Radioresistant Cervical Cancers Are Sensitive to Inhibition of Glycolysis and Redox Metabolism

The current standard of care for locally advanced cervical cancer is concurrent cisplatin chemotherapy with pelvic irradiation, which includes the administration of both external beam radiotherapy and intracavitary brachytherapy. Despite significant advances in radiation treatment delivery, more than 30% of patients fail this treatment. The prognosis of these patients is poor as there is currently no curative treatment for metastatic cervical cancer. Complete surgical resection including total pelvic exenteration has been used to salvage limited volume local pelvic recurrences but with significant treatment-related morbidity. We have previously reported that increased uptake of ¹⁸F-fluoro-deoxy-glucose (FDG) on pretreatment PET is prognostic for poor outcomes, and that cervical tumors with residual FDG uptake after standard-of-care chemoradiation have inferior long-term survival (1, 2). Cervical tumors with persistent FDG uptake after chemoradiation have altered expression of genes from the PI3K/AKT pathway, and increased expression of phosphorylated AKT is associated with poor outcomes after standard-of-care treatment (3). AKT inhibitors can be used to reduce cervical tumor cell glucose uptake and metabolism, which results in tumor cell death (4). Furthermore, redox signaling has been suggested to regulate both AKT activation and cervical cancer cell survival (5).

Enhanced glutathione (GSH) and thioredoxin metabolism (Trx) are two mechanisms by which cancer cells mitigate the redox stress resulting from increased steady-state levels of reactive oxygen species (ROS) produced by disruptions in oxidative metabolism (6–10). GSH and Trx neutralize hydroperoxides, and upregulation of their respective metabolic pathways occurs in many cancers (8, 9, 11). Glutathione peroxidase (GPx) enzymes inactivate H₂O₂ and other hydroperoxides using reducing equivalents derived from the conversion of GSH to glutathione disulfide (GSSG). GSSG is then recycled back to GSH by glutathione reductase (GR), consuming reducing equivalents from NADPH in the process (7–9, 12). The Trx system neutralizes H₂O₂ and hydroperoxides via the action of peroxiredoxins (Prx), which consumes reduced Trx(SH)₂ and releases oxidized TrxS₂ (Trx) (13) that is then reduced back to Trx(SH)₂ by thioredoxin reductase (TR); this process also consumes reducing equivalents from NADPH (7–9, 12). Given that both GSH- and Trx-dependent peroxide metabolisms require NADPH as the ultimate source of electrons, these pathways are inextricably linked to glucose metabolism, because the latter is required for the production of the majority of the NADPH pool (6, 9, 14).

Glucose deprivation and treatment with 2-deoxyglucose (2-DG) are known to selectively induce more oxidative stress in cancer cells compared with normal cells (8, 10, 15). 2-DG is transported into cells and phosphorylated to 2-DG-6-phosphate by hexokinase, after which it accumulates and inhibits downstream glycolytic enzymes (16). Previous studies in head and neck cancer cells have shown that 2-DG treatment promotes shunting of glucose into the pentose phosphate pathway in an effort to combat the increase in intracellular oxidative stress (17). Aside from inhibition of NAPDH production by targeting glucose metabolism, strategies using direct redox metabolism inhibitors, such as l-buthionine-sulfoximine (BSO), a γ-glutamyl cysteine ligase inhibitor, and auranofin (AUR), a TR inhibitor, alone or in combination have been shown to induce redox imbalances in several cancer cell types (7, 8, 11, 12, 18).

Our previous work demonstrated that highly glycolytic cervical cancers are resistant to standard-of-care therapy (cisplatin plus pelvic irradiation). The objective of the current study is to test whether inhibition of glycolysis, in combination with simultaneous inhibition of intracellular GSH- and Trx-mediated redox metabolism, is an effective alternative drug strategy for highly glycolytic and radioresistant cervical cancers. Because our ultimate goal is translation of these results into the clinic, we selected three drugs for preclinical studies that are already approved for human use: (1) 2-DG to inhibit glycolysis, (2) BSO to target GSH synthesis, and (3) AUR to target TR.

