Undermining Glutaminolysis Bolsters Chemotherapy While NRF2 Promotes Chemoresistance in KRAS-Driven Pancreatic Cancers

Pancreatic cancer ranks among the deadliest and most aggressive malignancies (1). Late-stage diagnoses limit surgical intervention, so radiation and chemotherapy become standards of care. However, given its resistance to established therapeutic regimens, including gemcitabine, a first-line chemotherapeutic agent for pancreatic cancer (2, 3), novel therapeutic approaches for patients with pancreatic cancer are desperately needed. Most pancreatic ductal adenocarcinomas (PDAC) harbor mutationally activated KRAS, most often at position G12D (4). Unfortunately, KRAS G12D itself is not yet druggable (5). Thus, alternative strategies are needed.

Oncogene-directed metabolic reprogramming is an emerging hallmark in cancer (6, 7). KRAS promotes metabolic rewiring, presenting opportunities for novel therapeutic strategies (4, 8, 9). To meet its increased anabolic need and support proliferation and survival, mutant KRAS upregulates glycolysis and alters glutamine metabolism (1, 4, 10, 11) by increasing glutamine uptake (12) and reprogramming glutaminolysis (1, 13).

Mutant KRAS confers chemoresistance by activating the transcription factor NRF2, encoded by NFE2L2, a critical contributor to the cellular stress response (14–16). Aberrant activation of NRF2 promotes pancreatic tumorigenesis. NRF2 regulates the expression of a battery of cytoprotective genes, including those involved in drug detoxification and redox balance (17, 18), and it plays a critical role in anabolic cancer metabolism by altering glucose and glutamine metabolism (19). Although activated NRF2 has recently been linked to upregulated glutamine dependency in cancer cells (20), the impact of NRF2 in KRAS-driven PDAC cell metabolic rewiring is less well characterized.

We investigated the role of NRF2 in PDAC by first analyzing NRF2 expression levels in pancreatic tissues and cell lines. We observed variable expression of NRF2 and identified elevated NRF2 levels in higher-grade pancreatic cancer tissues. We also show that PDACs with significantly elevated NRF2 expression are more chemoresistant. We provide evidence that elevated levels of NRF2 in PDACs are responsible for metabolic alterations—specifically, pathways that are regulated by glutamine metabolism. We perturbed glutamine metabolic pathways and showed that inhibiting glutaminolysis in PDACs can overcome gemcitabine resistance. Significantly, targeting glutamine metabolism via glutaminase inhibitors may be an effective strategy for treating PDACs harboring KRAS mutations.

All cell lines used in this study were obtained from the RAS Reagent Core (Frederick National Laboratory for Cancer Research, Frederick, MD). Cell line identities were confirmed by short tandem repeat analysis. Cell lines were tested for Mycoplasma, authenticated before and at several points throughout the study, and maintained under conditions described by the manufacturer. See Supplementary Methods for details.

For stable cell line generation, control cell lines were generated using Precision LentiORF RFP Positive Control (#OHS5833, Dharmacon), and NRF2-overexpressing lines were generated using Precision LentiORF NFE2L2 without Stop Codon (#OHS5900-202624558, Dharmacon) lentiviruses. Cells were expanded in selection media for a minimum of 1 week prior to further experimentation.

Nonessential amino acid (NEAA) solution (#11140) was purchased from Gibco. Erastin (5449) was obtained from Tocris Bioscience. AI-1(SML0009), DMKG/Dimethyl 2-oxogluterate (#349631), 5-Fluorouracil (5-FU #F6627), and capecitabine (SML0653) were purchased from Sigma-Aldrich. BPTES (S7753) and RSL3 (S8155) were obtained from Selleckchem. Gemcitabine (#3259) was obtained from Tocris Bioscience. CB-839 (#B0084-462472) was purchased from BOC Sciences.

Cells were plated at 70% confluence in medium containing 10% serum. Standard protocol was followed (21). In short, after 24 hours of plating, cells were transfected with shRNA or overexpressing plasmids using Lipofectamine 3000 (Invitrogen) according to the manufacturer's instructions. NFE2L2 mission shRNA plasmids (TRCN00007555 and TRCN00007558) and mission control shRNA plasmid (SHC002) were obtained from Sigma. For transient overexpression, NFE2L2-overexpressing plasmid was purchased from DNASU (# HsCD00079566) and control plasmid with a similar backbone (#V37006) was purchased from Invitrogen.

