G-quadruplex binding properties of a potent PARP-1 inhibitor derived from 7-azaindole-1-carboxamide

Poly ADP-ribose polymerases (PARP) are key proteins involved in DNA repair, maintenance as well as regulation of programmed cell death. For this reason they are important therapeutic targets for cancer treatment. Recent studies have revealed a close interplay between PARP1 recruitment and G-quadruplex stabilization, showing that PARP enzymes are activated upon treatment with a G4 ligand. In this work the DNA binding properties of a PARP-1 inhibitor derived from 7-azaindole-1-carboxamide, (2-[6-(4-pyrrolidin-1-ylmethyl-phenyl)-pyrrolo[2,3-b]pyridin-1-yl]-acetamide, compound 1) with model duplex and quadruplex DNA oligomers were studied by NMR, CD, fluorescence and molecular modelling. We provide evidence that compound 1 is a strong G-quadruplex binder. In addition we provide molecular details of the interaction of compound 1 with two model G-quadruplex structures: the single repeat of human telomeres, d(TTAGGGT)4, and the c-MYC promoter Pu22 sequence. The formation of defined and strong complexes with G-quadruplex models suggests a dual G4 stabilization/PARP inhibition mechanism of action for compound 1 and provides the molecular bases of its therapeutic potential.

www.nature.com/scientificreports/ showed a significant inhibition of the enzyme and ability to bypass the multidrug resistance mediated by Pgp.
In antitumor activity studies, Compound 1 exhibited in vivo higher activity against the MX1 human breast carcinoma growth in nude mice than the reference compound Olaparib in terms of tumor volume inhibition. Treatment was well tolerated, as all the treated animals survived without significant weight losses 22 .
In this paper we report a fluorescence, CD, NMR and molecular modelling study focused on the interaction of 1 with the single repeat sequence of human telomeres, d(TTA GGG T) 4 23 , and with the G-quadruplex found in the c-MYC promoter Pu22 sequence, whose overexpression is one of the most common aberration in a wide range of human tumors. In addition, the interaction with duplex oligonucleotides, d(CGT ACG ) 2 and with d(AAG AAT TCTT) 2 was also investigated by 1 H and 31 P NMR spectroscopy.

Results and discussion 1 H and 31 P NMR of 1 with double helix B-DNA d(CGT ACG ) 2 and d(AAG AAT TCTT) 2 . The NMR
spectra of both the self-complementary oligonucleotides d(CGT ACG ) 2 ("CG") and d(AAG AAT TCTT) 2 ("AATT") display signals in a region ranging from 12 to-13.5 ppm, which are characteristic of the NH imino protons of CG and AT base pairs. The presence of these signals confirms that both oligomers adopt, in Na + solution, a double helix conformation. For this reason, they were used as models for CG-and AT-rich sequences, respectively.
The phosphorus spectra of "CG" with 1 showed a low-field shift variation of G2pT3, T3pA4 and C5pG6 signals ( Fig. 2a; Table 1). It is known that 31P resonance is a sensitive probe to detect changes in the phosphoribose chain of the oligonucleotides due to the intercalation process; in fact, the chemical shift variation of the 31 P resonances reflects a deformation at the level of P-O(5′) and P-O(3′) bonds. As a consequence of the intercalation of a ligand into the oligonucleotide double helix, the alfa = O(3′)-P-O(5′)-C(5′) and zeta = C(3′)-O(3′)-P-O(5′) angles change from a gauche, gauche conformation (− 60° and − 90°) to a gauche, trans conformation (− 60°, + 180°). This conformational change is usually associated with a low-field shift up to 1.0-2.5 ppm for 31 P resonances 24,25 . In our case the lower Δδ values (− 0.2 ppm) found for G2pT3, T3pA4 and C5pG6 suggest that 1 is bound to  www.nature.com/scientificreports/ the oligonucleotide, however through a partial intercalation binding mode, at these three sites, with a possible exchange among them. The addition of 1 to "AATT" induced insignificant chemical shift variation of the phosphate signals in the 31 P NMR spectra ( Fig. 2b; Table 1). This is the proof that an intercalation process did not occur. Nevertheless, the A1pA2, A5pT6 and A4pA5 signals became very broad even at R = 0.25. The 1 H NMR titration with 1 gave different results for the two oligonucleotides. Broadening of the imino NH signals with no relevant chemical shift variation was observed for "AATT" resonances, whereas only the H1′ anomeric protons belonging to "AATT" tract were slightly perturbed (Δδ = −028/ −0.11 ppm) (Table S1). A line broadening was observed for the aromatic protons of A5 and T6, while the H2 proton of A4, which lies in the minor groove, splitted into two signals just at R = 0.25. This suggests the formation of both a free and a bound species; for R ≥ 1.0 the free species disappears ( Figure S1). The addition of 1 to a solution of "CG" presented a significant up-field chemical shift variation either for imino, aromatic or anomeric protons ( Figure S2). The most affected were the protons of the G2 and C5 units, in line with the interaction at these sites. The unique exception was represented by the T3 anomeric proton (Δδ + 0.19 ppm) (Table S1). A possible explanation of this deshielding effect is that the intercalation of the ligand at G2pT3 places the piperidine moiety in the minor groove at the level of T3.