Cervical cancer cell lines were obtained from the ATCC collection (2009–2010). Experiments were performed on cell lines between passages 10 and 30. Cells were maintained in IMDM media (Life Technologies) with 10% heat-inactivated FBS and incubated at 37°C in 5% CO₂. Mycoplasma testing was performed periodically by the iPSC Center at Washington University School of Medicine using MycoAlert Plus Kit (Lonza) to verify no infection. Last date of mycoplasma testing for cell lines used in this study was April 20, 2017. BSO, AUR, 2-DG, protease, and phosphatase inhibitor cocktails were purchased from Sigma. All drugs for cell culture were dissolved in normal saline except AUR, which was dissolved in 0.05% dimethyl sulfoxide.

Phosphorylation of the AMPK pathway and activation of autophagy with or without 2-DG+BSO+AUR were determined by Western blotting with primary antibodies against phosphorylated and total forms of ^Ser15p53, p53, ^{Thr269/Ser272}p62, p62, ^Thr389p70, p70, ^Ser317Ulk, Ulk, ^Ser62Myc, Myc, cleaved caspase 3 and PARP, ^Thr286CamKII, CamKII, ^Thr172AMPK, AMPK, ^Ser428LKB1, LKB1, ^{Thr183/Tyr185} JNK, JNK (1:1,000; Cell Signaling Technology), for total forms of thioredoxin reductase1 and 2 (TR1 and TR2), GAPDH, MAPLC3, and Bim (1:1,000; Cell Signaling Technology), and for Actin (1:1,000;Santa Cruz Biotechnology) and tubulin (1:10,000; Sigma). 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).

For cell viability assays, cells were treated with the glycolytic and redox inhibitors 2-DG (20 mmol/L), BSO (1 mmol/L) for 48 hours, and AUR (100 nmol/L) for 24 hours, respectively. Cell viability was tested using Alamar Blue from Life Technologies, according to the manufacturer's instructions.

For the clonogenic cell survival assays, cells were treated with or without drugs alone or in combination using the following concentrations: 2-DG (20 mmol/L), cisplatin (500 nmol/L), BSO (1 mmol/L) for 48 hours, and AUR (100 nmol/L) for 24 hours. For radiation experiments, 10 mmol/L-2-DG, 500 μmol/L BSO, and 50 nmol/L AUR were added simultaneously, incubated for 5 hours in the presence of drugs, and then irradiated using an RS2000 160kV X-ray Irradiator using a 0.3 mm copper filter (Rad Source Technologies). Cells were harvested after 24 hours by trypsinization including all floating cells, and 500 cells each were plated in a 6-well dish. The colonies were counted 10 days after staining with crystal violet.

FDG uptake assay was performed as described previously (3). Briefly, cells were seeded 48 hours prior to ¹⁸F–FDG labeling and maintained in standard tissue culture conditions. Cells were incubated in glucose-free media for 30 minutes prior to the addition of ¹⁸F–FDG. ¹⁸F–FDG (20 μCi) was added to the glucose-free medium for 1 hour. Cells were washed, harvested, and counted on a gamma counter. Data was normalized to total cell count.

Cells were grown in 100 mm dishes and treated with or without drugs as described above and then 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 assay kit (Sigma-Aldrich). Cells were processed as for the GSH assay by scraping into 150 μL of 5% 5-sulfosalicylic acid (Sigma-Aldrich). Total GSH and GSSG content was determined spectrophotometrically by NADPH recycling assay as described previously (7).

Levels of pro-oxidants were determined using the oxidation-sensitive (CDCFH₂, 10 μg/mL) and oxidation-insensitive (CDCF, 10 μg/mL) fluorescent dyes obtained from Molecular Probes. The cells were washed once with PBS and labeled with florescent dyes for 10 minutes at 37°C in PBS. At the end of the incubation, plates were read at Ex/Em: 492–495/517–527 nm. The mean fluorescent intensity (MFI) was plotted after correction for autofluorescence from unlabeled cells. Steady-state levels of superoxide were estimated using oxidation of the fluorescent dye, DHE (dihydroethidine), obtained from Molecular Probes. Briefly cells were incubated with DHE (10 μmol/L) at 37°C for 30 minutes in culture media and suspended in PBS. Samples were analyzed using a flow cytometer (λ_ex = 488 nm and λ_emission = 585 nm band-pass filter). The MFI was analyzed in each sample and corrected for autofluorescence from unlabeled cells (10).

For experiments using the thiol antioxidant, N-acetylcysteine (NAC), cells were treated with 10 mmol/L NAC for 5 hours following 2-DG and BSO for 24 hours followed by 2-DG, BSO, and AUR for another 24 hours. For overexpression of catalase, cells were infected with 100 MOI of adenovirus-expressing catalase (AdCAT) or empty virus (AdEmpty). After 48 hours, virus was removed and the drugs were added followed by colony-forming assay (CFA) as described above. Overexpression of catalase was confirmed by catalase activity assay as previously described (19). Activity was expressed in mκunits/milligram protein.

