Discovery of a dual inhibitor of NQO1 and GSTP1 for treating glioblastoma

The human glioblastoma cell line U87MG was stably transfected with vector control, pLHCX-EGFRvIII, pGFP-NQO1 and pRS-GSTP1, which were supplemented with various antibiotics. For EGFRvIII, 150 μg/mL of hygromycin was used; for pGFP-NQO1 and pRS-GSTP1, 0.7 μg/mL of puromycin was employed. The cells were supplemented with 10% fetal bovine serum (Hyclone, USA), penicillin (100 U/ml) and streptomycin (100 U/ml) (ABAM Life Technologies, USA) in a humidified incubator with 5% CO₂ at 37 °C.

The development of the uHTS enabled the discovery of genomic selective cancer cell growth inhibitors. The uHTS CellTiter Blue cell viability assay in 1536-well format rapidly screens for the compounds that selectively kill GBM cells with EGFRvIII cells. A parallel assay system was established for this purpose. The uHTS campaign was carried out at the Emory Chemical Biology Discovery Center (ECBDC) [22]. Screening data were analyzed using CambridgeSoft Bioassay software [23]. Z’ factors are greater than 0.5 across the screening plates, indicating a robust assay for uHTS. Z’ factor is calculated with the following equation: Z’ = 1-(3SD background + 3SD control)/(FI control – FI background). The background is defined by the average fluorescence intensity (FI) signal from wells with medium and without cells. The DMSO control is defined by the FI signal from wells with cells and with 1% vehicle (DMSO), but without compound. 0.5 < Z’ < 1 indicates a robust assay for HTS [24]. The effect of the compound on the cell growth and proliferation was expressed as % of control based on per plate and is calculated as the following equations:

For sphere formation experiments, cell numbers were calculated and cells were plated into 96-well plates at a density of 100, 50, 25, and 12 cells per well (24 wells for each density). Cells were kept in the incubator for 2 weeks before sphere formation was assessed and images were taken. Spheres larger than 10 cells in diameter were considered for analysis. The number of wells forming spheres was used as input to the Walter and Eliza Hall Institute Bioinformatics Division ELDA analyzer (https://​bioinf.​wehi.​edu.​au/​software/​elda/​) in order to obtain stem cell frequencies [25].

In order to produce alcohol or TIZ-agarose, 0.75 g lyophilized epoxy-activated agarose with a C₁₂ spacer was suspended in 10 ml distilled water and centrifuged at 500 rpm for 8 min. Washing in water was repeated twice, with one of which using the coupling buffer (0.1 M NaHCO3, pH 9.5). After the last wash, 15 mg TIZ or 200 μl alcohol was added and the coupling buffer was added to a maximum volume of 5 ml. The mixture was incubated overnight at 37 °C under slow but continuous shaking in order to allow the coupling of the epoxy group to TIZ or alcohol. The resulting column medium was then transferred to a chromatography column (Novagen, Merck, Darmstadt, Germany), and the column was washed with coupling buffer (20 ml). After that, it was washed with ethanolamine (1 M, pH 9.5) for room temperature overnight in order to block residual reactive groups. Finally, the column was extensively washed with PBS and PBS-DMSO (1:1) to remove unbound TIZ or alcohol. Protein extraction from U87MG/EGFRvIII cells was loaded with a flow rate of 0.1 ml/min. The column was washed with PBS until the baseline was flat. Proteins binding to alcohol or TIZ columns were incubated with 1 mM MNPC overnight. PBS followed by elution with a pH shift (100 mM glycine, pH 2.9) in order to remove nonspecifically bound proteins [26]. Silver staining was applied according to the method of the previous study [27].

The protein samples were in-gel digested with 10 ng/μl Glu-C. Then, the peptide samples were resuspended in loading buffer (0.03% trifluoroacetic acid, 0.1% formic acid, and 1% acetonitrile), then loaded onto a 20-cm nano-HPLC column and finally generated by a Dionex RSLCnano UPLC system (Thermo, USA). Peptides were ionized with 2.0 kV electrospray ionization voltage from a nano-ESI source on an Orbitrap Fusion mass spectrometer (Thermo, USA) [28].

