¹⁵N-labeled samples) and challenging owing to low sensitivity and severe spectral overlap^69–71. In such cases, having a site-specifically labeled sample even at low concentration (~0.3 mM in this study) provides a convenient methodology in a targeted approach towards identifying Watson-Crick/Hoogsteen bps in large nucleic acids molecules, potentially at lower external magnetic field strengths (say 400 MHz) and non-cryogenic probes.

Figure 2.

NMR characterization of site-specifically ¹³C/¹⁵N-labeled H′-DNA duplexes. (A) H′-DNA^A21, H′-DNA^A56T48 and H′-DNA^A9 sequences with ¹³C/¹⁵N-labeled nucleotides highlighted in color. (B) 1D ¹H NMR spectra showing imino proton resonances in site-labeled DNA overlaid with unlabeled H′-DNA measured at 25 °C. (C) Overlay of 2D [¹³C,¹H] TROSY⁷² and 2D [¹⁵N, ¹H] SOFAST HMQC^{73, 74} NMR spectra measured for the three site-specially labeled unbound H′-DNA sequences at 25 °C. Note that line broadening of A21-C1′H1′ relative to other resonance is not observed in 2D HSQC (not TROSY) spectra (see Figure S3).

Transient Hoogsteen bps in long naked DNA duplexes

Thus far, studies have examined transient Hoogsteen bps in relatively short (<15 bp) duplexes^{35, 38} using R_1ρ relaxation dispersion NMR experiments. To examine whether Watson-Crick-Hoogsteen exchange is detectable in the longer (34 bp) IHF DNA duplexes, we carried out R_1ρ RD measurements on A56-C1′ and T48-C1′ in H′-DNA^A56T48 in the absence of IHF protein (Table S1). The R_1ρ experiment^{33, 34, 75} measures the chemical exchange contribution (R_ex) to intrinsic transverse relaxation rate (R_2,int) of NMR resonances during a relaxation period when the spin is irradiated with an applied radiofrequency (RF) pulse with varying powers (ω_SL) and frequencies (ω_RF). Experiments were performed at T = 37 °C.

Indeed, we observed the expected RD at A56-C1′ (Figure 3). Fitting of the data to 2-state exchange model yields exchange parameters (p_B = 0.3±0.1%, k_ex = 1611±525 s⁻¹ and Δω = 4.2 ± 0.3 ppm) that are consistent with chemical exchange directed to Hoogsteen bps^{35, 38}. The measured p_B falls within the range (0.2% - 0.3%) measured in short (~12 bp) DNA duplexes containing the same CAA trinucleotide sequence context at 30.5 °C³⁵. The measured exchange rate (k_ex = 1611±525 s⁻¹) is only slightly slower (~3000 to 5000 s⁻¹) than values measured in shorter duplexes³⁵. As a negative control, we did not observe any RD at T48-C1′ (Figure 3) at 37 °C as expected for chemical exchange directed to Hoogsteen bps^{35, 39}. These results indicate that the Watson-Crick-Hoogsteen equilibrium is not significantly altered in longer DNA duplexes.

Figure 3.

Off-resonance relaxation dispersion profiles for unbound H′-DNA^A56T48. Shown are off-resonance profiles as a function of spin-lock offset (Ω 2π⁻¹ kHz, where Ω = Ω_obs − ω_RF) and power (ω_SL 2π⁻¹ Hz, color coded as indicated in the inset) measured at 37 °C. Error bars represent experimental uncertainties estimated from mono-exponential fitting of n = 2 (A56-C1′) and n = 2 (T48-C1′) independently measured peak intensities using a Monte Carlo–based method (Methods). The solid line represents a fit to two-state exchange³⁵.

Characterizing IHF-DNA complex formation

Two assays were used to assess IHF-DNA complex formation prior to NMR analysis. We observed the expected migration retardation in the electrophoretic mobility shift assay (EMSA)⁷⁶ when incubating the DNA duplexes with increasing IHF concentration (Figure S4). Saturation was observed at 1:1 molar ratio consistent with 1:1 binding stoichiometry (Figure S4A). The apparent dissociation constant (K_d^app = 1.0 ± 0.3 nM) measured using a fluorescence polarization (FP) assay⁷⁷ (Methods) employing a DNA duplex tagged with fluorescein-dT at the 5′-terminus of one strand (Figure S4B) is in good agreement with values previously reported for the same 34-bp H′-DNA with the IHF protein using gel mobility shift assays with ³²P labeling (K_d^app ~ 2.0 ± 0.5 nM)⁶⁸, and for a shorter (30 bp) H′-DNA (K_d^app ~ 1.65 ± 0.09 nM)⁷⁸ as well as for a longer (56 bp) H′-DNA duplex (K_d^app ~ 1.5 nM)⁷⁹ measured by EMSA.

