5-Fluorouracil Enhances the Antitumor Activity of the Glutaminase Inhibitor CB-839 against PIK3CA-Mutant Colorectal Cancers

It is well documented that cancer cells in vitro can utilize glutamine as an anaplerotic substrate of the tricarboxylic acid (TCA) cycle to generate ATP and precursors for the synthesis of lipids, nucleotides, and other macromolecules to sustain rapid tumor growth (1). Glutamine enters into cells by its transporter Slc1A5 (2). To enter the TCA cycle, glutamine is first converted to glutamate by glutaminases (GLS), then to α-ketoglutarate, a TCA cycle intermediate, by either aminotransferases or glutamate dehydrogenases (2). Our recent study shows that tumors in vivo also use glutamine as a fuel source to replenish the TCA cycle (3). In fact, targeting glutamine metabolism has shown promising results in preclinical models in a variety of tumor types including breast, kidney, lung, and pancreatic cancers, as well as acute myeloid leukemia, where all show in vivo sensitivity to the glutaminase inhibitor CB-839 (4–12). GLS1 and GLS2 are the two genes encoding glutaminases in human (13). GLS1 is expressed predominantly in kidney and cancers, whereas GLS2 is expressed in liver (13). CB-839 is a potent inhibitor of GLS1 (4).

PIK3CA, which encodes the catalytic subunit of PI3Kα (14, 15), is mutated in a wide variety of human cancers, including ∼30% of colorectal cancers (16, 17). However, an effective approach targeting PIK3CA mutations in patients with colorectal cancer is yet to be developed (18). We recently found that PIK3CA mutations render colorectal cancer cells dependent on glutamine (3, 19, 20). This dependency is associated with the upregulation of mitochondrial glutamate pyruvate transaminase 2 (GPT2; ref. 19), which converts glutamate to α-ketoglutarate. Moreover, we demonstrated that aminooxyacetate (AOA), a pan-aminotransferase inhibitor, suppresses xenograft tumor growth of PIK3CA-mutant, but not wild-type (WT), colorectal cancer (19).

Here we show that CB-839 preferentially inhibits PIK3CA-mutant colorectal cancers. Moreover, CB-839, in combination with 5-fluorouracil (5-FU), induces tumor regression of PIK3CA-mutant colorectal cancers in multiple xenograft models. These preclinical data prompted a phase I clinical trial to assess the safety and toxicity profile of a combination of CB-839 and capecitabine (an oral prodrug of 5-FU) in advanced solid tumor patients. Consistent with our preclinical data, an exploratory analysis of the clinical trial data shows a trend that patients with colorectal cancer whose tumors harbor PIK3CA mutations may derive a greater clinical benefit from this treatment strategy as compared with those with PIK3CA-WT tumors. Together, the data suggest that targeting glutamine metabolism may be an effective approach for treating patients with PIK3CA-mutant colorectal cancer.

Colorectal cancer cell lines, HCT116, DLD1, RKO, and SW480 were obtained from ATCC. These cell lines were cultured in McCoy's 5A medium containing 10% FBS, as described previously (21). The tissue culture cell lines are routinely checked for mycoplasma contamination. The cell lines were authenticated by the Genetica DNA Laboratories using STR profiling. CB-839 was kindly provided by Calithera Biosciences. Oxaliplatin, camptothecin, and regorafenib were purchased from Selleck Chemical. 5-Fluorouracil was purchased from Sigma-Aldrich.

Animal experiments were approved by the Case Western Reserve University Animal Care and Use Committee and performed in accordance with relevant guidelines and regulations.

Subcutaneous xenografts were established as described in ref. 22. Three million cells were injected subcutaneously into the flanks of 6- to 8-week-old female athymic nude mice. When tumors reached around 250 to 400 mm³, mice were randomly divided into groups (10 tumors for each group). Mice were treated with CB-839 (200 mg/kg) once or twice daily by oral gavage, 5-FU (30 mg/kg) once a day by intraperitoneal injection, or both CB-839 and 5-FU. Tumor volume was measured with electronic calipers, and volumes were calculated by the formula of length × width²/2. For patient-derived xenograft (PDX) models, tumors were first sliced into approximately 4 mm³ pieces and implanted subcutaneously into nude mice. When the average tumor volume reached around 250 to 400 mm³, mice were treated with the indicated drugs, and tumors were measured once per week.

Cells were plated in six-well plates at 2 × 10⁵ cells per well in complete medium. After 24 hours, drugs were added into the medium. Cells were treated by the drugs for 24 hours and then collected by trypsinization. Cells were fixed in 70% ethanol overnight at 4°C and stained with propidium iodide before flow cytometry analyses as described previously (23). Sub-G₁ phase cells were classified as cell death.

Cells were plated into six-well plates. Sixteen hours later, the indicated drugs or compounds were added into the medium. After 24 hours, cells were lysed with RIPA buffer [10 mmol/L Tris (pH 7.4), 150 mmol/L NaCl, 5 mmol/L EDTA, 0.1% SDS, 1% Triton-X100, 1 mmol/L DTT, 1 mmol/L PMSF, complete Protease Inhibitor Cocktail tablet (Roche)]. Then lysates were cleared by centrifugation (14,000 rpm, 10 minutes) and protein concentration was determined by the BCA Protein Assay Kit (Pierce). Equal amounts of total protein were used for immunoblotting as described previously (24). The following antibodies were used: Cell Signaling Technology: rabbit polyclonal anti-cleaved PARP (catalog no. 9544), rabbit McAb xCT/SLC7A11(catalog no. 12691), mouse McAb NQ01(catalog no. 3187); Novus Biologicals: Mrp5 (catalog no. NBP2–46467), and Mrp8 (catalog no. NBP1–59810); Sigma-Aldrich: mouse monoclonal Anti-β-Actin (catalog no. A5441); Santa Cruz Biotechnology: Nrf2, Lamin B2, and GAPDH.