Stable isotope labeling of cells was performed by substituting the culture media with labeling media consisting of glucose- and glutamine-free DMEM (Gibco; A1443001) supplemented with 10% dialyzed FBS, 2 mmol/L glutamine, and 10 mmol/L uniformly labeled ¹³C glucose (Cambridge Isotopes). 2-DG, BSO, and AUR were added to the labeling media at the time of substitution, and labeling and drug treatment were performed for 18 hours. Cells were then processed for LC/MS-based metabolomic profiling according to the protocol described in (20). Briefly, cells were washed with PBS and milliQ water, and metabolism was quenched with cold methanol. Cells were scraped and transferred to microfuge tubes. The methanol was evaporated using a speed-vac, leaving a dried cell pellet that was extracted with 2:2:1 methanol:acetonitrile:water (1 mL per 100-mm plate surface area equivalent). Metabolite extracts were concentrated using a speed-vac to remove the extraction solvent and reconstituted in 1:1 acetonitrile:water in a 10:1 original:final volume ratio. Five microliters of each concentrated metabolite extract were injected onto a 3 μm, 150 mm x 1 mm Luna NH2 column (Phenomenex) operated in hydrophilic interaction liquid chromatography (HILIC) mode on an Agilent 1260 high-performance liquid chromatography system. The mobile phases were A (20 mmol/L NH₄OAc/NH₄OH at pH 9.5 in 95% H₂O/5% acetonitrile) and B (95% acetonitrile/5% H₂O). A gradient from 100% B to 100% A was run over 40 minutes. Column eluate was injected into an electrospray source operating in negative ionization mode and mass analyzed in an Agilent 6530 or 6540 quadrupole time-of-flight mass spectrometer. Data were processed using MSConvert (21), XCMS (22), X¹³CMS (20), and Graphpad Prism 7.

All the in vivo studies were conducted according to the protocols approved by Washington University Division of Comparative Medicine and Institutional Animal Care and Use Committee. A total of 3.5 million SiHa and or CaSki cells were injected subcutaneously into the left flank of 6- to 8-week-old, female nude mice in a half matrigel, half serum-free IMDM mixture. After 2 weeks, initial tumor sizes were recorded using calipers and mice were grouped (N = 5 per treatment group) with 5 mm starting tumor volumes. For tumor growth delay studies in Caski tumors with drugs alone, treatment groups included sham injection, DG alone, BSO + AUR (BA), and DG + BSO + AUR (DBA). Mice were injected 3 times per week with 400 mg/kg 2-DG, 200 mg/kg BSO, and 1.5 mg/kg AUR over a period of 35 days. For tumor growth delay studies in SiHa tumors with drug treatment and concurrent tumor-directed radiation, single fraction radiation doses of 2 or 4 Gy were delivered after 1 week of treatment with DBA. Targeted radiation 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. DBA treatment was continued for 1 week after radiation treatment.

Tumor measurements were taken by caliper on a weekly basis. At the end of the experiment, animals were sacrificed, and tumors were excised and weighed. Intratumoral GSH, GSSG, and TR were quantified as described above.

To establish baseline sensitivities for cervical cancer cell lines to treatment, we tested the sensitivity of a panel of cervical cancer cell lines to radiation and cisplatin monotherapy (Fig. 1). Radiosensitivity was determined by CFA 48 hours after treatment with a single fraction of 2, 4, and 6 Gy radiation. CaSki was the most radioresistant cell line, whereas C33A was the most sensitive (Fig. 1A). Cisplatin sensitivity was determined by CFA 48 hours after treatment with 500 nmol/L cisplatin. CaSki was also the most resistant to cisplatin, and C33A was the most sensitive (Fig. 1B).

To test for the sensitivity to the glycolytic inhibitor 2-DG, a CFA was performed 24 hours after treatment with 20 mmol/L 2-DG (Fig. 1C). Radioresistant and chemoresistant CaSki cells were exquisitely sensitive to 2-DG monotherapy compared with the other cell lines. To further explore whether 2-DG sensitivity correlated with glucose uptake, we performed an FDG uptake assay (Fig. 1D). Consistent with the increased 2-DG sensitivity, CaSki had the highest FDG uptake in vitro followed by ME-180, C33A, and SiHa.