A standard immunoblotting method was followed (21). Briefly, cell lysates were collected using M-PER (Thermo Fisher Scientific, 78501) and proteins were separated on denaturing SDS-PAGE gels (Bio-Rad). Electrophoresed proteins were transferred to nitrocellulose membranes. After transfer, membranes were blocked in an isotonic solution containing 5% nonfat dry milk in TBS. Membranes were incubated with primary antibodies as described in the text, and depending on the origin of the primary antibody, either anti-mouse or anti-rabbit HRP (Pierce) conjugated IgG was used for detection using the ECL system (Pierce).

For immunofluorescence experiments, cells were permeabilized, incubated with indicated antibodies, and visualized by inverted microscope. For IHC, slides were stained on the Discovery Ultra System (Roche) following the manufacturer's instructions and scanned on the Digital Pathology Scanner (Leica). Further details are mentioned in the Supplementary Methods.

Cell proliferation assays were conducted under different drug or media conditions. At the indicated times, cells were washed with PBS, trypsinized, stained with crystal violet, and counted. Cell viability was assessed using the CellTiter-Glo assay (#G-8461, Promega) following Promega's instructions. IC₅₀ was calculated using GraphPad Prism.

Cells were trypsinized followed by washing and resuspension with D-PBS (#14190-144, Gibco). Samples were frozen and transferred to NIH-WCMC (metabolomics) and Lipotype (lipidomics) for detailed analysis.

The pancreas cancer data sets from Oncomine (https://www.oncomine.org/resource/) were used for various metabolic gene analyses. Parameters for each Oncomine data analysis are mentioned in the figures.

The Cancer Genome Atlas (TCGA)-pancreatic adenocarcinoma (PAAD) survival curves for NFE2L2 and NFE2L2/KRAS expression were assessed using the PROGgeneV2 webtool (http://genomics.jefferson.edu/proggene/). The data were presented with P values with limits at the 75th and 25th percentiles. The parameters used in the analysis are mentioned in the figures.

Metabolic genes upregulated in PAAD at later stage (stage II) or higher grade (grade 3) compared with stage I or grade 1, respectively, in RNA-sequencing data of TCGA PAAD patients. t test analysis was used to determine P values. Oncomine analysis was also used to determine the upregulated metabolic genes in higher grade PAAD tissues in copy number of TCGA PAAD patients. Threshold P values used in the analysis are mentioned in the figure.

For TCGA data clinical attributes and gene-set enrichment analysis (GSEA), we analyzed data on significance of pathways/terms from Gene Ontology Biological Process, and pathway databases using the GSEA method, which closely followed the previously described method (22).

Animal work was performed in accordance with the guidelines for animal welfare and protocols approved by the Institutional Animal Care and Use Committee. Twenty (n = 5 mice/group) nude mice (Jackson Laboratory) had 5 × 10⁶ PANC-1 or SU.86.86 or MIA PaCa-2 cells implanted in the left flank. Cells were resuspended in PBS then mixed with an equal volume of Matrigel (Corning 354234) before subcutaneous implantation into mice. The animals were randomized when tumors reached a mean volume of 100 mm³ and were assigned to the indicated groups. Treatment was continued until day 38 (PANC-1, MIA PaCa-2) or day 34 (SU.86.86).

Murine pancreatic adenocarcinoma cell line KPC (Pdx1-Cre^tg/+; Kras^G12D/+; Tp53^R172H/+) developed from spontaneous primary PDAC tumors in triple transgenic mice bearing a combination of genetically engineered Pdx-1-Cre^tg/+, Kras^SLS-G12D/+, and p53 R172H^LSL-R172H/+ alleles was provided by the NCI's Center for Advanced Preclinical Research. C57bl/6J mice (Jackson Laboratory) received a subcutaneous inoculum of KPC cells. Upon engraftment, animals were allowed to develop palpable lesions (<4 mm³ tumor volume) and then randomized into treatment groups and treated with drugs as indicated.

Data are expressed as mean ± SEM of at least three independent experiments. Statistical analysis was done using Microsoft Excel or GraphPad 8. Probability < 0.05 was considered the statistical significance threshold.