HCT116 and DLD1 PIK3CA WT or Mut cells were grown with or without glutamine for 24 hours. RNAs were extracted for microarray analyses. Input RNA was provided at 50 ng/μL and the labeling reaction was initiated with 150 ng of RNA samples for interrogation on the expression microarrays following the manufacturer's instructions. Samples were labeled robotically using the Affymetrix WT (whole transcript) labeling protocol and the Beckman Coulter Biomek FX^P Laboratory Automation Workstation; scripts were provided by Affymetrix to process up to 96 samples in batch. Samples were interrogated on the Human Gene Array 2.1 in the PEG format. Hybridization, washing, staining and data collection were carried out in the Affymetrix Gene Titan MC (multichannel) instrument. The Microarray data were deposited into GEO (accession no. GSE157024).

RNAs were extracted using either RNeasy Plus Mini Kits (Qiagen) according to the manufacturer's instructions. One microgram of each total RNA was used for reverse transcription using the Superscript III First-Strand Synthesis Kit (Thermo Fisher Scientific). The resulting cDNAs were diluted 5-fold with double-distilled water and were used as templates for reverse transcription PCR (RT-PCR). Primers used are listed below: for uridine phosphorylase 1 (UPP1) expression: forward: 5′-AACAGAGCAGGCAGTGGATA-3′; reverse: 5′-ATACGCCTGCTTGTCCTTCT-3′. For GAPDH: forward: 5′-GGAAATCCCATCACCATCT-3′; reverse: 5′-TGTCGCTGTTGAAGTCAGA-3′

RNAs were extracted using either RNeasy Plus Mini Kits (Qiagen) for cells or TRIzol (Thermo Fisher Scientific) for tumor samples according to the manufacturer's instructions. The Taqman assay system was used for qRT-PCR using UPP1 probes (Hs00370287; Applied Biosystems) with IQ super mix (catalog no. 170-8860; Bio-Rad). Expression levels of UPP1 in each sample was normalized to that of β2-microglobulin.

Reactive oxygen species (ROS) was measured by labeling cells with H2-carboxy-DCFDA (Invitrogen) according to the manufacturer's instructions. GSH/GSSG ratios were measured with a GSH/GSSG Ratio Assay Kit (Bio Vision) according to the manufacturer's instructions.

Cells were treated with the given chemicals overnight at the indicated concentrations. By the end of the treatment, cells were washed with ice-cold PBS once and then lysed in harvest buffer (10 mmol/L HEPES pH 7.9, 50 mmol/L NaCl, 0.5M sucrose, 0.1 mmol/L EDTA, 0.5% Triton X-100, and freshly added 1 mmol/L DTT). Cell lysates were incubated on ice for 5 minutes and nuclei were harvested by centrifuging at 500 × g for 5 minutes at 4°C. The supernatants were transferred to new tubes and spun down at 14,000 × g for 10 minutes. Nuclear pellets were washed twice with buffer A (10 mmol/L HEPES pH 7.9, 10 mmol/L KCl, 0.1 mmol/L EDTA, 0.1 mmol/L EGTA, and freshly added 1 mmol/L DTT), and lysed in buffer C (10 mmol/L HEPES pH 7.9, 500 mmol/L NaCl, 0.1 mmol/L EDTA, 0.1 mmol/L EGTA, 0.1% NP-40, and freshly added 1 mmol/L DTT) for 10 minutes on ice. The nuclear lysates were centrifuged at the highest speed for 10 minutes and supernatants were taken as nuclear fractions.

Two independent siRNA against Nrf2 were purchased from IDT (reference nos. hs.Ri.NFE2L2.13.1 and hs.Ri.NFE2L2.13.3). The siRNAs were transfected into HCT116 cells using Lipofectamine 2000 according to the manufacturer's instructions.

Chromatin immunoprecipitation (ChIP) assays were performed using a SimpleChIP Enzymatic Chromatin IP Kit (Cell Signaling Technologies; catalog no. 9003) and as described previously (25). Briefly, cells were cross-linked with 37% formaldehyde at a final concentration of 1% at room temperature for 10 minutes. Fragmented chromatin was treated with nuclease and subjected to sonication. ChIP was performed with an anti-NRF2 antibody (Cell Signaling Technologies, catalog no. 12721) or normal rabbit IgG (Cell Signaling Technologies, catalog no. 2729). After DNA purification, immunoprecipitated DNA was quantified by real-time PCR using iQ SYBR Green Supermix (Bio-Rad, catalog no. 1708880) with primers for UPP1 promoter (Forward: CTGGAGCATTGCGTTTGTC; Reverse: GACCCCTGGGAAGAGAGAAC); HMOX1 promoter (Cell Signaling Technologies, catalog no. 53538) and RPL30 exon 3 (Cell Signaling Technologies, catalog no. 7014). Fold enrichment was calculated on the basis of the threshold cycle (CT) value of the IgG control using the comparative CT method.