We next tested whether pretreatment with 2-DG would enhance the effects of cisplatin and radiation treatment (Fig. 1E and F). 2-DG enhanced the effects of cisplatin only minimally (Fig. 1E). In contrast, 2-DG significantly sensitized all cell lines tested to radiation, with the most dramatic effects observed in CaSki (P < 0.0001) after treatment with 2-DG and 2 Gy radiation (Fig. 1F).

BSO and AUR have been shown to inhibit GSH synthesis and inhibit TR activity (7, 23, 24). To characterize baseline levels of thiol redox pathway intermediates, intracellular GSH and GSSG levels were quantified in all cell lines tested (Fig. 2A–D). The total GSH content was high in CaSki and SiHa cells, whereas GSSG and %GSSG (% of total cellular GSH present in the form of GSSG) were found to be significantly higher in CaSki, consistent with a higher baseline level of oxidative stress in that cell line. Interestingly, TR activity was significantly higher in SiHa versus all other cells tested (Fig. 2D).

To examine if GSH depletion and TR inhibition would enhance the toxicity induced by 2-DG treatment, cells were treated with 20 mmol/L 2-DG and 1 mmol/L BSO for 48 hours with 100 nmol/L AUR added for the last 24 hours. CaSki and SiHa cells were significantly more sensitive to 2-DG+BSO+AUR (DBA) treatment than ME-180 and C33A, with significant decreases in clonogenic survival compared with treatment with 2-DG alone (Fig. 2E). In addition, C33A was more sensitive to BSO monotherapy compared with other cell lines (Fig. 2E).

To examine whether BSO and AUR treatment altered intracellular redox pools, levels of GSH, %GSSG, and TR activity were measured before and after treating cells with 2-DG+/-BSO+/-AUR (Fig. 2F–H). BSO monotherapy was effective in reducing total GSH content in all cell lines tested with residual GSH detected after BSO treatment in ME-180 cells (Fig. 2F). 2-DG+BSO (DB), BSO+AUR (BA), and 2-DG+BSO+AUR (DBA) significantly decreased total GSH content in all cell lines tested (Fig. 2F). The percentage of GSSG (% of total cellular GSH present in the form of GSSG) levels was significantly increased after BSO containing combination treatments in all cell lines tested, with the most significant increases in %GSSG seen in CaSki, SiHa, and C33A cells. (Fig. 2G). Increases in %GSSG in response to treatment were more modest in ME-180, consistent with posttreatment residual GSH pools in that cell line. We next tested whether baseline levels of GSH, %GSSG, and TR activity correlated with response to DBA treatment. There was a linear correlation between total GSH levels and sensitivity to DBA (Pearson r 0.963), whereas %GSSG levels and sensitivity to DBA showed a linear negative correlation (Pearson r 0.982). These results show that BSO was having the desired effect of depleting total GSH in all cervical carcinoma cells tested, with maximum increase in intracellular oxidative stress (estimated by increase in %GSSG) achieved with DB, BA, and DBA combinations. Overall, the simultaneous manipulations of GSH and TR using BSO and AUR combined with 2-DG were most effective at increasing metabolic oxidative stress in cervical cancer cell lines.

Consistent with these findings, the TR activity was significantly reduced by DBA treatment only in C33A cells (Fig. 2H). TR activity was decreased by BA treatment but not completely eliminated in SiHa cells (Fig. 2H). In contrast, DA and DB combination treatment increased TR activity in CaSki, the most sensitive cell line to 2-DG monotherapy. There was no significant difference in TR activity in ME-180 cells in response to BSO, AUR, DA, and DBA treatments (Fig. 2H). To test the specificity of AUR in target inhibition, we combined 2-DG along with TR1 and TR1 knockdown via siRNAs. A CFA was performed after CaSki (Supplementary Fig. S1A and S1B) and SiHa (Supplementary Fig. S1C and S1D) were transfected with TR1, TR2, and TR1+TR2 for 48 hours followed by 2-DG treatment for 24 hours. TR1, TR2, and TR1+TR2 knockdown decreased clonogenic survival in both CaSki (P < 0.01 Control vs. TR1, P < 0.01 Control vs. TR2, P < 0.01 Control vs. TR1+TR2) and SiHa (P < 0.001 Control vs. TR1/2, P < 0.01 Control vs. TR1+TR2; Supplementary Fig. S1A–S1D).