Experiments were performed on a Biacore X100 system (Biacore AB, Uppsala, Sweden). Recombinant human NQO1 and GSTP1 dissolved in 10 mM sodium acetate buffer (pH = 5.0) were covalently immobilized in the dextran matrix of a CM5 sensor chip with the Amine Coupling Kit using a standard primary amine coupling procedure. The compound MNPC was injected into the flow cells in running buffer at a flow rate of 30 μL/min for 120 s of association phase, followed by a 120-s dissociation phase and a 30-s regeneration phase. The surface of the sensor chip was regenerated via the injection of 10 μL of the regeneration buffer (5 mM NaOH). The association rate constant k_a and dissociation rate constant k_d were calculated and analyzed using the monovalent analyte model, and the equilibrium dissociation constant (KD) was calculated (KD = k_d/k_a) [29].

The enzymatic activity of GSTP1 and its mutations was measured through the increased absorbance at 340 nm, which derived from the conjugation of reduced glutathione (GSH) and 1-chloro-2,4-dinitrobenzene (CDNB). The conjugation was initiated by adding CDNB to the mixture of GSH and GSTP1 or its mutants, and OD was immediately and continually measured for 20 min at 25 °C. The final assay mixture included 1 mM CDNB, 1 mM GSH in the buffer of 100 mM potassium phosphate pH 6.5, 1 mM EDTA. The inhibitory potency of MNPC versus GSTP1 or mutants was determined by calculating the residual activity after incubated with MNPC for 30 min at room temperature. Parallel control was run to monitor spontaneous conjugation of GSH and CDNB in the absence of the enzyme, and absorbance change of the control was used for the correction of an enzymatic reaction. The inhibition rate in triplicate was calculated by GraphPad Prism software (GraphPad, San Diego, CA) to give IC₅₀ using nonlinear regression analysis.

Since Tris buffer is incompatible with MST labeling, GSTP1 and its mutants were performed an initial buffer exchange to 20 mM HEPES pH 7.6, 200 mM NaCl before labeling. Then, the proteins were fluorescently labeled using Monolith Protein Labeling Kit RED-NHS 2^nd generation dye (NanoTemper Technology, Munich, Germany) according to the protocol. MNPC was prepared with a series of concentrations (varying from 2 mM to 11 nM) at a dilution ratio of 1: 2. Six microliters of the fluorescence-labeled GSTP1 or mutations was mixed with 6 μl of variable concentration of MNPC and incubated for 10 min at room temperature. The mixtures were loaded into standard glass capillaries and analyzed on Monolith NT.115 at 25 °C, with 40% LED power and 100% Laser power in triplicate. Data were analyzed by NTAnalysis software to calculate the K_d values.

Human NQO1 (GenBank: NP_000894.1) and GSTP1 (GenBank: AAH10915.1) genes were cloned into pET28a with an N-terminal hexahistidine tag separated by a thrombin site and expressed in Escherichia coli strain BL21 (DE3). Bacterial culture was grown in LB medium with 35 μg/ml of kanamycin at 37 °C until OD₆₀₀ reached 0.6 to 0.8 and then induced by adding 0.4 mM isopropyl-L-thio-B-D-galactopyranoside (IPTG) for 16 h at 20 °C. Recombinant NQO1 proteins were purified as follows: after harvest by centrifugation, cells were lysed in 10% glycerol, 1% TritonX-100, 200 mM NaCl, 10 mM imidazole and 100 mM Tris (pH 7.6) supplemented with 1 mM phenylmethanesulfonyl fluoride (PMSF). Soluble protein was separated from the cleared cell lysate by centrifugation at 21,000 g 40 min, then submitted to Ni–NTA resin (Qiagen) with an elution buffer of 200 mM NaCl, 150 mM imidazole and 20 mM Tris (pH 7.6). Protein was then concentrated and loaded onto a Superdex 200 10/300 GL (GE Healthcare) pre-equilibrated with 200 mM NaCl, 20 mM Tris (pH 7.6). Recombinant GSTP1 protein was purified as described above, except with a slight difference in buffer composition. For GSTP1, β-mercaptoethanol was added to all buffers to a final concentration of 2 mM. The purity of NQO1 and GSTP1 was confirmed by SDS-PAGE and Coomassie blue staining.