The cellular thermal shift assay (CETSA) was performed as described (26). Briefly, for tissue culture cells, cells were plated in six-well plates at 2 × 10⁵ per well. When the cells reached ∼70% confluence, they were treated with DMSO or CB-839 (5 μmol/L) overnight, and then treated with DMSO or 5-FU (1 μmol/L) for 2 hours. Cells were harvested, washed with PBS, and resuspended in PBS supplemented with protease/phosphatase inhibitors. The cell suspension was heated at 52°C for 3 minutes in a thermal cycler followed by cooling for 3 minutes at room temperature. For Western blot analyses, the cells were lysed in kinase buffer [25 mmol/L Tris-HCl (pH 7.5), 5 mmol/L β-glycerophosphate, 2 mmol/L DTT, 0.1 mmol/L Na₃VO₄, 10 mmol/L MgCl₂] by using three cycles of freeze–thaw. The lysates were centrifuged at 20,000 × g for 20 minutes at 4°C to remove the precipitates. The supernatants were analyzed by Western blot analysis using a rabbit anti-TS antibody (Proteintech, catalog no. 15047-1-AP). For AlphaLISA assay, cells were lysed in CETSA Cell Lysis Buffer 2 (Perkin Elmer, catalog no. CETSA-BUF2-100ML) and AlphaLISA was performed according to the manufacturer's instructions (Perkin Elmer). A mouse (Santa Cruz Biotechnology, catalog no. sc-376161) and the rabbit anti-TS antibodies were used for the AlphaLISA.

For tumor tissues, frozen tumor tissues were homogenized in cold PBS containing protease/phosphatase inhibitors using tissue grinders followed by three cycles of freeze–thaw. The lysate was separated from the cell debris by centrifugation at 20,000 × g for 20 minutes at 4°C. The tissue lysates were heated at 48°C for AlphaLISA.

Immunofluorescent staining was performed as described previously (27). Briefly, cells were fixed with 4% paraformaldehyde for 15 minutes at room temperature in a multiwell plate. The plate was incubated with blocking buffer (1× PBS, 5% normal serum, and 0.3% Triton X-100) for 1 hour and then incubated overnight with phospho-histone H2A.X antibody (Cell Signaling Technology, catalog no. 80312) in antibody dilution buffer (1× PBS, 1% BSA, and 0.3% Triton X-100) at 4°C. After being washed three times with PBS, the cells were incubated with a secondary Alexa Fluor 488-conjugated antibody. The nuclei were labeled with 4′,6-diamidino-2-phenylindole (DAPI; Sigma, catalog no. D9542).

UPP1 knockout clones were generated using the CRISPR/Cas9 system. Two different guide RNAs were used to target the UPP1 gene. sgRNA 1 (5′-TGCCTCAGTTGGCGGAATGG) targets exon 3 of UPP1 genomic sequence and sgRNA2 (5′-TCAGGGCAAGGCCGTCTGGA) targets exon 8. A targeting plasmid expressing sgRNA1 or sgRNA2 and CAS9 protein was transfected into HCT116 cells. Single clones were picked up to screen for targeted clones by genomic PCR. Knockout clones were verified by sequencing both genomic DNA and RNA.

Cell pellets (approximately 10 million cells) frozen at −70°C were thawed at room temperature for 15 minutes. Then 500 μL of 80% methanol–water was added and vortexed for 5 minutes to resuspend the pellet, followed by incubation at −70°C for 30 minutes. This was repeated two times, and finally, the suspension was sonicated for 10 minutes to completely break the cells and release their contents. The suspension was centrifuged at 13,000 × g for 10 minutes. The supernatant was separated and dried using N2 at 35°C for 20 minutes. Finally, the residue was reconstituted in 100 μL water and subjected to LC/MS-MS analysis. For LC/MS-MS analysis, 2′-deoxyadenosine-13C10,15N5 5′-triphosphate (Sigma-Aldrich) was used as internal standard. The separation of analytes was achieved by a Shimadzu LC-20AD HPLC system with a Shimadzu SIL-20AC autosampler (Shimadzu) on a YMC-ODS-AM column (2.1 × 100 mm, 3.0 μm) using gradient mode of elution with mobile phase A: 5 mmol/L N,N-dimethyl hexylamine in water, 10 mmol/L ammonium bicarbonate, and B: 50:50 acetonitrile:water, 5 mmol/L N,N-dimethyl hexylamine, 10 mmol/L ammonium bicarbonate at a flow rate of 0.200 mL/min. The column was eluted by 10% B mobile phase for 1 minute, then from 10% to 25% B for 4 minutes, 25% B for 15 minutes, from 25% to 55% B for 1 minute, 55% B for 7 minutes, from 55% to 10% B for 1 minute, and then 10% B for 9 minutes. Quantitation of the analytes was accomplished by an AB Sciex API 3200 triple quadruple tandem mass spectrometer (AB Sciex), which was operated in the negative multiple-reaction-monitoring mode with mass transition 481.1 > 159.0 for dTTP.

