Glutaminase Inhibitors Induce Thiol-Mediated Oxidative Stress and Radiosensitization in Treatment-Resistant Cervical Cancers

This article is featured in Highlights of This Issue, p. 2407

Despite recent advances in screening and prevention, cervical cancer remains the fourth most common cancer diagnosis worldwide in women and a leading cause of cancer-related mortality (1, 2). The standard-of-care (SOC) treatment for locally advanced cervical cancer (LACC), pelvic irradiation with concurrent administration of cisplatin chemotherapy (CRT), has not changed in more than 30 years, and clinical outcomes have stagnated (3, 4). One third of patients with cervical cancer with LACC will fail SOC CRT and, although patient survival can be increased by salvage surgery or treatment with cisplatin and bevacizumab, there is currently no cure for metastatic or recurrent disease (5–7). Thus, the development of new biomarkers and treatment strategies, particularly for radiation (RT)-resistant LACC, is an unmet need.

The majority of cervical cancers are human papillomavirus (HPV)–positive, and persistent expression of HPV-related oncogenes E6 and E7 is necessary but not sufficient for tumorigenesis (8–10). The most common recurrent oncogenic mutations in cervical cancer occur within the PI3K pathway, which is mutated in 40% of cervical cancers and results in activation of AKT and mTOR (11). Previous studies demonstrated patients with cervical cancers harboring PI3K pathway mutations are resistant to SOC CRT, but alternative approaches have yet to reach the clinic (12–16). In addition to genomics, imaging metrics, particularly high tumor glucose uptake on pretreatment ¹⁸F-fluoro-deoxyglucose PET (FDG-PET) imaging, have been used to predict poor outcomes after SOC CRT in LACC (17–20), suggesting that tumor metabolism may be an effective target. Our group has recently reported combined inhibition of glucose and redox metabolic pathways in preclinical models of RT-resistant cervical cancers (21). Challenges in the clinical translation of these results include the need for a three-drug regimen, limited availability of targeted inhibitors of tumor glycolysis, and systemic toxicities associated with inhibition of glucose metabolism.

Cancers that exhibit high rates of glucose metabolism frequently depend upon exogenous glutamine. Glutamine is used by glycolytic cancer cells for energy production, to maintain redox homeostasis, and to generate other biosynthetic intermediates needed for cell growth (22). Inhibiting glutamine metabolism is an effective treatment strategy particularly for mTOR-activated cancers that depend upon glutamine for tumor cell survival (23–26). This approach has become more attractive clinically with the recent development of an orally bioavailable inhibitor of glutaminase, telaglenastat (CB-839), that is currently being tested in clinical trials alone and in combination with other agents (25–27). The purpose of the current study was to test whether radioresistant cervical cancers are dependent upon glutamine metabolism, and to determine the sensitivity of cervical cancer to single-agent and combination therapy with telaglenastat. As there are currently no reliable predictive biomarkers for telaglenastat response, a secondary goal was to determine whether metabolic reprogramming driven by PI3K pathway mutation predicts sensitivity of cervical cancer to telaglenastat.

Cervical cancer cell lines CaSki, SiHa, and C33A were obtained from the American Type Culture Collection (ATCC) and maintained in IMDM media (Life Technologies) with 10% heat-inactivated FBS. HCvEpC were cultured in HCvEpC media from Cell Applications, Inc. Mycoplasma testing was performed every 3 months to verify no contamination. Last date of mycoplasma testing for cell lines used in this study was January 20, 2020. Experiments were performed on all cell lines between passages 10 and 30. SiHa PTEN^−/− cells were generated by CRISPR nuclease–induced targeted double-strand break at the Genome Engineering Center at Washington University School of Medicine (St. Louis, MO). Approximately 1 × 10⁶ single cells were washed and resuspended in buffer with 1 μg gRNA (5′ AACTTGTCTTCCCGTCGTGT 3′), 1.5 μg Cas9, and 0.5 μg GFP expression construct then electroporated via 4D-Nucleofector (Lonza) using EN-158 program. Single-cell clones were screened for genetic knockout genotype using targeted deep sequencing analysis (28).

