Characterization and therapeutic suppression of KEAP1-NRF2-driven resistance to KRAS inhibitors in pancreatic and lung cancer

Abstract

The recent approval of KRAS inhibitors supports the therapeutic value of targeting mutant KRAS cancers. However, clinical efficacy is hindered by both primary and treatment-associated acquired resistance. We applied a CRISPR-Cas9 loss-of-function screen and identified loss of KEAP1 as a resistance mechanism to the KRAS^G12D-selective inhibitor MRTX1133 and the RAS(ON) multi-selective inhibitor RMC-7977 in pancreatic cancer models. RNA-sequencing analyses revealed a KEAP1^KO transcriptome that is distinct from the ERK-, MYC-, and YAP/TAZ-TEAD-dependent transcriptional programs that drive KRAS inhibitor resistance, demonstrating a distinct mechanism of resistance. We then established a PDAC KEAP1-deficient (PKD) gene signature that was enriched in patients and preclinical models insensitive to KRAS inhibitor treatment. Finally, we observed that KEAP1-deficient cells exhibited elevated glutamine metabolism, and combination treatment with the glutamine antagonist DRP-104 (sirpiglenastat) enhanced KRAS inhibitor suppression of pancreatic and lung tumors.

INTRODUCTION

Mutational activation of KRAS is associated with ~20% of all cancers, with the highest frequencies found in pancreatic ductal adenocarcinoma (95%; PDAC), and prevalent in lung adenocarcinoma (37%; LUAD) (1,2). Although KRAS was once considered undruggable (3), the recent approval of two KRAS inhibitors (sotorasib and adagrasib) selective for the glycine-12-cysteine (G12C) smoking associated mutation for KRAS^G12C-mutant non-small cell lung cancer (NSCLC) and colorectal cancer (CRC) has shattered that perception (4–7).

Following the successful development of KRAS^G12C-selective inhibitors, many inhibitors have been developed and entered clinical evaluation that target essentially all KRAS mutant alleles (3,8). Similar to the approved KRAS^G12C inhibitors, many are mutation-selective, especially for the most prevalent KRAS mutation in cancer, KRAS^G12D (9,10). Moreover, the first-in-class RAS(ON) multi-selective inhibitor daraxonrasib (RMC-6236) has shown excellent activity as second-line treatment in KRAS-mutant PDAC, comparable to responses to first-line chemotherapy but without the severe toxicity seen with standard-of-care cytotoxic drugs (11,12).

Despite the impressive clinical activities seen with KRAS inhibitors, depth and durability of response limit their long-term efficacy (3,13). Clinical evaluation of KRAS^G12C inhibitors in KRAS-mutant NSCLC, CRC, and PDAC patients have reported median overall response rates (ORR) generally below 50% with the duration of response (DOR) typically under 12 months, underscoring primary/innate resistance and the rapid emergence of acquired resistance, respectively (3). Acquired resistance also limits the long-term efficacy of daraxonrasib (14,15). Thus, a better understanding of the mechanisms underlying resistance to KRAS inhibitors is essential for developing more effective combination therapeutic strategies.

Evaluation of DNA alterations identified putative mechanisms of acquired resistance in only ~50% of patients who relapsed on KRAS^G12C and pan-RAS inhibitors, and most (>90%) mutations are in components in the RAS signaling network. Non-genetic transcriptional mechanisms have also been identified in preclinical studies (3). For example, prolonged treatment with KRAS inhibitors was shown to activate compensatory pathways, including TEAD and MYC transcription factors as well as the RAF-MEK-ERK signaling network, thereby promoting tumor survival and drug resistance (16–22).