This clinical trial was conducted with the approval of the institutional review board and according to good clinical practice with a primary objective of determining the MTD and dose-limiting toxicities (DLT) of oral CB-839 (provided by Calithera Biosciences) when administered with oral capecitabine. Dosing was based on a standard 3+3 dose escalation schedule. Response to therapy was assessed per RECIST 1.1 utilizing CT imaging obtained every 9 weeks. Patients were permitted to continue treatment until disease progression or development of unacceptable toxicity (Supplementary Fig. S6A). All patients provided written informed consent prior to participating in the study. The trial was registered in clinicaltrials.gov (NCT02861300).

Patients were eligible for study entry if they had an advanced solid tumor and had progressed on all standard lines of therapy, or if capecitabine was an appropriate treatment option for them. Patients with colorectal cancer specifically must have progressed on prior fluoropyrimidine-based chemotherapy. Patients otherwise must have been at least 18 years of age, had an ECOG performance status of 0 or 1, had normal bone marrow, renal, and hepatic function, had the ability to swallow pills and be able to understand and the willingness to sign consent. Patients were not eligible if they had ongoing treatment-related toxicities that were greater than grade 1 and they could not be receiving other investigational agents. Central nervous system involvement by their cancer or prior allergic reaction to either CB-839 or capecitabine was not permitted.

Toxicity was assessed utilizing the Common Terminology Criteria for Adverse Events (CTCAE) version 4.0. To be evaluable for a DLT, patients must have received at least 75% of the planned CB-839 and capecitabine during the first treatment cycle (21 days). A DLT was defined as any ≥ grade 3 nonhematologic toxicity thought to be treatment-related, grade 3 febrile neutropenia, ≥ grade 3 thrombocytopenia associated with ≥ grade 3 bleeding, grade 4 neutropenia or grade 4 thrombocytopenia occurring during the 21-day DLT window. At dose level four, one patient was replaced for not receiving enough study drug to be evaluable.

GraphPad Prism software was used to create the graphs. Data are plotted as mean ± SEM. We applied the t test to compare the means between two groups, assuming unequal variances. For xenograft growth, we carried out ANOVA for repeated measurements to test whether there is an overall difference in the tumor sizes by testing group differences as well as whether there was a difference in the development of tumor sizes over time between the two groups by testing the interaction between time and group. Kaplan–Meier analysis was used to assess differences in progression-free survival (PFS) stratified by PI3KCA mutation status for patients with colorectal cancer only, generating a log-rank P value as well as median survival time with 95% confidence intervals (95% CI). Patient response to therapy during the phase I clinical trial was defined per RECIST criteria. Patients were considered evaluable for response if they had measurable disease at the time of study entry, had received at least one cycle of therapy and had undergone a repeat disease evaluation with imaging. PFS was defined as the time from the beginning of treatment to RECIST evidence of progressive disease as determined by radiography or by clinical progression.

We have showed previously that PIK3CA mutations render colorectal cancers dependent on glutamine (19) and that CB-839 targets glutaminase in cultured colorectal cancer cells and xenograft tumors (28). Thus, we set out to determine the sensitivity of the isogenic PIK3CA mutant-only (Mut, with the WT allele being knocked out) and WT-only (WT, with the mutant allele being knocked out) colorectal cancer cell line pairs to CB-839 (Supplementary Fig. S1A). As shown in Supplementary Figs. S1B and S1C, HCT116 and DLD1 PIK3CA Mut cell lines are more sensitive to CB-839 than the corresponding isogenic PIK3CA WT cell lines. Consistent with our previous observation that KRAS mutations have no impact on glutamine dependency of colorectal cancer cells (19), isogenic HCT116 KRAS WT and mutant cell lines had similar sensitivity to CB-839 (Supplementary Figs. S1D and S1E). We next tested if CB-839 preferentially suppresses xenograft tumor growth of PIK3CA-mutant cells. As shown in Fig. 1A, CB-839 significantly inhibited xenograft tumor growth of HCT116 cells, which harbor a PIK3CA mutation, regardless of once or twice daily drug exposure. Moreover, CB-839 treatment also significantly inhibited tumor growth of a colon cancer PDX harboring a PIK3CA mutation (Fig. 1B). In contrast, CB-839 had no effect on PIK3CA-WT xenograft tumors established from SW480 colorectal cancer cells or a PIK3CA-WT colorectal cancer PDX model (Fig. 1C and D). Taken together, our data demonstrate that CB-839 preferentially inhibits the growth of PIK3CA-mutant, but not WT, colorectal cancers.

In an attempt to identify clinically useful drugs that might enhance the tumor inhibitory effect of CB-839 against colorectal cancer, we tested combinations of CB-839 with 5-FU, camptothecin, oxaliplatin, and regorafenib in HCT116 cells. As shown in Fig. 2A and Supplementary Fig. S2A, an additive effect on apoptosis, was only observed when CB-839 was combined with 5-FU.