L-buthionine sulfoximine (BSO), auranofin (AUR), N-acetylcysteine (NAC), protease, and phosphatase inhibitor cocktails were purchased from Sigma. Telaglenastat was obtained from Calithera Biosciences, Inc. Telaglenastat and AUR were dissolved and diluted in DMSO 0.002% and 0.01%, respectively. BSO was dissolved in normal saline. NAC was dissolved in 10% sodium bicarbonate solution.

Cell viability was assessed using Cell Titre (Promega). For CB-839 and redox inhibitor experiments, cells were treated with telaglenastat (200 nmol/L), AUR (25 nmol/L), and BSO (125 μmol/L) for 96 hours and then plated for clonogenic assay. For NAC rescue experiments, cells were treated with CB-839 (200 nmol/L) for 72 hours followed by 7 hours of 10 mmol/L NAC and after which fresh media were added to all the treatment groups for 24 hours and then plated for clonogenic assay. For RT sensitization, cells were pretreated with telaglenastat (500 nmol/L) for 48 hours, then irradiated using an RS2000 160kV X-ray Irradiator using a 0.3 mm copper filter (Rad Source Technologies), and 48 hours after RT, cells were harvested and plated for clonogenic assay. After 10 days, colonies were counted after staining with crystal violet. Assays were performed in triplicate and repeated 3 times. Clonogenic survival data were fitted using the Linear-Quadratic formula. T test was performed on the average of the three biological replicates for dose-modifying factors (DMF).

RNA was extracted from biological replicates of each cell line using the RNeasy Mini Kit (74104; Qiagen). Samples with 0.5 to 1 ug of RNA and RIN > 7 were used for library preparation and run on NovaSeq S1 flow cell. Read counts were calculated using featureCounts with default parameters, and differential expression analysis was performed on raw counts using edgeR (29, 30). Differentially expressed genes with a P value < 0.05 were visualized by heatmap. Gene expression data, read counts, and clinical data for TCGA-CESC (The Cancer Genome Atlas-Cervical Squamous Cell Carcinoma and Endocervical Adenocarcinoma) primary tumors (N = 304) were downloaded from NCI Genomic Data Commons. PTEN mutation and copy-number status were downloaded from cBioPortal (31).

Stable isotope labeling was accomplished using 10 mmol/L uniformly labeled ¹³C₆ glucose or 2 mmol/L ¹³C₅ glutamine to glucose- or glutamine-free DMEM (Gibco A1443001) supplemented with 10% dialyzed FBS (Cambridge Isotope Laboratories). Media were supplemented with 2 mmol/L glutamine (¹³C₆ glucose) and 10 mmol/L glucose (¹³C₅ glutamine). Cells were labeled for 18 hours followed by glutamine deprivation or telaglenastat treatment for 48 hours and then processed as described (32). Cells were washed with PBS and LC/MS-grade water, quenched with cold methanol, scraped, and transferred to microfuge tubes. Methanol was evaporated using a SpeedVac. Pellets were extracted with 2:2:1 methanol:acetonitrile:water (1 mL per 100 mm plate surface area equivalent). Extracts were reconstituted in 1:2 water:acetonitrile volume normalized to the dry mass of the cell pellet. For media studies, 800 μL of cold methanol:acetonitrile (1:1) was added to 200 μL of media. The solution was vortexed, sonicated, evaporated, and reconstituted in 100 μL of acetonitrile:water (2:1). The mixture was then sonicated and centrifuged, and the supernatant was taken for LC/MS analysis. The absolute concentrations of glucose and glutamine were quantified after culturing cells for 48 hours. Known concentrations of ¹³C₆ glucose or 2 mmol/L ¹³C₅ glutamine were spiked into media samples after extractions. Concentrations were determined by calculating the ratio between the fully unlabeled peak and the fully labeled peak from standards. Consumption rates were normalized to doubling time over the experimental time period. LC/MS analysis was performed by using an Agilent 6540 QTOF interfaced with an Agilent 1290 Infinity II LC system (Agilent Technologies) equipped with a Sequant ZIC-pHILIC column (100 × 2.1 mm, 5 μm; EMD Millipore). Mobile-phase solvents were composed of A = 20 mmol/L ammonium acetate and 5 μmol/L ammonium phosphate in water:acetonitrile (95:5) and B = 100% acetonitrile. The column was maintained at 40°C. The following linear gradient was applied at a flow rate of 225 μL/min: 0–1 minute: 89.5% B, 1–30 minutes: 89.5%–20% B, 30–33 minutes: 20% B. The column was re-equilibrated with 20 column volumes of 89.5% B. Injection volumes were 5 μL for all experiments. Data were collected with the following settings: gas, 200°C at 10 L/min; nebulizer, 44 psi at 2,000 V; sheath gas, 300°C at 12 L/min; capillary, 3,000 V; fragmentor, 175 V; skimmer, 65 V; and scan rate, 1 scan/second; mass range, 60–1,500 Da. The MS was operated in negative ionization mode for all samples analyzed. MS data were processed with XCMS, X¹³CMS, and MassHunter Profinder (Agilent Technologies; refs. 33, 34). The figures and plots were generated using Graphpad Prism.