2D-NOESY experiments did not show intermolecular Nuclear Overhauser Effect (NOE) contacts between the ligand and both oligonucleotides, probably due to a weak interaction or to a rapid exchange between different binding sites. Therefore, it was not possible to build a model for the complexes with 1.
Overall, these results show that 1 partially intercalates with a CG-rich sequence, whereas it gives a slight external interaction with an AT-rich sequence. In conclusion, the interaction of 1 with double helix oligonucleotides can be considered not relevant.
However, these findings cannot rule out an interaction with G-quadruplex DNA structures, whose stabilization has been found to activate PARP-1 enzyme 14 . Thus, we focused our study on the interaction of compound 1 with G-quadruplex structures of telomeres and proto-oncogenes.
Interaction of 1 with telomere d(TTA GGG T) 4 quadruplex. The titration of the oligonucleotide solution with 1 induced significant line broadening of G4 NH signal and less pronounced broadening of the aromatic resonances of the three guanines (Fig. 3a). An upfield shift variation was observed for the imino, aromatic and anomeric proton signals of all units, except for T7 (Table S2). Also the resonances of the ligand gave an upfield shift. They were partially overlapped to the signals of the oligonucleotide, however they were identified by a TOCSY experiment at 7.02 and 7.49 ppm (pyrrole moiety), at 7.09 and 7.38 ppm (phenyl and pyridine moieties). NOE contacts were found for A3 H8 and G4 NH with the pyrrole protons. Other NOE interactions involved NH and H8 of G6 with both pyrrole and aromatic protons of the ligand (Table 2) (Fig. 3b). These results indicate that 1 binds to the G-quadruplex at two sites: between A3 and G4 and over G6.
The interaction of 1 with the quadruplex d(TTA GGG T) 4 was also investigated using the molecular docking technique, followed by optimization via molecular dynamics (MD). The molecule was docked at sites A3-G4 and G6-T7. In both cases the ligand does not adopt a center-symmetrical stacking interaction with the upper and lower tetrads, but it is rather displaced towards two of the four residues ( Fig. 4).
At the AG intercalation site the 7-azaindole portion of the ligand is inserted between A3 and G4. In this intercalation site the complex is also stabilized by a hydrogen bond between the NH 2 group of A3 and the aromatic nitrogen of the pyrrole [2,3-b]pyridine moiety, at a distance of 2.98 Å. The piperidine ring is oriented outward from the quadruplex, with the quaternary nitrogen forming a hydrogen bond with N 7 of A3 (2.63 Å), a cation-π interaction with G4, and a ionic interaction with OP 2 A3 (Fig. 4B).
