NQO1 is a determinant for cellular sensitivity to anti-tumor agent Napabucasin

Abstract

Napabucasin (NAPA) is thought to be a potent cancer stemness inhibitor in different types of cancer cell lines. While it has shown promising activity in early phase clinical trials, two recent phase III NAPA clinical trials failed to meet the primary endpoint of overall survival. The reason for the failure is not clear, but a possible way to revive the clinical trial is to stratify patients with biomarkers that could predict NAPA response. Here, we report the identification of NAD(P)H dehydrogenase 1 (NQO1) as a major determinant of NAPA efficacy. A proteomic profiling of cancer cell lines revealed that NQO1 abundance is negatively correlated with IC₅₀; in vitro assays showed that NAPA is a substrate for NQO1, which mediates the generation of ROS that leads to cell death. Furthermore, activation of an NQO1 transcription factor NRF2 by chemicals, including an FDA approved drug, can increase the NAPA cytotoxicity. Our findings suggest a potential use of NQO1 expression as a companion diagnostic test to identify patients in future NAPA trials and a combination strategy to expand the application of NAPA-based regimens for cancer therapy.

Introduction

Napabucasin (NAPA) is an orally administered small molecule that was first isolated from the stem bulk of Tabebuia cassinoides (Lam.) in 1982 [1]. Recent studies have shown that NAPA can block cancer growth and metastasis without significant adverse effects in toxicology assessment [2-4]. Moreover, several phase I and II clinical trials for NAPA regimens have produced encouraging anti-tumor effects as monotherapy or in combination with other chemotherapeutics in different cancer types [5-7].

Previous studies in cell lines showed that NAPA appeared to be particularly effective in killing a subpopulation of cancer stem-like cells (CSC) in bulk tumors [2,3,8,9]. It is widely used in many studies for perturbation of INF-γ, IL11- or IL6-STAT3 pathways [3,10-12], as well as inhibition of stemness in cancer cells, including attenuating the expression of Oct4, Nanog and Sox2 [4]. To date, the underlying molecular mechanism remains controversial. While earlier studies suggest that the cell death depends on NAPA’s ability to inhibit pSTAT3 pathway [11,13-15], more recent studies showed that the reduction of STAT3 phosphorylation was just an inducible event of ROS when treated with NAPA [16]. It has been demonstrated that bioactivation of NAPA by oxidoreductases, such as NQO1, is critical for NAPA-mediated cancer cell death [16]. Moreover, the only two phase III NAPA clinical trials of advanced gastric cancer (GC) and colorectal cancer (CRC) failed to meet the endpoint of overall survival [5,17]. Interestingly, a retrospective analysis showed that CRC patients who expressed pSTAT3 had an increased overall survival in the NAPA treatment group than the placebo group (5.1 versus 3.0 months) [18], suggesting the necessity of identifying patients who would likely respond to NAPA.

Here we performed a proteomic study in cancer cell lines and identify NAD(P)H dehydrogenase 1 (NQO1) as a key determinant in mediating NAPA cytotoxicity. We carried out a detailed biochemical analysis on NQO1 activity and explored feasible means to enhance NAPA efficacy, paving ways for patient selection and possible combination regimens to promote NAPA-based regimens in future clinical trials.

Materials and methods

Cell cultures, siRNA silencing and transfections

All cell lines were obtained from Cobioer (Nanjing, China) and grown in recommended medium (Fisher Scientific, Waltham, MA) listed in Table S1. Cells were kept in corresponding medium containing 10% fetal bovine serum (Fisher Scientific, Waltham, MA) in a 5% CO₂ air incubator at 37°C. DharmaFECT transfection reagent (Dharmacon, UK) was used for all siRNA transfections. The siRNAs targeting NQO1 (siRNA-1: 5’-GCAGCCUCUUUGACCUAAA-3’, siRNA-2: 5’-CAAUUACAAAGCAGUUACU-3’, siRNA-3: 5’-CGAGUCUGUUCUGGCUUAU-3’), NRF2 (siRNA-1: 5’-GGGAGGAGCUAUUAUCCAU-3’, siRNA-2: 5’-AUUCAAUGAUUCUGACUCC-3’) and negative control sequence (5’-UUCUCCGAACGUGUCACGUTT-3’) were purchased from Genepharma (Shanghai, China) and transfected at 20 nM. In the determination of IC₅₀, cells were treated with NAPA or DMSO at indicated concentrations for 24 h followed by cell viability measurement using Cell Counting kit-8 (MedChemExpress, USA). MDA-MB-231 and PANC1 cells were transfected with lentivirus packaged human NQO1 cDNA purchased from Hanbio (Shanghai, China) and selected under blasticidin for the generation of NQO1 stable lines.

