A putative G-quadruplex structure in the proximal promoter of VEGFR-2 has implications for drug design to inhibit tumor angiogenesis

Tumor angiogenesis is mainly regulated by vascular endothelial growth factor (VEGF) produced by cancer cells. It is active on the endothelium via VEGF receptor 2 (VEGFR-2). G-quadruplexes are DNA secondary structures formed by guanine-rich sequences, for example, within gene promoters where they may contribute to transcriptional activity. The proximal promoter of VEGFR-2 contains a G-quadruplex, which has been suggested to interact with small molecules that inhibit VEGFR-2 expression and thereby tumor angiogenesis. However, its structure is not known. Here, we determined its NMR solution structure, which is composed of three stacked G-tetrads containing three syn guanines. The first guanine (G1) is positioned within the central G-tetrad. We also observed that a noncanonical, V-shaped loop spans three G-tetrad planes, including no bridging nucleotides. A long and diagonal loop, which includes six nucleotides, connects reversal double chains. With a melting temperature of 54.51 °C, the scaffold of this quadruplex is stabilized by one G-tetrad plane stacking with one nonstandard bp, G3–C8, whose bases interact with each other through only one hydrogen bond. In summary, the NMR solution structure of the G-quadruplex in the proximal promoter region of the VEGFR-2 gene reported here has uncovered its key features as a potential anticancer drug target.

The development of new blood vessels from pre-existing vasculature, known as angiogenesis, plays an important role in many physiological and pathological processes, including tumor growth (1). It is tightly regulated by a balance between pro-and antiangiogenic factors. During tumor growth, this balance favors angiogenic switch (2). Tumor angiogenesis is a key process for cancer progression as new blood vessels nourish the growth of tumor cells and facilitate their metastasis (3,4). Proangiogenic factors, including basic fibroblast growth factor (5), vascular endothelial growth factor (VEGF) (6 -8), epidermal growth factor (9), platelet-derived growth factor (10), and related receptors, including VEGF receptor 1 (VEGFR-1), 3 VEGFR-2, and VEGFR-3 (11,12), are then produced and secreted by the tumor, and the endothelial cells of nearby blood vessels are activated. These proangiogenic molecules are thus the targets for antiangiogenic drugs. VEGF-A is considered to be the most important factor implicated in tumor angiogenesis (13) through binding with high affinity to tyrosine kinase receptors expressed on the surface of endothelial cells such as VEGFR-1 and VEGFR-2, the receptor functionally more relevant in the transduction of proangiogenic stimuli coming from tumor cells (14 -16). The interactions between VEGF-A and VEGFR-2 lead to receptor autophosphorylation, and phosphorylated tyrosine residues activate signaling cascades, eventually resulting in cellular processes involved in angiogenesis (17) such as vesicle trafficking, cytoskeleton regulation, microtubule dynamics, cell polarity, and membrane transport (18). So far, a few drugs have been developed in preclinical models to target the VEGF/VEGFR pathway for anticancer therapy, including engineered proteins that mimic VEGF receptors, small-molecule inhibitors that preferentially target VEGFR-2, and antibodies against VEGF-A or its receptors.
G-quadruplexes are generally formed in DNA and RNA sequences containing repeated short guanine-rich tracts by the stacked interaction of successive G⅐G⅐G⅐G tetrads (G-tetrads) and stabilized by bound monovalent Na ϩ , K ϩ , or NH 4 ϩ cation. Its arrangement can be tetramolecular, bimolecular, or intramolecular as a result of the changes in strand polarities and the sequences and topologies of the loops (19,20). G-quadruplex structures are found in vitro in human telomere (21)(22)(23)(24) and the promoter regions of different oncogenes, such as c-MYC (25), c-KIT (26,27), VEGF (28), and BCL-2 (29,30). These  G-quadruplexes in gene promoters have physiological and structural characteristics that make them suitable drug targets. Their structural diversity further provides the capability to selectively design anticancer drugs. At the same time, with the application of stabilizing ligands representing a new class of anticancer agents, G-quadruplex structures can be visualized in human cells. Previously, a guanine-rich sequence was found within the proximal promoter region of VEGFR-2 and was suggested to form an antiparallel G-quadruplex structure (31). This G-quadruplex structure can be efficiently stabilized by the small molecule quarfloxin, which has progressed in clinical trials. Here, we report the NMR solution structure of a unique G-quadruplex formed in the proximal promoter region of VEGFR-2 that provides not only the molecular details of this G-quadruplex but also important insights into its loop conformations and interactions with core tetrad structures.