At the GT intercalation site, the 7-azaindole moiety of the ligand gives rise to π-π stacking interactions with G6 and T7. In this case, the complex is stabilized by a hydrogen bond between O 4 T7 and the CONH 2 group of the ligand, at a distance of 2.36 Å. The -CONH2 group itself is locked in position by an intra-molecular hydrogen Table 1. 31 P chemical shift assignments of phosphate in the free oligonucleotides and in the complex with 1 a,b . Oligonucleotide "CG" corresponds to duplex d(CGT ACG ) 2   www.nature.com/scientificreports/ bond with the pyridine nitrogen (2.24 Å). Again, the piperidine ring points outward from the quadruplex, with the quaternary nitrogen forming a hydrogen bond with N 3 of G6 (2.58A) and an ionic bond with OP 1 T7 (Fig. 4C).
The best docked conformations of the complexes at the A3G4 and G6T7 intercalation sites are in good agreement with the reported NOE contacts ( Table 2).

Interaction of 1 with G-quadruplex Pu22T14T23 sequence.
Pu22T14T23 gave high quality spectra in K + solution in comparison with wild-type sequence 17,26 . For this reason, it has been chosen as a good model to study ligand-quadruplex interaction. 1 H NMR titration experiments were performed by adding increasing amounts of ligand to Pu22T14T23 solution, with ratios R = [ligand]/[DNA] ranging from 0 to 2.5. The NMR data indicated a single G-quadruplex conformation for the complex, with each proton showing a single chemical shift value and NOEs characteristic of the three G-tetrad stacked structure (Table S3 and Table S4). The titration experiment proved a chemical shift perturbation of the imino protons (Fig. 5). The largest perturbations were observed for G18 and G22 (3′-end) and for G7, G11, G16 and G20 (5′-end). Less relevant perturbations were  www.nature.com/scientificreports/ observed for the imino protons of G8, G12, G17 and G21, in the central guanine tetrad. No relevant chemical shift variation was observed for the residues located in the loops of the G-quadruplex, such as A15, T14 and T19 (Table S3). This excludes the interaction of 1 with the groove. These findings suggest that 1 stacks on the 3′ and 5′ sites of Pu22T14T23. NOE contacts were found between the ligand and the imino NH protons of G13, G18 (3′-end), G7, G16 and G20 (5′-end) of the quadruplex (Fig. 6). The aromatic protons of the ligand could not be unambiguously assigned because of their overlapping. However, the resonances of H2 and H3 of the pyrrole moiety were identified by TOCSY experiments at 7.49 and 7.04 ppm, respectively. These protons show NOE contacts with the imino NH of G16 and G18 (H3) and with G7 and G20 (H2) ( Table 3). In addition an NOE interaction was detected between H8G5 and H3 of pyrrole moiety. Other resonances of the ligand were identified at 7.38 and 7.55 ppm, showing a NOE interaction with NH G13 of the quadruplex. No NOEs between the ligand and the flanking chains were detected.
In order to obtain a three-dimensional model congruent with the NOE contacts we performed a molecular docking experiment, followed by MD calculations (Fig. 7). At 5′-end, 1 is stabilized by an extensive network of π-π interactions involving the underlying 5′-end G-tetrad, with the pyrrole[2,3-b]pyridine moiety located near the center of the tetrad (Fig. 7a, b). The above cited aromatic rings interact with the π systems of G5, G11 and G16. They are held in place by the cation-π interaction between the potassium ion and the pyrrole moiety (4.96 Å). Also the central phenyl ring creates π-π interactions with G11 and G16, while the piperidine ring is oriented outside the system, towards G11, T14 and A15, without giving rise to observable interactions. The -CONH2 group is coplanar with the pyridine aromatic system and positioned above G16, without giving noteworthy interactions with Pu22.