The dysregulation of the KEAP1-NRF2 pathway has emerged to be a major factor limiting the efficacy of KRAS^G12C inhibitors (14,16,17,23). KEAP1, a component of Cullin 3-E3 ubiquitin ligase complex, regulates the expression of the NRF2 transcription factor by targeting it for ubiquitin-dependent proteasomal degradation (24–26). Under oxidative stress, KEAP1 undergoes conformational changes, leading to NRF2 stabilization and subsequent transcriptional activation of antioxidant and cell survival genes. In cancer cells, oncogenic mutations on KEAP1 impair its regulatory ability, causing elevated expressions of NRF2 and NRF2-regulated gene transcription (24,27). It has been widely reported that increased expression of NRF2 causes resistance to therapies against malignant tumors (28–31).

Although KEAP1 mutations are found in ~15% of NSCLC patients, it is present in only about 2% of PDAC patients. However, KEAP1 truncations were detected in tumors from PDAC patients who relapsed from daraxonrasib treatment (14). Moreover, KEAP1 inhibition is frequently found in malignant pancreatic cells through transcriptional or post-translational deregulation (31–33), and NRF2 overexpression has been observed as PDAC tumors progress to more aggressive stages (29,31). In a murine model of PDAC tumorigenesis, activation of Kras, Braf, or Myc induced Nrf2 expression, and genetic silencing of Nrf2 impaired tumorigenic growth (34). In human PDAC cells, activated NRF2 promoted chemoresistance to gemcitabine (29). Collectively, these findings suggest that NRF2 plays an important role in PDAC tumor progression and regulating its sensitivity to treatments. However, the detailed mechanism regarding how the KEAP1-NRF2 pathway affects sensitivity to KRAS inhibition remains to be determined.

In this study, we explored the role of the KEAP1-NRF2 signaling pathway in mediating resistance to KRAS inhibition in KRAS-mutant PDAC and LUAD. We defined a KEAP1^KO-dependent gene signature in PDAC and validated the association of this signature with reduced KRAS inhibitor sensitivity across PDAC and LUAD cancer cell lines, patient-derived xenograft (PDX) tumors, and patient clinical samples. Finally, we exploited the increased glutamine dependency caused by KEAP1 loss and determined that concurrent pharmacologic inhibition of glutamine metabolism enhanced KRAS inhibitor activity in KRAS-mutant PDAC and LUAD.

RESULTS

Loss-of-function CRISPR Screen Identified the KEAP1-NRF2 Pathway in Promotion of Resistance to RAS Inhibition

To identify mechanisms of resistance to inhibitors of mutant KRAS, we focused our study on the most prevalent KRAS mutation, G12D. We first evaluated the anti-tumor activity of two inhibitors, the pan-RAS inhibitor RMC-7977 (analog of daraxonrasib) and KRAS^G12D-selective inhibitor MRTX1133 (hereinafter referred to as RASi and G12Di, respectively) (9,19). We previously evaluated RASi specificity in a panel of matched mouse embryo fibroblasts (MEFs) with loss of all three Ras genes (RASless MEFs) (35), where proliferative capacity was restored with ectopic expression of human wild-type or mutationally activated KRAS or BRAF mutant genes (36). RASi exhibited comparable potent inhibition of KRAS effector signaling for all KRAS mutants evaluated as measured by inhibition of the phosphorylation (pERK) and activation of ERK (19), the key KRAS effector in PDAC. In contrast, G12Di exhibited potent inhibition of pERK in RASless MEFs expressing KRAS^G12D at 10 nM whereas pERK inhibition in KRAS WT cells was seen only at 500 nM (Supplementary Fig. S1A). An approximately 10-fold lower potency was seen against KRAS^G13D, 50-fold lower activity against KRAS^Q61H and limited to no activity was detected against other KRAS mutants. No activity was seen in BRAF^V600E cells, supporting the KRAS-selectivity of ERK inhibition.

We then characterized the sensitivity of a panel of KRAS^G12D-mutant PDAC cell lines (SW1990, Pa16C, Pa14C, and PANC-1) to RASi and G12Di. Both effectively reduced the levels of pERK as well as the activities of two ERK substrates, the phosphorylation of RSK (pRSK) and expression levels of total MYC, at comparably low nanomolar concentrations (Supplementary Fig. S1B and S1C).