NRF2 is induced by oncogenic KRAS and is critical for tumor progression (14, 15, 23). To investigate the connection between KRAS and NRF2 in PDAC, we used a pair of human pancreatic ductal epithelial stable cell lines (HPNE cells; human pancreatic nestin–expressing cells, ATCC CRL 4023), one expressing wild-type (WT) KRAS and the other a mutant KRAS G12D. As expected, we observed dramatic activation of NRF2 and known downstream effectors (17) of the NRF2 pathway (NQO1, GCLC, AKR1C1, and HMOX1) in the KRAS G12D HPNE compared with KRAS WT HPNE (Fig. 1A; Supplementary Fig. S1A). Oncomine analysis indicated a 2.2-fold induction of NRF2 expression in KRAS-mutant pancreatic cancer compared with KRAS WTs (Supplementary Fig. S1B). Further analysis through cBioPortal on PDAC samples revealed that 21% have elevated NRF2 mRNA expression and concurrent upregulation in KRAS expression (Supplementary Fig. S1C). Using TCGA-PAAD data, we evaluated the effect of combined NRF2-KRAS expression on patient survival. Patients with lower expression of the NRF2-KRAS gene set combination (Supplementary Fig. S1D and S1E) had longer survival. TCGA-PAAD data also demonstrate upregulation of NRF2 in mutant KRAS compared with WT (Supplementary Fig. S1F). Next, we chose a panel of six PDAC cell lines to check NRF2 basal levels. Surprisingly, we found differential expression of NRF2 in those lines (Fig. 1B and C). We also found, at least within this panel, that KRAS G12D–mutant cells (Supplementary Fig. S1G) have substantial NRF2 protein levels, which is significant, as G12D is the prevalent mutation in pancreatic cancer (4). We further examined the basal level of NRF2 in another tumor model (Kras^G12D/+; Tp53^R172H/+) KPC mouse pancreas tissue. NRF2 protein levels are also upregulated in KPC mouse PDAC tissues compared with normal pancreatic tissue (Fig. 1D). Taken together, these data reflect a variation of intensity in NRF2 expression among PDAC samples.

To further characterize levels of NRF2 expression in PDAC, we analyzed human pancreatic cancer tissue microarrays, including PDAC tissue samples and normal pancreas tissue (Fig. 1E, Supplementary Fig. S2A) with patient clinicopathologic features (Supplementary Table S1). The accumulated intensity of IHC staining of NRF2 revealed a wide range in all tissue groups. Further analysis demonstrated diverse staining scores according to accumulated intensity and revealed that the majority (63% of grade 3 and 58% of grade 2) of higher grade PDAC tissue samples showed intermediate and strong NRF2 staining compared with grade 1 samples (33% of mild staining). These data (Fig. 1F) illustrate the presence of NRF2 in poorly differentiated high-grade PDAC tissue samples (strong staining in grade 3: 26%), which are more aggressive. Among well-differentiated low-grade samples, we did not see strong staining of NRF2. Concurrently, 67% of grade 1 tissue samples in our study had weak staining scores. Surprisingly, certain normal pancreas tissues had moderately high levels of NRF2 based on staining intensity. This could be caused by several factors. Multiple studies have reasoned that NRF2 upregulation protects cells from stress stimuli including environmental agents/drugs, inflammatory stresses, and chronic exposure to carcinogens such as cigarette smoke. Moreover, NRF2 signaling is induced by medications for diabetes, pancreatitis, and other diseases (24, 25), which could have influenced NRF2 expression in some normal tissues. However, after thorough analysis of the human tissue microarray, our data indicate a possible correlation between NRF2 expression and higher grade PDAC tissue samples. Elevated NRF2 expression associated with tumor progression is also seen in TCGA-PAAD survival data. Kaplan–Meier analysis shows a significant difference in outcome when patients are separated by high and low NRF2 expression levels (Fig. 1G, Supplementary Fig. S2B). Therefore, NRF2 expression is associated with a poor clinical outcome. Oncomine analysis also validates elevated levels of NRF2 in PDAC compared with a normal pancreas (Fig. 1H). Taken together, our data demonstrate that NRF2 is expressed over a wide range that, in turn, correlates with the clinical status of KRAS-driven pancreatic cancer.