We then went on to test whether 5-FU enhances the efficacy of CB-839 on PIK3CA-mutant colorectal cancers in subcutaneous xenograft models. As shown in Fig. 2B to D, the combination of CB-839 and 5-FU induced regression of xenograft tumors established from HCT116 and RKO (another PIK3CA-mutant colorectal cancer cell line). Although CB-839 or 5-FU alone inhibited tumor growth to various extents, neither induced tumor regression (Fig. 2C and D). Moreover, the CB-839 plus 5-FU combination induced tumor regression of a PIK3CA-mutant colon cancer PDX that was derived from a metastatic liver lesion of a patient (Fig. 2E). Notably, a third of the initial tumor lesions in each of these xenograft models appeared to be completely eliminated by the drug combination and no evidence of tumor regrowth was seen in these mice one month after the treatment was stopped. Finally, although the drug combination did not induce regression of DLD1 xenograft tumors (also harboring a PIK3CA mutation), it potently inhibited the progressive growth of these tumors (Fig. 2F). To further ascertain that the drug combination preferentially induced tumor regression in PIK3CA-mutant tumors, we treated xenograft tumors established from the isogenic HCT116 PIK3CA Mut and WT cell lines. As shown in Supplementary Fig. S2B, the CB-839 and 5-FU combination induced regression of the HCT116 PIK3CA-mutant, but not the PIK3CA WT, tumors. Together, the data suggest the combination of CB-839 and 5-FU might be an effective treatment for PIK3CA-mutant colorectal cancers.

To determine the mechanism by which CB-839 enhances the tumor inhibitory effect of 5-FU in PIK3CA-mutant colorectal cancers, we profiled gene expression of the isogenic PIK3CA Mut and WT colorectal cancer cell lines. Given that CB-839 blocks glutamine utilization and that it mirrors glutamine deprivation, we performed microarray analyses of HCT116 and DLD1 PIK3CA Mut and WT cell lines cultured without glutamine for 24 hours. Figure 3A shows the volcano plots of gene expression changes in PIK3CA Mut cells versus WT cells. Gene ontology (GO) pathway analysis showed that genes involved in chromosome structure, cell cycle, DNA damage repair, thymidine metabolism, and others were enriched (Supplementary Fig. S3A). Interestingly, only 12 genes, including UPP1, were commonly upregulated in both HCT116 and DLD1 PIK3CA-mutant cell lines (Supplementary Fig. S3B). We turned our attention to UPP1 gene for the following reasons: (i) UPP1 gene expression is ranked as the second-highest upregulated gene in both HCT116 and DLD1 PIK3CA Mut cells versus PIK3CA WT cells; (ii) UPP1 facilitates the conversion of 5-FU to FdUTP (29); and (iii) PIK3CA^−/− ES cells are 10-fold more resistant to 5-FU (30). qRT-PCR analyses showed that both glutamine deprivation and CB-839 treatment-induced highest level UPP1 in PIK3CA Mut cell lines (Fig. 3B and C), although PIK3CA Mut cell lines had higher basal levels of UPP1 than their WT counterparts (Fig. 3C). Moreover, CB-839 treatment also induced UPP1 gene expression in HCT116 xenograft tumors (Fig. 3D). It is worth noting that there is no difference in UPP1 mRNA levels between colorectal cancers and normal colon tissues (Supplementary Fig. S3C).

To interrogate how CB-839 treatment induces UPP1expression in PIK3CA-mutant cells, we turned our attention to Nrf2 for the following reasons: (i) UPP1 was shown to be a transcriptional target of Nrf2 (31); (ii) Nrf2 is a latent transcription factor that is activated by oxidative stress through protein stabilization and nuclear translocation (32); and (iii) glutamine starvation increases ROS (33). Thus, we postulated that CB-839 treatment might increase ROS levels, which in turn activate Nrf2 and increase UPP1 transcription. Indeed, CB-839 treatment significantly increased ROS levels in PIK3CA WT and Mut cells derived from both HCT116 and DLD1 (Fig. 4A), although the PIK3CA Mut cells had higher basal levels of ROS than PIK3CA WT cells (Fig. 4A). Consistently, CB-839 treatment reduced the ratio of reduced versus oxidized forms of glutathione in the HCT116 and DLD1 cell lines (Fig. 4B). As expected, CB839 treatment increased the levels of nuclear Nrf2 in the HCT116 and DLD1 PIK3CA WT and PIK3CA Mut cell lines, but the nuclear levels of Nrf2 are higher in the PIK3CA Mut cell lines treated with CB-839 than in the PIK3CA WT cell lines (Fig. 4C). Consistent with a report that the PI3K–AKT pathway regulates Nrf2 nuclear translocation (34), Supplementary Fig. S4A shows that nuclear Nrf2 levels were markedly decreased in p110α lipid kinase-dead DLD1 mutant cells that we generated previously (19). To validate that Nrf2 modulates UPP1 gene expression, we knocked down Nrf2 with two independent siRNAs. As shown in Fig. 4D, knockdown of Nrf2 reduced CB-839-induced UPP1 gene expression, although knockdown of Nrf2 also decreased basal levels of UPP1 (Fig. 4D). Conversely, overexpression of Nrf2 increased UPP1 expression (Fig. 4E). Moreover, ChIP-qPCR analyses demonstrated that Nrf2 bound to a promoter region of UPP1 (Fig. 4F). Although NRF2 mutations are rare in colorectal cancers, an analysis of the pan-cancer The Cancer Genome Atlas data showed that NRF2 mutant colorectal cancers expressed significantly higher levels of UPP1 than NRF2 WT colorectal cancers (Fig. 4G). Taken together, these data suggest that CB-839 treatment induces higher levels of ROS in PIK3CA mutant cells, which in turn increases nuclear Nrf2 levels and thus upregulates UPP1 mRNA expression.

