From the Protein Data Bank (PDB) (27), we collected 11 X-ray crystallographic structures of EGFR harboring the resistance-triggering T790M mutation in complex with a ligand (PDB IDs: 3W2O, 3W2P, 3W2Q, 3W2R, 4I22, 4RJ4, 4RJ5, 4RJ6, 4RJ7, 4RJ8, 5HG7). The EGFR structures alone were extracted from their respective complex with the co-crystalized ligand and water molecules detached, and then converted from PDB format to PDBQT format for later use by the docking software. According to the observation that the geometry of the binding site is usually proportional to that of the bound ligand, the cubic search space was placed at the geometrical center of the bound ligand, with the length, width, and height configured to be 30% greater than that of the bound ligand. Then the search space was further stretched by 4 Å in all three dimensions to create sufficient room for the drug molecules to translate and rotate inside and achieve the optimal conformation upon binding.

From the ZINC database (28), we collected the structures of approved drugs worldwide from three catalogs, namely DrugBank-approved, FDA-approved drugs (via DSSTOX), and the NCGC Pharmaceutical Collection (NPC). These compounds, having undergone a filtering and curation procedure, constituted a non-redundant set of 3,167 drugs that have been approved for clinical use by US (FDA), UK (NHS), EU (EMA), Japanese (NHI), and Canadian (HC) authorities. Likewise, the drug compounds downloaded in Mol2 format were also converted to PDBQT format for use by the docking software.

Having retrieved and preprocessed the necessary 3D structures, we then performed the prediction of binding conformations and binding affinities of the 3,167 compounds docking against the 11 T790M-mutant EGFR structures with the free and open-source molecular docking software idock v2.2.1 (29). Software settings were tweaked to make the conformational searching process substantially more exhaustive than the default settings, i.e., grid maps of atomic free energy were calculated with a high density of 0.08 Å for each receptor structure, and 256 conformational searching tasks were executed for each compound to increase the chance of finding the optimal binding pose. In addition, although up to nine putative binding conformations will be generated for each input compound under the default settings, we set to output only the docked conformation with the best idock score (which is always the first outputted conformation among the nine) because such pose had been formerly evaluated to be most likely closest to the crystal pose (30).

Following docking, the compounds were sorted in ascending order of their predicted binding free energy. Meanwhile, the more accurate scoring function RF-Score v3 was executed to rescore all the compounds (30), providing an additional and more reliable estimation of intermolecular binding strength, given the assumption that the compounds were correctly docked. Consequently, the top-scoring compounds would be those predicted to harvest both a low idock score (in terms of binding free energy) and a high RF score (in terms of binding affinity). These top-scoring compounds were visually inspected in the context of the receptor using the convenient web-based visualizer iview (31). Taking commercial availability into consideration, we purchased and consequently validated in vitro some of the potential compounds indicated from the in silico screening results (Table 1).

Two NSCLC cell lines carrying specific EGFR mutations (HCC827: EGFR sensitizing E746_A750 deletion; H1975: resistance-causing EGFR T790M secondary mutation) were purchased from American Type Culture Collection (ATCC; Manassas, VA, USA). A normal human lung epithelial cell line BEAS-2B was also obtained from ATCC to investigate the potential toxic effect of the tested drugs. HCC827, H1975, and BEAS-2B cells were maintained in RPMI1640 medium supplemented with 10% fetal bovine serum, 100-U/mL streptomycin sulfate, and 100-U/mL penicillin G sulfate, and incubated at 37°C in 5% CO₂.

Growth inhibitory effects of various drug candidates on the cell lines were evaluated by the sulforhodamine B assay (32). Cells were seeded into 96-well tissue culture plates in 100 µL at a plating density of 3,000–5,000 cells/well, and allowed to incubate overnight. The cells were then treated with various drug candidates at a range of concentrations and allowed to incubate at 37°C in 5% CO₂ for 72 h. Each drug concentration was tested in quadruplicate and controls were tested in replicates of eight.