Oncogenic KRAS enhances chemoresistance by upregulating NRF2 signaling (15). Gemcitabine, which interferes with DNA synthesis, is used for treating patients with PDAC, although its efficacy is limited (2). It has been shown that long-term gemcitabine treatment leads to NRF2 activation, resulting in chemoresistance (26). Therefore, we examined whether differential expression of NRF2 in PDAC cells is linked to gemcitabine sensitivity. We measured the sensitivity of PDAC cells to gemcitabine and observed that cells with higher levels of NRF2 had higher IC₅₀ values (Fig. 2A; Supplementary Table S2A). Knocking down NRF2 using two different sequences of shRNAs (Supplementary Fig. S3A) elicited a significant increase in gemcitabine sensitivity (Fig. 2B; Supplementary Table S2B). We then used AI-1, a well-characterized, potent pharmacologic activator of NRF2 that covalently modifies Keap1 to activate NRF2 (27), in PANC-1 cells (Supplementary Fig. S3B). Remarkably, it increased NRF2 expression in PANC-1 cells (higher NRF2 basal levels) and increased gemcitabine resistance (Fig. 2C; Supplementary Table S2C). Overexpressing NRF2 in PDAC cells (Supplementary Fig. S3C), which have a low basal level of NRF2, increased gemcitabine resistance (Fig. 2D; Supplementary Table S2D). Taken together, our data indicate that NRF2 levels regulate the sensitivity of PDAC cells to gemcitabine. We then used AI-1 to activate NRF2 in BxPC-3 and MIA PaCa-2 cells (Supplementary Fig. S3D) and tested gemcitabine sensitivity. Cells became significantly more resistant to gemcitabine upon AI-1 treatment (Fig. 2E; Supplementary Table S2E). We also observed that knocking down NRF2 (Supplementary Fig. S3E) in MIA PaCa-2 (low NRF2 basal levels) further sensitizes the cells to gemcitabine (Fig. 2F; Supplementary Table S2F). These data collectively demonstrate NRF2′s critical role in conferring chemoresistance to PDAC cells.

We next sought to determine the metabolic effects caused by NRF2 in PDAC cells. After validating the dosage of AI-1 needed to activate NRF2 pathway downstream effectors (Fig. 3A), we performed metabolomic analysis (Supplementary Fig. S4A). Metabolite set enrichment analysis of upregulated metabolites revealed enrichments in protein biosynthesis, amino acid metabolism (aspartate, methionine, arginine, proline, alanine, and cysteine), ammonia recycle, urea cycle, and malate–aspartate shuttle, along with glutathione metabolism upon NRF2 activation (Fig. 3B, first panel). To better understand metabolic changes following NRF2 upregulation in PDAC cells, we also performed network analysis to identify key pathway clusters regulated by activated NRF2 (Fig. 3B, second panel; Supplementary Fig. S4B). As expected, the amino acids pool was one of the most enriched compound clusters upon NRF2 activation in BxPC-3 cells. In addition, we sought to determine the effect that similar treatment had on lipid composition. We performed a lipidomic analysis of BxPC-3 cells upon AI-1 treatment and, except for minimal induction in triacylglycerol level, we found no other significant changes among major lipid class compositions following NRF2 activation (Fig. 3C).

We also looked at NRF2 activation's impact on cell-cycle progression (Supplementary Fig. S5A and S5B) by analyzing flow cytometry. We found no significant changes. We performed a time course study of AI-1 in BxPC-3 cells, and within 6 hours, major metabolic signaling markers were dramatically induced (Fig. 3D). A similar phenomenon occurred when we elevated NRF2 level in MIA PaCa-2 cells using a pharmacologic (Fig. 3E) or genetic (Fig. 3F) approach. Oncomine analysis suggested that these metabolic genes are upregulated in human PDAC samples compared with their normal counterparts. Interestingly, analysis of TCGA-PAAD datasets also suggests that those metabolic genes induced by NRF2 activation are significantly upregulated at later stages or higher grades of pancreatic cancer (Table 1).

From our metabolomic data analysis, the common factor behind the metabolic changes appears to be glutamine metabolism. In this context, it is noteworthy that protein biosynthesis and amino acid synthesis, including other enriched pathways that were heightened upon NRF2 activation, are regulated by glutaminolysis (10). Likewise, major glutamine metabolic pathway intermediates such as GLS1, GOT1, GDH1, xCT, and GCLC were induced upon NRF2 activation (Fig. 3D–F).