Chemicals and reagents

Napabucasin (S7977), menadione (S1949), dimethyl fumarate (S2586), carnosic acid (S3838) and dicoumarol (S4299) were purchased from Selleck Chemicals and stock solutions were prepared at 50 mM in DMSO. β-lapachone was purchased from Sigma-Aldrich (L2037). All stock solutions were kept at -80°C before use. Antibodies were obtained from the following sources: Anti-NQO1 (#62262, Cell Signaling Technology); Anti-NRF2 (ab62352, Abcam); β-actin (#3700T, Cell Signaling Technology); anti-pSTAT3 (Y705, #9145T, Cell Signaling Technology) and anti-GAPDH (#2118s, Cell Signaling Technology), and used at manufacturer recommended conditions. For western blotting, proteins were detected using HRP-conjugated secondary antibodies (Cell Signaling Technology). NQO1 recombinant protein were purchased from Sigma-Aldrich (#D1315).

Cell viability and IC₅₀ measurement

Cells were seeded in a 96-well plate at corresponding density as recorded in Table S1 and allowed to attach overnight, confirming about 60% cell confluency when NAPA addition. For IC₅₀ determination, NAPA were added with at least six gradient concentrations and tested after 24 h treatment. Each concentration was replicated four times. For other cell viability assessments, cells were incubated with corresponding compounds at indicated concentration before CCK8 assays (MedChemExpress, USA) as described in the manufacturer’s protocol.

Protein preparation and LC-MS/MS analysis

Cells were scraped and lysed in 5 times volume of lysis buffer (1% sodium deoxycholate [W:V], 10 mM tris (2-carboxyethyl) phosphine, 40 mM 2-chloroacetamide (CAA), 100 mM Tris, pH 8.5) [19]. Lysate was subsequently boiled for 5 min at 95°C followed by 5 min sonication (3 s on and 3 s off, amplitude 25%) and centrifuged at 16,000 g for 10 min. Extract was digested using trypsin (V5280, Promega Corporation, USA) that cleaves at the C-terminus of the Arg or Lys residues at 37°C. Tryptic peptides were desalted in a home-made reverse-phase C18 column in a pipet tip. Peptide were eluted using a 50% acetonitrile buffer and dried in a vacuum concentrator (Thermo Scientific).

Dried peptide samples dissolved in 0.1% formic acid were loaded onto a home-made trap column (100 μm × 2 cm; particle size, 3 μm; pore size, 120 Å; Dr. Maisch GmbH) with a max pressure of 280 bar using Solvent A (0.1% formic acid in water), and separated on a home-made 150 μm × 30 cm silica microcolumn (particle size, 1.9 μm; pore size, 120 Å; Dr. Maisch GmbH) with a gradient of 5-35% mobile phase B (acetonitrile and 0.1% formic acid) at a flow rate of 600 nl/min for 150 min. LC-MS/MS was performed on an Orbitrap Fusion Lumos Tribrid mass spectrometer (Thermo Fisher Scientific, Rockford, IL, USA) coupled with an Easy-nLC 1000 nanoflow LC system (Thermo Fisher Scientific). Precursor scan was carried out in the Orbitrap by scanning m/z 300-1400 with a resolution of 120,000 at 200 m/z. The most intense ions selected under top-speed mode were isolated in the Quadrupole with a 1.6 m/z window and fragmented by HCD with normalized collision energy of 32%, then measured in the linear ion trap using the Rapid ion trap scan rate. Automatic gain control targets were 5e5 ions with a max injection time of 50 ms for full scans and 5e3 with 35 ms for MS/MS scans. Dynamic exclusion time was set as 25 s. Data were acquired using the Xcalibur software (Thermo Scientific).

