Human NSCLC cell line H1299 was acquired from the AOLUKEJI (Shanghai, China). The medium RPMI 1640 and fetal bovine serum were purchased from Gibco-BRL (Gaithersburg, MD, USA).

Tris, Tris–HCl, sodium acetate (NaAc), magnesium chloride (MgCl₂), dimethylsulfoxide (DMSO) and PEG4000 were all from Sigma-Aldrich Corp. (St. Louis, MO, USA). Crystal loop was supplied by Hampton Research Corp. (Aliso Viejo, CA, USA). Sodium chloride (NaCl), glycerol, hydrochloric acid (HCl), sodium hydroxide (NaOH) and ethyl alcohol were obtained from the Xi’an Chemical Reagent Company (Xi’an, China). NVP-AUY922 was from Target Molecule Corp. (Boston, MA, USA). DMSO was used to dissolve NVP-AUY922 and double distilled water (ddH₂O) was used for other chemicals.

First of all, quantitative IC₅₀ value was determined. RPMI 1640 with 10% (v/v) fetal bovine serum was used to cultured NSCLC cell line H1299 cells at 37°C in a humidified 5% CO₂ atmosphere incubator (Thermo Fisher Scientific Inc., Waltham, MA, USA). Briefly, cells were seeded in 96-well plates at 4 × 10³ per well in triplicates and were treated with NVP-AUY922 (100, 10, 1, 0.1, 0.01 μM) for 72 h after incubating for 24 h. OD450 was then detected after 2.5 h adding CCK-8 (7Sea, Shanghai, China) by ELISA plate reader (BioTek Instruments, USA).

Cell viability assay was followed and performed. H1299 cells were placed in 96-well plates as described above. Afterwards, cells were treated with 4 × 2.85 μM NVP-AUY922 for 24, 48, and 72 h. After incubating with CCK-8 for 2.5 h, OD₄₅₀ was valued to analyze the cell viability.

To monitor effects of NVP-AUY922 on cell cycle progression, cells were seeded in 6-well plates at 3 × 10⁵ per well. After treating with 4 × 2.85 μM NVP-AUY922, or the same amount of DMSO for 72 h at 37°C, cells were harvested and resuspended with 3 ml PBS/ethanol mixture (v:v = 1:2) and then preserved at 4°C. The samples were resuspended with 400 μl buffer after 5 min-centrifugation at 1,000g and washing in 2 ml PBS once. Lastly, 50 μl propidium iodide (PI) and the same volume of RNase were then added into the cells to be incubated for 30 min in the dark at 37°C (22). Flow cytometry (Becton Dickinson, NJ, USA) was used to analyze the distribution of cell cycle phases.

The apoptosis rate was detected by the Annexin V-FITC/PI Apoptosis Detection Kit (556547, Becton Dickinson, USA) according to the manufacturer’s instructions. Briefly, cells were seeded and incubated with NVP-AUY922 as is mentioned in Cell Cycle Analysis. Subsequently, cells were resuspended in 100 μl binding buffer after washing with 2 ml PBS. Ultimately, 5 μl Annexin V-FITC and 10 μl PI were added and incubated at 4°C for 30 min in the dark. Also, flow cytometry was adopted for cell apoptosis analysis.

The recombination plasmid, pET28a with the synthesized Hsp90^N gene (residues 9–236) was transferred into E. coli BL21 (DE3) (TIANGEN Biotech Corp., Beijing, China) for the target Hsp90^N expression. The protein was purified by Nickel affinity chromatography and molecular sieve chromatography, as previously reported (23).

For crystallization, NVP-AUY922 was mixed with Hsp90^N protein (20 mg/ml) with 5:1 molar ratio to incubate for 30 min at 4°C. The mixture was then centrifuged for 10 min at 3,000g to collect the supernate. Mixing with the same volume of precipitant (pH 8.5, 0.1 M Tris–HCl, 0.2 M MgCl₂, 25% PEG4000) (24) was conducted for 3–7 d at 4°C. Hanging drop vapor diffusion method was used. Co-crystallization was accomplished in a bath circulator controlled incubator (PolyScience 9712, PolyScience, USA).

