Homopurine guanine-rich sequences in complex with N-methylmesoporphyrin IX form parallel G-quadruplex dimers and display a unique symmetry tetrad

Abstract

DNA can fold into G-quadruplexes (GQs), non-canonical secondary structures formed by π-π stacking of G-tetrads. GQs are important in many biological processes, which makes them promising therapeutic targets. We identified a 42-nucleotides long purine-only G-rich sequence from human genome, which contains eight G-stretches connected by A and AAAA loops. We divided this sequence into five unique segments, four guanine stretches each, named GA1-5. In order to investigate the role of adenines in GQ structure formation, we performed biophysical and X-ray crystallographic studies of GA1-5 and their complexes with a highly selective GQ ligand, N-methyl mesoporphyrin IX (NMM). Our data indicate that all variants form parallel GQs whose stability depends on the number of flexible AAAA loops. GA1-3 bind NMM with 1:1 stoichiometry. The K_a for GA1 and GA3 is modest, ~0.3 μM⁻¹, and that for GA2 is significantly higher, ~1.2 μM⁻¹. NMM stabilizes GA1-3 by 14.6, 13.1, and 7.0 °C, respectively, at 2 equivalents. We determined X-ray crystal structures of GA1-NMM (1.98 Å resolution) and GA3-NMM (2.01 Å). The structures confirm the parallel topology of GQs with all adenines forming loops and display NMM binding at the 3’ G-tetrad. Both complexes dimerize through the 5’ interface. We observe two novel structural features 1) a ‘symmetry tetrad’ at the dimer interface, which is formed by two guanines from each GQ monomer and 2) NMM dimer in GA1-NMM. Our structural work confirms great flexibility of adenines as structural elements in GQ formation and contributes greatly to our understanding of the structural diversity of GQs and their modes of interaction with small molecule ligands.

Graphical Abstract

1. Introduction

G-quadruplex DNA (GQ) is a non-canonical DNA structure formed by G-rich DNA sequences. GQs have a unique tetra-stranded architecture composed of stacked G-tetrads. G-tetrads are formed by four guanines (G) organized via Hoogsteen hydrogen bonds in a square-planar geometry. GQs are stabilized via π-π stacking and ion coordination. Highly polymorphic, GQs vary in strand topology (parallel, antiparallel, and hybrid), molecularity, loop type, and number of G-tetrads.¹ They are also highly dynamic and can adopt a variety of conformations under appropriate conditions (buffer, DNA concentration, presence of co-solutes or ligands). Bases other than Gs can be part of the G-tetrad.² Bulges³ and loops^4,5 add other dimensions to GQ diversity. Structural elements such as Watson-Crick duplexes can be combined into one structure with GQs.^6,7 GQs form readily in vitro, and display high thermodynamic stability in physiological buffers.

Over 700,000 DNA sequences with GQ-forming potential have been identified within the human genome,^8,9 only a small fraction of which have been structurally characterized. Biological evidence suggests both positive and negative roles for GQs in human biology. GQ structures are proposed to serve as roadblocks during replication, transcription, and DNA repair causing DNA damage and genomic instability, linking GQs to cancer.¹⁰ The negative impact of GQs is counterbalanced by their protective roles at telomeres (as our laboratory demonstrated in yeast¹¹) and in defining DNA replication origins in vertebrates (80-90% of human replication origins are GC-rich¹²). Any cellular process that involves DNA in single-stranded form could potentially be regulated by GQs.

GQ-forming sequences with loops containing only thymines (T) or both T and adenines (A) have been thoroughly characterized. Contrary, G-rich purine-only sequences and the effect that adenines have on GQ secondary structures is underexplored. Structural work demonstrates great versatility of adenines. Their availability may lead to formation of A-only tetrads,^13-15 GAGA tetrads,¹⁶ or unique pentads,^17,18 hexads,¹⁹ and heptads,^20,21 where the traditional G-tetrad is coordinated in plane by 1-3 adenines at the groove positions. Biological evidence suggests that purine-only sequences are widely distributed throughout the human genome and are located near gene promoters, in telomeres, centromeres, triplet repeat disease sequences, or recombination hotspot sites. For example, GGA and GAA triplet repeats are associated with Friedreich ataxia, an autosomal recessive disease caused by repeat expansion, which likely involves the formation of non-canonical DNA structures.²² Long purine-rich sequences were shown to promote recombination²³ and block transcription.²⁴ Genomes other than human, such as the HIV genome, are also extremely adenine rich.²⁵

Here we characterize a purine-only 42-nucleotides sequence, named GA (see Figure 1A). Using the BLAST-like alignment tool (BLAT) by UCSC genome browser,²⁶ we found that the full GA sequence is located on human chromosomes 7 and shorter variants of it can be found on chromosomes 2, 4, 5, 6, 10, 11,13, and 15, indicating that GA is likely a common repetitive sequence in the genome. This sequence has eight guanine stretches connected by either A or AAAA loops. Bioinformatics studies suggest the most common loop length is one-nucleotide, and A is one of the most commonly occurring loop bases.^27,28 GA can be represented by five unique variants of four G-stretches each, GA1-5, Figure 1A. Our starting goals are to establish whether these variants fold into GQ structures and to determine their thermodynamic stability and 3-dimensional architecture. In the future, we would like to determine the behavior and structure of the full GA sequence to better understand the role that adenines play in its structure formation.

Figure 1.

(A) GA oligo and its shorter variants. Individual guanine stretches are colored differently for clarity. Guanines expected to participate in a G-tetrad formation are underlined. (B) Structure of NMM. Commercial NMM exists as a mixture of four regioisomers with N-Me group at one of the four core nitrogens (only one regioisomer is shown).

Given the potentially novel structures and important biological roles of purine-only DNA, understanding its interactions with small molecule ligands is important. N-methyl mesoporphyrin IX (NMM, Figure 1B) is a water-soluble porphyrin with excellent optical properties. The N-Me group at the center of the porphyrin ring causes modest non-planar distortion of the macrocycle that allows NMM to selectively interact with GQ DNA. Using competition dialysis, fluorescence enhancement, FRET melting, single-molecule FRET, and helicase inhibition assay, NMM was shown to bind and stabilize GQ DNA but not single-stranded DNA, double-stranded DNA or RNA, i-motif, Z-DNA, DNA-RNA hybrid, or triplex DNA (reviewed in Yett et all)²⁹. The structural basis of NMM selectivity is the presence of N-Me group that fits neatly into K⁺ channel of a parallel GQ at its 3’ G-tetrad side.^30,31 When GQs exist as an equilibrium mixture of multiple conformations, NMM converts the whole population to the parallel fold.^32,33

We characterized GA variants alone and in complex with NMM using biophysical methods and X-ray crystallography. We employed UV-Vis, circular dichroism (CD), and fluorescence spectroscopies as well as PAGE to determine folding, stability, homogeneity, and oligomeric state of the sequences and their binding affinity for NMM. We also report two crystal structures of GA1-NMM and GA3-NMM complexes at ~2 Å resolution. While both structures display parallel GQ fold and bind NMM at their 3’ G-tetrad, NMM’s binding pocket in the two structures differs. Very interestingly, the GQs dimerize via a previously unseen ‘symmetry tetrad’ formed by two guanines from each monomer.