Sequence isolation of VEGFR-17T G-quadruplex
Previously, four consecutive guanine repeats (Fig. 1A) in the G-rich region of VEGFR-2 promoter were suggested to form an intramolecular G-quadruplex in the presence of K ϩ solution (31). However, its imino proton NMR spectrum in Fig. 1B demonstrates very broad peaks, suggesting the presence of multiple G-quadruplex forms in the wildtype (WT) sequence of the proximal promoter of VEGFR-2. Its truncated forms, VEGFR-25mer and VEGFR-24mer, still have broad signals, although VEGFR-25mer has much narrower signals than those of VEGFR-24mer. To obtain the sequence of VEGFR-2 proximal promoter suitable for NMR G-quadruplex structure studies, we mutated the nucleotide residue flanking the fourth G-tract and got the sequence VEGFR-24T, which demonstrates more divergent signals than those of VEGFR-25mer and VEGFR-24mer. To investigate which guanines are essential to G-quadruplex formation, further single-site mutations from guanine to adenine (A) and to thymine (T) were carried out ( Fig. S1 and Table S1), indicating that only base G 17 is not involved in G-quadruplex formation. Mutants VEGFR-17T and VEGFR-17A have much better signal diversities in their imino proton NMR spectra than VEGFR-24T (Fig. 1). Obviously, VEGFR-17T looks more suitable for NMR studies than VEGFR-17A due to having fewer overlapping signals than VEGFR-17A. Moreover, its imino proton NMR spectrum displays almost no changes upon increasing the experimental temperature from 10 to 20°C (Fig. S2) and varying the concentrations of G-quadruplex and K ϩ ion (Figs. S3 and S4). Therefore, the VEGFR-17T sequence was chosen for NMR structure determination.

Folding topology of VEGFR-17T G-quadruplex
Generally, the circular dichroism (CD) spectrum of a parallel-type G-quadruplex always shows an absence of Cotton effect at around ϭ 290 nm but a strong positive band at 263 nm and a negative band at 240 nm. To probe the folding of VEGFR-17T G-quadruplex containing 5, 10, 20, and 100 mM K ϩ solution, respectively, CD spectroscopy was performed, and the spectrum exhibited positive absorptions at ϭ 290 and 263 nm as well as a smaller negative band at about 240 nm (Fig. S5), identical to the results reported previously (31), typical of the hybrid structure. Therefore, VEGFR-17T has identically mixed parallel/antiparallel folding in different K ϩ solutions.
NMR experiments were further carried out to obtain detailed information of the folding of VEGFR-17T G-quadruplex. The signals in the imino proton region of the one-dimensional 1 H NMR spectrum of VEGFR-17T display narrow spectral line widths (2-5 Hz for the sharpest peaks at 20°C), indicative of a monomeric intramolecular structure. This is supported by the concentration-independent and well resolved quality of 1 H NMR spectra for VEGFR-17T (Fig. S3) where the line widths of imino proton signals are almost identical at concentrations of 0.1, 0.5, and 1 mM VEGFR-17T NMR samples. The guanine imino and H8 protons of VEGFR-17T were unambiguously assigned using the site-specific low-enrichment 15 N-labeled and natural abundance VEGFR-17T samples through bond correlation strategies (32, 33) (Fig. 2). The nonexchangeable base and sugar protons in the VEGFR-17T sequence were assigned using standard protocols by two-dimensional (2D) 1 H-1 H NOESY, COSY, TOCSY, and heteronuclear 31 P-1 H correlation spectroscopy ( 31 P-1 H HETCOR) (