To determine the potential role of pro-oxidants in DBA-induced cell death, we quantified intracellular oxidation before and after treatment using 2′,7′-dichlorofluorescin diacetate (DCFDA) oxidation and dihydroethidium (DHE) as previously described (10). There was significantly higher DCFDA oxidation in SiHa and C33A after 2-DG+BSO+AUR treatment (Fig. 3A). When the cells were stained with CDCF, the oxidation-insensitive analogue of CDCFH_2, there was no significant difference in any of the cells confirming that the differences observed were indeed due to changes in levels of dye oxidation and not due to changes in cell size, probe influx, probe efflux, or ester cleavage (Fig. 3B). Only SiHa and ME-180 cells displayed significant increases in DHE oxidation after DBA treatment (Fig. 3C). These results suggest that in cervical cancer cells, treatment with DBA induces accumulation of pro-oxidants, presumably superoxide and hydroperoxides.

NAC, a thiol antioxidant, was used to determine whether 2-DG+BSO+AUR-induced cytotoxicity was related to thiol oxidation and depletion of intracellular thiol pools. NAC significantly inhibited the decrease in cell viability seen in SiHa cells following exposure to DBA, whereas no inhibition of toxicity was seen in similarly treated CaSki cells (Supplementary Fig. S2A and S2B). NAC significantly inhibited BSO+AUR (BA) treatment–related toxicity in both CaSki and SiHa cells. Furthermore, the addition of 2-DG to BSO+AUR increases cytotoxicity in a fashion that cannot be reversed by simply restoring intracellular-reduced thiol pools with NAC. Interestingly, the addition of NAC also did not inhibit 2-DG–associated toxicity in CaSki cells, the most sensitive cell line to 2-DG monotherapy. We also performed clonogenic cell survival assays after adenoviral-mediated overexpression of the antioxidant enzyme catalase in presence or absence of 2-DG+BSO+AUR (Supplementary Fig. S3A and S3B). Catalase expression in mκunits/milligram in the order of Ad empty and Ad-catalase is as follows: SiHa (8.0/316), CaSki (4/76), and C33A (6.0/116). Overexpression of catalase partially restored survival in response to DG, BSO, DB, and DBA treatments in SiHa cells, whereas effects of catalase overexpression were not significant in similarly treated CaSki cells (Supplementary Fig. S3). The varying ability of catalase overexpression to inhibit toxicity suggests cell line–specific differences in the relative abundance of hydroperoxides versus superoxide (Fig. 3A–C).

To determine effects of DBA treatment on glucose metabolic pathways, we performed metabolomic analysis of C33A, CaSki, ME-180, and SiHa cells grown in 10 mmol/L uniformly labeled ¹³C-glucose. Each untreated cell line showed similar rates of glucose entry into the tricarboxylic acid (TCA) cycle, as represented by citrate labeling (Fig. 4A). DBA treatment resulted in significant (P < 0.05) decrease in citrate labeling in CaSki, ME180, and SiHa cells (Fig. 4A). Upon DBA treatment, SiHa showed the lowest labeling in citrate (46%), followed by CaSki (57%), ME-180 (66%), and C33A (69%); the differences between SiHa and ME-180 or C33A were significant (P < 0.05) in pairwise comparisons. This ordering of citrate labeling parallels that of cell survival after DBA treatment, with SiHa and CaSki being exquisitely sensitive to the combined drugs and ME-180 and C33A being relatively resistant (Fig. 2A). Other evidence of decreased TCA cycle activity in DBA-sensitive cells after DBA treatment can be seen in the labeling of glutamate, which is derived from the TCA cycle intermediate 2-oxoglutarate (Fig. 4B; labeled fraction of glutamate in SiHa and CaSki is significantly lower than in either C33A or ME180), as well as the labeling pattern of the hexosamine biosynthesis pathway product UDP-N-acetylglucosamine (Fig. 4C), whose M+7, M+8, and M+13 isotopologues reflect the addition of a labeled cytosolic acetyl-CoA derived from labeled mitochondrial citrate to the M+5 (labeled ribose moiety), M+6 (labeled glucosamine moiety), and M+11 (both ribose and glucosamine moieties labeled) isotopologues, respectively. The fraction of UDP-GlcNAc in the M+7, M+8, and M+13 isotopologue forms decreased upon DBA treatment in all cell lines, but the decrease was greatest in SiHa and CaSki, indicating decreased availability of labeled acetyl-CoA and thus also decreased TCA cycle activity. DBA treatment also resulted in decreased ratio of the +10 to +5 isotopologues in NAD representing decreased flux from glucose to ribose of NAD in the sensitive cell lines CaSki and SiHa (Fig. 4D) and decreased lactate production (Fig. 4E).