Crystals of the NQO1 complex with MNPC were obtained by co-crystallization with the sitting drop vapor diffusion method. Purified NQO1 was concentrated to 12 mg/mL and then incubated with MNPC at a molar ratio of 1:3 over ice for 1 h. One microliter of NQO1-MNPC solution was mixed with 1 μL of mother liquor and further equilibrated with reservoir solution at 20 °C. Crystals appeared in a week, with a crystallization condition of 0.2 M lithium sulfate, 1.8 M ammonium sulfate, 0.1 M imidazole pH 7.0. The crystals were cryoprotected using the crystallization solution with 20% glycerol and then flash-frozen directly into liquid nitrogen.

The attempt was also made to obtain crystals of the GSTP1–MNPC complex. GSTP1 with a concentration of 10 mg/ml was used for crystallization, and the solution of the GSTP1–MNPC mixture was generated just as NQO1-MNPC. Crystals appeared in one day or two in the condition of 0.1 M MES PH5.4, 30% PEG8000, 10 mM DTT, 20 mM CaCl₂, and grew in a week to the maximum size at 20 °C. After crystals grew to the full size, the crystallization condition was supplemented with MNPC of final concentration 3 mM. After soaked for 4 h, crystals were then flash-frozen in liquid nitrogen until data collection.

Diffraction data were collected at the Shanghai Synchrotron Radiation Facility (SSRF) at beamline 17U1, 18U1 and 19U1. The data were measured from a single crystal maintained at 100 K at a wavelength of 0.9789 Å, and the reflections were indexed, integrated, and scaled by HKL2000 [30]. The structure of NQO1–MNPC complex was solved by molecular replacement using the program PHASER in the PHENIX package [31] with the search model of PDB ID 2F1O for NQO1, 3GUS [32] for GSTP1, followed by repeated cycles of model building with　Coot [33] and refinement with REFMAC [34] and PHENIX, yielding the published NQO1–MNPC complex structure (PDB ID 6LLC). The solvent, ligand and inhibitor were built into the density in later rounds of the refinement. Data collection and refinement statistics are shown in Extended Table 2.

ICM 3.8.2 modeling software on an Intel i7 4960 processor (MolSoft LLC, San Diego, CA) was used to perform molecular docking. GSTP1 model was obtained for protein data bank (PDB ID 3GUS) (Federici et al., 2009). MNPC was input as the 3D compound and calculated according to the internal coordinate mechanics (Internal Coordinate Mechanics, ICM) [35].

In vitro assessment was implemented for cell proliferation using the 3-(4, 5- dimethylthiazol- 2-yl)-2,5-diphenyltetrazolium bromide (MTT) assay. After different treatments with specific time periods, 10 μL of MTT (5 mg/mL in PBS, Sigma, USA) was added to each well and the plates were incubated for 3 h at 37 °C. The resulting formazan product was dissolved with DMSO. The absorbance at a wavelength of 490 nm was recorded using a microplate reader (BioTek Instruments Inc., USA).

The DCFH-DA method was employed to detect the levels of intracellular reactive oxygen species (ROS). After various treatments, the cells were collected and then incubated with 5 μM DCFH-DA (ROS dye, #C6827, Invitrogen, USA) for 1 h at 37 °C. The fluorescence intensity was measured by the microplate reader (BioTek Instruments Inc., USA) with settings at excitation and emission equal to 485/535 nm.

After collecting the supernatants, the cytokine concentrations were measured using LDH assay kits (Promega Corporation, USA). Protein carbonyl level and GSH/GSSG ratio were measured from cell homogenates using Protein Carbonyl Assay Kit (#ab126287, Abcam, Cambridge MA, USA) and GSH/GSSG-Glo™ (Promega Corporation, USA), respectively.

After different treatments with the conditions described, the cells were harvested and the total proteins were extracted. Equal amounts of the proteins were loaded on SDS-PAGE and western blot assays were analyzed. Primary antibodies were from CST, USA, and the following targets were used: p-EGFR (#2236); EGFR (#2232); NQO1 (#3187); GSTP1 (#3369); cleaved caspase 3 (#9664); β-actin (#3700).