Peptide identification and protein quantification

MS raw files were searched against the human National Center for Biotechnology Information (NCBI) Refseq protein database (updated on 04-07-2013, 32015 entries) by Mascot 2.3 search engine (Matrix Science Inc) implemented on the Firmiana proteomics workstation [20]. The mass tolerance was 20 ppm for precursor and 0.5 Da for product-ions collected by Fusion. Up to two missed cleavages were allowed. The minimal peptide length was seven amino acids. The search engine set cysteine carbamidomethylation as a fixed modification and N-acetylation, oxidation of methionine as variable modifications. The charges of precursor ions were limited to +2, +3, +4, +5 and +6. All identified peptides were quantified in Firmiana with peak areas derived from their MS1 intensity. Peptide FDR was adjusted to 1%. The data were also searched against a decoy database so that peptide identifications were accepted at a false discovery rate (FDR) of 1% using percolator validation based on the q-value. Proteins that had at least one unique peptide and two high-confidence peptides (mascot ion score > 20) were considered in this study. Label-free protein quantifications were calculated using intensity-based absolute quantification (iBAQ) algorithm [21] based on the area under the curve (AUC) of precursor ions. The fraction of total (FOT) was used to represent the normalized abundance of a protein across experiments. iBAQ value was converted to FOT value of each protein divided by the sum of all iBAQ values of all proteins identified in one experiment. For easy visualization of low abundant proteins, the FOT values were then multiplied by 10⁵ to obtain final iFOT numbers.

Biochemical and functional assays

NQO1 enzyme activity was measured using a commercial NQO1 Activity Assay kit (ab184867, Abcam) as described by the manufacturer’s protocol. The enzyme activity of recombinant human NQO1 protein or equal amount of cell lysates was determined by following the reduction of substrates (Menadione, Napabucasin or β-lapachone) with cofactor NADH (250 μM) and the simultaneous reduction of WST1 (tetrazolium salt) to formazan (yellow), which leads to the increased absorbance at 440 nm. Dicoumarol was used as an inhibitor for NQO1 activity and used as blank when co-treated with NAPA.

For apoptotic and cell cycle analysis, cells were seeded and allowed to attach overnight before NAPA treatment. Cells were incubated with NAPA at indicated concentrations for 6 hours and replaced with fresh culture media. We detected the cell viability within 24 h after incubation by Annexin V staining using Annexin V-FITC Apoptosis Detection Kit (#C1062M, Beyotime) or Cell Cycle and Apoptosis Analysis Kit (#C1052, Beyotime). Cell cycle distribution was analyzed on FACSAria (BD Biosciences, San Jose, CA) and calculated in FlowJo. Oxygen consumption rate was monitored by a Seahorse XF24 bioanalyzer (Seahorse Bioscience, North Billerica, MA, USA) as reported previously [22]. ROS generation assessment was performed using a DCFDA (2’,7’-dichlorofluorescein diacetate) Cellular ROS Detection Assay Kit (ab113851, Abcam). Cells were treated with NAPA for 30 min after pre-incubation with DCFDA dying solution for 45 min in the dark. The product DCF (2’,7’-dichlorofluorescein) was excited by the 488 nm laser and detected at 535 nm on a fluorescence plate reader.

Correlation, overall survival and statistical analysis

Spearman’s correlation coefficient calculation and elastic net analyses [23] were performed for protein iFOT values from the ten cell lines described in Figure 1A with the corresponding IC₅₀ values. Correlation analysis was achieved by using the corrplot and glmnet packages in R software (version 3.5.1). Kaplan-Meier survival analysis was performed using TCGA (The Cancer Genome Atlas), European Genome-phenome Archive (EGA) and Gene Expression Omnibus (GEO) datasets via the portal of Kaplan-Meier plotter as described in previous studies [24,25]. Graphs were plotted as means with bars denoting S.D, unless otherwise noted. Curve fitting, IC₅₀ calculation and two-tailed Student’ t-tests were performed with GraphPad Prism version 7.0.

Figure 1.