Cells were grown in 100 mm dishes and treated with or without drugs as described, harvested by scraping, and frozen as a dry pellet. Cell pellets were lysed in 50 mmol/L potassium phosphate buffer (pH 7.8) containing 1.34 mmol/L diethylenetriaminepenta-acetic acid, centrifuged at 5,000 rpm for 5 minutes, and then the supernatant was assayed using a TR kit (Sigma-Aldrich; CS0170). For glutathione (GSH) assay, cells were harvested by scraping using 150 μL of 5% 5-sulfosalicylic acid (Sigma-Aldrich). Total GSH and percent glutathione disulfide (GSSG) content was determined spectrophotometrically by NADPH recycling assay as described (35).

Steady-state levels of reactive oxygen species (ROS) were estimated using oxidation of the fluorescent dye, 2′,7′-dichlorodihydrofluorescein diacetate (H2DCFDA), normalized to the oxidation insensitive analog, 2′,7′–dichlorofluorescin diacetate (DCFDA), obtained from Molecular Probes (Eugene). Cells were incubated with H2DCFDA (10 μmol/L) or DCFDA at 37°C for 30 minutes in culture media without washing. Samples were analyzed using a flow cytometer (λ_ex = 495 nm and λ_emission = 529 nm band-pass filter). The mean fluorescence intensity (MFI) was analyzed in each sample and corrected for autofluorescence from unlabeled cells (36). MFI/cell results from H2DCFDA labeling were normalized to replicate samples labeled with the oxidation insensitive dye (DCFDA) under each treatment condition.

Western blotting was performed with primary antibodies against phosphorylated and total forms of p70s6k ^Thr389 and Akt^Ser473 (1:1,000; Cell Signaling Technology), for total forms of PTEN (1:1,000; Cell Signaling Technology), GAPDH and Akt (1:1,000, Cell Signaling Technology), and for total forms of p70s6k and Actin (1:1,000, Santa Cruz Biotechnology) and PC (1:500, Thermo Fisher). Blots were probed with horseradish peroxidase–conjugated anti-rabbit (Cell Signaling Technology) or anti-mouse polyclonal IgG secondary antibodies (Santa Cruz Biotechnology) for 1 hour at room temperature. For detection, Amersham ECL select (GE-healthcare) was used according to the manufacturer's protocol. Images were acquired using Chemidoc Imaging systems (BioRad).

All in vivo studies were conducted according to the protocols approved by Washington University Division of Comparative Medicine and Institutional Animal Care and Use Committee. For tumor generation, 3.5 × 10⁶ SiHa and or CaSki cells were injected subcutaneously into the left flank of 6- to 8-week-old female nude mice (Crl:NU(NCr)-Foxn1 nu, Charles River Laboratories) in a half matrigel, half serum-free IMDM mixture. Initial tumor sizes were recorded using calipers, and mice were grouped (N = 5 per treatment group) with 5 mm tumor diameters. For tumor growth delay studies in CaSki tumors, treatment groups included: vehicle, telaglenastat at 50, 100, and 200 mg/kg. For telaglenastat treatments, mice were treated 3 times per week via oral gavage over a period of 37 days. For tumor growth delay studies in SiHa tumors, single fraction RT doses of 4 Gy were delivered 1 week after initiating treatment with telaglenastat. Targeted RT delivery was performed using the Xstrahl Small Animal Radiation Research Platform (SARRP) 200 (Xstrahl Life Sciences). Mice were placed on the irradiation platform one at a time and fitted with a nose cone for isoflurane anesthesia. CT images imported into Muriplan were used to select an isocenter. The tumor was then irradiated using anterior–posterior-opposed beams using the 10 mm x 10 mm collimator at a dose rate of 3.9 Gy/min. Telaglenastat treatment was continued for 4 weeks after RT treatment. Tumor measurements were quantified by caliper measurement on a weekly basis for each animal. Statistical comparisons were performed using Restricted Maximum Likelihood (REML) mixed effect model with volume as response variable, treatment groups as covariates and replicates, and day as random effect.