After different treatments, cell lysates were incubated in 200 μl assay buffer (20 mM citric acid, 60 mM Na₂HPO₄, 1 mM EDTA, 0.1% CHAPS and 1 mM DTT, pH 6.0) containing 20 μM AEP substrate Z-Ala-Ala-Asn-AMC (Bachem, USA). AMC released by substrate cleavage was quantified by measuring every 10 min at 460 nm in a fluorescence plate reader at 37 °C in kinetic mode for the total time of 1 h.

A total of 1 × 10⁴ cells were seeded onto the upper part of a transwell chamber (BD Bioscience, USA) containing a membrane filter for migration assays. Serum-free medium was added to the upper well, and medium containing 10% FBS was added to the lower well. After different treatments and incubation at 37 ̊C with 5% CO₂, the filters were stained with crystal violet. Five random fields were counted per chamber by using a microscope.

NQO1 activity was measured essentially as described previously. Assays were performed in 25 mM Tris–HCl, pH 7.4, 0.7 mg/mL BSA, 250 mM sucrose, 0.2 mM NADPH and 40 μM dichlorophenolindophenol (DCPIP) in the presence of NQO1 protein with 10, 5, 2.5, 1.25, 0.625, 0.3125, 0.15625 and 0 μM of MNPC. It is analyzed for the reduction of DCPIP measured at 600 nm [36].

After different treatments, the cells and the tumor tissues were processed with Apo-Direct TUNEL Assay (Roche Applied Science, Germany) following the manufacture’s instruction. The slides were photographed with a fluorescence microscope (Nikon, Japan).

The NQO1-siRNA(#sc-37139), GSTP1-siRNA (#sc-72091) and nontargeting siRNA (#sc-37007) as control were purchased from Santa Cruz Biotechnology. HA-NQO1 plasmid was purchased from Addgene. Myc-DDK-GSTP1 (#RC203086) was purchased from Origene. The U87 MG and U87MG/EGFRvIII cells were transfected with 20 nM siRNA or 2 μg plasmid using the Lipofectamine 3000 and P3000 (#L3000075, Invitrogen, USA) according to the manufacture’s protocol.

Animals were housed, maintained and treated at Emory University in accordance with protocols approved by the Institutional Animal Care and Use Committee (IACUC) at Emory University. For xenograft animal models, U87MG/EGFRvIII cells (2 × 10⁶) in 100 μl of PBS were inoculated subcutaneously into 6-week-old nude mice. Tumor growth was assessed every 3 days via size and body weight measurements. The total tumor volume (TV) was calculated according to the following formula: TV (mm³) = a * b²/2, where “a” is the minimum diameter and “b” denotes the maximum diameter. The mice were euthanized after 28 days. For the intracranial model, mice were placed in a stereotaxic instrument, and cells injection (1 × 10⁵) in 2 μl was performed stereotaxically at coordinates anteroposterior (AP) -2.0 mm and mediolateral (ML) + 0.7 mm relative to bregma, and dorsoventral (DV) -3.0 mm from the dural surface. The needle remained in place for 5 min before it was removed slowly. The mice were placed on a heating pad until it began to recover from the surgery. After the surgery for 8 days, intraperitoneal (i.p.) injection was used at dose levels of 3 mg/kg, 10 mg/kg or control agent. This treatment was performed every 2 days for a total of 10 times. After the drug treatment, mice were euthanized and the tumor volumes were analyzed by MRI [37].

The tumors and primary organs from the nude mice of the above models were fixed in 10% formalin overnight and were then embedded in paraffin. Sections were prepared, and H&E staining was conducted to detect any histological changes of the tumors and organs. The expression of Ki67 in the tumor tissue slices was assessed using a technique that has been reported previously. Photographs were taken using a fluorescence microscope (Nikon, Japan).

Bioinformatic data analysis was obtained from the TCGA data portal (https://​cancergenome.​nih.​gov/​dataportal/​data/​about), UALCAN (https://​ualcan.​path.​uab.​edu/​index.​html) [38], GlioVis (https://​gliovis.​bioinfo.​cnio.​es)[39], respectively.