We next examined the effect of NRF2 activation on the DNA damage pathway caused by gemcitabine. By measuring the levels of well-known DNA damage stress response pathway markers (28), we demonstrated that NRF2 activation can reverse DNA damage induced by gemcitabine (Fig. 3G), which establishes NRF2′s comprehensive role in PDAC metabolic signaling. These results show that activated NRF2 plays a significant role in pancreatic cancer metabolic signaling pathways.

After exploring NRF2′s role in PDAC metabolism and observing the correlation between glutamine metabolism and NRF2 upregulation, we sought to elucidate the role of metabolic pathways in gemcitabine-resistant patients with PDAC. On the basis of Oncomine analysis, we identified the upregulation of GLS1 (glutaminase1) in gemcitabine-resistant samples (Supplementary Fig. S6A). To satisfy our curiosity, we further analyzed TCGA data for patients with gemcitabine-treated PAAD and compared relative metabolic pathway enrichments between gemcitabine-treated patients with responsive and resistant groups (Supplementary Table S3; Supplementary Fig. S6B–S6E). It appears that higher expressions of NRF2 and GLS1 mRNA were detected in a small subset of patients grouped under patients with gemcitabine-resistant pancreatic cancer, with trivial significance (Supplementary Fig. S6F). However, the nucleotide biosynthetic pathway was significantly enriched in patients with progressive disease compared with patients from complete response in patients of all stages as well as stage 2–only patients (Fig. 4A; Supplementary Table S3). Given that KRAS mediates PDAC tumor survival by regulating nucleotide synthesis (8) and that glutamine is a central precursor for protein and nucleotide biosynthesis (10, 29), we theorized that targeting glutamine metabolism would enhance sensitivity to chemotherapy.

Glutaminolysis plays diverse roles in cellular metabolism. Once inside the cell, glutamine is deaminated to glutamate by GLS and then converted to α-ketoglutarate and nonessential amino acids by GOT. α-Ketoglutarate plays a critical role in anaplerotic entry of glutamine into the tricarboxylic acid (TCA) cycle. Glutamate is also exported to the extracellular space by the xCT antiporter concurrent with the exchange of cystine to support the redox balance inside the cell (10). The conversion of α-ketoglutarate from isocitrate is mediated by IDH1, whereas a multistep conversion of heme from α-ketoglutarate (30) is followed by the generation of biliverdin mediated by HMOX1. The interplay between genes induced by activated NRF2 (indicated in Fig. 3D–F) and glutaminolysis is summarized in Fig. 4B. As expected, glutamine deprivation affects PDAC cell proliferation (Fig. 4C; Supplementary Fig. S7A), which could be rescued (Fig. 4D; Supplementary Fig. S7B) by adding glutaminolysis intermediates (α-ketoglutarate and nonessential amino acids) or by using erastin, an xCT (glutamate–cysteine antiporter) inhibitor. The combination of both α-ketoglutarate and nonessential amino acids rescued proliferation in glutamine-starved PANC-1, indicating that generating both α-ketoglutarate and nonessential amino acids in the transaminase reaction is critical for cell proliferation. These data elucidate PDAC cells' dependence on glutamine metabolism, specifically the utilization of glutamine anaplerosis in the TCA cycle and on cystine import, which occurs upon glutamate export. The ability of erastin to rescue the effect of glutamine deprivation potentiates the critical role played by cystine in glutamine dependence of PDAC cells.