We determined the sensitivity to glutamine starvation of a panel of cervical cancer cell lines, including those with known PI3K pathway alterations (Fig. 1A; Supplementary Table S1). Normalized surviving fractions following glutamine deprivation demonstrated decreased survival in CaSki and C33A cells, whereas SiHa cells were relatively resistant. To test whether cervical cancer cell dependence on glutamine could be directly influenced by single-gene mutations in the PI3K pathway, we generated PTEN-deleted cells (SiHa PTEN^−/−) from parent SiHa cells using CRISPR/Cas9 (Supplementary Fig. S1B and S1C). In contrast to parental SiHa cells, SiHa PTEN^−/− cells demonstrated complete attenuation of PTEN and express high levels of phosphorylation of PI3K pathway targets including pAKT (Supplementary Fig. S1D). Interestingly, SiHa PTEN^−/− cells were more sensitive to glutamine starvation than the parent SiHa cells (Fig. 1A), suggesting that PTEN deletion and resulting PI3K pathway activation are associated with increased dependency on glutamine metabolism in vitro. To demonstrate the prevalence of PTEN alterations in human cervical cancer tumors, we analyzed whole transcriptome and exome sequencing data from TCGA cervical cancer (TCGA-CESC) cohort (N = 304). We observed 35% of primary cervical cancers harbor genomic alterations that decrease PTEN expression (Fig. 1B and C). To examine the direct effects of PTEN loss on gene expression of metabolic pathways, we performed whole transcriptome sequencing of our engineered SiHa cells (Supplementary Tables S2 and S3). Compared with parental cells, SiHa PTEN^−/− cells displayed decreased expression of multiple genes related to glutamine metabolism in addition to several key glycolytic enzymes (Fig. 1D and E).

Relative rates of glutamine consumption were compared in cervical cancer cells with and without PI3K pathway mutations. As demonstrated in Fig. 2A, CaSki cells consume high levels of glutamine, whereas SiHa cells consume less. Compared with parental SiHa cells, SiHa PTEN^−/− cells consume higher levels of glutamine and glucose (Fig. 2A and B). We compared glucose consumption rates in exhausted media after culturing cells in the presence and absence of glutamine deprivation (Fig. 2C). CaSki cells consumed less glucose from the media when glutamine was limiting, whereas SiHa cells were able to maintain baseline levels of glucose consumption. Compared with parental SiHa cells, SiHa PTEN^−/− cells consumed slightly less glucose at baseline but retained the ability to metabolize glucose even in the setting of glutamine deprivation. We performed cell proliferation assays at different time points after the initiation of glutamine starvation (Fig. 2D and G). All three PTEN-mutated or null cell lines (CaSki, C33A, and SiHa PTEN^−/−) displayed significant decreases in cell proliferation after day 4, compared with day 8 in SiHa cells. We compared the relative sensitivity of cervical cancer cell lines to glucose versus glutamine deprivation. Surprisingly, cervical cancer cells were more sensitive to glutamine starvation (Gluc+Gln−) versus glucose (Gluc−Gln+) starvation (Fig. 2H–K). When glutamine starvation was combined with glucose starvation (Gluc- Gln-), CaSki cells demonstrated increased dependency on glutamine relative to glucose (Gluc+, Gln−; Fig. 2H), whereas SiHa cells continued to display intermediate sensitivity to both glucose and glutamine starvation (Fig. 2J). Compared with parental SiHa cells, SiHa PTEN^−/− cells displayed increased dependency upon glutamine relative to glucose (Fig. 2K), suggesting that inhibition of glutamine metabolism may be a more effective metabolic treatment strategy than inhibition of glycolysis in PI3K pathway–activated cervical cancers.