Data visualization and analysis were performed with GraphPad Prism 6 (GraphPad Software Inc., La Jolla, CA, USA). Statistical analysis was performed using either Student’s t test or one-way ANOVA. Significant difference among groups was assessed as *p < 0.05; **p < 0.01; ***p < 0.001.

To search for the pharmacological agents for potential GBM therapeutics, we employed an ultra-high-throughput screen (uHTS) in a 1536-well plate format and testing funnels to screen and identify the small molecules, which differentially inhibited the growth of U87MG/EGFRvIII cells against U87MG (PTEN mutant) parental cells. While the precise origin of U87MG cells is in the debate, recent studies demonstrate that it is GBM-derived and thus appropriate for these screening studies [40]. After screening more than 38,000 compounds, we obtained 15 positive hits with different chemical backbones that showed > 50% inhibition of cell viability in U87MG/EGFRvIII cells but < 25% inhibition in U87MG cells (Fig. 1a). After the structure–activity relationship (SAR) analysis, we found that most compounds contained two aromatic rings bridged through an amide group. Namely, these compounds possessed N-phenylbenzamide or N-phenylthiophene/furan-2-carboxamide moieties. The representative compound MNPC (5-methyl-N-(5-nitro-thiazol-2-yl)-3-phenylisoxazole -4-carboxamide) displayed an IC₅₀ of 1.32 μM in inhibiting U87MG/EGFRvIII cell proliferation (Fig. 1b and c). SAR studies also demonstrated that 3-phenyl-5-methyl isoxazole group and the nitro group on MNPC could be replaced without substantially affecting the inhibitory activity (Additional file 1: Supplementary Fig. 1), suggesting that these groups are not essential for the inhibitory function. On the other hand, compared to oxazole, thiazole substitution increased the inhibitory effect (Fig. 1d), indicating that a size increase in X position elevates its binding to the target. Hence, the MNPC was selected for further study for its anticancer functions. To determine whether MNPC exerts similar effects on primary GBM cells, we performed limiting dilution sphere-forming assays on cultures with EGFRvIII mutation as well as two other EGFR wild-type cultures. MNPC diminished sphere formation and dramatically suppressed the growth of spheres in the EGFRvIII mutant cultures (Fig. 1e-g and Additional file 1: Supplementary Fig. 2a) with little or no effect on the EGFR wild-type cultures (Additional file 1: Supplementary Fig. 2b and c). Hence, MNPC selectively blocks the proliferation of a standard cell line and primary GBM cells bearing the EGFRvIII mutation.

Since MNPC was the most active compound against EGFRvIII-expressing GBM in the cell growth inhibition experiments, we sought to identify the cellular targets of this compound. Further SAR study revealed that nitazoxanide (NTZ) and tizoxanide (TIZ), two FDA-approved drugs for treating antiparasitic and cryptosporidium infection, displayed anti-proliferative activities in U87MG/EGFRvIII cells to a comparable extent as MNPC (Additional file 1: Supplementary Fig. 2d and e). This finding supported the hypothesis that 5-nitro-thiazol-4-carboxamide is the crucial pharmacophore for MNPC to exert its selective anticancer activity. To determine candidate protein targets that bind both TIZ and MNPC, we subjected U87MG/EGFRvIII cytoplasmic extracts to an affinity column prepared by covalently coupling TIZ to the epoxy-agarose via its phenol group with subsequent elution with free MNPC solution. Two major proteins with the molecular weight of 24 kDa and 29 kDa, as visualized by silver staining, were specifically observed in the MNPC fraction but not in the control fraction. Proteomic analysis of the eluted proteins revealed their identities: NQO1 and GSTP1 (Fig. 2a and b). To determine whether MNPC, indeed, specifically binds to NQO1 and GSTP1, we performed the BIACORE binding assay by immobilizing MNPC to the chip and found that MNPC exhibited high affinity to NQO1 and GSTP1 recombinant proteins, respectively (Fig. 2c).