Solution structure of VEGFR-17T G-quadruplex
The three-dimensional structure of the VEGFR-17T quadruplex was calculated on the basis of NMR restraints (Table 1) using program XPLOR (34). Finally, 20 refined structures of the VEGFR-17T G-quadruplex with the lowest energy are shown in Fig. 5A. Ribbon and schematic views of a representative refined structure of the VEGFR-17T G-quadruplex are shown in Fig. 5, B and C. VEGFR-17T G-quadruplex contains four loops; two of them, loops 1 and 3, include six ( 3 GTACCC 8 in loop 1) and three bases ( 17 TGC 19 in loop 3), respectively, whereas loop 2 is only composed of 1 nucleotide (nt) (T 12 ). The V-shaped loop 4 (from G 20 to G 21 ) does not contain any nucleotides (i.e. 0 nt) in the sequence. This characteristic, as well as the 6-nt-long loop 1 in the structure, leads to a relatively unstable conformation of VEGFR-17T, consistent with its small melting temperature (i.e. T m ϭ 54.51°C) measured by differential scanning calorimetry (DSC) assay (Fig. S7).
The three G-tetrads of VEGFR-17T G-quadruplex are linked with three parallel G-strands (G 9 -G 10 -G 11 , G 13 -G 15 -G 16 , and G 21 -G 22 -G 23 ) and one antiparallel G-strand (G 20 -G 1 -G 2 ) that are connected by two side loops and two roof loops, consistent with the results from the CD spectrum. The first loop is of the double-chain-reversal type, connecting two antiparallel G-strands G 9 -G 10 -G 11 and G 20 -G 1 -G 2 . Among the six bases in loop 1, G 3 and C 8 form a nonclassical G-C bp, interacting with each other through only one hydrogen bond between the G 3 carbonyl group and C 8 amino group. The NMR signal of the imino proton (i.e. H1) of G 3 is absent (Fig. 2), indicating that the nitrogen (N1) of G 3 is not involved in the hydrogen bond. Both G 3 and C 8 bases have stackedinteractions with the G-tetrad G 2 ⅐G 21 ⅐G 9 ⅐G 13 (Fig. 5D), supported by the NOEs observed between protons H5, H6 of base C 8 , and H1 of bases G 9 and G 21 , that enhance the stability of the G-quadruplex conformation.

Structure of G-quadruplex formed in VEGFR-2 promoter
When the amino group of G 3 is removed by replacing G 3 by inosine (i.e. I 3 ), the signals of the imino protons of bases G 2 , G 13 , G 9 , and G 21 were disappeared or shifted to overlap with other signals (Fig. S8), suggesting that this G-tetrad is unstable. When base G 3 is replaced with adenine (i.e. A 3 ), folding of the whole G-quadruplex collapses, indicating that the carbonyl group of G 3 is involved in a hydrogen bond with the amino group of base C 8 and stabilizes folding of the whole VEGFR-17T G-quadruplex.

The first guanine is positioned in the central G-tetrad of VEGFR-17T
In contrast to the fact that the first guanine is always part of a terminal G-tetrad in almost all unimolecular G-quadruplexes (19,35), in VEGFR-17T, base G 1 is positioned in the central G-tetrad; thus, G 20 and G 1 -G 2 , which are part of the same strand, are not connected with each other, relative to the unbroken linkage for the other three G 9 -G 10 -G 11 , G 13 -G 15 -G 16 , and G 21 -G 22 -G 23 strands of the VEGFR-17T G-quadruplex. Currently, there are only two reported structures of G-quadru-plexes in which base G 1 is located in the central G-tetrad (36,37) (Fig. S9), one that formed in the human CHL1 intronic region and the other that formed in d(G 3 T 4 G 4 ) 2 . All three of these G-quadruplexes have a V-shaped loop (i.e. 0 nt) spanning three G-tetrad planes due to the position of the G 1 base, indicating a possible relationship in G-quadruplex folding between the V-shaped loop and G 1 position in the central G-tetrad.