Cells growth in 96-well plates were treated with either gefitinib or the repurposed drug alone or their combination in a fixed ratio of each drug in increasing concentrations. The drugs were added either simultaneously (24 h) or the tested drug (24 h) followed by gefitinib (24 h) or vice versa. The ratio of the two drugs in combination was initially determined according to the relative ratio of the IC₅₀ values of the drug alone. For the three positive drug candidates, the range of concentration tested was as follows: indacaterol (10 µM–10 nM, in 1:1 serial dilution), canagliflozin (10 µM–10 nM, in 1:1 serial dilution), and cis-flupenthixol (2 µM–2 nM, in 1:1 serial dilution). Gefitinib was tested in the range of 10 µM–10 nM (1:1 serial dilution). The median-drug effect analysis method was used to evaluate the nature of the drug combination (33). Combination index (CI) was then calculated to assess the outcome of the drug combination at different fraction of cells affected (Fa) as described previously (33).

Inhibition of tyrosine kinase signaling by the drug candidates identified was examined in a cell-free system by assessing the phosphorylation of a poly-EY (4:1 Glu, Tyr) peptide substrate (for EGFR) with recombinant kinases EGFR^wt, EGFR^L858R, or EGFR^L858R+T790M as described previously (34). Inhibition of the recombinant kinases by the drug candidates was evaluated by using the ADP-Glo Kinase assay kit according to the manufacturer’s instruction (Promega, Madison, WI, USA). Briefly, the drug candidates in a range of different concentration (indacaterol: 10 nM–10 µM; canagliflozin: 10 nM–10 µM; cis-flupenthixol: 2 nM–2 µM) were allowed to incubate with 4 ng of the recombinant kinases and 0.2 µg/mL of the poly-EY substrate at room temperature for 60 min. Also, 5 µL of ADP-Glo reagent was then added and incubation continued at room temperature for another 40 min. Afterward, 10 µL of kinase detection reagent was added and the mixture was allowed to incubate at room temperature for 30 min, before the measurement of luminescence by GloMax 20/20 Luminometer (Promega).

Non-small cell lung cancer (HCC827, H1975) cell lines were treated with the tested drug candidates for the designated time (4 h). The cells were then harvested for Western blot analysis in lysis buffer (0.05-M HEPES pH 7.4, 0.15-M NaCl, 2-mM EDTA, 10% v/v glycerol, 1% v/v Triton X-100) supplemented with protease and phosphatase inhibitor cocktail (Thermo Scientific). Whole cell lysates were separated by SDS-PAGE and subjected to immunoblot analysis with the respective antibodies [total EGFR, phosphor-EGFR (Y845), phosphor-ERK1/2 (Thr177/Thr160), ERK1/2, phosphor-Akt, total Akt, and GAPDH (Santa Cruz Biotechnology, Santa Cruz, CA, USA)]. Primary antibody incubation was carried out at 4°C overnight in 5% bovine serum albumin/phosphate-buffered saline-Tween 20. Afterward, the membranes were incubated with HRP-conjugated donkey anti-mouse/anti-rabbit secondary antibody at room temperature for 1 h, and developed using the WesternBright Quantum chemiluminescence detection system (Advansta Corporation, Menlo Park, CA, USA). Anti-GAPDH antibody was used as the loading control (Santa Cruz Biotech, Santa Cruz, CA, USA). Digital chemiluminescence images were captured and quantified by using the FluorChem Q Imaging System (Alpha Innotech Corporation, Santa Clara, CA, USA).

Cells were grown on a 60-mm tissue culture dish at a density of about 5.0 × 10⁵ cells/well. They were treated for 48 h with the repurposed drug or its combination with 10-µM gefitinib. At the end of the treatment, both floating and attached cells were collected and washed twice with ice-cold phosphate buffer solution. The extent of apoptosis was determined by using the APC annexin-V apoptosis kit (BD Bioscience, San Jose, CA, USA) according to the manufacturer’s instructions. Cells positive for both annexin V and 7-AAD were considered apoptotic.

All experiments were repeated at least three times. The results were conveyed as mean ± SD. One-way analysis of variance (one-way ANOVA) with Bonferroni’s multiple comparison test was used to compare the differences among various treatment groups. A confidence level of p < 0.05 was considered significant.