Next, we obtained the co-crystals of MNPC with NQO1 (PDB ID 6LLC) and GSTP1 (PDB ID 6LLX). The crystal structure of MNPC in complex with NQO1 was solved at 2.5 Å (Fig. 2d and e) and revealed that MNPC bound with NQO1 at the same active site as the known inhibitor dicoumarol (PDB ID 2F1O) [41], where MNPC deeply resided in the pocket and was oriented above the isoalloxazine ring of bound cofactor FAD by π-π stacking interaction. The active pocket of NQO1 was formed by two enzyme monomers with head and tail inlaid to each other, in which several aromatic amino acids composed a relatively hydrophobic pocket to contain FAD and MNPC. Phe178 and the isoalloxazine ring of FAD formed π-π stacking sandwich with the benzene ring of MNPC. Tyr126, Tyr128, M131, Phe236 from one protein monomer and Trp105, Phe106, Met154, Tyr155 from another monomer formed strong hydrophobic interactions with MNPC. Interestingly, the hydrogen bonding between His161 with the amide carbonyl group of MNPC and the hydrophobic interaction between M131 with the thiazole ring of MNPC locked MNPC in a spatial orientation in which the plane of thiazole ring was perpendicular to the plane of the benzene ring and FAD. Attempts at co-crystallization between MNPC with GSTP1 revealed some electron density in the putative inhibitor binding site. However, it could not be fitted for the whole MNPC molecule. Therefore, we performed the molecular docking and subsequent mutagenesis analysis to elucidate the binding mode of MNPC with GSTP1. From the generated docking model, we found that MNPC was located in the active pocket of GSTP1, closely adjacent to the substrate GSH. It formed two hydrogen bindings with Arg100 and Tyr108 and one hydrogen bonding with GSH. MNPC also exhibited the π-π stacking with Phe8 and Tyr108 (Fig. 2f). To further verify the docking results, site-directed mutagenesis was conducted to generate several mutants (F8A, R100A, V104A, Y108A, F8A/Y108A), and inhibition of enzymatic activity against these mutants was measured. As expected, inhibitory activities of MNPC against mutants F8A, R100A, Y108A, F8A/Y108A were much weaker than that of wild-type GSTP1 and V104A mutant, with IC₅₀ of 2.42 ± 0.36 μM, 2.37 ± 0.38 μM, 25.74 ± 1.41 μM, 4.01 ± 0.60 μM, respectively, while MNPC had an IC₅₀ of 0.40 ± 0.39 μM and 0.65 ± 0.19 μM against wild-type GSTP1 and V104 mutant, respectively (Additional file 1: Supplementary Table 1, Additional file 1: Supplementary Fig. 2f and 2g). Importantly, the inhibition of MNPC decreased the most with Y108A mutant, which was consistent with the docking results that Tyr108 formed both hydrogen bonding and hydrophobic interaction with MNPC. To further verify these observations, a microscale thermophoresis (MST) assay was employed to measure the binding ability of MNPC with GSTP1 and its mutants. While K_d values of wild-type GSTP1 and V104A mutant were 17.60 ± 0.80 μM and 24.40 ± 1.95 μM, respectively, other mutants showed no detectable binding. Taken together, the data show that Tyr108 is the most important amino acid for binding of GSTP1 with MNPC, while Phe8 and Arg100 also participate in the interaction between MNPC and GSTP1. Val104 is not involved in the interaction and its mutant could then act as a negative control.