To determine whether glutamine deprivation would overcome chemoresistance in PDAC, we tested the effect of glutamine limitation on a chemoresistance marker, glutathione peroxidase 4 (GPX4), which lies downstream of the glutamate–cysteine antiporter system and is also regulated by NRF2 (31, 32). By controlling glutathione synthesis, NRF2 is involved in this process (33). Surprisingly, glutamine deprivation reduced GPX4 levels (Fig. 4E). We also examined the effect of glutamine restriction on another marker of chemoresistance, stress-granule formation, which is upregulated by the production of a prostaglandin, 15d-PGJ2, in mutant KRAS cells (34, 35). In agreement with earlier findings (36), we found that 15d-PGJ2 activates NRF2 and forms stress granules (Fig. 4F). We then knocked down NRF2 and observed that ablating NRF2 decreased stress granule index (Fig. 4G), stress granule number, and area covered by stress granules (Supplementary Fig. S7C). Therefore, 15d-PGJ2–induced stress granule formation is controlled by NRF2 (Fig. 4H; Supplementary Fig. S7D). Remarkably, glutamine starvation restricts the capacity of 15d-PGJ2 to form stress granules in PDAC cells (Fig. 4I), which has been confirmed by decreased stress granule number, stress granule index, and stress granule area upon glutamine limitation (Fig. 4J; Supplementary Fig. S7E). Consistent with a previous report (37), gemcitabine alone could not trigger stress granule formation, but 15d-PGJ2 was able to form stress granules under gemcitabine treatment, which could be restricted by glutamine deprivation (Fig. 4K). Taken together, our data indicate that, by perturbing glutamine metabolism, we can restrain chemoresistance in PDAC cells and control pancreatic cancer cell proliferation.

Recent studies suggest that exploiting glutamine dependence could lead to new therapeutic strategies (38, 39). Therefore, we evaluated the effect of glutamine starvation in combination with gemcitabine on PDAC cell survival. Glutamine starvation in combination with gemcitabine significantly reduced clonogenic survival versus gemcitabine alone (Fig. 5A). As anticipated, glutamine deprivation sensitizes PDAC cells to gemcitabine; remarkably it could be rescued by adding α-ketoglutarate and nonessential amino acids (Fig. 5B; Supplementary Fig. S8A). This signifies the role of glutamine and its contribution to nucleotide biosynthesis through anaplerotic reactions, rationalizing the sensitization of cells to gemcitabine during glutamine starvation. While combining gemcitabine with glutamine restriction, SU.86.86 cells create a synthetic lethal phenotype (Fig. 5C), a result not seen in noncancerous cells (Fig. 5D; Supplementary Fig. S8B). Even though glutamine deprivation can be achieved in cell culture, this is clearly not a viable option in vivo. However, interfering with glutaminolysis is a possible approach. Various combinations of chemotherapeutic agents (such as gemcitabine, capecitabine, and 5-FU, which interferes with DNA synthesis) sensitize PDAC cells to pharmacologic inhibitors (glutaminase inhibitors: CB-839, BPTES) of glutamine metabolism (Fig. 5E). Next, we examined the impact of a potent GPX4 inhibitor, RSL3, in PANC-1 cells and found that enhanced gemcitabine sensitivity was detected after adding RSL3 (Fig. 5F). Because both glutamine metabolism and NRF2 affect GPX4, we checked the effect of a GPX4 inhibitor on gemcitabine sensitivity in PDAC cells. These data suggest that NRF2 and glutaminolysis regulate chemoresistance in conjunction with GPX4. Notably, some of these small-molecule agents targeting metabolism are approved or involved in clinical trials for the treatment of patients with cancer (40). To evaluate the combination of a glutaminase inhibitor, CB-839, with gemcitabine in PDAC in vivo, we followed the drug treatment dosing scheme shown in Fig. 5G. This scheme is used in multiple preclinical models of pancreatic ductal adenocarcinoma, including the genetically engineered KPC mice allograft model of PDA (Fig. 5H and I), MIA PaCa-2 (Fig. 5J and K), SU.86.86 (Fig. 5L and M), and PANC-1 (Fig. 5N and O) cell xenograft model. This combination caused tumor regression (Fig. 5H, J, L, and N) without adverse events or marked differences in mean body weight (Fig. 5I, K, M, and O). While it was important to examine the effect of treatments on NRF2 level in in vivo models, we could not detect any significant changes in NRF2 expression upon combination therapies in MIA PaCa-2 and PANC-1 xenograft models (Supplementary Fig. S8C and S8D). Kaplan–Meier analysis shows that this drug combination has significant antitumor effects. And, while neither CB-839 nor gemcitabine alone impacted survival, we found that the combination notably improved median survival over vehicle-treated animals or any single drug–treated cohort (Fig. 5P). CB-839 is being used in clinical trials for multiple cancers, including in combination with another chemotherapeutic agent (capecitabine: NCT 02861300), and it is ready for a basket trial in patients with NRF2 aberrant tumors (41). Our finding (summarized in Fig. 6) that disrupting glutamine metabolism potentiates the efficacy of chemotherapy represents an attractive approach to treating patients with RAS-driven pancreatic cancer.