Hsp90^N-NVP-AUY922 complex crystal image of was obtained by a stereomicroscope (M165, Leica Microsystems, Germany). Mounted with cryo-loop (Hampton Research Corp., Aliso Viejo, CA, USA), the complex crystals were flash-frozen in liquid nitrogen for XRD after soaking briefly in the cryoprotectant solution (pH8.5, 0.1 M Tris–HCl, 0.2 M MgCl₂, 25% PEG4000, and 20–25% glycerol). All data sets were acquired at 100 K on Macromolecular Crystallography Beamline17U1 (BL17U1) using an ADSC Quantum 315r CCD detector at Shanghai Synchrotron Radiation Facility (SSRF, Shanghai, China) (25) and the diffraction data were auto-processed by aquarium pipeline (26). The complex crystal structure was analyzed by molecular replacement method using the PHENIX software (27) according to the reported Hsp90^N-FS23 (PDB ID, 5CF0) (28). The program Coot was used to rebuild the initial model (29). CCP4MG software was used for graphic reconstruction and molecular interaction analysis (30).

The ligand NVP-AUY922, dissolved in DMSO, and the target Hsp90^N were mixed at a molar ratio of 1:5 in a 20 μl TSA dye reaction buffer with 20 mM Tris–HCl, 150 mM NaCl and 10% glycerol (pH 7.5). The melting-curve (25 to 95°C with a ramp rate of 1°C/min) was determined on ABI 7500 system (7500, ABI Corp., USA). Fluorescence was stimulated by environmentally-sensitive TSA dye binding to the exposed hydrophobic regions during protein unfolding as it is heated. Binding capacity of protein–ligand can be closely evaluated by the melting temperature (Tm) and melting temperature shift (ΔTm) which was calculated based on the melt curves (31).

ITC was performed to further illustrate molecular interaction mechanism of Hsp90^N-NVP-AUY922 using an ITC-200 calorimeter (Malvern Instrument Ltd., UK). All measurements were carried out in PBS (pH 7.5) at 25°C. Briefly, fresh-purified Hsp90^N protein was concentrated to 50 μM. The ligand NVP-AUY922 was prepared in a concentration of 500 μM. Sample was degassed before use. For each titration in the cell, 2 μl aliquot of NVP-AUY922 was added into Hsp90^N solution at a 200 s-interval and stirring speed of 750 rpm. The experimental data was fitted to a bimolecular binding model by adjusting three parameters, namely, stoichiometry (n), enthalpy (ΔH°) and association constant (Ka) using Microcal Origin software embedded in the equipment. The thermodynamic parameters ΔG° (free energy) and ΔS° (entropy) were obtained according to the formula, −RT ln Ka = ΔG° = ΔH° − TΔS° (32).

Based on the analysis of the Hsp90^N-NVP-AUY922 crystal structure, TSA and ITC, we designed a series of novel NVP-AUY922 derivatives. Evaluation of the derivatives was performed through molecular docking on software SYBYL-X 2.0 (Tripos Associates Inc., St. Louis, MO, USA).

First, a library of novel NVP-AUY922 derivatives was built and a geometric and force field optimization was carried out over 10,000 times. Second, the protein optimization was performed, such as adding hydrogenation and charge. Lastly, the molecular docking evaluation was performed by Surflex-Dock module of SYBYL software. Further, complex 3D structure reconstruction was simulated by CCP4MG software and molecular interaction mechanism was also analyzed (30).

All data were exhibited as the mean ± standard deviation (SD). The difference significance between two variables was evaluated by Unpaired Student’s t-test, and P <0.05 was considered having statistical significance. Data analysis was performed using GraphPad Prism 5.01 (GraphPad Software, San Diego, USA) and SPSS 13.0 (International Business Machines Corporation, USA). Flow cytometry (Becton Dickinson, NJ, USA) was adopted for cell apoptosis analysis.

AUY-922, as a positive prodrug control, performed favorable anti-NSCLC activity on NSCLC cell lines H460, A549 and H1975 (33). Herein, the effects of NVP-AUY922 on cell viability, cell cycle progression and cell apoptosis of another NSCLC cell line H1299 were analyzed.

NVP-AUY922 exhibited favorable inhibitory activity against H1299 cells (IC₅₀, 2.85 ± 0.063 μM). As shown in Figure 1A , NVP-AUY922 significantly inhibited H1299 cell viability at 24 h (P <0.01), 48 h (P <0.001) and 72 h (P <0.0001), and its inhibitory activity was increased with extended treat time.