To explore the anti-proliferative effects of MNPC, we treated U87MG/EGFRvIII cells with different doses of MNPC (0, 0.5, 1, 2 and 5 μM) for 5 days and found that it dose-dependently inhibited the growth of the cultures (Fig. 3a). Cell proliferation and LDH assays demonstrated MNPC barely affected the cell proliferation of normal cells (Additional file 1: Supplementary Fig. 3a and b), indicating that the toxicity is negligible. We found that MNPC had no effect on migration (Additional file 1: Supplementary Fig. 3c), nor inhibited a key enzyme AEP (asparagine endopeptidase) (Additional file 1: Supplementary Fig. 3d) in migratory processes [42]. Since NQO1 and GSTP1 were the major binding proteins of MNPC, we conducted in vitro reductase enzyme activity assays with purified recombinant proteins and found that MNPC inhibited NQO1 and GSTP1 activities with IC₅₀ values of 0.926 and 0.797 μM, respectively (Fig. 3b and c). NQO1 acts as an antioxidant enzyme by regenerating antioxidant forms of ubiquinone and vitamin E quinone [43]. We also examined the cellular ROS concentrations and cytotoxicity after MNPC treatment. MNPC treatment resulted in elevated ROS levels and induced LDH release and carbonyl expression in U87MG/EGFRvIII cells, while downregulating GSH/GSSH ratios in a dose-dependent manner (Fig. 3d-g). Hence, MNPC may regulate oxidative stress and modulate redox homeostasis via inhibiting both NQO1 and GSTP1. Notably, the protein levels of NQO1 and GSTP1 were not changed upon MNPC treatment. Thus, MNPC inhibited GSTP1 and NQO1 enzyme activities without promoting a reflex upregulation of GSTP1 and NQO1 levels that could potentially mediate resistance. MNPC treatment-induced caspase 3 activation in U87MG/EGFRvIII cells (Fig. 3h), indicating that the escalated oxidative stress triggers apoptosis. The results were also confirmed by the TUNEL assay (Fig. 3i). In sum, these findings suggest that reducing oxidative stress by reductive enzymes like NQO1 and GSTP1 is critical for GBM cell proliferation.

To validate the roles of NQO1 and GSTP1 in the specific oxidative defense of EGFRvIII-expressing cells, we used knockdown and overexpression strategies. Knocking down either NQO1 or GSTP1 selectively decreased the cell proliferation in U87MG/EGFRvIII cells but not U87MG cells, and knockdown of both enzymes in combination produced an even greater effect (Fig. 4a and Additional file 1: Supplementary Fig. 4a). As expected, the ROS levels and LDH levels were preferentially increased in U87MG/EGFRvIII but not U87MG cells upon NQO1 and GSTP1 deletion (Fig. 4b and c). Protein carbonyl expression and GSH/GSSG ratios confirmed these findings (Fig. 4d and e). NQO1 and GSTP1 knockdown activated apoptotic caspase 3 in U87MG/EGFRvIII cells (Additional file 1: Supplementary Fig. 4b), fitting with MPNC’s pro-apoptotic effect. Overexpression of NQO1 and GSTP1, on the other hand, resulted in diminished ROS and enhanced cell growth in EGFRvIII-expressing but not EGFR-WT U87MG cells (Fig. 4 and Additional file 1: Supplementary Fig. 4). The studies thus far indicate that NQO1 and GSTP1 preferentially regulate the oxidative stress and cell viability in U87MG/EGFRvIII in comparison with U87MG cells.

To further examine the effects of NQO1 and GSTP1 on U87MG/EGFRvIII cancer cell proliferation in vivo, we constructed stable knockdown cell lines that expressed their specific shRNAs. The transfection efficiency was validated by fluorescence microscopy and Western blotting, as presented in Fig. 5a and b. Next, U87MG/EGFRvIII cells stably transfected with sh-NQO1 or sh-GSTP1 or both were subcutaneously inoculated into nude mice. Representative photographs of mice from each group were taken at the endpoint (Fig. 5c). As shown in Fig. 5d, knockdown of either NQO1 or GSTP1 separately significantly decreased tumor volumes and weight, and combined knockdown resulted a much greater effect. H&E staining of tumor slices in dual knockdown groups revealed tumor tissue damage and morphology differences such as sparser cellularity as compared to other groups. Immunohistochemical (IHC) staining for Ki67 showed an almost 40% reduction in the number of proliferating cells in the dual inhibition group compared to the control group. TUNEL assay indicated extensive apoptosis in tumor tissues, suggesting that depletion of either NQO1 or GSTP1 triggers apoptosis, with the combined knockdown producing more extensive apoptotic cell death than the other groups (Fig. 5f). Thus, NQO1 and GSTP1 played a critical role in EGFR-vIII-positive GBM cell survival in vivo, supporting that they are the promising targets for GBM therapy.