2. Material and Methods

2.1. DNA, ligand, and buffers

Lyophilized oligonucleotides were purchased from Integrated DNA Technologies (IDT; Coralville, IA) with standard desalting purification. DNA was hydrated in doubly distilled water to 1-2 mM and stored at 4 °C. Extinction coefficients for all sequences were obtained using IDT’s OligoAnalyzer and DNA concentration was determined from UV-vis spectra collected at 90 °C. The full list of DNA sequences along with their thermodynamic parameters can be found in Table 1. Extinction coefficients and molecular weights can be found in Table S1.

Table 1.

DNA sequences studied in this work and their thermodynamic parameters in 100K buffer. T_m is a melting temperature; ΔH is an enthalpy change associated with melting transition; ΔT_m is an increase in T_m upon NMM binding; ΔΔH is an increase in ΔH upon NMM binding.

DNA	Sequence	loops	Tm, °C	ΔH, kJ/mol	ΔTm, °C	ΔΔH, kJ/mol
GA1	GAGGGAGGGAGGGAAAAGGGGA	A, A, AAAA	63.7 ± 1.1	164 ± 3	14.6 ± 1.1	81 ± 5
GA2	GAGGGAGGGAAAAGGGGAGGGA	A, AAAA, A	63.5 ± 0.8	158 ± 10	13.1 ± 0.8	63 ± 10
GA3	AGGGAAAAGGGGAGGGAAAAGGGGA	AAAA, A, AAAA	47.5 ± 0.7	143 ± 11	7.0 ± 1.7	14 ± 18
GA4	AGGGGAGGGAAAAGGGGAGGGGA	A, AAAA, A	66.9 ± 0.5	220 ± 17	ND	ND
GA5	GGAGGGAAAAGGGGAGGGGAGGG	AAAA, A, A	58.8 ± 0.8	197 ± 20	ND	ND

^*ND – not determined

NMM stock was prepared in doubly distilled water and its concentration was determined via UV-Vis scans using an extinction coefficient of 1.45 × 10⁵ M⁻¹cm⁻¹ at 379 nm³¹. Here we used the following buffers. One contained 10 mM lithium cacodylate (LiCaco) pH 7.2 supplemented with either 10 (10K buffer), 50 (50K), or 100 (100K) mM KCl. Another contained 10 mM LiCaco pH 7.2, 95 mM LiCl, and 5 mM KCl (5K buffer). Titrations were done in 5K buffer; the rest of biophysical experiments were done in 50K and 100K buffers; crystallization trials were completed in 10K and in 100K buffers. To induce GQ formation, DNA was diluted to the desired concentration in 10 mM LiCaco pH 7.2 and heated at ~95°C for 3 minutes. At this point, KCl was added to the desired concentration and the samples were heated for an additional 3 minutes. The samples were cooled slowly to room temperature over 4 hours and equilibrated at 4 °C overnight. Such sample preparation was used for all but titration experiments (see Sections 2.4.2).

2.2. Circular dichroism (CD) scans

All CD experiments were conducted on an Aviv 435 circular dichroism spectrophotometer equipped with a Peltier thermocontroller (± 0.3 °C error) in 1 cm quartz cuvettes in 100K buffer. CD scans were taken at 20 °C from 220 to 330 nm with 1 s averaging time, 2 nm bandwidth, and 1 nm step. Five scans were collected and averaged. CD data were processed as described in our earlier work.³¹

2.3. Native polyacrylamide gel electrophoresis (PAGE)

PAGE samples contained 3 μg DNA in 10 μL of 100K buffer and were weighted down with 7% w/v sucrose prior to loading. To prepare crystal samples, 5-10 crystals were harvested, washed in the crystallization condition, and dissolved in 10-15 μL of 10K or 100K buffer. Fifteen percent native polyacrylamide gels were made with 10 mM KCl and 1×Tris-Borate-EDTA (TBE). Running buffer contained 10 mM KCl and 1×TBE. Gels were pre-migrated at 150 V for at least 30 min, loaded with 10 μL of each sample, and allowed to run for 150 min at 150 V at room temperature. A tracking dye was used to monitor gel progress and an oligothymidylate ladder consisting of dT₁₅, dT₂₄, dT₃₀, and dT₅₇ was used as a length marker. DNA bands were visualized using Stains-All and the resulting gel was captured using a smart phone camera.

2.4. UV-Vis spectroscopy

UV-Vis experiments were carried out with an Agilent Cary 3500 UV-Vis spectrophotometer equipped with a Peltier temperature controller (± 0.5 °C error). Data were collected using 1 nm intervals, 0.02 s averaging time, 2 nm spectral bandwidth, automatic baseline correction, and 220-330 nm range for DNA, and 350-480 nm for NMM.

2.4.1. Melting experiments

UV-Vis melting experiments were conducted from 20-90 °C with a 1 °C step, 0.5 or 0.2 °C/min temperature rate, 2 s averaging time, and 2 nm spectral bandwidth. The data were collected at 260, 295, and 335 nm. The signal at 295 nm reports on GQ unfolding; the signal at 260 nm reports on duplex DNA unfolding; and the signal at 335 nm was collected to monitor instrument stability. Both 260 and 295 nm signals were corrected for the 335 nm absorbance. Melting temperatures, T_m, were determined via two methods. The first method involves taking the first derivative of the 295 nm absorbance signal and finding the temperature at the trough or peak through a visual inspection (associated with ± 0.5 °C error). The second method assumes a two-state model for GQ folding with constant ΔH applicable to fully reversible melting transitions, i.e. when melting and cooling curves are nearly superimposable.³⁴ Hysteresis was determined as the difference between T_m determined from the melting and cooling curves. As hysteresis did not exceed 3 °C, all systems studied are considered reversible, and the reported thermodynamic data were obtained using the two-state model (i.e., the second method).

2.4.2. UV-Vis titrations

One thousand μL of NMM sample was prepared in a 1 cm methylmethacrylate cuvette to target an absorbance of ~0.5 (3-4 μM NMM). DNA samples were annealed in 5K buffer for 5 min, cooled to room temperature for 4 hours, and equilibrated at 4 °C overnight. DNA target concentration was 170-470 μM to achieve [DNA]/[NMM] final ratio of at least five. To maintain constant porphyrin concentration in the cuvette throughout the titration, DNA was annealed at double the target concentration and was diluted 1/1 (v/v) with 2×NMM sample which was also used to prepare NMM solution in the cuvette. DNA was added to NMM in increasing increments, equilibrated for 2 min, after which the UV-Vis spectra were collected. The titration stopped when no changes were observed for three additions. The volume of DNA added (~100 μL total), λ_max, and absorbance at λ_max were monitored. The total volume of DNA added was typically less than 10% of the total volume of NMM sample. UV-Vis spectra were processed using Singular Value Decomposition (SVD)^35,36 to generate binding curves. Binding constants were calculated by direct fit of the binding curves in Graphpad Prism software, as described in our earlier report³¹ with minor modifications. All SVD decompositions had two dominant components, v1 and v2, followed by a large gap between v2 and the rest of the components. Binding constants were generated by a global fit to v1 and v2. The NMM concentration was fixed to the value determined experimentally via UV-Vis spectroscopy. Reported K_a values represent the average of at least three trials with associated errors. We estimated binding energies at 293 K, ΔG₂₉₃, using the following relationship, ΔG = −RTlnK_a.