As can be seen from Figures 1B – D , after treated with NVP-AUY922 for 72 h, the percentage of H1299 cells in G1, S and G2 phases was obviously changed (NVP-AUY922, G1,↓33.79%; S,↑0.56%; G2,↑33.24% vs Control), indicating that when treated with NVP-AUY922, H1299 cells were notably arrested at G2 phase.

Effects of NVP-AUY922 on H1299 cell apoptosis was also analyzed by the annexin V and PI double staining kit by flow cytometry. Briefly, during early apoptosis, Annexin V can bind to phosphatidylserine (PS) that exists on the external leaflet of the plasma membrane. While, early apoptotic cells exclude PI, late apoptotic cells and necrotic cells can be stained positively (34). Cell percentage at different apoptosis stages are shown in Figures 1E – G . Compared with the control, NVP-AUY922 elevated both the early apoptosis rate and later apoptosis rate of H1299 cells obviously (NVP-AUY922, normal cell rate,↓39.00%; early apoptosis rate,↑14.60%; late apoptosis rate,↑21.80% vs Control).

The thermal stability of proteins with or without ligands was detected by TSA according to the Tm and ΔTm calculated by the protein melting curve, which closely reflect the protein–ligand binding force. The affinity between target Hsp90^N and ligand can be indicated by the absolute value of ΔTm, while |ΔTm| >3 is usually considered as a favorable ligand. The results showed that after binding with NVP-AUY922, the thermal stability of Hsp90^N was shifted by approximately 15.56°C (ΔTm, 15.56 ± 1.78°C), suggesting a strong binding force between the Hsp90^N and NVP-AUY922 ( Figures 2A – C ).

The dissociation constant of Hsp90^N-NVP-AUY922 was further determined by ITC at 25°C in PBS. As can be seen in Figure 2D , a perfect S-shaped curve was established by nonlinear data fitting. The negative peaks in the raw ITC titration data revealed that the binding process of Hsp90^N-NVP-AUY922 was exothermic at 25°C. The stoichiometric ratio was 0.78 ± 0.04, suggesting a 1:1 binding mode between Hsp90^N and NVP-AUY922. The Kd values derived here was 5.10 ± 2.10 nM, further indicating a strong binding force of Hsp90^N-NVP-AUY922, as shown in Table 1 . The thermodynamic signature was dominated by a large enthalpy change, indicating favorable thermodynamic interactions established in the Hsp90^N-NVP-AUY922 binding process.

As is shown in Figure 3 , the empty ATP-binding site of Hsp90^N showed continuous and strong electron density, which was well matched with NVP-AUY922 molecule structure ( Figures 3A, C, D ). A major part of NVP-AUY922 was buried in the ATP-binding pocket of Hsp90^N, while the morpholine ring was extending out of the pocket. As shown in Figure 3B , the complex hydrogen bond networks were responsible for strong binding force of Hsp90^N-NVP-AUY922. Three direct hydrogen bonds were formed between NVP-AUY922 and residue D93 (2.7 Å), residue K58 (3.0 Å) and residue G97 (3.0 Å) of Hsp90^N, respectively. Furthermore, one hydroxy group of the isoxazole ring of NVP-AUY922 formed four water-mediated hydrogen bonds with residues D93, T184, and G97. The other hydroxy group of phenyl ring formed three water-mediated hydrogen bonds with residues L48 and S52. The carbanyl group of carboxamide moiety formed three water-mediated hydrogen bonds with residues K58 and D54. In addition to hydrogen bonds, hydrophobic effects, derived from isopropyl group of the benzene ring in NVP-AUY922 with surrounding residues of Hsp90^N, also played an important role for their powerful binding force. The results suggested a competitive ATP-binding inhibition of NVP-AUY922 against Hsp90^N, which inactivated the molecular chaperone function of Hsp90^N, resulting in inhibiting or killing cancer cells.

Hsp90^N protein, expressed in E. coli BL21 DE3, was purified by immobilized Ni+ affinity chromatography, and followed by gel filtration chromatography. Hsp90^N was co-crystallized with NVP-AUY922 by hanging-drop vapor diffusion method. The complex crystals of Hsp90^N-NVP-AUY922 were obtained at 4 °C after growing for 3–5 d. Crystals were rhombus, with the average dimension of approximately 120 μm × 70 μm × 50 μm ( Supplementary 1, Figure S1 ).