At 3′-end, the complex is also stabilized by a dense network of π-π interactions involving all the guanines of the tetrad: specifically, the pyrrole[2,3-b]pyridine moiety interacts with the π systems of G18 and G13 (Fig. 7c,  d). The phenyl group of 1 forms π-π interactions with the G13 unit as well. This is complemented by a further π-π interaction with the G9 aromatic system. The CONH 2 group is oriented towards G22 and T23, forming two www.nature.com/scientificreports/ hydrogen bonds: one with O6G22 (2.84 Å) and the other with O4T23 (3.02 Å). The piperidine ring is oriented outside the system, in the area between G9, G13 and A25, and does not present particular interactions with Pu22. In both 3′-end and 5′-end positions, the piperidine ring is arranged along the main groove of Pu22, with a docking score difference in favour of the complex in 3′ (ΔE = 5.77 kcal/mol). The best docked conformations of the complexes at 5′ and 3′ are in good agreement with the reported NOE contacts ( Table 3). The flanking chains at the two terminals in the complex, even though no NOE interactions with the ligand were detected, showed significant chemical shift variations, especially in the segment T23-A24-A25 at the 3′-end. Many resonances move upfield: T23 Me (Δδ = −0.28), T23 H6 and H1′ (Δδ = −0.11 and −0.30); H8 of A24 and A25 (Δδ = −0.38 and − 0.32). Other resonances move low-field, the most significant being the anomeric protons H1′′of A25 and A24 (Δδ + 0.57 and + 0.25, respectively), and a slight deshielding of G9 NH (Δδ + 0.10). This finding suggests that the three units T23, A24 and A25 do not prevent the binding. The ligand, positioned on the 3′-end tetrad, changes the architecture of the tail of this terminal. Actually, the structure of the free nucleotide Pu22T14T23 shows that A25 folds back to form a base pair with T23, thus protecting the external 3′-end G-tetrad. The entry of the ligand breaks the Hoogsteen-type H-bond between T23 and A25 27 , pushing away T23 toward the top of G22, thus experiencing the stacking effect of this guanine. This is highlighted by the up-field shift of T23 protons, being also in line with the model. The A25 unit is no more folded over the G9 aromatic moiety, as in the free nucleotide 27 . This justifies the deshielding of G9 NH and of the anomeric H1′ protons of both A25 and A24. The aromatic portion of these units is slightly shielded. This may be explained with the increased flexibility of the tail. At 5′-end the ligand induces small conformational changes as shown by the slightly low-field shift of the T4, G5 and A6 aromatic protons (Δδ = +0.09, − 0.15 ppm), which indicates that also this tail is pushed away from the tetrad. Similar experiments were also performed using ABT-888 as a ligand, in order to find possible interactions with Pu22T14T23 and to compare the results with those above described for 1. The titration of Pu22T14T23 with ABT888 did not show any chemical shift variation ( Figure S3) and no NOE contact was found in the NOESY spectra. Consequently, we must conclude that the Pu22T14T23 quadruplex of the c-MYC promoter is not a target for ABT-888.
CD and fluorescence studies. Both DNA G-quadruplex sequences, d(TTA GGG T) 4 and Pu22T14T23, used for NMR investigation, form stable parallel G-quadruplex structures at the experimental conditions used for CD and fluorescence studies (Fig. 8). The intramolecular G-quadruplex formed by Pu22T14T23 showed a  www.nature.com/scientificreports/ Tm value around 85 °C ( Figure S4), which indicated a high thermal stability of this structure. On the other hand, the human telomeric sequence formed an intermolecular G-quadruplex structure. Therefore, the kinetics of the folding/unfolding process depends so strongly on DNA concentration that determination of the right Tm value would need an extremely small heating rate, due to the presence of hysteresis. Consequently, the midpoint of the transition determined in the conditions used in this work should be named as T 1/2 28 , being its value 50 °C for d(TTA GGG T) 4 .
CD-monitored titrations of both G-quadruplexes with the ligand 1 were carried out to determine potential structural and, global changes due to the interaction. Titrations were carried out at 15 °C, where both G-quadruplex structures are the major species. No clear changes were observed that could be related to dramatic structural modifications of the G-quadruplex (Fig. 8). Upon addition of the ligand, no variations could be observed in Tm value of the G-quadruplex formed by Pu22T14T23, in agreement with the high stability of this structure.   www.nature.com/scientificreports/ The fluorescence spectra of ligand 1 in presence of both G-quadruplex sequences are shown in Figure S5. The ligand showed strong fluorescence in potassium phosphate buffer solution without oligonucleotides. The addition of the G-quadruplex structures induced a decrease of the fluorescence signal intensity. From the titration curves, an estimation of the stoichiometry and the binding constants (Kb) relative to the interaction were done with the EQUISPEC program, which is based on the multivariate analysis of the whole spectra measured along the titration.