2.5. Fluorescence titrations

Fluorescence experiments were conducted on a Photon Technology International QuantaMaster 40 fluorometer at 20 °C. Data were collected using an emission range of 560 - 720 nm with 2 nm slit widths, 0.5 nm step size, and 0.5-1.0 s integration time. The isosbestic points determined through UV-vis titrations, 395-398 nm, were used as the excitation wavelengths. NMM was placed in methylmethacrylate fluorescence cuvettes (~1.0 μM, 1500 μL) and titrated with small increments of 140-270 μM DNA targeting a final [DNA]/[NMM] ratio of at least eight. DNA was prepared as described in section 2.4.2. The volume of DNA added (~150 μL total), λ_max, and intensity at λ_max were monitored throughout the titration. The data were analyzed in the same way as UV-vis titration but using a single SVD component, v1, which adequately described the variation. Reported K_a values represent the average of at least three trials with associated errors. Fluorescent enhancement was determined as described previously.³⁷

2.6. X-ray Crystallography

2.6.1. General

Crystallization was achieved at room temperature using the hanging-drop vapor diffusion method. DNA samples at 1.0 mM were annealed with 1.0-1.5 mM NMM in 10 mM LiCaco pH 7.2 at 90 °C for 3 min at which point 10 or 100 mM KCl was added and the sample was annealed for another 3 min. The samples were allowed to slowly cool to room temperature and equilibrate overnight at 4 °C.

Initial screening was performed on a TTP Labtech Mosquito Liquid Handling robot system equipped with a humidity chamber using commercial Natrix (Hampton Research) and Helix (Molecular Dimensions)³⁸ screens, along with a homemade Amber screen.³⁹ Ninety-six well trays were set up using drops of equal volumes of sample and crystallization condition (either 0.1 or 0.2 μL) and 100 μL of well condition. Optimization was performed manually in 24-well trays by combining 1 μL DNA sample with 1 μL crystallization condition into wells containing 400 μL of crystallization condition.

Diffraction data were collected at the Advanced Photon Source 24 ID-C synchrotron facility at both the native wavelength (λ = 0.98Å, 12622 eV) and the cobalt wavelength (λ = 1.61Å , 7725 eV). Raw diffraction data were processed using RAPD software provided by the beamline or XDS⁴⁰, Pointless⁴¹, and Aimless⁴². All structures were solved through molecular replacement (MR) in PHENIX⁴³. The initial solution after MR was improved by extensive manual model building in COOT⁴⁴, followed by refinement in PHENIX.

2.6.2. GA1+NMM

For the original screening, DNA was prepared in a 100K buffer. Natrix2-3 yielded clustered, flaky crystals within 1 week. Through optimization in 24-well plates, the best crystals were obtained at 0.04 M sodium cacodylate pH 5.5, 0.012 M NaCl, 0.08-0.1 M KCl, 45% MPD, and 0.002 M hexamminecobalt(III) chloride. The optimized crystals were thin, well-ordered rhombuses. The crystals were harvested and flash frozen in liquid nitrogen without additional cryo-protectant.

Diffraction data were collected on 14 crystals to a maximum resolution of 1.98 Å in space group F222 at the native (λ = 0.98Å) and cobalt (λ = 1.61Å) wavelengths. For MR we used a partially refined GA3+NMM structure (see below) which was cropped to only include the GQ core and an additional guanine stacking on the tetrad, making one of the G stretches a four-guanine run. The asymmetric unit contains one DNA chain, one NMM molecule, three potassium ions, one cobalt(III) ion, and four waters. Two adenine bases in the AAAA loop (A₁₄ and A₁₅) and the 3’-A overhang were not built due to weak electron density caused by high flexibility of this region. NMM has clear density, but no further than the first carbon beyond the porphyrin ring. Therefore, we did not build the ethyl groups and propanoic side chains of NMM and its exact orientation is unknown. The Fo-Fc difference map contains a green electron density aligned with the K⁺ channel near NMM and may suggest that NMM should be rotated 180° around the central GQ axis so that its N-Me group fits into this green density. However, the density reappears after each round of NMM rotation and refinement of the entire structure. We have also modeled one cobalt(III) ion using a cobalt anomalous map but without amine ligands due to insufficient electron density. Data collection and refinement statistics are shown in Table 2. The structure was deposited into the PDB database and assigned PDB ID: 8EBO.

Table 2.

Data collection and refinement statistics

	GA1+NMM (outer shell)	GA3+NMM
Resolution Range	30.44-1.98	56.02 - 2.01
Highest resolution shell	2.03-1.98	2.06 - 2.01
Space group	F222	P22121
Unit cell dimensions
a, b, c (Å)	42.05, 47.47, 119.6	30.58, 40.93, 56.02
α, β, γ (°)	90, 90, 90	90, 90, 90
Unique reflections	4296 (279)	5027 (370)
Redundancy	5.8 (6.3)	5.4 (5.3)
Completeness (%)	99.2 (99.8)	99.6 (97.7)
I/sigma	15 (0.7)	8.8 (0.9)
R-merge (%)	0.049 (2.074)	0.081 (1.433)
Rwork/Rfree (%)	0.2314/0.2789	0.2448/0.2797
Copies in ASU	1	1
Number of atoms	476	521
DNA (no hydrogens)	435	472
NMM	33	33
Solvent	4	5
Potassium	3	4
Cobalt	1	1
Ammine	0	6
Overall B-factor (Å2)	95.625	86.87
RMS deviations
Bond length (Å)	0.014	0.011
Bond angles (°)	1.934	1.47
PDB ID	8EBO	8EDP

2.6.3. GA3+NMM

For the original screening, DNA was prepared in a 100K buffer. Red cubic, rhombic, or square prism-shaped crystals grew within a week in Helix1-10, Helix1-45, and Helix2-16 (Table S2). Helix1-10 condition yielded the best crystals upon optimization in a 24-well tray. Diffraction quality crystals grew from 0.05 M MES 6.5, 0.3 M KCl, 40% MPD, and 5 mM hexamminecobalt(III) chloride crystallization condition using 1 mM DNA sample in either 10K or 100K buffer within 3 weeks, although additional aging for 6 months did not affect the quality of their diffraction. 10K buffer produced higher quality crystals. The crystals were harvested and flash frozen in liquid nitrogen without additional cryo-protectant.

Datasets were collected to a maximum resolution of 2.01 Å in space group P22₁2₁ at the native (λ = 0.98 Å) and cobalt (λ = 1.61 Å) wavelengths. The structure was solved by MR in PHENIX using a search model of only the GQ core and partially truncated loops (truncated thymines from the first two TTA loops while the last loop was kept intact) of a parallel GQ formed by human telomeric repeats in complex with NMM (PDB ID 40GF). NMM and waters were also removed. The asymmetric unit contains one DNA chain, one NMM molecule, four potassium ions, one hexaamminecobalt(III) ion, and four waters. The bases of all adenines in the first loop (A₅-A₈) as well as 5’- and 3’-A overhangs were not built due to high conformational freedom and weak electron density. For the first 4A-loop, the sugar-phosphate backbone was built, and in some cases (A₅ and A₈), partial bases (an atom to a half of the base) were built due to clearly defined density around part of the bases. These partially built bases indicate the direction of the bases. For the overhangs, only phosphates were built. NMM is built the same way as in GA1-NMM. X-ray data collection and refinement statistics are summarized in Table 2. The structure was deposited into the PDB database and assigned PDB ID: 8EDP.