Identification of proteins correlated with NAPA sensitivity. (A) IC₅₀ spectrum of NAPA in cancer cells. (B) NAPA dose response curves of sensitive and insensitive cell lines whose proteomic profiles were measured. Each dot represents the means ± S.D. from 6 independent measurements. (C) Number of proteins identified in each dataset and the data quality filters used. (D) Correlation between cell sensitivity to NAPA and protein expression levels obtained from ten cell lines described in (A). (E) Heatmap of proteins with greater than 500/1000 sampling times in elastic net analysis.

Results

A proteomic approach to identify proteins that mediate NAPA sensitivity

We hypothesized that the abundance of proteins that mediate NAPA action is correlated with the cell sensitivity. To obtain NAPA cytotoxicity in cells, we first measured the IC₅₀ values for a panel of 51 cancer cell lines across different cancer types (Figure 1A and Table S1). The dynamic range of the IC₅₀ values varied from 0.3 to 11 μM. Regardless of the cancer types, cells were relatively susceptible to NAPA-mediated cytotoxicity, as the IC₅₀ values of 31 out of the 51 cell lines were below 1 μM (Figure S1A) and the IC₅₀ values of the top sensitive cell lines were all nearly 0.4 μM (e.g. AGS, HGC27, FaDu and OVCA429). However, the most resistant cell line (MDA-MB-231) showed an IC₅₀ value of 11 μM.

To find proteins that may mediate cellular sensitivity to NAPA, we performed proteomic profiling on 10 representative cell lines, including 6 insensitive cells (MDA-MB-231, HeLa, A2780, PANC1, MCF7 and SK-O-V3) and 4 least sensitive cells (AGS, HGC27, FaDu and OVCA429) (Figure 1B). Three biological replicates were performed using a one-shot, label-free quantitative MS method [26]. In total, we identified 9613 proteins with 1% global protein false discovery rate (FDR) in the 10 cell lines with high repeatability (Figure S1B). For high reliability, we selected 7218 proteins that were detected with at least 2 unique peptides and Mascot ion scores greater than 20 with 1% FDR at the peptide level for further analysis (Figure 1C and Table S2). Protein quantification was performed as previously described with the iBAQ algorithm followed by a normalization to iFOT (fraction of total iBAQ) [21].

We calculated the association between the abundance of the 7218 proteins and the IC₅₀ values of the 10 cell lines using two algorithms, namely Spearman’s correlation coefficient and the elastic net model analysis [23]. Using appearance frequency of > 500 times out of 1000 reiterations for the elastic net and the coefficient < -0.7 for the Spearman’s as cutoffs, we found 6 proteins including HIST1H1B, MBNL1, CCDC134, NQO1, AP1B1, and SEC23A among the top candidates whose abundance was negatively correlated with the IC₅₀ values (Figure 1D, 1E and Table S2), suggesting that they may mediate cellular sensitivity to NAPA.

NAPA induces inhibition of cancer cell growth in an NQO1-dependent manner

Among proteins negatively correlated to IC₅₀, NQO1 caught our attention. NQO1 is a cytoplasmic protein and a 2-electron reductase; its known substrate - menadione (MENA) [27] resembles NAPA in chemical structures (Figure 2A).

Figure 2.

Efficacy of NAPA-induced cell death depends on NQO1 protein levels. (A) Schematic of ROS generation during menadione (MENA) reduction by NQO1, which has similar structure as NAPA. (B) Cell viability analyses of control and NQO1 transient knockdown HeLa cells. Results represent means ± S.D. from 4 independent measurements (*P < 0.05, **P < 0.01, and ***P < 0.001, Student’ t-test). NC, negative control. Left penal: Western blotting of NQO1 in control and NQO1 knockdown HeLa cells. Right penal: the cell viability of HeLa cells treated with NAPA at 5 μM for 24 h. (C) Flow cytometry analysis of cell cycle distribution in cell lines (MDA-MB-231, HeLa and HGC27) treated with NAPA at 1 μM for 6 h. Left penal: cell cycle distribution. Right penal: relative protein level of NQO1 in cell lines (MDA-MB-231, HeLa and HGC27). (D) Cell viability analyses of NQO1-overexpressing PANC1 (middle) or MDA-MB-231 (right) and corresponding control lines. Right panel: western blotting of NQO1 in PANC1 and MDA-MB-231 cell lines overexpressing GFP or NQO1. (E) Western blotting of NQO1 and pSTAT3 (Y705) in 10 cell lines used in this study. (F) Cell viability of OVCA429 treated with NAPA or with DIC. Results of (B, D and F) represent means ± S.D. from 4 independent measurements (*P < 0.05, **P < 0.01, and ***P < 0.001, Student’ t-test).