In PDAC cells, mutant KRAS transcriptionally activates NRF2, a principal modulator of cellular redox, resulting in a reduced intracellular environment (42). NRF2 is also activated following MAPK inhibitor treatment, suggesting a role for NRF2 in RAS pathway drug sensitivity (14, 43). Recently, Papagiannakopoulos and colleagues (20, 44) reported that hyperactivation of NRF2 leads to increased glutamine dependency. Our data, which confirm these reports, show the induction of key proteins involved in glutamine metabolism following NRF2 upregulation in PDAC cells. Interestingly, some of those upregulated signaling nodes are associated with PDAC chemoresistance (45), suggesting that targeting glutamine metabolism could be an effective strategy to combat chemoresistance. Although activated NRF2 was not identified as a predictive marker, our data indicate that the NRF2-driven metabolic state makes PDAC cells sensitive to glutaminase inhibition, which is not validated in NRF2-low cells as they are also sensitive to CB-839. While other factors independent of NRF2 level (13, 34, 46) could influence sensitivity to the CB-839–gemcitabine combination in KRAS-mutant NRF2-low cells, it is also possible that NRF2 and KRAS regulate a convergent set of metabolic pathways such that NRF2 provides an additional benefit to NRF2-high cells and imparts glutaminase dependency. Further studies in genetically engineered mouse models of PDAC will be required to understand the complex mechanism of gemcitabine sensitivity with glutaminase inhibition in KRAS-driven PDAC cells specifically in the light of variable NRF2 level.

In addition to activating the NRF2 detoxification system, oncogenic KRAS alters glucose and glutamine metabolism to support PDAC cell proliferation (1, 9). Glutamine is a critical nutrient for cell growth and survival. It supports cell proliferation with the necessary intermediates for biosynthesis by providing carbon and nitrogen backbones for nucleotide and amino acid synthesis (10, 29). Interestingly, our data indicate that the production of nonessential amino acids along with α-ketoglutarate is critical for cell growth, whereas cystine import at the expense of glutamate also plays a significant role in PDAC cell proliferation. Surprisingly, we found that glutamine restriction not only affects PDAC cell proliferation but also regulates the level of GPX4, a protein responsible for cancer cell survival during therapeutic treatment (47). It has also been reported that GPX4 is transcriptionally modulated by NRF2 and is regulated by glutathione metabolism (31, 32). Our data demonstrating downregulation of GPX4 upon glutamine restriction signify the impact of targeting glutaminolysis to combat chemoresistance. Recent work from the Vander-Heiden group revealed that cancer cells' glutamine dependence is driven by xCT in combination with physiologic concentration of cystine (46). Consistent with their findings, our data suggest that activated NRF2 induces xCT to increase glutamine dependence, signifying xCT's role in facilitating the effects of glutamine restriction in PDAC cells.

We also found that a glutamine deficit can constrain the assembly of stress granules, which form in response to chemotherapy and in turn provide cytoprotection against stress stimuli (37, 48). A recent study revealed that KRAS upregulates stress granule formation, which contributes to chemoresistance. Mutant KRAS induces 15-d-PGJ2 accumulation, which mediates upregulated stress granule formation, and KRAS-dependent stress granule induction confers chemotherapy resistance (34). Prostaglandin 15d-PGJ2 is known to induce the antioxidant system that regulates NRF2-mediated pathway activation (36, 49). Here we report that glutamine restriction reduces the capacity of 15d-PGJ2 to form stress granules in KRAS-driven PDAC cells. Future studies of glutamine metabolism's role in regulating 15d-PGJ2–mediated NRF2 antioxidant pathway activation and cytoprotective stress granule accumulation will elucidate the complexity of chemoresistance in KRAS-mutant cancer cells.