To elucidate the mechanisms underlying resistance to KRAS inhibitors in PDAC cells, we carried out a loss-of-function CRISPR-Cas9 screen using a focused library targeting components of major oncogenic signaling networks (37) (Fig. 1A). We determined enrichment of sgRNAs targeting multiple genes encoding tumor suppressors, several of which were previously associated with acquired resistance to KRAS^G12C inhibition in patients (e.g., PTEN, APC, and RB1) (16,17,22,38) (Fig. 1B). STK11 co-mutation was associated with early disease progression and shorter survival in NSCLC patients treated with G12C inhibitors (23,39). Conversely, we found depletion of sgRNAs targeting genes with gain-of-function mutations identified in patients who relapsed on KRAS^G12C-selective inhibitor (G12Ci) treatment (e.g., EGFR, ERBB2, BRAF, MYC, FGFR2 and PIK3CA (16,17,22,38). The identification of genes associated with relapsed patients support the validity of the screen to detect bona fide mechanisms of resistance.

Figure 1.

CRISPR screen on PANC-1 PDAC cell treated with G12Di identified the KEAP1-NRF2 pathway promoting resistance to KRAS inhibition. A, The loss-of-function CRISPR screen was performed using the druggable genome sgRNA library. The sgRNA library was transduced to cells by lentivirus and propagated growth medium supplemented with G12Di or vehicle (DMSO) control, for 4 weeks. The genomic DNA was isolated and sequenced by next generation sequencing (NGS). The NGS results were further analyzed by MAGeCK to determine the Beta scores for each gene. B, The snake plot highlights genes causing PANC-1 to be resistant or sensitized to G12Di identified in the CRISPR screens. Y axis represents the difference in Beta scores between G12Di- and DMSO-treated cells (Δ Beta Score). C, Resistant or parental PDAC cell lines were treated with G12Di (0 – 333 nM) or RASi (0 – 333 nM) for 5 days, and cell viability was determined by the addition of Hoechst. Normalized cell counts compared to the control conditions, which were normalized to 1.0, were plotted, and error bars denoted standard error of at least three independent biological replicates. D, Representative western blots probing KEAP1, NRF2, NQO1, and vinculin in Pa14C or PANC-1 cells which are resistant to G12Di (G12Di-R), RASi (RASi-R), or their parental counterpart. The relative levels of KEAP1, NRF2, and NQO1 expressions are denoted underneath the designated blot images.

Loss-of-function mutations in KEAP1 were associated with shorter survival in NSCLC patients on G12Ci therapy (23,39), and we observed that KEAP1 knockout was the most potent resistance driver. Conversely, NFE2L2 (encoding NRF2/NFE2L2) knockout caused increased sensitivity to KRAS inhibitor treatment. These observations support a role for KEAP1 loss and NRF2 activation in causing resistance to KRAS inhibition.

To determine a role for KEAP1-NRF2 in acquired resistance, we cultured sensitive cell lines with each inhibitor to select for outgrowth of resistant subpopulations (Fig. 1C; Supplementary Fig. S1D). Whereas the GI₅₀ values for each parental cell line for both inhibitors were ~30 nM or less, the resistant cell populations exhibited very limited growth suppression at 100 nM or more of each inhibitor (Supplementary Fig. S1E). Western blot analysis determined that three of the four resistant cells exhibited lower KEAP1 and higher NRF2 expression levels, as well as increased expression of the NRF2 transcriptional target NQO1 encoded protein (40) (Fig. 1D; Supplementary Fig. S1F). Together with the findings from our CRISPR screen, these results further support a role for the KEAP1-NRF2 signaling network in modulating sensitivity to KRAS inhibitors in KRAS-mutant PDAC.