A V-shaped loop in VEGFR-17T spans three G-tetrad planes
The VEGFR-17T G-quadruplex contains one V-shaped loop, which contains 0 nt, and spans three G-tetrad planes. This kind of V-shaped loop has also been observed within a G-quadruplex scaffold linked by two adjacent G-tetrads (38), the dimeric G-quadruplexes formed by d(G 3 T 4 G 4 ) 2 (37), the dimeric Gquadruplex adopted by d(GLGLT 4 GLGL) 2 where L is a locked nucleic acid (39), the G-quadruplex in the CHL1 intronic sequence (36), and the G-quadruplex in the hCEB1 minisatellite G-rich sequence (40) as shown in Fig. 6. Among these G-quadruplexes, VEGFR-17T is unique with a 6-nt loop in which base G 20 belongs to the top G-tetrad of one strand (this strand is broken by G 1 ), whereas residue G 21 is in the bottom G-tetrad of an adjacent strand. Thus, it seems that G 1 is partially intercalated between bases G 20 and G 21 within the folding topology of the VEGFR-17T G-quadruplex, similar to that observed in the CHL1 G-quadruplex. In contrast, the CHL1 G-quadruplex contains four loops, each of which includes fewer than three bases. The G-quadruplex formed by d(G 3 T 4 G 4 ) 2 contains one V-shaped loop (i.e. G 8 -G 9 ) only in one strand, whereas the G-quadruplex formed by d(GLGLT 4 GLGL) 2 includes four specific V-shaped loops, i.e. G 1 -L 2 in one strand and G 9 -L 10 in another strand, all spanning three G-tetrad planes. The bases G 1 , L 10 , L 2 , and G 9 alternately form two diagonal columns, G 1 -L 10 -L 2 -G 9 , of the G-quadruplex.

A 6-nt-long and diagonal loop destabilizes VEGFR-17T G-quadruplex
Generally, due to the conformational flexibility, the longer the loop contained in a G-quadruplex, the more unstable the G-quadruplex. For examples, the G-quadruplex form observed in the hCEB1 minisatellite G-rich sequence (40) contains a 4-nt side loop ( 5 CTGA 8 ), which leads to instability of this G-quadruplex with a T m value of only 46.18°C. The G-quadruplex adopted by an oligodeoxynucleotide sequence containing a 5-nt diagonal loop exhibits low stability with a T m value of 58°C (41). This accounts for why there are very few reported structures of G-quadruplex with a loop longer than 6 nt in the Protein Data Bank. However, when the loop is anchored by some interactions between loop and G-tetrads, the G-quadruplex will become stable. As shown in Fig. 7, although the G-quadruplex formed in the human CEB25 minisatellite locus has a 9-nt side loop ( 10 TGTAAGTGT 18 ) (42), its structure is still stable with a T m value of 76.5°C in a 90 mM K ϩ solution because the conformation of this long loop is fixed bystacking interactions between G-tetrads and Watson-Crick bp A 2 ⅐T 18 and potentially A 1 ⅐G 17 bp. Another example is the G-quadruplex formed in the human BCL2 promoter region (i.e. Bcl2mid) (43) with a 7-nt-long loop ( 10 AGGAATT 16 ). The structure of this G-quadruplex is stable with a T m value of 65°C due tostacking

Structure of G-quadruplex formed in VEGFR-2 promoter
interactions between bp A 10 ⅐T 15 in this loop and G-tetrad G 1 ⅐G 9 ⅐G 17 ⅐G 21 . The current VEGFR-17T G-quadruplex contains a 6-nt loop ( 3 GTACCC 8 ) connecting two diagonal columns (G 20 -G 1 -G 2 and G 9 -G 10 -G 11 ). Although bp G 3 ⅐C 8 in this loop hasstacking interactions with G-tetrad G 2 ⅐G 13 ⅐G 9 ⅐G 21 , this bp has only one hydrogen-bond interaction between them and thus is a nonclassical Watson-Crick bp. Therefore, the VEGFR-17T G-quadruplex is more stable than the G-quadruplex formed in the hCEB1 minisatellite G-rich sequence, but less stable than those formed in the human BCL2 promoter region and CEB25 minisatellite locus.