2.6.4. Analysis of crystallographic data

G-tetrad planarity and distances were calculated via SVD with an in-lab MATLAB script³¹. DNA backbone torsion angles were calculated using 3DNA⁴⁵. Helical twist and groove widths (via an average of C3’-C3’ sugar atom distances) were calculated using the program Advanced Structural Characteristics of G-quadruplexes ASC-G4 (http://tiny.cc/ascG4). The grooves of the regular G-tetrads were given double weight while the groove for the symmetry tetrad was given a single weight. Therefore, the average groove widths are reported for the dimer (and not the monomer). RMSD was calculated in PyMol. B-factors were calculated both manually from the PDB and with Baverage in CCP4i. Metal geometry was checked using Metal Binding Site Validation Server (https://cmm.minorlab.org/).⁴⁶ NMM stacking distances were calculated using the distance between centroids with an in-lab MATLAB script. NMM pyrrole angles were calculated from the angles between normal vectors from SVD-calculated planes.

3. Results and Discussion

In this work, we performed thorough biophysical characterization and structural study of DNA sequences derived from a purine-only GA oligo and their complexes with NMM. GA is a 42-nucleotides sequence with 33.3% adenine and 66.7 % guanine content found on chromosome 7. GA can be represented by five unique sequences, GA1-5, where four GGG or GGGG stretches are connected by A or AAAA loops, Figure 1A. All oligo sequences but GA3 have two A loops and one AAAA loop; GA3 has two AAAA and one A loop. To improve homogeneity of the oligos and to increase our chances for crystallization, we have also explored variants of each GA1-5 oligo with different 5’- and/or 3’-overhangs, Table S1.

3.1. Biophysical studies demonstrate that GA sequences form stable, parallel GQs

We characterized GA sequences (Table 1) and their variants (Table S1) using circular dichroism (CD), UV-Vis melt, thermal difference spectra (TDS), and PAGE. Results for GA1-5 are shown in Figure 2 and S1; results for their variants are shown in Figures S2-S5. TDS shows that all GA sequences fold into GQs, as indicated by the characteristic trough at ~295 nm.⁴⁷ CD signatures reveal the parallel topology of GA GQs, indicated by the peak at ~263 nm and the trough at ~240 nm (Figure 2A). Similar CD signatures were observed for (GGA)_n (n = 4, 7, and 8) and (GGGA)₅ DNA.^20,21,48 Thermodynamic stability of GA GQs was measured via UV-vis melting studies. Representative melting curves are shown in Figure S6; results of data analysis are shown in Figure 2C and Table 1. In all cases, melting displays low hysteresis of 1 ± 1 °C, suggesting a reversible melting process and monomolecular structures. GA3 with two long AAAA loops displays the lowest T_m of 47.5 °C. The rest of the sequences with one AAAA loop display stabilities above 58.0 °C in 100K buffer. In line with our data, a G-rich sequence with three one-nucleotide A loops (i.e. no long loops) displayed stability of 82.2 °C in the presence of 50 mM KCl;⁴⁹ (GGA)₄ and (GGA)₈ displayed stability > 80 °C in the presence of 100 mM KCl.^20,21 Robust stability of all GA sequences was very encouraging for our crystallization studies.

Figure 2. GA1-5 sequences form stable parallel GQs in 100K buffer with and without NMM.

(A) CD scans collected at 20 °C. (B) CD scans for GA1 alone and with 2 eq. NMM collected at 20 °C. (C) Stability of GA1-5 defined by T_m values. (D) Stability of GA1-3 alone and in complex with 2 eq of NMM. All samples were prepared at ~4 μM DNA. (E) Fifteen percent native gel prepared in TBE buffer supplemented with 10 mM KCl. DNA samples were prepared at ~50 μM alone or in the presence of 2 eq. of NMM. Size markers correspond to dT_n sequences. On PAGE, some samples contained faint higher order species. These bands were not reproducible and may be related to sample age or storage conditions.

We utilized native PAGE to examine the purity, homogeneity, and oligomeric state of GA GQs. Representative PAGE is shown in Figure 2E for GA1-3 and in Figures S2D-S5D for other variants. GA1 displays one fast migrating band with some smearing around the dT₁₅ marker, indicating that GA1 is monomolecular. The smear likely suggests that GA1 folds into several conformations with similar topology. GA2 and its variants (see Figure S3D) display two bands, likely a monomer and a dimer. Finally, GA3 displays a single slower moving band around the dT₂₄ marker. The slower mobility of GA3 may be a result of dimerization or the presence of two flexible AAAA loops in a monomer. The reversibility of its thermal melting, Figure S6, argues for the monomolecular nature of GA3. GA4 and GA5 are less homogenous - GA4 forms three well-defined species likely including a monomer and a dimer, Figure S4D. The folding and homogeneity of GA5 is highly dependent on the presence of 5’- and 3’-overhangs and the number of species range from one to three–all displaying slower mobility, suggesting that GA5 is at least a dimer, Figure S5D.

3.2. NMM stabilizes GA GQs and preserves their parallel conformation

We characterized the interaction of three GA sequences, GA1-3, with NMM. We chose NMM because of its exceptional selectivity and preference for parallel GQ fold.^33,37 Addition of 2 eq. of NMM did not alter the parallel topology for any of the sequences (Figure 2B, S1B, and S7), confirming NMM’s preference for the parallel GQ fold. Similarly, native PAGE demonstrates that the addition of NMM only caused small change in the mobility of the major band for GA1-3, Figure 2E. Interestingly, NMM eliminated the higher order species seen for GA2 and its variants (Figure 2E and S3D) and increased homogeneity of GA1 (Figure 2E). Two equivalents of NMM stabilized GA1, GA2, and GA3 by 14.6, 13.1, and 7.0 °C, respectively (Figure 2D, Table 1). These values compare favorably with the ability of NMM to stabilize other parallel GQ structures.²⁹ Combining our biophysical data, we conclude that NMM preserves and stabilizes GA GQs and is an excellent candidate ligand to attempt crystallization of its complex with GA sequences.

3.3. UV-Vis and fluorescent titrations reveal moderate binding of NMM to GA1-3

We utilized UV-Vis and fluorescence titrations to investigate the binding affinity and stoichiometry of GA1-3 in complex with NMM (Figure 3, S8, and S9). We used 5K buffer in place of 100K so that we can compare binding constants to values for other GQ DNA published previously by our laboratory^33,37. In UV-Vis titration, the Soret band of NMM displays ~21 nm red shift and 19-28% hypochromicity (Table 3). While the red shift is similar to that observed for other DNA-NMM complexes, the hypochromicity is higher²⁹ and may indicate a more efficient π-π stacking interactions.

Figure 3. Determination of K_a for GA3-NMM complex via UV-vis and fluorescence titrations in 5K buffer.

(A) Representative UV-Vis titration of 3.3 μM NMM with 470 μM GA3 GQ to a final [DNA]/[NMM] of 13.6 at 20 °C. (B and D) Fit of titration data using 2DNA: 1NMM binding model with NMM concentration fixed to value measured via UV-vis. The 95% confidence interval is shown as dashed lines. Data points result from SVD analysis of UV-vis data presented in A or fluorescence data presented in C. (C) A representative fluorescence titration of 1.0 μM NMM with 170 μM GA3 to a final [DNA]/[NMM] of 21.6 at 20 °C.