KEAP1 Loss Induces NRF2-Dependent RAS Inhibitor Resistance in PDAC and LUAD

To validate the observations from the CRISPR screen, we applied CRISPR knockout of KEAP1 in a panel of four KRAS^G12D-mutant human PDAC cell lines (SW1990, Pa14C, Pa16C, and PANC-1) (Fig. 2A). An increase in the steady-state levels of NRF2 was seen in KEAP1 knockout (KEAP1^KO) relative to control cells stably infected with a lentivirus vector expressing Cas9 and sgRNA targeting eGFP. Increased expression of NRF2 transcription gene target encoded proteins (NQO1 and SLC7A11) was also seen, indicating activation of NRF2 transcriptional activity in the KEAP1^KO cell lines.

Figure 2.

The knockout of KEAP1 drove resistance to KRAS inhibition by upregulating NRF2 expression. A, Representative western blots probing KEAP1, NRF2, NQO1, SLC7A11, and vinculin in PDAC cell lines (SW1990, Pa16C, Pa14C, and PANC-1) where KEAP1 gene was knocked-out using two different sgRNA sequences targeting KEAP1 (KEAP1–1 and KEAP1–2), or control sgRNA targeting eGFP (eGFP). The relative levels of NRF2, NQO1, and SLC7A11 expressions are denoted underneath the designated blot images. B, PDAC cell lines with KEAP1 knockout (KEAP1–1, red; and KEAP1–2, green), or control cells (eGFP, blue), were treated with G12Di or RASi (0 – 300 nM) for 5 days, and cell viability was determined by the addition of Calcein-AM assay. Normalized cell counts compared to the control conditions, which were normalized to 1.0, were plotted, and error bars denoted standard error of at least three independent biological replicates. C, GI₅₀ values for PDAC cells with KEAP1 (KEAP1–1, KEAP1–2) or eGFP knockout treated with G12Di or RASi. D, Representative 2D clonogenic growth assays show long-term cell proliferation of Pa14C cells upon knockout of KEAP1 (KEAP1–1 and KEAP1–2) or eGFP. Cells were cultured with G12Di (0 – 100 nM) for 14 days, and they were stained at the end of the experiment with crystal violet and imaged using a Typhoon FLA9500 biomolecular imager. E, PDAC cell lines (Pa14C and PANC-1) cells with knockout of KEAP1 using KEAP1–1 sgRNA (KEAP1, red), double knockout of KEAP1 and GFP (KEAP1/GFP, orange), double knockout of KEAP1 and NRF2 (KEAP1/NFE2L2, magenta), and control cells with eGFP knockout (eGFP, blue) were treated with G12Di (0 – 300 nM) for 5 days, and cell viability was determined by the addition of Calcein-AM assay. Normalized cell counts compared to the control conditions, which were normalized to 1.0, were plotted, and error bars denoted standard error of at least three independent biological replicates. Concentrations of G12Di inhibit 50% of viability were defined as GI₅₀ and are shown in the corresponding colors underneath each plot. F, Tumor size measurements of mice bearing KP allografts with knockout of Keap1 (KPK, solid dots) or neomycin resistant gene (KPNeo, hollow dots). Mice were treated with either vehicle (blue) or G12Di (10 mg/kg, red) and monitored for 17 days. G, Tumor size measurements of mice bearing KP allografts ectopically expressing wild type Keap1 (Keap1^WT, hollow dots) or KEAP1^R470C (KEAP1^R470C, solid dots). Mice were treated with either vehicle (blue) or G12Di (10 mg/kg, red) and monitored for 13 days. H, Tumor size measurements of mice bearing KP allografts with knockout of Keap1 (KPK, solid dots) or neomycin resistant gene (KPNeo, hollow dots). Mice were treated with either vehicle (blue) or G12Di (30 mg/kg, red) for 9 days. The tumor growth was monitored for 34 days or until they reach the sacrificing endpoints.