We determined whether glutamine starvation resulted in decreased availability of GSH and increased levels of baseline intracellular oxidative stress in cervical cancer cells with and without PI3K pathway activation. As shown in Fig. 3A, glutamine deprivation resulted in decreases in total reduced GSH (Fig. 3A) with concomitant increases in percent glutathione disulfide (%GSSG), a reliable surrogate measure of intracellular oxidative stress (Fig. 3B). Glutamine deprivation was associated with significant increases in intracellular oxidation as evidenced by increases in H2DCFDA oxidation (Fig. 3C) in CaSki cells under conditions of glutamine deprivation, but not in SiHa cells. Consistent with these observations, intracellular NADPH/NADP ratios increased only in CaSki cells as a compensatory mechanism to deal with ROS accumulation under conditions of glutamine deprivation (Fig. 3D). These effects were most pronounced in CaSki versus SiHa cells, an effect related to increased reducing capacity via alternative pathways such as thioredoxin in SiHa cells (Fig. 3E). As previously reported by our group (21), upregulation of thioredoxin reductase (TXNRD) is an alternative mechanism used in some cervical cancer cells to maintain redox balance, an observation that becomes clinically relevant when selecting drug strategies designed to increase oxidative stress. Importantly, overexpression of TXNRD and other genes within this pathway is not a cancer cell line–specific phenomenon, but rather a relevant potential treatment resistance mechanism observed in human cervical as well as other cancers. As shown in Supplementary Fig. S2A–S2C, we have identified overexpression of TXNRD and related transcripts within TCGA cervical cancer data set (37–39). Oxidative stress parameters were similar in SiHa cells with and without PTEN deletion, suggesting that PI3K pathway signaling is less relevant to the maintenance of redox balance when TXNRD is overexpressed. These results demonstrate that glutamine deprivation is associated with increased baseline oxidative stress in cervical cancer cells, but the extent of this particular effect is maximized in highly glycolytic cervical cancer cells that rely on the GSH system to maintain redox balance (i.e., CaSki cells). Increases in intracellular oxidative stress induced by glutamine deprivation may be limited by increased expression of redox metabolic pathways other than GSH (i.e., thioredoxin reductase in SiHa cells and some human cervical tumors).

We next wanted to determine whether inhibition of glutaminase with telaglenastat would be selectively cytotoxic for glutamine-dependent cervical cancers and determine to what extent this sensitivity was dependent upon mutations in the PI3K pathway. To begin, we modeled dose–response curves by performing cell viability assays for telaglenastat monotherapy in a panel of cervical cancer cell lines including CaSki, SiHa, and SiHa PTEN^−/− cell lines (Fig. 4A). In general, sensitivity to telaglenastat monotherapy was more pronounced in cervical cancer cell lines with PI3K pathway mutations compared with those without, including HPV-negative lines HT-3 and C-41, with the exception of C33A cells which demonstrated intermediate sensitivity (Fig. 4A; Supplementary Fig. S2D). Interestingly, sensitivity to telaglenastat monotherapy was most pronounced in CaSki and SiHa PTEN^−/− cells relative to parent SiHa cells, even though there was relatively little difference in oxidative stress parameters under conditions of glutamine deprivation in SiHa versus SiHa PTEN^−/− cells, suggesting that metabolic outputs from glutamine other than GSH may contribute to telaglenastat sensitivity (Figs. 3 and 4A). Quantification of total GSH, %GSSG, and TR activity before and after telaglenastat treatment confirmed increases in oxidative stress with associated decreases in the reducing capacity through GSH after telaglenastat treatment in all cell lines tested (Fig. 4B–D). Importantly, the cytotoxic effects of telaglenastat could be rescued by the addition of the thiol antioxidant, NAC (Fig. 4E and F), confirming that the mechanism of action of telaglenastat includes treatment-related increases in thiol-dependent oxidative stress.

The efficacy of telaglenastat monotherapy in glutamine-dependent cervical cancer tumors in vivo was tested using CaSki tumor implants grown in the flanks of immune-compromised mice (Fig. 4G and H; Supplementary Fig. S8A–S8C). Treatment with telaglenastat monotherapy at 200 mg/kg in this model resulted in tumor regression with no observable toxicities and no evidence of tumor regrowth after 37 days (Supplementary Fig. 8B). These experiments were repeated with telaglenastat dose de-escalation, with telaglenastat administered at 50 and 100 mg/kg every other day. Importantly, telaglenastat using this dose schedule, which includes significantly lower doses and less frequent administration than has been used in previous preclinical studies with other tumor models, resulted in tumor regression, suggesting that telaglenastat monotherapy (50 mg/kg) is highly potent and effective in appropriately selected glutamine-dependent cervical tumor models (Fig. 4G). SiHa xenograft tumor regression was observed with telaglenastat monotherapy at a concentration of 100 mg/kg after day 15 (Fig. 4H).