Table 3.

Binding parameters for GA-NMM complexes obtained via UV-Vis and fluorescent titrations in 5K buffer. Graphical representation of this data can be found in Figure S10.

	GA1		GA2		GA3
Method	UV-vis	FL	UV-vis	FL	UV-vis	FL
Red Shift, nm	21.0 ± 0.5	-	21.3 ± 0.2	-	21.4 ± 0.4	-
Hypochromicity, %	20.9 ± 2.4	-	19 ± 4	-	28 ± 6	-
Isosbestic point, nm	395.6 ± 0.6	-	395.6 ± 0.3	-	397 ± 1	-
FL enhancement	-	27 ± 4	-	33 ± 16	-	27 ± 4
Ka, μM−1 (1: 1NMM)	0.19 ± 0.05	0.4 ± 0.2	1.4 ± 0.1	1.0 ± 0.2	0.24 ± 0.03	0.4 ± 0.2
Ka, μM−1 (1:2NMM)	0.07 ± 0.02	0.18 ± 0.07	-	0.36 ± 0.03	0.095 ± 0.01	0.2 ± 0.1
Ka, ÅM−1 (2:1NMM)	0.8 ± 0.2	1.1 ± 0.6	-	3.7 ± 1.4	0.96 ± 0.01	1.1 ± 0.8

To validate our UV-vis titration results, we conducted fluorescence titrations (see Figures 3C, S8 and S9) relying on the “light-switch” property of NMM (i.e., increased fluorescence in the presence of GQs). The titrations display fluorescence enhancement of ~30, lower than the previously reported values of 40-70 for a variety of parallel GQs³⁷ and could result from a weaker binding affinity or a shorter lifetime of the complexes.

Both UV-vis and fluorescence titrations were analyzed using SVD.^35-36 The preferred binding model and the values of K_a are presented in Table 3. The values of K_a agree well between the two methods. GA2, which binds the strongest among all constructs, displays K_a of 1-1.4 μM⁻¹ and demonstrates an unambiguous 1:1 binding stoichiometry in UV-vis experiments, but in fluorescent experiments all three binding models (1:1, 2:1, and 1:2) fit well. Similarly to the last case, the data for GA1 and GA3 fit well to all three binding models with the 1:1 model yielding slightly better fits for GA1. Therefore, all three K_as are reported in Table 3. Crystal structures of GA1-NMMM and GA3-NMM (Section 3.4.) display 1:1 binding stoichiometry, thus associated K_a is 0.2-0.4 μM⁻¹. Parallel GQs display great variation of their binding affinity for NMM with K_a range of 0.1-70 μM⁻¹ (Figure S10).²⁹ The association of GA1 and GA3 with NMM is at the lower end of this range, while association of GA2 falls in the middle. Interestingly, despite weaker association for GA1 and GA3, we acquired diffraction quality crystals of GA1-NMM and GA3-NMM complexes, but not GA2-NMM. In the latter case, we did obtain crystals, but their diffraction quality was low. Using binding constants from the UV-vis titrations, we estimated binding energies, ΔG₂₉₃, to be −7.1 ± 0.1, −8.2 ± 0.1, and −7.2 ± 0.1 kCal/mol, for GA1-3, respectively.

3.4. Crystal structure of GA1-NMM and GA3-NMM complexes

We successfully crystallized GA1-NMM and GA3-NMM complexes and solved the structures in space groups F222 and P22₁2₁ to 1.98 and 2.01 Å, respectively.

3.4.1. Overall architecture GA-NMM complexes

GA1-NMM and GA3-NMM complexes fold into parallel intramolecular GQs with three Hoogsteen-bonded G-tetrads and three propeller loops (Figures 4A and 5A). Every guanine in the structure participates in the formation of a G-tetrad and every adenine built forms a loop (note, 3’ adenines in the overhang are disordered and not observed). Both structures can be classified as VIII-1a parallel GQs with propeller loops according to da Silva,⁵⁰ or simply parallel GQ. One asymmetric unit (ASU) contains one GA monomer and one NMM ligand that stacks at the 3’ G-tetrad, yielding a 1:1 GQ:NMM ratio. Both GQs dimerize at the 5’ interface (Figure 4B and 5B) observed in other DNA-NMM structures.^30,31 Excitingly, we report for the first time that the 5’-5’ crystallization dimers contain seven G-tetrads, and not an even number, six in this case. Instead of dimerizing via π-π stacking of the terminal 5’ G-tetrads, the two monomers each contribute two guanines which form an interlocked G-tetrad, termed ‘symmetry-tetrad’ as its bases relate to each other via symmetry operations. Both structures are further stabilized by six potassium ions (per dimer) in the central channel. GA1 has a single adenine bulge at the 5’ end, while GA3 has a 5’-A overhang. Both structures contain highly flexible 3’-A overhangs whose bases were not built.

Figure 4. Crystal structure of GA1+NMM complex.

(A) Schematic representation of the folding topology. All bases are numbered, and arrows indicate chain progression. Blue and green rectangles indicate anti and syn conformations of guanine bases, respectively. The symmetry-generated bases are in grey. (B) Cartoon representation of the GA1 dimer with bases shown as filled rings and potassium and cobalt ions as purple and blue spheres, respectively. DNA strands from different ASUs are shown in different colors. NMM is shown in red. (C) GA1+NMM surrounded by electron density at I/σ = 1.0. DNA is colored by nucleotide with guanine in teal and adenine in orange. Due to high flexibility of loops, some bases have little electron density associated with them and some were not built. (D) Representative crystals of GA1-NMM.

Figure 5. Crystal structure of the GA3+NMM GQ.

(A) Schematic representation of the folding topology. (B) Cartoon representation of the GA3 dimer. Ammines are shown as yellow spheres. (C) GA3+NMM surrounded by electron density at I/σ = 1.0. (D) Representative crystals of GA3+NMM.

All modeled bases in GA1 adopt the anti glycosidic conformation save G₁ of the symmetry tetrad and loop nucleotides A₆ and A_17-18, which are syn. All modeled bases in the GA3 are anti, as expected for parallel GQs.⁵¹ Torsion angle analysis of the DNA backbone indicates a small number of outliers, predominantly associated with flexible adenines (Figure S11). The distances between tetrads are 3.37 ± 0.03 Å in GA1 and 3.39 ± 0.06 Å in GA3, see Table S3, in line with 3.4 ± 0.1 Å expected for parallel GQs, based on a survey of 310 crystallographically determined base-stacking geometries.⁵² The out-of-plane deviations are 0.65, 0.74, 0.80, and 1.83 Å for the symmetry tetrad and tetrads 1-3, respectively, in GA1. In GA3, the same tetrads have deviations of 1.02, 1.09, 1.39, and 1.82 Å, Table S3. When compared to other dimeric parallel GQ structures from our lab (Tel22-NMM, PDB ID 4FXM; T7-NMM, 6P45; and TET26, 6W9P, 7JKU, and 7LL0), the deviations for GA1-NMM are consistent, but middle and 5’ G-tetrads in GA3-NMM are significantly more distorted. Even the symmetry tetrad, which is expected to display the highest degree of planarity, deviates by 1.02 Å in GA3-NMM. We cannot explain this unexpected behavior by the high thermal motion of the GA3 GQ. The average B-factor for all guanines in GA1-NMM is 85 Ų and in GA3-NMM is 68 Ų (Figure S12) indicating that the latter structure is more ordered. On an absolute scale, the average B-factors for both GA1 and GA3 are somewhat high, leading to the similarly high R_free values compared to the resolution. As expected, the loops, overhangs, and bulges combined have higher average B-factors than the GQ core (121 vs 85 Ų for GA1 and 130 vs 68 Ų for GA3).