We found that KEAP1^KO cells exhibited variable resistance to both G12Di and RASi with significantly higher GI₅₀ values relative to eGFP^KO control cells (Fig. 2B and 2C). KEAP1^KO cells also exhibited several fold reductions (1.8- to 2.6-fold) in G12Di induced cell death (Supplementary Fig. S2A). KEAP1^KO cells exhibited reduced sensitivity to G12Di treatment, compared to eGFP^KO control cells in long-term clonogenic growth assays (Fig. 2D; Supplementary Fig. S2B and S2C). Similarly, KEAP1^KO cells showed reduced sensitivity to G12Di treatment when cultured in three-dimensional (3D) non-adherent spheroid cultures (Supplementary Fig. S2D).

To determine whether the resistance induced by the loss of KEAP1 is dependent on NRF2 expression, we generated KEAP1 and NFE2L2 double knockout (KEAP1/NFE2L2^KO) PANC-1 and Pa14C cells together with control KEAP1/GFP^KO knockout cell lines. Compared with KEAP1/GFP^KO cells, the KEAP1/NFE2L2^KO cells showed decreased expression of NQO1, verifying reduction in NRF2-mediated transcriptional activity (Supplementary Fig. S2E). The KEAP1/GFP^KO control cells exhibited similar resistance to G12Di as KEAP1^KO cells, whereas KEAP1/NFE2L2^KO cells showed comparable sensitivity as control PDAC cells (Fig. 2E). We concluded that resistance to RAS inhibition driven by KEAP1 loss was dependent on NRF2 expression.

We also assessed the consequences of pharmacologic induction of NRF2 expression using two chemically distinct small molecules, AI-1 and bardoxolone methyl (CDDO-me), which block KEAP1 interaction with NRF2, stabilizing NRF2 expression (41,42). Treatment with either AI-1 or CDDO-me caused increased expression of NRF2 (Supplementary Fig. 2F) and reduced sensitivity to G12Di (Supplementary Fig. 2G). Taken together, these observations suggest that NRF2 activation is both necessary and sufficient to support resistance to RAS inhibition.

We then extended our analyses of KEAP1 loss in driving resistance to RAS inhibition in LUAD. We first knocked out KEAP1 in the KRAS^G12V-mutant LUAD cell line NCI-H441 by CRISPR-Cas9 editing. Similar to what we observed in PDAC cell lines, KEAP1^KO resulted in upregulation of NRF2 expression and transcriptional activity, and resistance to RASi (Supplementary Fig. S2H and S2I). Second, using murine Kras^G12D/Trp53^−/− (KP) cells with Keap1 knockout (KPK) (43), we established KPK cells ectopically expressing wild-type Keap1 or an empty vector control. Restoration of Keap1 expression caused a five-fold increased sensitivity to G12Di (Supplementary Fig. S2J). Together, these observations support KEAP1 loss as a mechanism of resistance to RAS inhibitors in LUAD.

We then investigated the impact of Keap1 loss on response to G12Di in syngeneic immune competent tumor models established by KPK or KP cells with a control CRISPR-Cas9 knockout (KPNeo) (43,44). Tumor growth was monitored in mice treated with G12Di or vehicle. We found that Keap1 loss significantly reduced tumor response to G12Di (Fig. 2F). Similarly, we treated KP cells with ectopic expression of either wild type (KPK^WT) or a dominant negative Keap1^R470C mutant with reduced ability to degrade NRF2 (KPK^R470C) with G12Di and observed that tumors expressing Keap1^R470C grew significantly faster than those expressing Keap1^WT when treated with G12Di (Fig. 2G). Finally, we tested the durability of G12Di response by treating mice bearing KPK or KPNeo allografts with G12Di or vehicle for 9 days, after which we stopped the treatment (Fig. 2H). While G12Di caused robust tumor suppression in both models, KPK tumors rebounded after day 20 and grew rapidly. However, the KP control tumors remained suppressed and showed limited signs of relapse until day 30. We conclude that loss of KEAP1 can drive resistance to RAS inhibition in both PDAC and LUAD in vitro and in vivo.