To investigate whether NAPA and NQO1 has a direct relationship in cell death, we measured changes in NAPA sensitivity in HeLa cell when NQO1 was knocked down (KD). We observed that the cytotoxicity was significantly attenuated in NQO1-KD cells using small interfering RNAs (siRNA) of three independent sequences (Figure 2B). Dicoumarol (DIC) is an inhibitor of NQO1 activity [28]. Consistently, inhibiting NQO1 activity with dicoumarol (DIC) greatly reduced the cell toxicity and the proportion of Annexin V-positive cells upon NAPA treatment (Figure S2A and S2B). Similar effect was also observed in other NQO1-expressing cell lines, including FaDu, MIA PaCa2, A549 and HCT116 (Figure S2C-G). Moreover, NAPA treatment caused a profound S phase accumulation in a NAPA-sensitive HGC27 cells that expressed high level of NQO1 (Figures 2C and S3), but had no obvious change in cell cycle distribution in NAPA-resistant MDA-MB-231 and HeLa cell lines that expressed no detectable or lower level of NQO1, respectively. Conversely, stable expression of NQO1 in MD-MB-231 and PANC1 cells sensitized both cell lines to NAPA (Figure 2D). Together, these findings verified that the NAPA-induced cell death is dependent on NQO1.

To determine whether STAT3 phosphorylation is required NQO1-mediated cell death, we examined the effect of NAPA sensitivity in OVCA429 cells which expresses high level of NQO1 with no detectable pSTAT3 at Y705 (Figure 2E). Inhibition of NQO1 activity by DIC resulted in an increased viability compared to NAPA treatment alone (Figure 2F). These results suggest that NQO1 is a major determinant for NAPA sensitivity, which does not require STAT3 phosphorylation.

NAPA is a potent ROS-inducing agent as an NQO1 substrate

NQO1 performs a futile two-electron oxidoreduction using NAD(P)H to generate an unstable hydroquinone form of the quinone-like compounds, such as MENA (a paraquinone) [27,29], a nature substrate (vitamin K3) and β-lapachone (LAPA, a naphthoquinone), an NQO1 bio-activated compound [28,30]. The substantial amount of reactive oxygen species (ROS) generated during this process is considered to be the main cause of cell death [30] (Figure 2A). Given the similar chemical structures of NAPA, MENA, and LAPA, we asked whether NAPA underwent the same reduction as MENA and LAPA did in the present of NQO1. We carried out in vitro reduction assays with purified recombinant NQO1 protein or cell lysates using NAPA, MENA and LAPA as substrates. NAPA was reduced by NQO1 with a higher rate compared to the other two substrates (Figures 3A, 3B and S4). Thus, NAPA is a substrate for NQO1.

Figure 3.

NAPA-induced ROS formation is mediated by NQO1. (A) In vitro oxidoreduction reaction catalyzed by recombinant human NQO1 using MENA (200 nM) or NAPA (200 nM) as substrates. (B) In vitro oxidoreduction reaction rate using cell lysates from HGC27 or HeLa cell lines as NQO1 donor and NAPA (200 nM), MENA (200 nM) or LAPA (200 nM) as substrates with or without DIC. (C) Oxygen consumption rate (OCR) in HGC27 cells when treated with NAPA at indicated concentrations. (D) ROS generation in MDA-MB-231, HGC27 and HeLa cell lines treated with NAPA for 30 min. All results in (A-D) represent means ± S.D. from 4 replicates (*P < 0.05, **P < 0.01, and ***P < 0.001. Student’ t-test).

To monitor the cellular redox state after NAPA treatment, we next measured the oxygen consumption rate (OCR) for a sensitive cell line (HGC27) in response to NAPA. NAPA-treated HGC27 cells showed a drastic OCR increase and reached peak levels within 30 minutes (Figure 3C). Consistently, a substantial increase in superoxide formation monitored with the DCFDA (2’,7’-dichlorofluorescein diacetate) staining was also observed in HGC27 cells treated for 30 min (Figure 3D), while a moderate increase was measured in the relatively insensitive HeLa cells. In contrast, the ROS content was almost stable in the NQO1-undetectable MDA-MB-231 cells. These results demonstrated that NAPA is a compound that can be bio-activated by NQO1 to generate ROS.