Parallel GQs are characterized by four medium groves. In both structures, grooves calculated using only traditional G-tetrads are 14.5 Å (Table S4), consistent with values in other parallel GQs.^30-32 The values calculated for the symmetry tetrad are somewhat smaller for GA1, 13.92 ± 0.09 Å and bimodal for GA3 with the values of 13.5 and ~16 Å, likely due to the unusual geometry of the symmetry tetrad (see Section 3.4.4).

3.4.2. Central ion channel

The central ion channel is highly important for GQ stabilization.⁵³ The GA1 and GA3 dimers were modeled with six K⁺ positioned in the central channel, Figures 4B and 5B. K⁺ ions are relatively equidistant from each other, with an average distance of 3.3 ± 0.1 Å and 3.4 ± 0.2 for GA1 and GA3 dimers, respectively, Figure S13. On average, the K-O distances are similar and equal to 2.7 ± 0.2 Å and 2.8 ±0.1 Å for GA1 and GA3, respectively. Both K-O and K-K values are similar to those observed in other parallel GQ structures.³² Because cobalt ions were introduced to the system through the crystallization condition (which contained 2-5 mM hexamine cobalt(III) chloride) and crystals did not grow in the absence of cobalt (we tested only for GA3-NMM), it is possible that Co³⁺ coordinates one or more G-tetrads. To obtain reliable data about cobalt locations, we collected the anomalous data and built anomalous maps for both structures, Figure S14. The anomalous maps clearly show that there is no cobalt in the central ion channel. In addition, Metal Binding Site Validation Server⁴⁶ verified that the likely identity of all metal ions in the central cavity is potassium based on the M-O distances and preferred coordination (square antiprism).

3.4.3. Loops and bulge in GA1 and GA3

GA1 has two A and one AAAA loop. GA3 has one A and two AAAA loops. In addition, the 5’-G₁A₂ in GA1 is arranged such that G₁ participates in a G-tetrad and A₂ points outwards, forming a bulge, Figure 6A. In both structures, the single-A loops are oriented outward; the AAAA loops are expected to display high conformational freedom. Resolved adenine nucleotides in the loops are stabilized by the π-π stacking with adenines belonging to the symmetry-generated partners.

Figure 6. Structural features involving adenine bases.

(A) A₂ bulge in GA1. The bulge adenine is green, G₁, G₃-G₅ are blue and other bases are grey. (B) Infinite π-π stack of A₁₀, A₁₆, and A₁₇ bases in GA1. Bases from different ASUs are colored in different colors. (C) Four-adenine propeller loop A₁₆-A₂₁ in GA3. Bases that form the π-π stack are shown in green; A₁₈ points outward and is shown in blue.

In GA1 there is an intricate infinite π-π stacking network that engages A₁₆ and A₁₇ from the A₁₄-A₁₇ loop as well as A₁₀ single nucleotide loop, Figure 6B. The network consists of A₁₀-A₁₀-A₁₇-A₁₆-A₁₆-A₁₇ motif that requires contributions from four different ASUs and repeats indefinitely. An individual ASU contributes A₁₀ and A_16-17 to two different infinite π-π stacks. The first two bases in the A₁₄-A₁₇ loop in GA1 were not built but are likely oriented outwards. Finally, the A₆ single nucleotide loop is located near the A₂’-bulge (note ‘ indicates symmetry-generated copy) but due to weak electron density for both A₂ and A₆, their probable interactions are not resolved. A₂, A₆, and A_14-15 display the highest B factors (Figure S12) due to their great flexibility and lack of stabilizing interactions.

In GA3, the single-nucleotide A₁₃ loop and the second A₁₇-A₂₀ loop were fully modeled. The bases of all adenines in the first loop (A₅-A₈) were not built due to a lack of DNA/ligand interactions and high conformational freedom of this loop. The A₁₇-A₂₀ loop is arranged such that three adenines form a π-π stack of A₁₇-A₂₀-A₁₉ that is nestled in the cavity of the loop, Figure 6C. Interestingly, A₁₉ is additionally stabilized by stacking with NMM from a symmetry-generated molecule (see Section 3.4.6.), A₁₈ of the A₁₇-A₂₀ loop points outwards and was modeled to stack with A₁₃’ loop and likely with A₈’ from the A₅-A₈ loop. However, the density for A₁₃ and A₁₈ is weak, and the A₈ nucleotide was not fully built.

In addition to π-π stacking, the loops are further stabilized by Co³⁺and K⁺ ions. In GA1, Co³⁺ coordinates to the phosphates of A₁₃ and A₁₇ from the same monomer as well as A₁₃’ and A₁₇’ of the monomer in another ASU. In GA3, a hydrated K⁺ ion at the symmetry position stabilizes the loops by interacting with phosphates of A₆ nucleotides from two different ASU.

3.4.4. GA1 and GA3 form dimers through a unique 5’ symmetry tetrad

5’-5’ crystallization dimers traditionally contain an even number of tetrads. Here we report the first 5’-5’ crystallization dimers with an odd number of tetrads. The odd number originates from the unique G-tetrad formed at the dimer interface and named here the ‘symmetry tetrad’. This tetrad consists of two nucleotides from one monomer interlocking with the symmetry-generated nucleotides from another monomer (Figure 7). The two nucleotides are positioned adjacent to each other in the GA1 structure (Figure 7A) and diagonal from each other in GA3 (Figure 7B). The observed difference originates from the location of participating guanines. Both sequences have a total of 14 guanines: 12 participate in the formation of traditional G-tetrads, and two contribute to the symmetry tetrad. In GA1 the symmetry tetrad is formed by G₁ from the 5’-overhang and G₁₈, the first guanine in the only four-nucleotide G-stretch. In GA3, both are first guanines in the four-nucleotide G-stretches, G₉ and G₂₁. The parallel GQ conformation, dimerization via 5’-interface, and location of those guanines dictate their adjacent or diagonal position. Both symmetry tetrads are nearly identical (RMSD for their bases is below 0.4 Å) (Figure S15D) but are oriented differently in both structures, Figure S15B. Very interestingly, in GA3, the nucleotides of the symmetry tetrad are stabilized by hexamminecobalt(III) which bonds directly to G₂₁ and G₉’ (hydrogen bonding distance is ~2.6 Å between amine and phosphate). This essential interaction may explain why GA3-NMM crystals did not grow in the absence of hexamminecobalt(III). In addition, hexammine cobalt(III) interacts directly with a water molecule and with G3' (3.69 Å) and A17' (3.82 Å) either directly or via an unbuilt water molecule.

Figure 7. Symmetry tetrad in (A) GA1 and (B) GA3.

Nucleotides in different ASU are colored in blue and green, potassium is a purple sphere. Traditional G-tetrad is colored in gray and is part of the ‘green’ ASU. (C) Schematics of the symmetry tetrad and a traditional G-tetrad. In the latter, guanines are held together by eight Hoogsteen hydrogen bonds.