To determine whether DBA treatment stimulated nutrient stress–induced cell signaling pathways, a series of Western blots were performed. DBA treatment induced AMPK and JNK activation as evidenced by increases in pAMPK and pJNK expression in DBA-treated CaSki and SiHa cells (Fig. 5A). This effect was not dependent on expression of LKB1 as SiHa cells are LKB1 negative (Fig. 5A). DBA treatment was also found to promote accumulation of pro-cell death BH3 only protein Bim (Fig. 5A). DBA treatment induced autophagy in CaSki cells but not SiHa cells as evidenced by MAPLC3 cleavage (Fig. 5A). CaSki cells have higher Myc and ^Ser62 Myc levels at baseline and after DBA treatment (Fig. 5B).

Knockdown via siRNA was used to test whether AMPK signaling was required for DBA-induced cell death in CaSki and SiHa cells (Fig. 5C and D). Interestingly, AMPK knockdown rescued DBA-induced cell death in LKB1-negative SiHa cells (Fig. 5C), but not in LKB1-intact CaSki cells. To test whether cell death induced by DBA treatment was due to autophagic cell death in CaSki cells, we used an orthogonal strategy with drugs and siRNA knockdown. Chloroquine, an autophagy inhibitor, significantly inhibited DBA-induced cell death in CaSki confirmed by CFA and cell viability (Fig. 5E; Supplementary Fig. S3C). Beclin1 knockdown also rescued CaSki cells significantly from DBA-induced cell death (DBA alone = 27%, siRNABeclin1+DBA = 45%, P < 0.0003; Fig. 5F). The number of punctate CaSki cells (cleaved LC3) after DBA treatment was 50% compared with control cells (Fig. 5G), and this punctate pattern was not observed in SiHa cells as evidenced by no cleaved fragment of LC3 (Fig. 5A). DBA treatment activated AMPK autophagic cell pathway in CaSki as seen by ^ser317Ulk phosphorylation in contrast to the ^ser757Ulk in SiHa (Fig. 5B). AMPK activates p62, which is a marker for autophagic cell death (25). DBA-treated CaSki cells displayed activation of p62, but similar results were not observed in SiHa (Fig. 5H). DBA treatment induced PARP and caspase 3 cleavages in CaSki but not in SiHa cells (Fig. 5B). We observed DBA treatment–induced p53 phosphorylation (Fig. 5H) resulting in G₀–G₁ cell-cycle arrest in CaSki cells and cell death as evidenced by higher sub-G₁ fraction (Fig. 5I). Taken together, these results suggest that oxidative stress in DBA-sensitive CaSki cells results in AMPK, p53, and JNK activation, BIM accumulation, and activation of the caspase and PARP-dependent autophagic form of cell death. In contrast, in SiHa cells, this effect is LKB1-independent, AMPK-dependent, nonautophagic form of cell death.

To determine whether our in vitro observations with 2-DG, BSO, and AUR treatment could be verified in in vivo, mice with CaSki xenografts were treated with 2-DG, BSO+AUR, and 2-DG+BSO+AUR for 35 days 3 times a week. Drug treatment did not result in weight loss or behavioral changes during the period of study. CaSki tumor growth delay in response to 2-DG only was observed after 28 days (Fig. 6A). Both BA and DBA treatments were effective in reducing CaSki tumor volumes (Fig. 6A). To assess whether 2-DG, BSO, and AUR combination altered GSH/GSSG levels in vivo, tumors were harvested at the end of the experiment and GSH/GSSG levels, %GSSG, and TR activity were quantified. Similar to our in vitro results, %GSSG increased in response to 2-DG and BA treatments (Fig. 6B). 2-DG and BA treatments decreased GSH levels and increased TR activity in CaSki tumors (Fig. 6C and D). In order to test the radiosensitizing potential of our new drug strategy, CaSki cells were treated with 2-DG, BSO+AUR, and 2-DG+BSO+AUR and after 24 hours were irradiated using single fraction radiation dose of 2 Gy. 2-DG, BA, and DBA radiosensitized CaSki cells as evidenced by reduced surviving fraction in Fig. 6E.