NQO1 is a potential response biomarker for NAPA therapy

The above results suggest that NQO1 could be a response biomarker for NAPA cancer therapy. To explore the translational potential, we analyzed the correlation between NQO1 mRNA expression and patients’ overall survival (OS) utilizing publicly available data. Kaplan-Meier survival analysis using the Cancer Genome Atlas (TCGA), European Genome-phenome Archive (EGA) and Gene Expression Omnibus (GEO) data sets revealed lower OS with higher NQO1 expression in more than half of the cancer types, such as hepatocellular carcinoma (HCC) and PDAC [24,25] (Figure S5A-C); but the opposite was observed for other cancer types including gastric [31] and ovarian cancers [32] (Figure S5D and S5E). These analyses suggest that NAPA therapy could only be beneficial for selected and sub-populations with high level of NQO1.

Because NAPA-induced cell death depends on NQO1 activity, we next analyzed the NQO1 protein levels in patients’ tumors and tumor nearby tissues from the National Cancer Institute’s Clinical Proteomic Tumor Analysis Consortium (CPTAC) and the China Human Proteome Project (CNHPP) studies. Two independent gastric cancer datasets [33,34] showed that a higher expression of NQO1 (T/N ≥ 3) in tumors than the paired nearby tissues (Figure 4A) was seen in only 5.6% and 6.0% of the total patients. Similarly, only 16.7%, 18.0% and 7.3% of ovarian [35], breast [36] and colorectal cancer patients [37], respectively, exhibited > 3 times higher NQO1 in tumors than the nearby tissues (Figure 4B). In contrast, 40.2% of the HCC patients exhibited significantly higher NQO1 levels in liver tumors compared to the nearby tissues [38] (Figure 4C). These data suggest that the great variation in NQO1 differential expression in different types of cancer makes it necessary to determine the applicable cancer types and to stratify patients for increased success of NAPA therapy.

Figure 4.

Ratios of NQO1 protein level in tumors and nearby tissues (T/N) in datasets from the CPTAC and the CNHPP studies. A. Ratios of NQO1 in gastric cancer patients from the 2 datasets. B. Ratios of NQO1 in patients with ovarian, breast or colorectal cancer. C. Ratios of NQO1 in patients with hepatocellular carcinoma (HCC). NA, HCC patients with undetectable NQO1 in both tumor and nearby normal tissues.

NRF2 activators enhances NAPA sensitivity

To expand the application scope of the NAPA therapy to NQO1-low tumors, we explored the strategy of combined regimens to increase NQO1 expression that could facilitate NAPA cytotoxicity. NQO1 expression is mainly regulated by the transcription factor NRF2 [39-42]. We asked whether manipulation of NRF2 activation can lead to elevated NAPA efficacy. As expected, knocking down NRF2 led to a decreased NQO1 expression and increased cellular survival upon NAPA treatment (Figure 5A), indicating that modulating NRF2 expression can impact NAPA sensitivity.

Figure 5.

Manipulating of NAPA-induced cytotoxicity by modulating NQO1 expression levels. A. NAPA-induced death in HeLa cells transfected with control or NRF2-siRNA. Left penal: western blotting of NRF2 in control and NRF2 knockdown HeLa cells. Right penal: the cell viability of HeLa cells treated with NAPA at 5 μM for 24 h. B. NQO1 expression in HeLa cells treated with NRF2 activator DMF or CA. C. Impact of NRF2 activators DMF (left) or CA (right) on NAPA-induced cell death. Results in all plots represent means ± S.D. from 4 independent measurements (*P < 0.05, **P < 0.01, and ***P < 0.001. Student’ t-test).