KEAP1 Loss Drives a Distinct Transcriptome Independent of Oncogenic KRAS

Patients who relapsed on G12Ci therapy exhibit mutational activation of ERK signaling (16,17,22,38), and we showed that ERK reactivation alone was sufficient to drive RAS inhibitor resistance (19,45,46). Therefore, we examined if KEAP1 loss-driven G12Di resistance involved activation of ERK. Control (eGFP^KO) and KEAP1^KO matched pairs of PANC-1 and Pa14C cell lines were treated with G12Di, and the suppression of ERK signaling was monitored (Supplementary Fig. S3A). Comparable suppression of pERK and the ERK substrate RSK (pRSK) was seen after 4- and 24-hour G12Di treatment regardless of the KEAP1 status. We conclude that KEAP1 loss-induced resistance was not due to an impaired ability of G12Di to inhibit ERK.

To assess a molecular basis for KEAP1 loss-mediated resistance against G12Di, we conducted RNA-sequencing (RNA-seq) to profile gene expression changes in four PDAC cells depleted of KEAP1. We applied a doxycycline (dox)-inducible CRISPR-Cas9 system (47) to acutely (5 days) knockout KEAP1 in four KRAS^G12D-mutant PDAC cell lines. KEAP1 knockout caused elevated expression of NRF2 and NRF2 transcriptional targets NQO1 and SLC7A11 (Supplementary Fig. 3B) and promoted resistance to G12Di to similar levels seen with stable KEAP1 knockout (Supplementary Fig. S3C).

Our RNA-seq analyses detected a total of 17,674 genes (with read counts > 20) across all four cell lines. Among these, 1,248 genes were found significantly upregulated (PDAC-KEAP1^KO UP, log₂ fold change (FC) > 0, adjusted p value (adj. p) < 0.05) and 1,016 genes significantly downregulated (PDAC-KEAP1^KO DN, log₂FC < 0, adj. p < 0.05) upon KEAP1 knockout (Fig. 3A; Supplementary Table S1).

Figure 3.

RNA sequencing identified unique KEAP1-loss transcriptome independent of KRAS. A, Volcano plots showing the differentially expressed mRNA transcripts from the RNA-seq analysis in PDAC cells following induced KEAP1 knockout, the PDAC-KEAP1^KO transcriptome. A total of 1,248 significantly up-regulated (red) and 1,016 significantly down-regulated (blue) genes were identified. Representative NRF2 transcriptional targets, including SLC7A11, GCLC, NQO1, and GCLM were found significantly upregulated in the KEAP1-loss transcriptome. B, Barcode plots illustrating the differential expression (log₂FC) in the PDAC-KEAP1^KO transcriptome of genes within each signature (left panel). Significantly upregulated and downregulated genes within each signature are shown in red and blue, respectively. The number of genes (N) in each set is labeled in each box (right panel) Normalized enrichment scores (NES) for the corresponding signatures of the PDAC-KEAP1^KO transcriptome in (A). Adjusted P values are indicated within the bars. C, Venn diagram showing the numbers of overlapping genes within ROMERO, IBRAHIM, CHORLEY, SINGH signatures, and the PDAC-KEAP1^KO UP signature from (A). D, Dot plot showing normalized enrichment scores (NES) of 50 Hallmark gene sets from GSEA on the PDAC KEAP1^KO transcriptome. E, Volcano plots showing the differentially expressed mRNA transcripts from RNA-seq analysis on KEAP1^KO PDAC cells treated with G12Di at their GI₅₀ concentrations for three days. 65% of PDAC-KEAP1^KO UP genes remain significantly upregulated (red). F, Volcano plots showing the differentially expressed mRNA transcripts from RNA-seq analysis on KP mouse lung cancer cells treated with NRF2 activator KI696 (500 nM) for 6 days. A total of 2,533 significantly up-regulated (red) and 1,966 significantly down-regulated (blue) genes were identified. G, Volcano plots showing the differentially expressed mRNA transcripts from RNA-seq analysis of KP cells pre-treated with NRF2 activator KI696 (500 nM) for 4 days and G12Di (10 nM) for 2 days. Genes upregulated in (F) remain significantly upregulated under KRAS inhibition (red). H, GSEA performed using the 50 Hallmark gene sets, as well as PDAC-KRAS UP and DN signatures previously reported by our group. Analyses were conducted on PDAC cell panels (PDAC) and KP LUAD cells (mKP) with or without KEAP1 knockout (KEAP1^KO) or NRF2 activation (NRF2^act), and with or without G12Di treatment.