To determine all the metabolic contributions of glutamine in cervical cancer cells with and without PI3K pathway activation, we performed isotope-tracing studies with uniformly labeled ¹³C glutamine in CaSki, SiHa, and SiHa PTEN^−/− cells. As shown in Fig. 5A and B and Supplementary Fig. S3, cervical cancer cells generated glutamate and the tricarboxylic acid (TCA) cycle intermediate α-ketoglutarate from exogenous glutamine as evidenced by incorporation of labeled carbon into the m+5 fraction of each metabolite. The generation of labeled glutamate and α-ketoglutarate from exogenous ¹³C₅ glutamine was increased in CaSki cells relative to SiHa cells, and in SiHa PTEN^−/− cells relative to parental SiHa cells, suggesting that PI3K pathway activation facilitates this process. As expected, cervical cancer cells also generated GSH from exogenous glutamine, as evidenced by the incorporation of labeled carbon from ¹³C₅ glutamine into the m + 5 fraction of GSH (Fig. 5C). Similar to the effects on TCA cycle intermediates, the generation of labeled GSH from exogenous ¹³C₅ glutamine was increased in CaSki cells relative to SiHa cells, and in SiHa PTEN^−/− cells relative to parental SiHa cells, suggesting that PI3K pathway activation facilitates this process.

In our previous experiments, we demonstrated that glutamine deprivation results in increased oxidative stress in cervical cancer cells via decreases in the availability of GSH, an effect which is most prominent in cervical cancer cell lines such as CaSki that do not upregulate alternative pathways for maintaining redox balance (Fig. 3). To examine the effects of glutamine deprivation on the metabolic fates of glucose, we performed LC/MS-based isotope-tracer studies with uniformly labeled glucose (¹³C₆ glucose) performed in the presence and absence of glutamine deprivation (Fig. 5E and H; Supplementary Figs. S4–S7). Glutamine deprivation resulted in increased glucose carbon utilization through TCA cycle via pyruvate anaplerosis. This was evidenced by increased labeling of malate (Fig. 5E) and citrate (Fig. 5F; m+ 3) fractions in all three cell lines from ¹³C-glucose under conditions of glutamine starvation. Interestingly, SiHa cells displayed an increased capacity to divert glucose carbon into the TCA cycle via pyruvate anaplerosis as evidenced by higher m+3 fractions (Fig. 5E and F), an effect that was associated with increased expression of pyruvate carboxylase and successful rescue of clonogenic survival with the addition of pyruvate or α-ketoglutarate in SiHa cells (Supplementary Figs. S6 and S7). Glucose flux to the pentose phosphate pathway was also affected by glutamine starvation, as demonstrated by decreased ribose labeling in ADP (m+5) in CaSki (Fig. 5G) and SiHa cells under conditions of glutamine starvation. Consistent with this observation, we also saw decreases in the concentrations of purine nucleotides with glutamine deprivation (Fig. 5H). These results suggest that inhibition of glutamine metabolism could be combined with other strategies in addition to those that target redox balance, such as drugs that target mitochondrial metabolism or DNA synthesis and repair, as means to achieve therapeutic synergy.

To determine whether telaglenastat monotherapy was sufficient to eliminate reducing capacity via GSH, we performed clonogenic cell survival assays in CaSki, SiHa, and SiHa PTEN^−/− cells with telaglenastat or telaglenastat combined with BSO, an inhibitor of the rate-limiting step of GSH synthesis. As we previously identified upregulation of redox metabolic pathways other than GSH (TR) in SiHa cells, we also included AUR, an inhibitor of thioredoxin reductase (TR), in these experiments. As shown in Fig. 6A–C, treatment with telaglenastat alone reduced cell survival in CaSki and SiHa PTEN^−/− cells, but not in SiHa parental cells, an effect that may be due to increased capacity for maintaining redox balance via the thioredoxin system in this cell line (Figs. 3E and 4D). The addition of BSO increased telaglenastat cytotoxicity in all three cell lines, with relatively little additional benefit derived from the addition of AUR, even in SiHa parental cells. Interestingly, telaglenastat and AUR were more effective than telaglenastat alone in CaSki and SiHa PTEN^−/−, but not in SiHa parental cells, an effect that may be related to treatment-induced TR upregulation.