A total of 3,167 drugs approved for clinical use by worldwide authorities were individually docked to an ensemble of 11 crystal structures of the T790M-mutant EGFR kinase domain, and ranked in ascending order of their estimated binding free energy. The docking results (predicted binding poses of the compounds) can be openly visualized by accessing http://istar.cse.cuhk.edu.hk/idock/iview/?3W2R-dbap+fda+npc(The PDB ID 3W2R can be changed to 4I22, 4RJ8, or 5HG7 to view the other docking results using the corresponding protein conformation). The top-scoring compounds were manually assessed based on in silico estimations of binding strength, appropriate molecular weight and other drug-like properties, complementary matching of molecular shape, and some sense of intuition from a computational chemist’s experience. Based on commercial availability, eight high-scoring compounds (Table 1) were shortlisted and purchased for subsequent wet-laboratory investigations.

The anticancer activity of the drug candidates was tested in two NSCLC cell lines harboring EGFR sensitizing mutation (HCC827: E746_A750 deletion) or resistance-causing EGFR secondary mutation (H1975: L858R/T790M). HCC827 is extremely sensitive to erlotinib (first-generation EGFR TKI; IC₅₀ = 6.1 ± 1.8 nM), whereas H1975 is more than 1,000-fold resistant to erlotinib (IC₅₀ = 8.2 ± 2.1 µM) (Figure 1). Among the drug candidates tested, indacaterol, canagliflozin, and cis-flupenthixol were found to exhibit anticancer activity in both HCC827 and H1975 cells (Figure 1). Importantly, indacaterol and canagliflozin were both significantly more potent in inhibiting the growth of H1975 than HCC827 (indacaterol: 14.3 ± 2.2 µM in H1975 versus 41.5 ± 3.9 µM in HCC827; canagliflozin: 25.5 ± 3.1 µM in H1975 versus 45.4 ± 2.9 µM in HCC827). The other five drug candidates tested did not affect cell proliferation of HCC827 and H1975. Representative dose–response curves for two negative drug candidates (silfenafil and ketanserin) are shown in Figure 1. To assess the potential toxic effect of the drug candidates in normal lung tissue, cytotoxicity in a normal lung epithelial cell line BEAS-2B was also evaluated. While erlotinib was found to have an IC₅₀ of around 20 µM, indacaterol, canagliflozin, and cis-flupenthixol only reduced cell viability by less than 20% at their highest concentration tested (indacaterol: 100 µM; canagliflozin: 100 µM; cis-flupenthixol: 20 µM).

Since the positive drug candidates were identified by idock to interact with T790M EGFR, we hypothesize that their anticancer activity was contributed by their EGFR kinase inhibitory activity. The inhibitory effect of the positive drug candidates on the kinase activity of wild-type, L858R and L858R/T790M EGFR recombinant protein were evaluated in cell-free system. All of the three positive drug candidates were found to inhibit L858R and L858R/T790M EGFR to different extent (Table 2), presumably leading to the anticancer activity. Consistent with the preferential anticancer activity of indacaterol and canagliflozin in H1975 cells (harboring EGFR L858R/T790M resistance-causing mutation) over HCC827 cells (harboring the sensitizing EGFR E746_A750 deletion), the two drug candidates were found to inhibit EGFR L858R/T790M more significantly than EGFR L858R (Table 2). While the adverse effects of EGFR TKIs are known to be due to the inhibition of wild-type EGFR, the IC₅₀ of the three positive hints in wild-type EGFR were at least two times (for cis-flupenthixol) higher than that in EGFR L858R/T790M.

The inhibition of EGFR autophosphorylation by the three positive drug candidates were also examined in HCC827 and H1975 cells. Consistent with the data in the cell-free system, both indacaterol and canagliflozin were found to inhibit EGFR phosphorylation more potently in H1975 than in HCC827 cells (Figure 2; Figure S1 in Supplementary Material; albeit more obvious for indacaterol). On the other hand, cis-flupenthixol was found to inhibit EGFR phosphorylation similarly in both H1975 and HCC827 cells (Figure S2 in Supplementary Material). The phosphorylation status of Akt and ERK was chosen as marker to illustrate the inhibition of the EGFR downstream survival signaling pathways. Parallel to the inhibition of EGFR autophosphorylation by indacaterol and canagliflozin, these two positive drug candidates were also found to inhibit the phosphorylation of Akt and ERK more remarkably in H1975 than in HCC827 (Figure 2; Figure S1 in Supplementary Material).