We first performed HOMER (Hypergeometric Optimization of Motif EnRichment) analyses to assess the transcription factor binding motifs present in the PDAC-KEAP1^KO UP and DN gene sets (48). Supervised promoter motif-enrichment analyses for the top 500 UP gene set identified the DNA motif for NFE2L2/NRF2 as the top motif (Supplementary Fig. S3D), and this motif corresponds to the antioxidant response element (ARE) (49). In contrast, no ARE motifs were identified in the top 10 scoring motifs for the PDAC-KEAP1^KO 500 DN genes (Supplementary Fig. S3E). Given that we determined that KEAP1 loss-associated resistance to RASi was NRF2-dependent (Fig. 2E), we speculated that the PDAC-KEAP1^KO UP but not DN gene set comprises the significant genes involved in driving RASi resistance.

We next compared our PDAC-KEAP1^KO UP/DN 2,264-gene signature with eight published gene signatures previously reported by KEAP1 loss and/or NRF2 activation in rodent and/or human species (Supplementary Table S2) (43,50–55). Although gene set enrichment analysis (GSEA) found enrichments in six of these KEAP1-NRF2 signatures in our PDAC-KEAP1^KO RNA-seq, only our previously established ROMERO signature (43) and the 14-gene LOU signature showed substantial overlaps with our gene set (49 and 93%, respectively). In contrast, fewer than 30% of genes within the remaining five signatures were significantly upregulated in our gene set (Supplementary Fig. S3F; Fig. 3B).

Reflecting the distinct nature of our PDAC-KEAP1^KO UP gene signature, we found only nine genes overlapping with the ROMERO, SINGH, IBRAHIM, and CHORLEY signatures (Fig. 3C). In contrast, 1,089 genes in our PDAC-KEAP1^KO UP gene set (87%) have not been identified previously in any of these published gene sets. The lack of overlap with previous signatures may reflect the distinct experimental approaches for activating NRF2 as well as the different cell models studied, with none KRAS-mutant or pancreatic cell models.

We performed overrepresentation analysis (ORA) using the mSigDB 50 Hallmark gene sets and found that the p53, hypoxia, mTORC1 signaling, reactive oxygen species, and glycolysis pathway gene sets were highly enriched in our PDAC-KEAP1^KO UP signature as well as in the published KEAP1-NRF2 gene sets. The convergence of these signaling pathways underscores a conserved core of NRF2-mediated functions, highlighting the robustness of our signature in capturing KEAP1-NRF2 oncogenic function (Supplementary Fig. 3G).

We next evaluated the 1,089 unique genes within the PDAC-KEAP1^KO UP gene set by ORA using Reactome, Kyoto Encyclopedia of Genes and Genomes (KEGG), and Gene Ontology (GO) gene sets. There were nine pathways that were found significantly enriched, with five of these related to lipid metabolism or biosynthesis (Supplementary Fig. 3H). Moreover, HOMER analysis performed on these 1,089 unique genes no longer showed any enriched transcription binding motifs, including NFE2L2/NRF2 (Supplementary Fig 3I). These results suggest that secondary gene targets regulated by the KEAP1-NRF2 pathway were captured in our transcriptomic analysis. These analyses highlight that our PDAC-KEAP1^KO model captures KEAP1-NRF2-associated genes distinct from the core NRF2 gene targets.