The primary treatment strategy for locally advanced cervical cancer is pelvic RT, a treatment which is also known to increase tumor cell oxidative stress. In order to determine whether telaglenastat could serve as a RT modifier in vitro, we performed clonogenic cell survival assays with increasing single fraction doses of RT in the presence and absence of telaglenastat in CaSki, SiHa and SiHa PTEN^−/− cells (Fig. 6D–F). These results demonstrate that telaglenastat enhances the effect of RT with a DMF of 1.9, 1.2, and 1.7 in CaSki, SiHa, and SiHa PTEN^−/− cells, respectively. The RT response of the corresponding normal cervical epithelial cell HCvEpC line was not modified by telaglenastat at any of the doses tested (Fig. 6G). Finally, in an in vivo experiment with SiHa xenograft tumors in immune compromised mice, we observed increased tumor control in response to tumor directed RT when administered with concurrent telaglenastat, suggesting that telaglenastat can function as an RT modifier in vivo (Fig. 6H; Supplementary Fig. S8D and S8E).

The purpose of the current study was to test whether RT-resistant cervical cancers are dependent upon glutamine metabolism, and to determine whether metabolic reprogramming driven by PI3K pathway mutations predicts sensitivity of cervical cancer to telaglenastat. Using a combination of in vitro and in vivo model systems, including PTEN-deleted cells (SiHa PTEN^−/) that we engineered from parent SiHa cells using CRISPR/Cas9 technology, we demonstrate that PTEN is frequently altered in human cervix tumors, preclinical models of PI3K activated tumors are uniquely sensitive to glutamine deprivation (Figs. 1 and 2), and that glutamine deprivation and treatment with telaglenastat result in increases in intracellular oxidative stress in models of treatment refractory cervical cancer (Figs. 3 and 4). PTEN is a regulator of glutamine metabolism, and our results confirm that PTEN loss leads to increased dependency on glutamine with resultant increased sensitivity to telaglenastat in cervix cancer (40). Importantly, we show that inhibition of glutaminase is selectively cytotoxic for PI3K-activated cervical cancers, and that this effect is due to resultant decreases in intracellular reducing capacity derived from GSH (Figs. 4 and 6I). As such, telaglenastat is particularly potent and effective even as monotherapy in appropriately selected glutamine-dependent PI3K activated cervical cancer models (Fig. 4G). It is important to note that although GSH levels are decreased, they are not completely eliminated by telaglenastat treatment in our models, allowing for synergistic activity with other agents, such as BSO, that target the GSH system (Fig. 6A–C).

Our results support published data that link telaglenastat efficacy to increased oxidative stress in other cancers. Biancur and colleagues reported the accumulation of ROS in response to telaglenastat treatment and synergistic response with BSO in pancreatic ductal adenocarcinoma (PDAC; ref. 41). Similarly, telaglenastat combined with β-lapachone results in oxidative imbalance and DNA damage–induced cell death in KRAS mutant/NQO1 overexpressing PDAC models (42). In contrast to pancreatic and lung cancers, KRAS, LKB1, and NRF2/KEAP1 are not frequently mutated in human cervix cancer (12, 43–45). Our results demonstrate for the first time that mutations in the PI3K pathway play a critical role in maintaining redox homeostasis and glutamine dependency in cervical cancer, and suggest that activation of the PI3K pathway by PTEN alterations could be used as a biomarker to predict sensitivity to telaglenastat. These results are consistent with existing evidence for increased telaglenastat sensitivity observed in preclinical models of PIK3CA-mutated colon cancer (46, 47).