Symmetry tetrad differs greatly from the traditional G-tetrad where the Watson-Crick face of one base hydrogen bonds to the Hoogsteen face of the adjacent base, relating the guanine bases through rotational symmetry, Figure 7C. The symmetry tetrad is, instead, governed by reflective symmetry because it contains guanines from two different ASUs. Therefore, its bases cannot be manipulated independently and changing the orientation of one base flips the orientation of its symmetry-generated partner. Adjacent location of two guanines in GA1 ASU can theoretically allow hydrogen bonding interactions between them. However, the observed electron density dictates the coordination of the bases in the non hydrogen bonding conformation, Figure 7A. In GA3, the symmetry relations and diagonal positioning of the bases preclude hydrogen bonding between bases in the symmetry tetrad, Figure 7B.

To verify the correctness of our model we removed the symmetry relations that force guanines into observed configuration by reprocessing the data for GA3-NMM in the lower space group (P12₁1, Figure S16). The lower space group yielded two copies of DNA in the ASU, dimerized via the 5’ interface tetrad with all bases independent from each other. We manipulated guanines such as to maximize hydrogen bonding (Figure S16B-D), however, the observed electron density dictated the orientation of guanines observed in our original structure, Figure 7B.

The traditional G-tetrad is a robust building block because its bases are engaged in four hydrogen bonds, two on their Watson-Crick side and two on their Hoogsteen side, leading to eight hydrogen bonds in total, Figure 7C. Bases in non-guanine tetrads, such as cytosine,¹⁴ adenine,^13-15 and uracil tetrads¹³ (non-G tetrads were recently reviewed⁵⁴) are connected by a single hydrogen bond, leading to four hydrogen bonds in total and lower stability of the arrangements. The lack of hydrogen bonding in our symmetry tetrad questions its validity, but its existence in two different structures supports its credibility. In addition, a uracil tetrad was reported without any hydrogen bonds between its bases. Even more so, this tetrad did not benefit from the π-π stacking interactions and was stabilized only by the interaction with Ba²⁺ and Na⁺ ions on both sides.¹³ We observe similar interactions between the symmetry tetrads and two K⁺, Figure S13. In addition, our symmetry tetrad is stabilized on each side by π-π stacking with traditional G-tetrads. There is efficient 5-ring overlap and partial 5/6-ring overlap (the latter is only present in GA3), Figure 7A-B, commonly observed at other dimeric interfaces.⁵²

Additional support for the validity of our symmetry tetrad comes from the molecular dynamic study.⁵⁵ The study indicates that electrostatic interactions, particularly those minimizing repulsion of sugar-phosphate backbone and not hydrogen bonding interactions, play a key role in stabilization of G-tetrads and GQs. Minimization of repulsion in GA-NMM structures is accomplished via right-handed rotation of G-tetrads. We observe helical twist between traditional 1-2 and 2-3 G-tetrads in GA1 to be 32 ± 2° and 28 ± 2°, respectively, and 30.6 ± 0.4° and 28.2 ± 0.6°, respectively, in GA3 (Table S5). These values are within the reported range for parallel GQs.⁵⁶

Literature search yielded examples of interlocked dimers where the interlocked tetrad is formed by the three bases from one monomer and one base from another. Examples include the NMR structure of GGGAGGTTTGGGAT,¹⁷ HIV aptamer,¹⁸ and DNA sequences with short G2 stretches.⁵⁷ In all examples, the interlocked tetrad adopts configuration of a traditional G-tetrad. The configuration of the dimer interface reported in this work is seen for the first time.

3.4.5. Biological relevance of the crystallization dimer

To confirm the presence of the symmetry tetrad in dilute conditions (< 50 μM DNA, 100K buffer), we generated mutants with a disrupted or eliminated symmetry tetrad, Table S7. Specifically, we either mutated G₁/G₁₈ in GA1 or G₉/G₂₁ in GA3 to adenine (M mutants) or deleted these bases (D mutants). In addition, we attempted to disrupt the symmetry tetrad by forcing its guanines to participate in the formation of a traditional G-tetrad. For example, G₂₁ from the symmetry tetrad in GA3 is part of the G₂₁-G₂₄ four guanine stretch, so deleting G₂₄ (and for simplicity also A₂₅) will lead to a three-guanine stretch that includes G₂₁ where all guanines are needed to form a traditional G-tetrad.

We characterized all mutants, in complex with NMM, via biophysical methods and PAGE, Figure S17-18. For GA1-NMM, the mutant data indicate that removal of the symmetry tetrad has little effect on the GQ fold, stability, or PAGE mobility suggesting that the symmetry tetrad is likely a result of crystal packing. Bases involved in the symmetry tetrad in the crystalline state, in dilute solution could form part of the 5’-overhang (G₁) and extend the A₁₄-A₁₇ loop by an extra G₁₈.

The data for GA3-NMM are a little more complicated, but the take-home-message is the same: the symmetry tetrad results from crystal packing, and under dilute conditions G₉ and G₂₁ likely form part of two long loops (now AAAAG loop vs AAAA loop in the crystal structure). PAGE data in Figure 2E indicate that GA3-NMM may form a dimer in dilute solutions. However, the mobility of the mutants is largely unchanged as compared to GA3-NMM, suggesting that retardation of the GA3-NMM band is likely due to the presence of two (and not one) long and flexible AAAA loops and not due to dimerization.

The final piece of evidence that dimerization results from the crystal packing forces comes from the native PAGE where dilute DNA samples were compared to the crystallization and crystal samples, Figure 8. All GA1-NMM samples migrated at the same rate, indicating that the dimer observed in the crystal is unstable and dissociates during electrophoresis. Very interestingly, the same data for GA3-NMM demonstrates that dimerization indeed occurs in the crystal but not in the crystallization or dilute sample. The PAGE data allow us to visualize the GA3-NMM dimer present in the crystalline state and confirm that slower mobility of GA3-NMM as compared to GA1-NMM is due to the presence of two long loops in the former.

Figure 8. Demonstration of dimer formation in GA3-NMM via native PAGE.

Fifteen percent gel was prepared in 1 × TBE supplemented with 10 mM KCl. Size markers correspond to dT_n sequences. The DNA samples were prepared the following way: lane 1 – concentrated DNA sample used to grow crystals deposited in lane 3; lane 2 - 50 μM DNA in 100K buffer; lane 3 – crystals washed and dissolved in 100K or 10K buffer. The light smears observed in lanes 1 are likely due to high concentration of DNA that causes formation of small amount of higher order species.

3.4.6. NMM binding facilitates intermolecular assembly in GA1 and GA3

In both GA1-NMM and GA3-NMM structures, the porphyrin ring of NMM has clearly defined electron density. The N-methyl group of NMM points into the ion channel. However, due to resolution constraints, there is no density beyond the first carbon for all peripheral substituents, Figure 9. As a result, we are unable to determine the exact orientation of NMM and to clarify whether each of its four isomers (with N-Me group at each of the pyrrole nitrogens, see Figure 1) interacts with GA DNA. Similar situation was observed in our previous high resolution structure of Tel22-NMM, although this structure contains a very clear density for the N-Me group.³⁰ NMM alone or bound to GQ DNA is non-planar with the N-Me bearing pyrrole being distorted out of plane toward 3’ G-tetrad.^30-31 This non-planar deviation breaks the aromaticity of the NMM macrocycle and allows for a selective interaction with a GQ as we discussed earlier.³⁰ The pyrrole is bent out-of-plane by 44.8° in Tel22-NMM,³⁰ 43.2° in GA1-NMM and 30.6° in GA3-NMM.