To test whether activation of NRF2 can increase the NAPA cytotoxicity, we co-treated NQO1-low HeLa cells with NRF2 activators dimethyl fumarate (DMF) [40] or carnosic acid (CA) [43]. DMF is an FDA-approved drug for treating psoriasis and multiple sclerosis. Both DMF and CA led to increased levels of NQO1 (Figure 5B) and increased NAPA sensitivity in HeLa cells in a time-dependent manner, and the effect was reversed by DIC inhibition (Figure 5C). These results suggest that combinational treatment with NRF2 activators may boost NAPA’s efficacy, which could find potential applications to treat tumors with low NQO1 expression in the clinics.

Discussion

In this study, we applied an unbiased proteomic approach to identify proteins that mediate NAPA cytotoxicity and identified NQO1 as one of the potential NAPA sensitivity markers. Further biochemical characterization demonstrated that NAPA is a substrate of NQO1 which produces ROS for cell death. The cell line-based biochemistry studies provide invaluable mechanistic insights that can be applied in the designing more effective strategies in clinical applications.

Our finding that NAPA is a substrate of NQO1 has several ramifications for the NAPA therapy. First, it raises the possibility that NQO1 should be used as a biomarker for patient selection in future clinical trials to increase the likelihood of success. Prior to our study, NAPA was touted as a potent cancer stemness inhibitor and pSTAT3 was reported as a biomarker for NAPA treatment [2,11,13]; however, its molecular target(s) remains unclear. Moreover, we did not find correlation between activated STAT3 and NAPA sensitivity. Since NQO1 is a NAPA target, NAPA treatment should be considered as a targeted therapy just like kinase inhibitors or other antibody-based therapies. We propose that a companion diagnostic test to identify NQO1-high expressing tumors is a necessary component in future design of NAPA trials.

Second, since normal gastric mucosa express high levels of NQO1 [26], it provides a possible explanation for the gastrointestinal side effects caused by oral administration of NAPA, and suggests that intravenous injection might be a better delivery method.

Third, a proteomic study on diffuse-type gastric cancer showed that for most patients, NQO1 expression level in the tumor is lower than that in the nearby tissues [33], suggesting that NAPA treatment for gastric cancer may not be effective. This may partly explain the failed clinical trial for gastric cancer [17]. In contrast, NQO1 levels are found to be elevated in comparison to nearby tissues in early-stage HBV-positive liver cancer [38], suggesting that HCC may be the right cancer type for NAPA treatment. This prediction is consistent with the high response rate in a Phase II clinical trial with liver cancer [44]. Previous study has demonstrated that NQO1 activation of drugs tends to be response in patients with extremely higher level of NQO1 in tumors than the nearby tissues [45,46]. Therefore, we anticipate that more rational NAPA trials with a companion diagnostic test before therapy could identify the favorable patients as well as the right cancer types.

Forth, understanding the molecular mechanism of NAPA action allowed us to recommend possible combination regimen to boost the expression of NQO1. Our data in cancer cell lines showed that treatment with NRF2 activators, such as DMF and CA, could increase the expression of NQO1 protein to facilitate NAPA’s sensitivity, suggesting the potential to repurpose of these FDA-approved drugs in future NAPA clinic trials.

Finally, our proteomic characterization allowed the identification of proteins with high positive or negative correlation with NAPA IC₅₀ values, suggesting the additional signaling pathways that mediate NAPA sensitivity and resistance. Further validation may reveal other mechanistic aspects of the NAPA actions and guidance for effective therapy in future clinical applications.

While this study was in progress, Chang and co-workers independently reported NAPA as an NQO1-bioactivatable small molecule [47]. They suggested that a connection may exist between NQO1 expression in cancer cells and pSTAT3 in tumor cells and the tumor microenvironment. In contrast to previously reported inhibition of pSTAT3, this work suggests that NQO1 expression in tumor cells correlates with increased secretion of multiple cytokines, which promotes an immunosuppressive microenvironment. Therefore, the exact mechanism and the contribution of pSTAT3-depenent NAPA cytotoxicity remains unclear. It would be interesting to compare NQO1 and pSTAT3 as biomarkers retrospectively using samples obtained in previous clinical trials.

Data availability

MS raw files and searching output data have been deposited into ProteomeXchange Consortium (http://proteomecentral.proteomexchange.org) via the iProX partner repository [48] with the accession number IPX0002073000.

Disclosure of conflict of interest

None.