VEGFR-17T G-quadruplex is a potential drug target to inhibit tumor angiogenesis
In short, in the presence of K ϩ , the proximal promoter of VEGFR-2 can form a G-quadruplex. Our NMR data indicate that the G-quadruplex of VEGFR-17T has a melting temperature of 54.51°C. It has a unique fold with three striking features. Its first guanine is positioned in the central G-tetrad plane, it has a V-shaped loop spanning three G-tetrad planes, and it has a 6-nt loop connecting two antiparallel columns. By single-site mutations of G 1 , G 2 , and G 11 (in the sequence of VEGFR-24T) into adenine, a new, parallel G-quadruplex (confirmed by their CD spectra, shown in Fig. S5) with two G-tetrads is generated with a melting temperature of nearly 80°C (Fig. S7), indicating that they are much more stable than the VEGFR-17T G-quadruplex. The imino proton NMR spectra of these mutants display only eight signals (Fig. S10), revealing that they are identical to one another, all with only two G-tetrad planes. This stable G-quadruplex is one of multiple forms of WT G-quadruplex formed in the proximal promoter of VEGFR-2. Therefore, the VEGFR-17T G-quadruplex may be a potential anticancer drug target, which can be efficiently stabilized by small molecules. These molecules can inhibit VEGFR-2 expression, which results in turning off signaling components that further medi-ate cellular events leading to endothelial cell proliferation, migration, and differentiation. This assumption had been confirmed by previously reported data (31) in which the angiogenic process is strongly inhibited in vitro and in vivo by ligands targeting G-quadruplex formed in the proximal promoter of VEGFR-2, resulting in impairment of endothelial cell function.
In conclusion, we have determined the solution structure of a unique G-quadruplex form adopted by the proximal promoter of VEGFR-2 that can be potentially stabilized by G-quadruplex ligands, thereby revealing a promising way to block VEGFR-2 expression as a target for anticancer therapy.

CD spectroscopy
CD spectra were recorded at 25°C on a Jasco-815 spectropolarimeter using a 1-cm-path length quartz cuvette with a reaction volume of 350 l. The DNA concentration was 20 -30 M. The DNA oligonucleotides were prepared in a pH 6.8 buffer containing 20 mM KH 2 PO 4 and 80 mM KCl, heated to 98°C for 5 min, and cooled to room temperature overnight. An average A, characteristic guanosine imino-H8 NOE connectivity patterns around a G ␣ ⅐G ␤ ⅐G ␥ ⅐G ␦ tetrad as indicated by arrows. B, guanosine imino-H8 connectivities observed for the G-tetrad G 20 ⅐G 23 ⅐G 11 ⅐G 16 , G 1 ⅐G 22 ⅐G 10 ⅐G 15 , and G 2 ⅐G 21 ⅐G 9 ⅐G 13 planes. C, interstrand NOEs between imino H1 and aromatic H8 of unimolecular guanine bases within the same layer of G-tetrads mentioned in B.

Structure of G-quadruplex formed in VEGFR-2 promoter
of three scans was taken, and the spectrum of the buffer was subtracted.

Differential scanning calorimetry
To probe the stability of VEGFR-17T, VEGFR-1A, VEGFR-2A, and VEGFR-11A, DSC measurements were carried out using an automated VP-CAP-DSC microcalorimeter (Malvern Inc.). The experiments were performed at single-strand DNA concentrations in the range of 50 -70 M. Scans were performed at 1.5°C/min in the 20 -110°C temperature range. A buffer-buffer scan was subtracted from the buffer-sample scans, and linear-polynomial baselines were drawn for each scan. Baseline-corrected thermograms were normalized with respect to the single-strand molar concentration to obtain the corresponding molar heat capacity curves. The enthalpy for the overall unfolding of G-quadruplex structures was estimated by integrating the area under the heat capacity versus temperature curves. T m values were obtained as the temperatures corresponding to the maximum of each thermogram peak. Entropy values were calculated by integrating the curve Cp/T versus T (where Cp is the molar heat capacity and T is the temperature in Kelvin), and the free-energy values were computed by the equation ⌬G ϭ ⌬H Ϫ T⌬S.