In order to test whether targeting glycolysis and intracellular redox metabolism with DBA could function as a radiosensitizer, we performed CFAs after 2 Gy radiation +/- DBA with the SiHa cell line (Fig. 7A). Radioresistant SiHa cells were sensitized to 2 Gy by the addition of DBA. To determine whether radiosensitization could be observed in vivo, we treated SiHa xenografts with DBA for 2 weeks with concurrent single fraction dose of 2 and 4 Gy tumor-directed irradiation using the SARRP system. (Fig. 7B and C). We observed that DBA treatment significantly reduced tumor growth after single fraction doses of both 2 and 4 Gy (*DBA vs. 2 Gy + DBA, P < 0.01, DBA vs. 4 Gy + DBA, P < 0.0001). The combination of 4 Gy and DBA treatment was most effective at limiting SiHa tumor growth in vivo (Fig. 7C and D).

In this preclinical study, we tested whether inhibition of glycolysis, with or without simultaneous inhibition of intracellular redox metabolism, was an effective treatment strategy for highly glycolytic and radioresistant cervical cancers. 2-DG was used to inhibit glycolysis, BSO to target GSH synthesis, and AUR to target TR. BSO and AUR have been studied in other cancer types as a means to inhibit intracellular redox metabolism (7, 12, 26). Our results show that the cisplatin-resistant and radioresistant cervical cell line CaSki is highly sensitive to inhibition of glycolysis with 2-DG monotherapy. Addition of BSO and AUR sensitized SiHa cells to 2-DG treatment, indicating that the redox potential generated by glycolysis was dampened by inhibition of glutathione and thioredoxin pathways (27). Using detailed analysis of glutathione and thioredoxin activity, we were able to demonstrate direct inhibition of the redox pathways by BSO and AUR treatment. Due to the high baseline TR activity, SiHa survived better after 2-DG+AUR and BSO+AUR-only treatment groups.

Although the in vitro evidence for our strategy is compelling, we wanted to test whether our results could be validated in vivo. In support of our in vitro observations, we found that CaSki xenografts responded well to drug treatment alone with metabolic inhibitors 2-DG, BSO+AUR, and 2-DG+BSO+AUR. Tumors samples collected at the end of the experiment demonstrated inhibition of glutathione and thioredoxin pathways by BSO and AUR, respectively, in CaSki xenografts. The drug strategy was well tolerated in animals, supporting the possibility of transitioning this approach to the clinic. We wanted to test whether DBA treatment could enhance radiosensitivity and potentially be integrated or replace cisplatin treatment as part of the standard of care. Importantly, we were able to demonstrate radiosensitization of SiHa with DBA treatment both in vitro and in vivo. In vivo, DBA treatment resulted in significant tumor growth delay at clinically relevant radiation doses, suggesting that this drug strategy could be feasible and effective at radiation doses currently used in the clinic. 2-DG and BSO have already been used in humans to treat other malignancies (28, 29), and AUR is an FDA-approved drug (30, 31). Our preclinical results form a strong rationale for testing 2-DG+BSO+AUR as an alternative drug strategy for highly glycolytic, radioresistant cervical tumors.

Mechanistically, we wanted to determine whether DBA treatment was stimulating cell signaling pathways known to sense nutrient stress. Oxidative stress is known to activate AMPK in cancer cells (32), and we observed high levels of activated AMPK activation after DBA treatment in DBA-sensitive cells (33). We also observed DBA-induced phosphorylation of JNK in DBA-sensitive cell lines, and these data support the finding that JNK induces AMPK activation under conditions of glucose deprivation (34). This was true in both LKB1-null (35) and -intact cell lines (36), suggesting that AMPK may be activated by DBA treatment via LKB1-independent pathways (37–39). Concannon and colleagues reported that AMPK activates BH3-only protein Bim during stress-induced cell death (40), and we observed Bim accumulation in DBA-sensitive cells. Intriguingly, activation of AMPK in DBA-sensitive cell lines was not associated with an increase in glucose-derived mitochondrial energy production, as entry of glucose into the TCA cycle was decreased in CaSki and SiHa upon 2-DG+BSO+AUR treatment (Fig. 4A). This suggests that AMPK modulates these cells' response to drugs through a PGC-(peroxisome proliferator–activated receptor gamma coactivator) 1α-independent mechanism (41, 42). DBA induced LC3 cleavage and p62 phosphorylation (Fig. 5A and H) in CaSki cells, suggesting cell death may be occurring via autophagic cell death (43). Chloroquine, a known inhibitor of autophagy, and knockdown of proautophagy gene BECLIN1 partially rescued CaSki cells from 2-DG+BSO+AUR-induced cell death (44).