Figure 9. NMM binding pocket in GA1-NMM.

(A) Monomers from two different ASU do not align (as indicated by the position of their ion channels) due to offset in NMM-NMM interaction. For clarity, loops and 3’-overhangs have been removed. (B) Top- and side-view of overlap between NMM and a 3’ G-tetrad. Electron density for NMM is displayed at I/σ = 1.0. (C) Top- and side-view of NMM dimer.

NMM interacts with the 3’ G-tetrad of each GA1 and GA3 as was observed for other GQ-NMM structures.^30,31 The distance between NMM and the 3’ G-tetrad is 3.62 Å in GA1 and 3.54 Å in GA3, similar to distances observed for other planar ligands bound to human telomeric DNA (3.4-3.6 Å for berberine, metal–salphen, naphthalene diimide, and NMM).^30,58-60 Unlike other cases, in GA1, the other face of NMM stacks with a symmetry-generated copy of itself, forming a novel NMM dimer, Figure 9. The offset of the NMM-NMM dimer precluded an accurate centroid-based measurement of distance. However, there are significant stacking distances ranging from 3.5-3.8 Å, Table S6. This dimerization aids in formation of an infinite assembly of GQ dimers linked to each other via NMM dimers, Figure S19. NMM molecules in the NMM dimer are misaligned such that only the planar portions of ligands stack, leading to an efficient overlap of only one pyrrole in each molecule, Figure 9C. This offset, coupled with the central location of the N-Me group, leads to misalignment of two GQ monomers connected to the same NMM dimer, Figure 9A.

In GA3-NMM complex, NMM stacks with A₁₉ from a symmetry-generated molecule (Figure 10A-B), leading to intermolecular assembly of GQ dimers (Figure S19B). Interaction of NMM with 3’ G-tetrad on one side and with a base or base pair on another is observed in two other parallel GQ-NMM structures. Specifically, NMM bound to (TGGGT)₄ GQ interacts with a thymine from a propeller loop³¹ (Figure 10C) and bound to the human telomeric GQ, interacts with a reverse Watson-Crick base pair (Figure S10D).³⁰ Note, in all three cases, there is no significant overlap between π character of NMM and that of the base or base pair, although NMM-A₁₉ distance is short (3.25 Å) and consistent with distances of other NMM-base interactions, Table S6.

Figure 10. NMM binding pockets in GA3-NMM, Tel22-NMM, and (TGGGT)₄-NMM crystal structures.

(A) Association of two GA3 molecules via their interaction with NMM. For clarity, loops and 3’-overhangs have been removed. (left) Top- and (right) side-view of overlap between NMM and a loop nucleotide in GA3 (B), (TGGGT)₄ (PDB ID 6P45) (C), and Tel22-NMM (PDB ID 4G0F) (D). Adenines are colored green, guanines are blue, thymines are teal, phosphates are yellow, and NMM is red.

4. Summary

In this work, we carried out extensive biophysical and X-ray crystallographic characterization of five variants of the G-rich homopurine sequence found on chromosome 7 and other human chromosomes. Each sequence contains four GGG/GGGG guanine stretches connected by A or AAAA loops. We demonstrated that all sequences fold into parallel GQs with stability > 59 °C for sequences with one AAAA loop and 47.5 °C for GA3 which contains two AAAA loops. We characterized GA1-3 in the presence of NMM and showed that NMM does not change the conformation of these GQs but stabilizes them by 7-15 °C. While GA1 and GA3 bind NMM modestly with K_a of ~0.3 μM⁻¹, both produce diffraction quality crystals in the presence of NMM. Two more structures of GQ-NMM complex exist, one of Tel22-NMM that is characterized by even lower K_a of ~0.1 μM⁻¹,³³ and another of T1-NMM with an exceptionally high K_a of 50 μM⁻¹,³¹ Figure S10. Unfortunately, our crystallography work did not help answer the question about structural features which govern the strength of GQ binding to NMM. Likely, higher resolution structures would be necessary to do so.

We observed two novel structural features. The first is a symmetry tetrad present in both GA1 and GA3. The tetrad is formed by two G-bases from each monomer, which relate to each other via symmetry operations. These tetrads result from the crystal packing forces. In dilute conditions, the bases that form the symmetry tetrad likely form overhangs or loops. Nevertheless, the observation of the symmetry tetrad demonstrates a novel way in which GQs interact and adopt a more stable structural arrangement. The second feature is a NMM dimer seen in GA1-NMM. In all published structures to date^30,31 (and here), NMM interacts with GQs via 3’ G-tetrad on its one face and uses its N-Me group as a “key” to the tetrad’s ion channel “lock”, leading to a very selective binding. The other face of NMM typically stacks with adenine (GA3-NMM), thymine³¹ or T-A reverse Watson-Crick base-pair³⁰, but in GA1-NMM, the ligand interacts with itself, producing the NMM dimer. We have suspected dimerization of NMM based on some of our biophysical work for both NMM alone and in complex with GQs, but we observed the dimer structurally for the first time here.

Purine rich sequences (including homopurine sequences) are widely distributed throughout the human genome and are located near gene promoters, in telomeres, centromeres, triplet repeat disease sequences, or recombination hotspot sites. Similarly, purine rich RNA has biological importance. While typically in its double-stranded form, biological processes (e.g. replication) can lead to local unfolding of the double-stranded DNA freeing the purine rich strand which may adopt transient non-canonical structure with a regulatory function. Understanding the conformation of such non-canonical structures will allow us to learn more about their biological roles as well as design selective ligands to control their presence and stability. Guanines are more likely to form G-tetrad due to robustness of such a structural feature. On the other hand, adenines are more flexible. When investigating G-rich purine-only sequences, we expected to see A-tetrads, GA-tetrads, or pentads, hexads or heptads with GA sheared base pairs in combination with traditional G-tetrads. However, we observed that all guanines participate in G-tetrad formation, while all adenines form one- and four-nucleotides loops. All short loops point away from the GQ core as is typical for one-nucleotide loops. The adenines in AAAA loops participate in stabilizing stacking interactions with adenines from the symmetry-generated monomers which, in the GA1 case, leads to the extended crystal packing. Thus, our work confirmed great flexibility of adenines in the non-canonical DNA structures. Loops, grooves, and G-tetrads can all be sites of protein binding in the biological function of the specific purine-rich sequences. In addition, or alternatively, these regions can bind small molecule ligands that can control their presence and stability. A well-known GQ-binding ligand NMM binds via 3’ terminal G-tetrad to homopurine GQs. This binding mode was expected based on the previously published crystal structures of NMM in complex with Tel22³⁰ or (TGGGT)₄³¹. The work presented in this paper expands our knowledge about potential structures adopted by homopurinic sequences, GQ diversity, and ligand binding.

Highlights.

• Homopurine sequence found on human chromosome 7 folds into parallel G-quadruplexes

• NMM, a small-molecule ligand, binds and stabilizes these GQs

• GQ-NMM structures reveal a unique symmetry tetrad at the dimer interface

• The symmetry tetrad is composed of two Gs from each monomer

• NMM forms a dimer that binds to the 3’ G-tetrad of two GQ monomers on each side

• Adenines form loops stabilized via extensive π-π stacking