NMR spectra indicate that A21–T48 forms a predominantly Watson-Crick conformation in the DNA-IHF complex.

Next, we characterized IHF-DNA complex formation using NMR. The imino protons belonging to the DNA duplex experienced chemical shift perturbations and line-broadening upon addition of IHF to a 1:1 ratio (Figure 4A). Similar changes were also observed for aromatic and aliphatic resonances consistent with the DNA undergoing changes in structure upon complex formation (Figure 4B). Binding of IHF to DNA is slow on the NMR time scale as evidenced by observation of two sets of resonances that belong to free and bound DNA in spectra with excess DNA. These results establish the utility of using site-specifically labeled DNA to study relatively large DNA-protein complexes by NMR.

Interestingly, the addition of IHF resulted in similar NMR chemical shift perturbations at A56 and A21 (Figure 4B). This indicates that the two adenine nucleotides experience similar changes in their environment upon complex formation in solution, as expected for two pseudo-symmetrically related sites (Figure 1B). Nevertheless, the A56–T13 and A21–T48 chemical shifts in the IHF-DNA complex are not identical, as expected given differences in their sequence and structural contexts (Figure 1B). For example, A56 is located within the consensus sequence while A21 is not, and in the X-ray structure, the A56-T13 uniquely forms H-bonds with a nearby Arg63 residue (Figure 1B).

We observed the T48-H3 imino resonance in the 2D ¹⁵N-¹H SOFAST HMQC spectra of IHF-DNA complex (Figure 4C) with ¹H and ¹⁵N chemical shift (T48-H3 13.8 ppm and T48-N3 158.8 ppm) that fall within the region expected for a Watson-Crick bp. We did not observe any evidence for an upfield shifted thymine imino resonance (~1.5–2.0 ppm in ¹H and ~2 ppm in ¹⁵N) that would be expected for an A-T Hoogsteen bp^{11, 22, 35}. The fact that the T48-H3 imino is observable and not significantly upfield shifted also argues in favor of a stable Watson-Crick conformation and against a high population of transient Hoogsteen conformation that are in fast exchange with the Watson-Crick on the NMR timescale. A predominant A21-T48 Watson-Crick bp is in good agreement with the prior NMR studies on a smaller DNA construct containing the consensus sequence of the H1 binding site⁶⁰. In contrast, the control site A9 showed distinct and generally smaller chemical shift perturbations (Figure 4B), consistent with the crystal structure showing weaker distortions at this site (Figure S1).

Next, we carried out ¹⁵N-edited NOESY experiments (Figure 4D) to directly obtain structural information regarding the nature of base pairing at A21–T48 in the DNA-IHF complex. If A21–T48 adopts a Hoogsteen conformation, we would expect to observe an NOE cross peak between T48-H3 and A21-H8^{11, 23, 35}. This NOE cross-peak has previously been reported for a variety of Hoogsteen bps in DNA duplexes, including in modified m¹A–T Hoogsteen bp in duplex DNA^{11, 23, 35} and A–T Hoogsteen bps in DNA-echinomycin complexes^{22, 80}. We did not observe this NOE in the IHF-DNA complex (Figure 4D and Figure S3). Instead, we observed the A21H2-T48H3 NOE as expected for Watson-Crick pairing (Figure 4D).