It has been reported that the cystine/glutamate antiporter xCT/SLC7A11, an Nrf2 transcriptional target (35), drives glutamine dependency in lung cancer (36). We thus examined protein levels of SLC7A11 and other Nrf2 targets in the isogenic HCT116 and DLD1 PIK3CA Mut and WT cell line pairs. SLC7A11 protein levels were not upregulated in the PIK3CA Mut cell lines compared with its isogenic WT counterparts (Supplementary Fig. S4B). Moreover, an xCT inhibitor, sulfasalazine, did not suppress the growth of HCT116 colorectal cancer cells (Supplementary Fig. S4C) even at 100 μmol/L concentration. In contrast, CB-839 markedly inhibited HCT116 cell growth at low μmol/L concentrations (Supplementary Fig. S4C). The data suggest that differential sensitivity of PIK3CA-mutant versus WT colorectal cancers to CB-839 is not due to the cystine/glutamate antiporter xCT/SLC7A11.

Multidrug resistance-associated protein (Mrp) transporters, including Mrp3 and Mrp4, are reported as the transcriptional targets of Nrf2 (37), whereas an upregulation of Mrp5 or Mrp8 is known to cause 5-FU resistance in cancer cells (38, 39). We thus set out to determine if CB-839 treatment caused changes in Mrp5 and Mrp8, thereby modulating 5-FU sensitivity. As shown in Supplementary Fig. S4D, CB-839 treatment had no impact on Mrp5 and Mrp8 protein levels. The data suggest CB-839 treatment is unlikely to affect the export of 5-FU or its metabolites.

UPP1 catalyzes the reversible phospholytic cleavage of uridine and deoxyuridine to uracil and ribose- or deoxyribose-1-phosphate, thereby facilitating the conversion of 5-FU to FdUMP (29). We thus used a cellular thermal shift assay (40) to test if CB-839 treatment enhanced engagement of FdUMP with its target thymidylate synthase (TS), a major mechanism of 5-FU cytotoxicity (41). As shown in Fig. 5A and B, CB-839 treatment significantly increased the binding of TS to FdUMP in HCT116 cells in culture. Moreover, CB-839 treatment dramatically increased the binding of TS to FdUMP in xenograft tumors established from both HCT116 and DLD1 cells (Fig. 5C–F). Given that TS converts dUMP to dTMP to produce cellular dTTP, we set out to determine how the combination of CB-839 and 5-FU impacts cellular dTTP levels. As shown in Fig. 5G, either drug alone significantly reduced amounts of dTTP. However, the combination of CB-839 and 5-FU resulted in near-complete depletion of cellular dTTP (Fig. 5G). Because depletion of cellular dTTP can cause DNA damage (42), we determined if the combination of CB-839 and 5-FU induced more DNA damages than the single drug alone by staining drug-treated cells with an anti-p-H2A.X antibody. As shown in Fig. 5H and Supplementary Fig. S5A, the combination of CB-839 and 5-FU induced significantly more p-H2A.X positive cells than the single drug alone. Given that incorporation of 5-FU into DNA and RNA is a potential mechanism for the tumor inhibitory effect of 5-FU, we set out to determine if CB-839 treatment increased 5-FU incorporation into DNA and RNA. As shown in Supplementary Fig. S5B, CB-839 treatment actually decreased 5-FU incorporation into DNA and had no impact on 5-FU incorporation into RNA (Supplementary Fig. S5C), thereby ruling out those as the mechanisms by which CB-839 enhances 5-FU activity. Although we attempted to determine if the combination of CB-839 and 5-FU depletes dTTP levels in xenograft tumors, technical limitations of our mass spectrometry method did not allow this analysis on tumor samples. Together, these data suggest that CB-839 treatment increases ROS level, induces nuclear translocation of Nrf2, and upregulates UPP1 gene expression. This in turn facilitates the conversion of 5-FU to FdUMP and enhances inhibition of TS, thereby leading to depletion of cellular dTTP (Fig. 5I).

To test whether UPP1 plays a critical role in the enhancement of 5-FU activity by CB-839, we set out to knockout (KO) UPP1 in HCT116 cells with CRISPR/Cas9 genome editing technology by two different strategies (Fig. 6A): (i) targeting the starting ATG codon in exon 3 and (ii) targeting the substrate-binding site (amino acids Q217 and R219) encoded by exon 8. Although the two UPP1 KO cell lines grew as well as the parental cells in tissue culture, they grew significantly slower as xenograft tumors (Fig. 6B). Importantly, both UPP1 knockouts abolished the additive tumor inhibitory effect of CB-839 and 5-FU (Fig. 6C). Surprisingly, the UPP1 KO cell lines were also insensitive to CB-839 treatment alone (Fig. 6C). Taken together, our data suggest that the upregulation of UPP1 by CB-839 treatment plays a key role in enhancing the tumor inhibitory effect of 5-FU.

On the basis of our preclinical data, we conducted a phase I clinical trial to assess the safety and toxicity profile of CB-839 and capecitabine in humans. Specifically, the objective of the study was to determine the MTD and any DLTs of the combination of CB-839 and capecitabine in patients with advanced solid tumors.

Eligible patients were treated with CB-839 by mouth twice daily continuously and with capecitabine by mouth twice daily on days 1 to 14 of a 21-day treatment cycle (see study design and eligibility criteria in the materials and methods section). Dosing was based on a standard 3+3 dose escalation schedule with CB-839 ranging from 400 mg by mouth twice daily to 800 mg by mouth twice daily, as a previous phase I clinical trial shows that CB-839 as a single agent is well tolerated with a biologically active dose of 600 mg orally twice daily (43). Capecitabine dosing ranged from 750 mg/m² twice daily (on days 1–14) to 1,000 mg/m² twice daily (on days 1–14), a dose level commonly used against a variety of cancers (Fig. 7A; Supplementary Fig. S6A).