To further increase the translational relevance of our findings, we have combined telaglenastat with RT treatment, the SOC for locally advanced cervical cancer, and shown that telaglenastat can be used as a RT modifier both in vitro and in vivo (Fig. 6D–F). Previous reports suggested telaglenastat enhanced RT sensitivity in lung cancer cells and glutamine-dependent, IDH1-mutant glioma cells (48, 49). An ongoing phase Ib clinical trial in patients with IDH-mutated diffuse or anaplastic astrocytoma is determining the best dose of telaglenastat in combination with radiotherapy and temozolomide (50). Our results support investigating telaglenastat in combination with definitive RT in cervical cancer in the context of clinical trials. Importantly, although we show that the RT-modifying effect of telaglenastat is likely due to increased thiol-mediated oxidative stress in cervical cancer, telaglenastat has other metabolic effects that may contribute to RT sensitivity. Using stable isotope labeling, we show that cervical cancer cells use glutamine not only for maintaining GSH levels, but also for replenishing TCA cycle intermediates needed for energy balance (Fig. 5A–D), and using ¹³C-labeled glucose, we demonstrate that some cervical cancer cell lines have increased capacity for pyruvate anaplerosis (Fig. 5E and F; Fig. 6I), but all cell lines demonstrate decreased nucleotide production in the setting of glutamine deprivation. These results suggest that, in addition to increasing oxidative stress, telaglenastat also has the capacity to limit DNA repair via decreased nucleotide production, which supports the rationale combination of telaglenastat with a variety of DNA-damaging agents and drugs that induce redox stress. Work is ongoing in our lab to determine to what extent the RT-modifying properties of telaglenastat are dependent upon decreased capacity for DNA repair, and to optimize drug combination schedules with SOC chemotherapies including cisplatin.

In summary, we report for the first-time preclinical data to support the use of telaglenastat for RT-resistant cervical cancers, and demonstrate that PI3K pathway mutations including PTEN alterations may be used as a predictive biomarker for telaglenastat sensitivity. Here, we show that patients with cervix cancer with PI3K pathway mutant, glutamine-dependent tumors could potentially benefit from telaglenastat and the rationale combination of telaglenastat with RT and BSO with or without AUR. Given that telaglenastat (registrational phase II clinical trial drug), BSO, and AUR are FDA-approved drugs, if additional experiments in other xenograft models of cervical cancer confirm our findings, then they can be translated into clinical trials for patients with PI3K-activated cervical cancer.

B.E. Rogers reports grants from Washington University during the conduct of the study. G.J. Patti reports grants from R35ES028365 and grants from R24OD024624 during the conduct of the study. J.K. Schwarz reports grants from NIH (R01 CA181745), Goldman Sachs Philanthropy Cancer Research Fund, Siteman Investment Program, and 2019 AACR-Bristol-Myers Squibb Mid-Career Female Investigator Grant outside the submitted work, and nonfinancial support from Calithera Biosciences (Telaglenastat CB-839 [drug only] was provided as part of a Material Transfer Agreement [MTA] from Calithera Biosciences) during the conduct of the study. No potential conflicts of interest were disclosed by the other authors.

R. Rashmi: Conceptualization, data curation, formal analysis, investigation, visualization, methodology, writing-original draft, writing-review and editing. K. Jayachandran: Data curation, investigation, visualization. J. Zhang: Resources, software, formal analysis, visualization, methodology. V. Menon: Data curation, validation. N. Muhammad: Methodology. M. Zahner: Data curation, investigation, visualization, methodology. F. Ruiz: Data curation, visualization. S. Zhang: Data curation, investigation, visualization. K. Cho: Data curation, visualization, methodology. Y. Wang: Data curation, visualization. X. Huang: Data curation, software, visualization. Y. Huang: Data curation, formal analysis, validation, visualization, methodology. M.L. McCormick: Investigation. B.E. Rogers: Resources. D.R. Spitz: Resources, supervision, funding acquisition, methodology, writing-review and editing. G.J. Patti: Resources, software, formal analysis, supervision, funding acquisition, methodology, writing-review and editing. J.K. Schwarz: Conceptualization, resources, formal analysis, supervision, funding acquisition, methodology, writing-original draft, project administration, writing-review and editing.

This work was supported by NIH R01CA181745 to J.K. Schwarz; R35ES028365 to G.J. Patti; and Radiation and Free Radical Research Core P30CA086862 and P01CA217797 to D.R. Spitz and M.L. McCormick.

We would like to thank Cedric Mpoy for technical assistance.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.