Using NMR chemical shift fingerprinting to resolve Watson-Crick and Hoogsteen bps

The aromatic and aliphatic ¹³C/¹⁵N/¹H chemical shifts of A21 undergo large perturbations on complex formation that are inconsistent with the perturbations expected for formation of Hoogsteen bps in unbound DNA duplexes^{11, 35, 39}. For example, formation of Hoogsteen bps is expected to induce 2–3 ppm downfield shift in A21-C1′³⁵, but a 0.6 ppm upfield shift is observed upon IHF-DNA formation (Figure 4B). Formation of Hoogsteen bps lead to very minor changes in A21-C2 chemical shift³⁵, yet large changes (downfield shifted by 1.4 ppm) are observed on complex formation. Many of the ¹H chemical shifts are also inconsistent with expectations of Hoogsteen formation. For example, A-H8 and A-H1′ are typically upfield shifted by ~0.5 and ~ 0.1 ppm, respectively, when forming m¹A–T Hoogsteen bps^{11, 23, 35}; but downfield shifts are observed (by 0.14 and 0.88 ppm, respectively) on complex formation (Figure 4B).

The above analysis is however complicated by the fact that chemical shift perturbations likely include other contributions such as bending of the DNA and changes in stacking of bases upon binding resulting in altered ring current effects and possibly interactions with the IHF protein. The Hoogsteen chemical shift signatures may also differ in the highly bent IHF-DNA as compared to relaxed duplex DNA. To further understand the observed chemical shift perturbations on complex formation, we used automated fragmentation quantum mechanics/molecular mechanics (AFQM/MM)⁶³ to compute chemical shifts based on the crystal structure of the IHF- NA complex for models in which the A56 –T13 adopts a Watson-Crick (native) versus Hoogsteen (modeled) conformations. We then compared these predicted values with those measured experimentally using the site-labeled NMR samples. The AFQM/MM approach was recently shown to accurately predict chemical shifts in DNA duplexes with Hoogsteen and Watson-Crick bps³⁹. The chemical shifts measured at the control site A9–T60 was used as an internal reference. The difference in chemical shift between values predicted for a given site (A56 or A21) and for the corresponding nucleus in A9 (Δω = ω_{A56 (or A21)} − ω_A9) were predicted computationally and then compared to the experimental data (Figure 5). This approach is clearly limited by the accuracy of the predicted chemical shifts including the modeling of the complex using a single structure rather than an ensemble of conformations as done previously³⁹. Nevertheless, better agreement was observed between measured chemical shifts for both A21 and A56 and values calculated for the distorted Watson-Crick bp as in the X-ray structure as compared to the modeled Hoogsteen bp (Figure 5). While the agreement is somewhat better for the Hoogsteen form for AC8, and no distinctions can be made for AH1′ and AH2, AC1′, AC2 and AH8 resonances suggest better overall agreement with the Watson-Crick conformation The chemical shift data therefore indicate that for the intact IHF-DNA complex in solution, the predominant conformation adopted by both A21–T48 and A56–T13 is similar to that observed for the distorted A56–T13 Watson-Crick bp in the X-ray at the site without the nick. Indeed, this observed result is consistent with the X-ray structure of DNA in complex with Hbb, an IHF-family bacterial DNA-bending protein. Although Hbb recognizes a different DNA sequence from IHF, the two protein-DNA complexes present similar overall highly bent DNA structures upon binding and two distorted A-T Watson-Crick bps are observed at the two sharp kinks in the Hbb-DNA complex determined without any nicks (PDBID: 2NP2⁸¹).

Figure 5.

Comparison of measured and calculated chemical shifts for Watson-Crick and Hoogsteen models in the IHF-DNA complex. Shown is the difference between chemical shifts, Δω = ω_Aix − ω_A9x in which i refers to different nucleotides (A21 or A56) and x to different type of nuclei (e.g. C1′, C8, C2). The residue A9 is used as an internal reference. Shown are AFQM/MM predicted values at A56 for 100% Hoogsteen (gray), 100% Watson-Crick (orange), and values measured by NMR for A21(cyan) and A56 (green).

7-deazaadenine substitutions minimally affect the IHF-DNA binding affinity

Our results indicate that predominant conformation for A21-T48 is Watson-Crick. If this were the case, selective destabilization of the Hoogsteen bp relative to the Watson-Crick should minimally affect the IHF-DNA binding affinity. 7-deazapurines have previously been shown to selectively destabilize Hoogsteen bps by knocking out one hydrogen bond (N7---H3-N3) while not affecting Watson-Crick type hydrogen bonding^{82, 83}. The modification destabilizes Hoogsteen bps relative to Watson-Crick bps by at least ~1 kcal/mol (i.e. 10-fold reduction in transient Hoogsteen population), resulting in undetectable RD in naked DNA duplexes³⁶. 7-deazapurine was also shown to significantly inhibit DNA replication by the low-fidelity Y-family polymerase iota (Polɩ) that replicates DNA via Hoogsteen base-pairing by suppressing the formation of Hoogsteen bps³⁰.