In both cases, the Kb values obtained for the interaction of ligand 1 with Pu22T14T23 and d(TTA GGG T) 4 considering a 1:1 stoichiometry are in the order of 10 6.1 M −1 , which suggests a relatively strong interaction between this ligand and both structures. It is known that forward titrations of a ligand with increasing concentrations of DNA favours the formation and detection of the 1:1 complex, whereas reverse titrations favour the formation of complexes with higher stoichiometries 29 . In a similar way to the NMR studies, an additional reverse titration was carried out, where the Pu22T14T23 sequence was titrated with ligand 1 (Figure S6). In this case, a 1:2 (DNA:ligand) stoichiometry (overall Kb value equal to 10 12.8 M −2 ) fitted better the experimental fluorescence data than the 1:1 stoichiometry.
Alternatively, titration curves were studied in PBS buffer obtaining a Kb value for Pu22T14T23 similar to that obtained in K buffer. However, the binding constant for d(TTA GGG ) 4 is in the order of 10 5.1 M −1 ( Figure S7).

Conclusions
In this study we have investigated the specific DNA binding mode of a PARP-1 inhibitor (1) derived from 7-azaindole-1-carboxamide. The results provide evidence that 1 lacks the typical features of DNA intercalators. In fact, the interaction with a model of duplex DNA, as d(CGT ACG ) 2 , showed a partial intercalation mode at three different sites, whereas there is an external slight interaction with d( AAG AAT TCTT) 2, a model of AT-rich sequence.
On the contrary, 1 binds both the G-quadruplexes, d(TTA GGG T) 4, a model of human telomere sequence, and Pu22T14T23, a model of the c-MYC promoter Pu22 sequence. In the first case, 1 is located between A3 and G4 units and over the G6 residue, while it forms a 2:1 complex with Pu22 quadruplex, with the two molecules located over the external tetrads at 5′ and 3′-end. In both cases, the Kb values obtained for the interaction of ligand 1 with Pu22T14T23 and d(TTA GGG T) 4 are in the order of 10 6.1 M −1 , which indicates a significant interaction between this ligand and both structures. The strong interaction between G-quadruplex model sequences and compound 1 validates its potential therapeutic action, based on a synergistic action of PARP-1 inhibition with G-quadruplex affinity.    2 have previously been reported 30,31 . Proton resonance assignments of the free d(TTA GGG T) 4 and Pu22T14T23 sequences were performed on the basis of previous assignments 17 . The protons of the double helix and G-quadruplex oligonucleotides in the complexes were assigned using standard procedures by NOESY and TOCSY experiments (Table S1 and Table S2). Ligand protons were assigned by an integrated series of 2D experiments such as ROESY, TOCSY and COSY (Table S3). Phase sensitive NOESY spectra of the complexes were acquired at 25 °C and 15 °C in TPPI mode, with 2048 × 1024 complex FIDs. Mixing times ranged from 50 to 400 ms. TOCSY spectra were acquired with the use of a MLEV-17 spin-lock pulse (60 ms total duration). All spectra were transformed and weighted with a 90° shifted sine-bell squared function to 4 K x 4 K real data points.

Reagents
CD and fluorescence experiments. CD spectra were recorded on a Jasco J-810 spectropolarimeter equipped with a Peltier temperature control unit (Seelbach, Germany). The DNA solution (Pu22T14T23 or d(TTA GGG T) 4 was transferred to a covered cell and ellipticity was recorded with a heating rate of approximately 0.4 °C·min −1 . Simultaneously, CD spectra were recorded every 5 °C from 210 to 320 nm. The spectrum of the buffer was subtracted. Each sample was allowed to equilibrate at the initial temperature for 30 min before the melting experiment began. In all experiments, the concentration of DNA was kept constant (2 µM) whereas the concentration of the considered ligands was increased. The medium consisted of 25 mM KH 2 PO 4 and 70 mM (Pu22T14T23) or 150 mM (d(TTA GGG T) 4 ) KCl 32 . Molecular fluorescence spectra were measured with an JASCO FP-6200 spectrofluorimeter. The temperature was controlled by means of a water bath. The fluorescence spectra were acquired using a quartz cuvette with a 10-mm path length. In the fluorescence measurements, both the excitation and emission slits were 10 nm, and the scan speed was 250 nm/min. Measurements were taken at 308 nm excitation wavelength. The medium consisted of 25 mM phosphate buffer (pH 6.9) and 70 mM KCl or PBS. In all experiments, the concentration of ligand was kept constant (3 µM), whereas the concentration of the considered DNA sequence was increased. The determination of the stoichiometries and the calculation of the binding constants was done from the fluorescence data recorded along titrations of ligands with DNAs by using the EQUISPEC program 33 . This program is based on the multivariate analysis of the whole spectra measured along the titration.