NMR data collection and spectral analysis
NMR experiments were performed on 600-and 800-MHz Varian and Bruker spectrometers at 10 and 20°C, respectively. Resonances of DNA were assigned unambiguously using sitespecific low-enrichment 15 N-labeled samples and throughbond correlations at natural abundance (32,33). The NMR experiments for samples in water solution were performed with watergate or jump-and-return water suppression techniques. The acquisition data points were set to 2048 ϫ (250 -512) (complex points). All spectra were processed with the program NMRPipe (44). The 45°or 60°shifted sine-squared functions were applied to 2D 1 H-1 H NOESY and TOCSY spectra. Fifthorder polynomial functions were used for the baseline corrections. The final spectral sizes are 2048 ϫ 1024. The 1 H chemical shifts were referenced to 2,2-dimethylsilapentane-5-sulfonic acid. Peak assignments and integrations were achieved using SPARKY (46). The NOE peaks were integrated using the peak fitting function and volume integration of SPARKY. The 31 P NMR spectrum was collected on a DNA sample at 1.5 mM in D 2 O at 20°C and was referenced to an external standard of 85% H 3 PO 4 , including the one-dimensional protondecoupled phosphorus spectrum and two-dimensional 31 P-1 H HETCOR. Assignments of the individual 31 P resonance were accomplished by a combination of 2D 1 H-1 H NOESY, COSY, TOCSY, and 31 P-1 H HETCOR spectra.

Solution structure determination
The distances between nonexchangeable protons were estimated based on the NOE cross-peak volumes at 50-, 200-, and 250-ms mixing times and divided into strong (1.8 -2.9 Å), medium (1.8 -3.5 Å), and weak (1.8 -6.0 Å) groups, respectively, using the proton H5-H6 distance (2.45 Å) in the cytosine base as a reference. Exchangeable proton restraints are based  Guanosine bases in the G-tetrad core are colored cyan (anti) and magenta (syn). Bases in loops are in yellow for G 3 -C 8 bp, violet for 4 TACC 7 , green for T 24 , and blue for 17 TGC 19 . For clarity, all bases are only shown in ribbon in B. The orientation of G-quadruplex in A is similar to that shown in B. C, schematic structure of VEGFR-17T G-quadruplex. The anti and syn guanines are colored cyan and magenta, respectively, and G 3 -C 8 bp is colored in yellow. D, the nonclassical G 3 -C 8 bp (in yellow) hasstacked interactions with the Gtetrad G 2 ⅐G 21 ⅐G 9 ⅐G 13 . The hydrogen bond between G 3 and C 8 bases is displayed as a dashed line.

Structure of G-quadruplex formed in VEGFR-2 promoter
on NOESY data sets at two mixing times (50 and 250 ms) in H 2 O. Cross-peaks involving exchangeable protons were classified as strong (strong intensity at 50 ms), medium (weak intensity at 50 ms), and weak (observed only at 250 ms) NOEs. The G-tetrads within the G-quadruplex were restrained with distances corresponding to ideal hydrogen-bond geometry. Each individual hydrogen bond was restrained using two distance restraints (heavy atom-heavy atom and heavy atom-proton, respectively). Hydrogen bond distance restraints were also applied to the carbonyl oxygen (in residue G 3 ) and the -NH 2 group (in residue C 8 ) based on single-site mutation of residue G 3 into adenine (i.e. A 3 ) and inosine (i.e. I 3 ). The aromatic rings of guanines in each G-tetrad were also restrained into one plane during calculation. 24 dihedral angle restraints were used to restrict the glycosidic torsion angle () for the experimentally assigned syn configuration, i.e. G 1 , G 9 , and G 20 tetrad guanines (60 Ϯ 35°) as well as for the experimentally assigned anti configuration bases, i.e. G 2 , G 3 , T 4 , A 5 , C 6 , C 7 , C 8 , G 10 , G 11 , T 12 , G 13 , A 14 , G 15 , G 16 , T 17 , G 18 , C 19 , G 21 , G 22 , G 23 , and T 24 (240 Ϯ 70°).
The structural calculations of VEGFR-17T were carried out using a standard simulated annealing protocol implemented in the program XPLOR-2.37 (NIH version) (45). A total of 618 NOE distance restraints (Table 1), of which 216 are from interresidue NOE interactions, were incorporated into the NOErestrained structure calculation. A total of 10 iterations (50 structures in the initial 10 iterations) were performed. 100 structures were computed in the last five iterations, and 20 conformers with the lowest energy and minimal restrain violations are used to represent the three-dimensional structures. In the ensemble of the simulated annealing 20 structures, there was no distance constraint violation greater than 0.3 Å and no torsion angle violation greater than 5°.