We have observed that perturbing glutamine metabolism reinforces chemotherapy treatment and shown that a first-in-class small-molecule glutaminase inhibitor, CB-839, which is currently in a phase II clinical trial (NCT03263429), enhances the efficacy of gemcitabine and that this combination enhances tumor regression in a mouse model. Importantly, the doses used in this combination treatment are clinically tolerated. In addition, combining CB-839 with gemcitabine shows that utilizing chemotherapy could mitigate the adaptiveness of PDAC toward glutamine metabolism, substantiating recent findings from Kimmelman and colleagues (38). Consistent with their report, we demonstrated that glutaminase inhibition combined with chemotherapy delays tumor growth, even though single treatments of CB-839 or gemcitabine were ineffective. By using multiple preclinical models including KPC allograft and three different xenograft models, we have provided evidence that interfering with glutamine metabolism could overcome resistance against chemotherapies, suggesting that a clinical trial using this combinatorial therapy is merited in patients with pancreatic cancer. While our preclinical studies are encouraging and equally promising, they are preliminary in nature given their dependence on subcutaneous xenografts and allograft models. Although those models were employed as the first step in validating cell line results, which allowed for rapid screening to asses safety and efficacy of drug combinations, they have limitations. These include exhibiting a less-functional immune system, displaying a poor microenvironment, and lacking cellular diversity (50). Notably, a differential functional dependence on glutamine metabolism in variable environmental contexts has been indicated recently (46). Future work in genetically engineered PDAC mouse models will more adequately reflect the clinical application of this combination treatment.

Overcoming resistance to existing therapies is one of the major challenges for PDAC patient treatment. By combining glutamine metabolism inhibitors with chemotherapeutic drugs, we exploit the reliance of PDAC on glutamine and increased glutamine dependency upon NRF2 activation (Fig. 6). This study highlights therapeutic opportunities created by the metabolic reprogramming liability in pancreatic cancer cells and provides a possible therapeutic intervention for patients with chemoresistant pancreatic cancer. Future studies exploring this finding in a larger clinical context are warranted. It is conceivable that the metabolic needs associated with KRAS could be exploited to overcome chemotherapy resistance in a wide array of RAS-driven pancreatic cancers.

F. McCormick is an NCI Ras Initiative Scientific Director at Frederick National Laboratories for Cancer Research/Leidos Biomedical Research, Inc., a consultant at Amgen, Pfizer, BridgeBio Pharma, Inc., is a cofounder/consultant at Daiichi Sankyo and Aduro BioTech, Inc., is a consultant and reports receiving commercial research grant from Gilead Sciences, and has ownership interest (including patents) in BridgeBio Pharma, Inc. No potential conflicts of interest were disclosed by the other authors.

The content of this publication does not necessarily reflect the views or policies of the Department of Health and Human Services (HHS), nor does mention of trade names, commercial products, or organizations imply endorsement by the U.S. Government.

Conception and design: S. Mukhopadhyay, P.P. Adiseshaiah, F. McCormick

Development of methodology: S. Mukhopadhyay, D. Goswami, P.P. Adiseshaiah, W. Burgan, F. McCormick

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Mukhopadhyay, D. Goswami, W. Burgan, T.M. Guerin, S.V. Kozlov, F. McCormick

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Mukhopadhyay, D. Goswami, M. Yi, S.V. Kozlov, F. McCormick

Writing, review, and/or revision of the manuscript: S. Mukhopadhyay, D.V. Nissley, F. McCormick

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Mukhopadhyay, F. McCormick

Study supervision: S. Mukhopadhyay, D.V. Nissley, F. McCormick

This work was supported by grants from the NCI, HHS-NIH, under contract no. HHSN261200800001E. We thank Drs. Elda Grabocka, Jefferson University, for sharing details on stress granule assay, Gina M. DeNicola, Moffitt Cancer Center, for providing technical advice with NRF2 shRNAs and NRF2 antibody optimizations, and Pankaj K. Singh, Eppley Institute, for kindly sharing the gemcitabine-treated stage II pancreatic cancer patients' classification TCGA metafile. We thank Dr. Sung Eun Kim/UCSF for critical reading of the manuscript. We would like to thank Dr. Brynmor A. Watkins/Pharma Models, LLC, for assistance with the mouse study, Allan Wang/US Biomax for assistance with the TMA study, and Sili Fan/UC Davis for assistance with metabolomic data analysis. We also would like to thank members of the Frederick National Laboratory (Drs. Carissa Grose, Oscar Cheng, and Robert Stephens), NCI SPGM, NIH West Coast Metabolomics Center, and Lipotype GmbH for their 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.