Molecular modelling studies. The ligand 1 that was the object of this study was optimized as previously described 17 while the coordinates for the Pu22T14T23 and d(TTA GGG T) 4 starting models were obtained from the NMR structure deposited in the Protein Data Bank (accession code: 2L7V for Pu22T14T23 and 1NZM for d(TTA GGG T) 4 ) 26,34 . The GROMACS package 35 with a modified version of the 53A6 GROMOS force field 36 was used to perform energy minimizations and molecular modeling calculations, while molecular docking experiments were conducted using the AutoDock 4.2 software 37 . The molecular docking calculations were performed using the Lamarckian Genetic Algorithm 38 , and the AutoDock Toolkit (ADT) 39 was used to further process the ligand and the Pu22T14T23 and d(TTA GGG T) 4 models. In ADT, the Gasteiger-Marsili charges 40 were added to the ligand, while the phosphorus atoms in the DNA were parameterized using the Cornell parameters. The solvation parameters were added to the system by means of the Addsol utility of AutoDock. For each docking run, the initial population consisted of 100 randomly placed individuals, with a maximum number of 250 energy evaluations and a mutation rate of 0.02, a crossover rate of 0.80, and an elitism value of 1. For the local search, 250 independent docking runs were carried out for the ligand by applying the so-called pseudo-Solis and Wets algorithm with a maximum of 250 iterations per local search. The system in the actual docking process was represented by grid maps calculated with Autogrid, centered between the two K+ ions and with a grid dimensions of 80 × 80 × 80 Å (spacing of 0.01 Å). The docking results were scored by using an in-house version of the simpler intermolecular energy function based on the Weiner force field, and the results differing by less than 1.0 Å in positional root-mean-square deviation (rmsd) were clustered together and represented by the most favorable free energy of binding. The best poses obtained in the docking phase were equilibrated through 5.0 ns of molecular dynamics using the OpenCL version of the GROMACS package running on a dual-Xeon workstation (8 core) equipped with an NVIDIA GPU containing about 5000 CUDA cores. The boxes were generated by placing the systems in the centre of a box with boundaries at 2.0 nm apart from all atoms. 3′ and 5′ terminal nucleotide topologies were modified according to Ricci et al. 41 counterions (K+ ions) were random placed and SPC water molecules were added to the systems. The full solvated systems were simulated through 100 ps of position restrained molecular dynamics, followed by a heating ramp of short (100 ps) consecutive simulations at 50, 100, 150, 200, 250, and 300 K. The production simulations consisted of 5 ns of partially restrained MD at 310 K (time step of 0.002 ps). Constraints were calculated using the Lincs 42 and SETTLE 43 algorithms, while Lennard-Jones interactions were calculated using a two-range switch interaction (cut-off radius of 0.9 and 1.1 nm). A Berendsen thermostat was applied 44 , (coupling time of 0.1 ps) and the electrostatic interactions were calculated using PME 45,46 , (Coulomb cut-off radius of 1.2 nm).
Molecular graphics in Figs. 4 and 7 obtained with UCSF ChimeraX, developed by the Resource for Biocomputing, Visualization, and Informatics at the University of California, San Francisco, with support from National Institutes of Health R01-GM129325 and the Office of Cyber Infrastructure and Computational Biology, National Institute of Allergy and Infectious Diseases 47,48 .