Meanwhile, the complex crystal structure of Hsp90^N-NVP-AUY922 was determined by molecular replacement method, and Hsp90^N-FS23 (PDB ID 5CF0) (25) was taken as the analytic model. The coordinates and structure factors have been deposited in PDB (PDB ID 6LTI). Data for collection and refinement statistics are described in Table 2 . A 1.59 Å resolution limit diffraction data was collected with a final R_free of 0.20 and R_work of 0.18, co-crystal space group of I 222, and unit cell parameters of a = 66.18 Å, b = 89.29 Å, c = 99.56 Å; α = β = γ = 90.00°. Residues Glu16–Lys224 were observed in the refined model structure with no electron density for N-terminal residues Asp9–Glu15 and C-terminal residues Glu225–Glu236. The missing residues were conceivable for the N- and C-termini disordered electron density.

A series of novel NVP-AUY922 derivatives was designed according to the complex crystal structure and molecular interaction analysis, as is shown in Figure 4A . For the new derivative design, it was crucial to maintain the special molecular conformation of NVP-AUY922. First, hydrogen bonds played a key role and introduction of hydrophilic groups could be a favorable strategy. For instance, the 2,4-Dihydroxy-5-isopropyl-phenyl ring moiety was varied to an amino group or sulfydryl group, the isoxazole ring was modified to a 1H-pyrazole ring and the carbanyl group of formamide group was changed by a sulfonyl group. Second, hydrophobic effects were another specific factor contributed to their affinity. Therefore, transformation of hydrophobic functional groups was considered as appropriate structural modification scheme, followed by isopropyl group on the 2,4-Dihydroxy-5-isopropyl-phenyl ring moiety was replaced by a dimethylamino, trifluoromethyl, ethane or tert-butyl group, respectively. Last, considering the further drug metabolism, a thiomorpholine, piperazine or 1-methyl-piperazine was used to substitute the morpholine ring.

Based on the scheme, thirty-two novel NVP-AUY922 derivatives were successfully designed and evaluated by molecular docking with Total score and Cscore. The molecular structures, Total scores and Total score increments compared with NVP-AUY922 for new derivatives are presented in Supplementary 2 Table S1 . Total score, evaluates the binding force between target and ligand, is the negative logarithmic value (−log) of the dissociation constant (K _d) of target-ligand, with effects of hydrophobic interaction, polarity, repulsion, entropy change, hydrogen bonds, collision, and so on, to be considered. Higher values indicate stronger binding force of target-ligand. Cscore is a comprehensive score of D-score, PMF-score, G-score, Chem-score, and Total score. The max score is five points. Usually, derivatives with high scores (Total score ≥6, Csore ≥4) were recognized as favorable ligands. Twenty-eight novel NVP-AUY922 derivatives showed increased binding force with the target Hsp90^N shown as high Total scores, which manifested a feasible design scheme by molecular docking evaluation. The molecular structures and the Total score increment of top ten new derivatives are shown in Figure 4B .

A15 perfomed the top optimal representative (Total score, 11.9). Thereby, simulated complex 3D structure of Hsp90^N-A15 was built, and analyzed their molecular interaction mechanism in detail. A15 was accommodated appropriately in the ATP-binding pocket of Hsp90^N ( Figure 4C ). Compared with NVP-AUY922, the 5-amino-2-tert-butyl-phenol group of A15 was obviously shifted from the interior of cavity to the exterior. Meanwhile, the branched chain of pyrazole ring was moved oppositely from the external space into the internal pocket. Besides, the piperazine ring of A15 was shrunk gently into the pocket ( Figures 4C – E ), while the morpholine ring of NVP-AUY922 was extended out of the cavity ( Figure 4A ).

Three hydrogen bonds between A15 and residues D93 (2.9 Å), L107 (2.7 Å) and G135 (3.0 Å) of Hsp90^N were formed, respectively. Meanwhile, the piperazine ring together with adjacent aromatic ring formed hydrophobic interactions with residue G135, the pyrazole ring with branched chain arranged hydrophobic contacts with residues S52, M98, and V186, the 5-amino-2-tert-butyl-phenol group formed hydrophobic contacts with residues D54 and A55 ( Figure 4D ). Therefore, the hydrogen bonds and hydrophobic interactions played an important role for the intense binding of Hsp90^N-A15. The results would provide promising lead compounds for anti-NSCLC new drug development based on NVP-AUY922.