Sustained high-level AMPK activation was observed in both CaSki and SiHa in response to DBA treatment, but with different downstream effects. CaSki is MYC transformed and SiHa is not (Fig. 5B). Myc is known to promote lipid, nucleotide, and protein synthesis by utilizing the citric acid cycle to serve biosynthetic processes, which simultaneously leads to ATP production and the activation of cellular energy sensing protein, AMP-activated protein kinase (AMPK). Cells with normal growth control can stop cell proliferation to replenish ATP reservoirs, whereas MYC expression prevents a similar break by blocking the cell-cycle exit. The relentless cell-cycle activation, accompanied by sustained metabolic stress and AMPK activity, switches the energy-saving AMPK to proapoptotic AMPK. Previously, Jones and colleagues have demonstrated that glucose deprivation–induced AMPK activation directly phosphorylates N-terminal Ser15 of p53, leading to initiation of G₀–G₁ cell-cycle arrest (45). Our study supports this finding that AMPK leads to ^Ser15p53 phosphorylation leading to a G₀–G₁ cell-cycle arrest in CaSki cells (Fig. 5H and I). The persistent AMPK activity in CaSki ultimately results in caspase- and PARP-dependent autophagic cell death (Fig. 5A–I). MYC-transformed CaSki cells experience energy crisis due to persistent cell-cycle activation, which results in activation of energy sensor AMPK, ultimately leading to autophagic cell death (46). This could be one of the reasons that CaSki was rescued by autophagy inhibitors and not by AMPK knockdown. In contrast in SiHa cells, which are not MYC-transformed, DBA treatment results in a caspase/PARP and p53-independent nonautophagic cell death.

Here, we demonstrate that four different cervical cancer cell lines with varying mutation states in the PI3K/Akt/PTEN pathway all exhibit similar rates of glucose flux into lactate (Fig. 4D and E). However, treatment with DBA differentially decreases this flux in DBA-sensitive cells (SiHa and CaSki). Furthermore, these DBA-sensitive cells are unable to process glucose through the TCA cycle (Fig. 4A), with subsequent downregulation of pathways that are dependent on TCA cycle products such as glutamate and UDP-GlcNAc synthesis (Fig. 4B and C). Overall, the metabolic effect of DBA treatment in DBA-sensitive MYC-transformed cells is one of activation of AMPK and autophagy, which may be a consequence of their decreased glucose metabolism and exhaustion of ATP reservoirs.

G.J. Patti is a consultant/advisory board member for Cambridge Isotope Laboratories. No potential conflicts of interest were disclosed by the other authors.

Conception and design: R. Rashmi, X. Huang, A.E. Elhammali, D.R. Spitz, J.K. Schwarz

Development of methodology: X. Huang, J.M. Floberg, A.E. Elhammali, M.L. McCormick, J.K. Schwarz

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): R. Rashmi, X. Huang, J.M. Floberg, A.E. Elhammali, M.L. McCormick, G.J. Patti, D.R. Spitz, J.K. Schwarz

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): R. Rashmi, X. Huang, J.M. Floberg, A.E. Elhammali, M.L. McCormick, G.J. Patti, D.R. Spitz, J.K. Schwarz

Writing, review, and/or revision of the manuscript: R. Rashmi, X. Huang, J.M. Floberg, A.E. Elhammali, G.J. Patti, D.R. Spitz, J.K. Schwarz

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M.L. McCormick

Study supervision: G.J. Patti, J.K. Schwarz

This work was supported by NIH R01CA181745 (to J.K. Schwarz), Resident Research Seed Grant 531448 from the American Society for Radiation Oncology (to J.M. Floberg), Research Medical Student Grants from the Radiological Society of North America (RSNA) RMS1612 (to X. Huang) and RMS1408 (to A.E. Elhammali), and NIH R01ES022181 and R21CA191097 (to G.J. Patti). The Small Animal Radiation Research Platform was purchased with the assistance of NIH S10 OD020136 (to D. Hallahan). The work at the University of Iowa done by the Radiation and Free Radical Research Core Lab was partially supported by NIH R01 CA182804 and NIH P30 CA086862 (to D.R. Spitz). We would like to thank Michael Zahner and 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.