A total of 16 patients were enrolled in the study. Patient demographics are presented in Supplementary Table S1. Although patients with any solid tumor were permitted, 12 of 16 patients had a colorectal or appendiceal primary malignancy. The median age of participants was 68.5 years (range: 42–79) and the mean number of treatment cycles received was 5.6 (range: 1–17). Although not required for study participation, among the colorectal cancer/appendiceal patients, 7 were noted to have a PIK3CA mutation, 9 harbored a KRAS mutation, and 5 had both PIK3CA and KRAS mutations (Supplementary Table S2).

Toxicity was assessed utilizing the CTCAE version 4.0 and toxicities for patients on the study are shown in Supplementary Table S3. No DLTs were observed for any patient on the trial and the fourth and final dose level of CB-839 800 mg by mouth twice daily continuously and capecitabine 1,000 mg/m² orally twice daily on days 1 to 14 of a 21-day treatment cycle was recommended to be used in further phase II trials of this drug combination.

Response to therapy was assessed per RECIST 1.1 with CT imaging obtained every 9 weeks (Supplementary Fig. S6A). Of the 16 enrolled patients, 14 were evaluable for response. There were no complete responses or partial responses (Supplementary Fig. S6B). Ten and four patients had stable disease and progressive disease as their best response, respectively. Time-on-treatment (time from initiation of study treatment to coming off study) for each patient is shown in Fig. 7A. The median time on treatment for all patients was 17.9 weeks. For patients with PIK3CA-mutant colorectal cancer, the mean time-on-treatment was 19.9 weeks versus 16 weeks for patients with wild-type PIK3CA colorectal cancer. PFS among all patients was 18.75 weeks. In an exploratory Kaplan–Meier analysis (Fig. 7B), median PFS for patients with colorectal cancer with PIK3CA-mutant tumors was 24.8 (95% CI, 18.8–NA) weeks versus 16 (95% CI, 9.1–NA) weeks for patients with PIK3CA-WT colorectal cancer (log rank P-value = 0.196). However, the median PFSs of the patients on first-line fluoropyrimidine-based chemotherapy were similar between the two groups (Supplementary Fig. S6C). Note that median time-on-treatment is reduced in the PIK3CA-mutant group due to inclusion in the time on treatment analysis of one patient who elected to terminate the study after 4 weeks for reasons unrelated to disease progression, and who is thus censored in the analysis of PFS. Because KRAS mutations often co-occur with PIK3CA mutations, we performed Kaplan–Meier analysis according to KRAS mutation status. As shown in Supplementary Fig. S6D, the median PFS for the patients with colorectal cancer with KRAS-mutant colorectal cancer was 19 weeks versus 30 weeks for patients with KRAS-WT colorectal cancer (log rank P-value = 0.8), suggesting KRAS mutations do not sensitize colorectal cancer to the combinational drug treatment.

Our data show that the combination of CB-839 and 5-FU induced tumor regression in three different PIK3CA-mutant colorectal cancer xenograft models. One-third of tumors in these models were cured. These results suggest the hypothesis that the drug combination may be an effective treatment for patients with colorectal cancer whose tumors harbor PIK3CA mutations. This is further supported by the results of our initial phase I trial, where an exploratory subgroup analysis of PIK3CA-mutant patients with colorectal cancer suggests a trend of clinical benefit as compared with PIK3CA-WT patients (time-on-treatment of 19.9 weeks vs. 16 weeks and exploratory PFS of 24.8 weeks vs. 16 weeks). Given that previous studies demonstrated that PIK3CA mutant patients with colorectal cancer have either worse survival than or no difference from PIK3CA WT patients with colorectal cancer after first-line treatment (5-FU-based therapy), the observed suggestive clinical benefit of PIK3CA-mutant patients with colorectal cancer is unlikely to be due to differential response to 5-FU between PIK3CA mutant and WT patients (44, 45). A limitation of this phase 1 study is the small size of the cohort that assessed only 12 patients with colorectal/appendiceal cancer in total and treated most patients at lower doses of the capecitabine/CB-839 combination. However, the promising nature of the results warrant our now ongoing phase II trial to more fully evaluate the activity of this approach, specifically in PIK3CA-mutant patients with colorectal cancer treated at maximal drug levels (NCT02861300).