Consistent with the anticancer activity found for indacaterol, canagliflozin and cis-flupenthixol in NSCLC cells, these drugs were also found to induce apoptosis in a concentration-dependent manner in the EGFR TKI-resistant H1975 cells (Figure 3). Since the key apoptotic effector PUMA is required for EGFR TKI inhibition-mediated apoptosis (35), the induction of PUMA by the repurposed drugs were also evaluated by Western blot analysis. Consistent with the more remarkable inhibition of EGFR in H1975 cells than in HCC827 cells by indacaterol and canagliflozin, both drugs were also found to give rise to more pronounced induction of PUMA in H1975 cells than in HCC827 cells (Figure 2; Figure S1 in Supplementary Material).

Combinations of gefitinib (a representative first-generation EGFR TKI) with the most potent positive drug candidate (indacaterol) were evaluated in H1975 cells. Equipotent concentration ratio of gefitinib (10 nM–10 µM) and indacaterol (10 nM–10 µM) was used in the combination treatment. The concentrations were chosen so that each drug in the combination contributed equally to the overall anticancer effect. To determine whether the sequence of drug addiction affects the resulting drug combination effect, H1975 cells were treated concomitantly (for 24 h) with gefitinib and indacaterol, or 24 h with indacaterol, followed by drug-free washout and treatment with 24-h gefitinib, or vice versa. The CI was adopted as a measure of combination effect. The effects of the drug combinations are shown in the CI-fraction affected (Fa) plot (Figure 4A). For all drug combination sequence, a CI remarkably <1 was observed over a wide range of growth inhibition levels. At 50% growth inhibition level, a CI of around 0.3 was obtained, indicating a strong synergistic effect.

H1975 cells were treated with a combination of gefitinib (4 µM) and indacaterol (5 µM) concomitantly for 48 h, after which the extent of apoptosis was measured. While both gefitinib and indacaterol alone at the concentration tested only caused mild apoptosis (<10%), their combination was found to dramatically increase the proportion of apoptotic cells [27.8 ± 2.5% for drug combination versus 6.9 ± 1.2% (gefitinib alone) or 9.7 ± 1.8% (indacaterol alone); p < 0.01] (Figures 4B,C).

The activation of EGFR and its downstream signaling molecules (Akt and ERK) were also evaluated after concomitant combination of gefitinib (10 µM) and indacaterol (10 µM) for 4 h. While both gefitinib and indacaterol were only able to inhibit phosphorylation of EGFR, Akt, and ERK mildly, the drug combination was found to inhibit these signaling molecules more profoundly than the drugs alone (Figure 5).

Figure 6A plots the macromolecular surface of EGFR L858R/T790M as well as the putative binding pose of indacaterol predicted by the docking software. Carbons, oxygens, nitrogens, and sulfurs were, respectively, rendered in gray, red, blue, and yellow. It can be seen that the indacaterol molecule was buried deep down in the binding pocket, with an oxygen atom acting as a hydrogen-bond acceptor and creating a hydrogen bond with the donor atom of the backbone of “gatekeeper” MET790. We hypothesize that the formation of the hydrogen bond might possibly permit the indacaterol molecule to be inserted into the binding cavity and therefore explain its specificity to EGFR T790M expressing cancer cells.

Figure 6B shows the same predicted conformation of indacaterol upon binding to EGFR L858R/T790M, but with the macromolecular surface hidden in order to better inspect the intermolecular interactions, depicted by dashed lines. In addition to the hydrogen bond with MET790, indacaterol was estimated to establish two hydrogen bonds with ASP855 (cyan color), a salt bridge with ASP855 (purple color), a parallel displaced π stacking with PHE723 (pink color), a cation–π interaction with PHE856 (pink color), and some hydrophobic contacts with LYS745, LEU747, LEU788, and ARG858 (green color).