We examined the impact of 7-deazaadenine (7dA) substitutions at the kink sites on the IHF-DNA binding affinity. Using the FP assay, the 7dA21 substitution only decreased the binding affinity ~2-fold (K_d = 2.6 ± 0.6 nM for H′-DNA^7dA21 compared to 1.0 ± 0.3 nM for the control DNA) (Figure 6). In comparison, the 7dA substitution at the pseudo-symmetry related site (7dA56) had no measurable effect on the binding affinity (K_d = 1.0 ± 0.4 nM). These results argue against a predominantly Hoogsteen conformation in the complex (>99%) at A21 but do not rule out having a small portion of Hoogsteen population in equilibrium with the major Watson-Crick conformation. The small differences observed for the two sites (A21 and A56) are also consistent with small differences in NMR chemical shifts at these two bps in the IHF-DNA complex (Figure 4B).

DISCUSSION

Resolving Hoogsteen versus Watson-Crick bps is not always straightforward by X-ray crystallography. Here, we demonstrated an NMR approach for examining Watson-Crick to Hoogsteen dynamics in large protein-DNA complexes. Application to the IHF-DNA complex indicates that A21-T48 adopts a predominantly Watson-Crick conformation and that the Hoogsteen bp observed in the crystal structure most likely arises due to the introduction of a nick near the kinked site. In this regard, it is interesting to note that Hoogsteen bps have been observed in structural contexts in which the nucleotide is not chemically linked to both neighbors, including bps at duplex termini, and also replication bps in the polymerase active site during replication^{10, 49}. The NMR strategy presented here could facilitate studies of Hoogsteen bps in such contexts and possibly illuminate their biochemical consequences.

In a recent study, it was shown that the binding of the echinomycin antibiotic to DNA increased the Hoogsteen population to 8%²². Our data cannot rule out the possibility that the transient Hoogsteen population for A21-T48 in the IHF-DNA complex is also elevated by comparable amounts. While we observed line broadening for A21-C1′ and A56-C1′ relative to A9 and T48 in the IHF-DNA complex, further studies are needed to resolve whether these differences reflect μs-to-ms chemical exchange at A21-C1′ and A56-C1′ or faster ps-to-ns dynamics that lead to narrowing of A9–C1′ and T48-C1′. TROSY based NMR RD experiments combined with the site-specific labeling strategy used here may help extend the methodology to allow the study of transient Hoogsteen bps in large protein-DNA complexes^{72, 84–86}. These studies could illuminate the DNA bending motions reported previously in the IHF-DNA complex⁸⁷ as well as other dynamics that could impact the kinetic rates of DNA and IHF assembly^{87, 88}.

While our data suggest that the bps in the IHF-DNA complex form a predominantly Watson-Crick conformation, recent solution state studies on p53-DNA complex provide strong evidence for a predominantly Hoogsteen conformation which was observed in crystallographic structures^{42, 89}. These crystal structures did not feature any nicks in the DNA. In particular, inosine or 2-oxo adenine substitutions that weaken or enhance the Hoogsteen bp formation, as verified by X-ray crystallography, had more dramatic (~6-fold) effects on DNA-protein binding affinities⁸⁹. This strongly suggests that Hoogsteen bps can form in protein-DNA complexes under solution conditions where they can contribute to shape recognition.

The NMR strategy presented in this work opens the door for characterizing the dynamics of DNA in protein-DNA complexes. Indeed, very few studies^{60, 90, 91} have examined DNA dynamics at atomic detail when in complex with proteins and the potential role of these fluctuations in the stability and function of the complex. This approach can be applied to systems as large as 50 kDa but could potentially be extended to much larger complexes with the use of solid-state NMR. The site-labeling approach also provides an important avenue for cataloguing valuable chemical shift structure relationships that can be harnessed in the future to aid chemical-shift-based characterization of DNA structure and dynamics.