Mechanistically, CB-839 treatment upregulates the expression of UPP1, an enzyme facilitating conversion of 5-FU to active compound FdUMP, which in turn enhances the tumor inhibitory effect of 5-FU. We provide here three lines of evidence to support this notion: (i) CB-839 enhances the binding of FdUMP to TS in both culture cells and in xenograft tumors (Fig. 5); (ii) knockout of UPP1 abrogates the tumor inhibitory effect of combined CB-839 and 5-FU (Fig. 6); and (iii) the combination of CB-839 and 5-FU depletes cellular dTTP, a major product of TS (Fig. 5). It is worth noting that knockout of UPP1 slows down xenograft tumor growth of HCT116 cells, which may be due to the role of UPP1 in the salvage nucleotide synthesis pathway (41). Moreover, knockout of UPP1 does not have a significant impact on 5-FU treatment (Fig. 6C), which may be because 5-FU can still be converted to FdUPM by thymidine phosphorylase (Fig. 5I). Interestingly, UPP1 knockout also abrogates the effect of CB-839 (Fig. 6C). We postulate that knockout of UPP1 may activate a pathway(s) that compensates the impact of glutaminase inhibition. Nonetheless, our data are consistent with previous studies showing that UPP1 knockout mouse embryonic stem cells are resistant to 5-FU (30), whereas overexpression of UPP1 enhances the cytotoxicity of 5-FU (46). Our studies reveal that CB-839 treatment increases ROS levels, which lead to nuclear translocation of Nrf2 and increased transcription of UPP1 (Fig. 4). Here we focused the studies on the tumor-intrinsic effect of the combination of CB-839 and 5-FU. However, it is possible that the drug combination may also impact tumor microenvironments, especially the immune cells, as well, which warrants further investigation.

Finally, our study provides further evidence supporting the notion that PIK3CA-mutant colorectal cancers are more dependent on glutamine. Our previous isotope-tracing studies in cultured cells and xenograft tumors showed that PIK3CA-mutant colorectal cancers utilized more glutamine to replenish the TCA cycle than the isogenic PIKCA WT colorectal cancers (3, 19). However, tumors established from the isogenic PIK3CA-mutant and WT HCT116 cell lines use similar amounts of glucose to fuel the TCA cycle (3). We previously reported that PIK3CA mutations render colorectal cancers dependent on glutamine through the upregulation of GPT2 (19). We further demonstrated that aminooxyacetate, which blocks the second step of glutamine metabolism, suppresses xenograft tumor growth of PIK3CA-mutant, but not WT, colorectal cancer (19). We show here that blocking the first step of glutamine metabolism by CB-839 also preferentially inhibits xenograft tumor growth of PIK3CA-mutant colorectal cancers. This is again consistent with what we observed in patients during our clinical trial. Interestingly, loss of PTEN, the enzyme that catalyzes the opposite reaction of PIK3CA/p110α, also renders breast cancers dependent on glutamine (47). These observations indicate that the PI3K pathway plays a critical role in modulating glutamine metabolism in certain cancer types and support proceeding to phase II human trials for further evaluation of glutamine metabolism as a therapeutic target for human PIK3CA-mutant cancers.

J.S. Barnholtz-Sloan reports grants from NIH during the conduct of the study. S. Vinayak reports other funding from Oncosec (direct funding to institution for clinical trial), Pfizer (direct funding to institution for clinical trial), and Seattle Genetics (direct funding to institution for clinical trial) outside the submitted work. N.J. Meropol reports grants from NIH during the conduct of the study, other compensation from Flatiron Health (employment, equity interest) and Roche (equity ownership) outside the submitted work, and a patent for US Patent Office 20020031515 issued (Methods of therapy for cancers characterized by over expression of the HER2 receptor protein; expires May 14, 2021). J.R. Eads reports grants from NIH and Stand Up 2 Cancer during the conduct of the study; personal fees from Pfizer (honoraria for giving a talk), Ipsen (advisory board), Lexicon (consulting), Novartis (advisory board), Advanced Accelerator Applications (advisory board), grants from Incyte (research funding), and personal fees from Bristol Meyers Squibb (husband is an employee) outside the submitted work. Z. Wang reports grants from NCI and SU2C during the conduct of the study and a patent for 10221459 issued. No potential conflicts of interest were disclosed by the other authors.

Y. Zhao: Conceptualization, data curation, investigation, writing-original draft, writing-review and editing. X. Feng: Data curation, investigation. Y. Chen: Data curation, investigation, methodology. J.E. Selfridge: Data curation. S. Gorityala: Investigation. Z. Du: Investigation. J.M. Wang: Data curation, formal analysis. Y. Hao: Investigation. G. Cioffi: Formal analysis. R.A. Conlon: Resources, writing-review and editing. J.S. Barnholtz-Sloan: Formal analysis. J. Saltzman: Recruited patients. S.S. Krishnamurthi: Recruited patients. S. Vinayak: Recruited patients. M. Veigl: Data curation, investigation. Y. Xu: Data curation, investigation. D.L. Bajor: Recruited patients. S.D. Markowitz: Conceptualization, resources. N.J. Meropol: Conceptualization, writing-review and editing, recruited patients. J.R. Eads: Conceptualization, data curation, supervision, writing-original draft, project administration, writing-review and editing, recruited patients. Z. Wang: Conceptualization, formal analysis, supervision, funding acquisition, writing-original draft, project administration, writing-review and editing.

The authors would like to thank Drs. Lewis Cantley, John Pink, and Robert B. Diasio for suggestions and discussions. This work was supported by NIH grants R01CA196643, UH2CA223670, P50CA150964, and P30 CA043703. This work was also supported by a Stand Up to Cancer Colorectal Cancer Dream Team Translational Research Grant (Grant No. SU2C-AACR-DT22-17). Stand Up to Cancer is a division of the Entertainment Industry Foundation. Research grants are administered by the American Association for Cancer Research, a scientific partner of SU2C. This research was further supported by the Gene Expression and Genotyping Facility of the Case Comprehensive Cancer Center (P30 CA043703). Calithera Biosciences provided CB-839 for preclinical and clinical studies as well as financial support for patient enrollment.

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.