To gain insight into the roles of NQO1 and GSTP1 in human GBM, we analyzed a dataset from The Cancer Genome Atlas (TCGA) featuring expression and the corresponding clinical data of selected GBM patient samples. Notably, NQO1 and GSTP1 expression levels were significantly increased in the samples of high grade to low grade (Additional file 1: Supplementary Fig. 5A). Consistent with our findings in primary GBM cells, a significant positive correlation was observed between the expression of NQO1 and GSTP1 (Additional file 1: Supplementary Fig. 5B). When compared with normal brain tissues, GSTP1 was highly expressed in GBM patient samples (Additional file 1: Supplementary Fig. 5C), with no significant sex difference (Additional file 1: Supplementary Fig. 5D). However, NQO1 expression only slightly changed in different ages of patients (Additional file 1: Supplementary Fig. 5E). Interestingly, both NQO1 and GSTP1 expression levels were higher in colon and lung cancers as well as compared to the normal tissues (Additional file 1: Supplement Fig. 6A and B), which was also reported in a previous study [44, 45], though no significant influence on stomach cancer was found (Additional file 1: Supplement Fig. 6C). Hence, these findings suggested that NQO1 and GSTP1 were upregulated in the tumor tissues of GBM patients, and their expression was correlated.

To further investigate the cell proliferation and ROS status relationship among NQO1, GSTP1 and MNPC, we overexpressed NQO1, GSTP1 or both, treated the cells with MNPC or vehicle in U87MG or U87MG/EGFRvIII cells. As described above, both NQO1 and GSTP1 overexpression enhanced cell proliferation, which was reversed by MNPC (Fig. 6a). Accordingly, overexpression of NQO1 and GSTP1 significantly repressed the ROS levels (Fig. 6b), LDH levels (Fig. 6c) and carbonyl expression (Fig. 6d) in U87MG/EGFRvIII cells but not U87MG cells (Additional file 1: Supplementary Fig. 7A-C), while MNPC antagonized these effects. Moreover, MNPC also reversed the upregulated GSH/GSSH levels induced by NQO1 and GSTP1 (Fig. 6e). Immunoblotting analysis indicated that MNPC increased cleaved-caspase 3 concentrations when both NQO1 and GSTP1 were highly expressed (Fig. 6f), results validated with the TUNEL assay (Fig. 6g). On the other hand, MNPC had no effect on cell number (Fig. 7a), ROS levels (Fig. 7b), LDH level (Fig. 7c), carbonyl expression (Fig. 7d), GSH/GSSH ratios (Fig. 7e) or apoptosis (Fig. 7f and g) in cells in which NQO1, GSTP1 or both were depleted in U87MG/EGFRvIII cells (Fig. 7), supporting the hypothesis that both NQO1 and GSTP1 were the principal cellular targets of MNPC.

To assess the therapeutic potential of MNPC, we used a murine intracranial xenograft model [37], inoculated with U87MG/EGFRvIII cells (Additional file 1: Supplementary Fig. 8a). Seven days following tumor injection, the animals were treated with vehicle or different doses of MNPC (3 or 10 mg/kg) every 2 days for a total of 10 doses. MNPC treatment at the higher dose resulted in enhanced mouse survival (Fig. 8a), body weight (Fig. 8b) and reduced tumor growth (Fig. 8c and Additional file 1: Supplementary Fig. 8B). Next, we examined the ROS levels and signaling cascades in the tissue samples collected from tumors extracted from MNPC or vehicle-treated mice. MNPC dose-dependently increased ROS levels and active apoptotic caspase 3 and apoptosis (Fig. 8d and f). H&E staining of major organs demonstrated no detectable toxicity, including kidney, liver, spleen, heart and lung (Fig. 8e), suggesting that MNPC does not have detectable systemic toxicity. We also tested the CBC (complete blood chemistry) from the serum, ALT and AST from the whole blood and found that the side effect of MNPC is modest (Additional file 1: Supplementary Fig. 8C). Together, the data support that MNPC significantly blocks tumor growth by suppressing cell proliferation and inducing apoptosis via escalating ROS in GBM in vivo.