To assess the biological processes regulated by KEAP1 in PDAC, we evaluated our PDAC KEAP1-dependent transcriptome by gene set enrichment analyses (GSEA) with the 50 Hallmark gene sets. As expected, the REACTIVE_OXYGEN_SPECIES_PATHWAY was among the top upregulated activity (Fig. 3D).

To evaluate whether the RAS inhibition can suppress KEAP1^KO transcriptome, we performed RNA-seq on these PDAC cells cultured with doxycycline to induce KEAP1 knockout and treated with G12Di. Interestingly, only 11 genes (< 1%) within the 1,248-gene PDAC-KEAP1^KO UP signature were suppressed by G12Di, while 813 genes (65%) of the signature remained upregulated under G12Di treatment (Fig. 3E). Moreover, we extended our RNA-seq analysis again using KP mouse lung cancer cells cultured in 3D spheroid and treated with a NRF2 inducer KI696 (KP-NRF2^act) to induce NRF2 expression (56). We detected a total of 13,182 genes (11,079 genes mapped with human homologs), with 2,553 upregulated and 1,966 downregulated genes (Fig. 3F). Similarly, 76 % (1,931) gene expressions that were upregulated upon KI696 treatment sustained upon treated with G12Di, with only 9% (232) gene exhibited decreased expressions (Fig. 3G). Overall, these observations suggest that KEAP1-NRF2 regulated signaling pathways that are independent of KRAS-regulated signaling networks.

We next performed GSEA with the MSigDB 50 Hallmark gene sets plus our previously established PDAC-KRAS UP/DN gene signature on PDAC-KEAP1^KO or KP-NRF2^act treated without or with G12Di, and control PDAC or KP cells treated with G12Di as controls (45,57). Trends for gene sets showing either upregulation or downregulation were generally consistent across all PDAC and KP models (Fig. 3H, and Supplementary Fig. S3J). As expected, the most significant activities upregulated in KEAP1^KO cells were the Reactive Oxygen Species Pathway and the Xenobiotic Metabolism gene sets, well-established NRF2-driven activities (24,27), and these pathways remain enriched even when cells were treated with G12Di. Importantly, although G12Di cause a significant downregulation of the PDAC-KRAS UP gene set, KEAP1 loss or NRF2 activation cannot restore its expression, consistent with our observation that KEAP1^KO was not able to restore the activation of ERK from G12Di (Supplementary Fig. S3A).

Furthermore, we performed ORA on genes that are significantly upregulated by more than 50% (adj. p < 0.05, FC > 1.5) for each cell line (Supplementary Fig. 3K), and we determined that 29 gene sets that were significantly upregulated in at least three of four cell lines. These gene sets included the Reactome KEAP1-NFE2L2 pathway, Nuclear events mediated by the NFE2L2 pathway, and NFE2L2 regulating antioxidant/detoxification enzymes gene sets, which is consistent with our findings that KEAP1 depletion activated NRF2-dependent gene transcription.

The sustained activation of the Reactive Oxygen Species Pathway in KEAP1^KO or NRF2^act cells treated with G12Di (Fig. 3H) identified one mechanistic basis for resistance to RAS inhibition. A previous study found that treatment of BRAF-mutant lung or KRAS-mutant pancreatic cancer cell lines with the MEK inhibitor trametinib (MEKi) induced accumulation of ROS, but this effect was attenuated with KEAP1 loss (30). Therefore, we determined whether KRAS-ERK inhibition causes oxidative stress by accumulating ROS, and whether KEAP1 knockout would counter ROS activation. We treated control (eGFP^KO) or KEAP1^KO PANC-1 and Pa14C cells with G12Di or MEKi, and the level of ROS was determined using the CellROX Green fluorogenic probe. Both G12Di and MEKi treatment increased ROS levels in control cells but not KEAP1^KO cells, except for KEAP1^KO PANC-1 cells which exhibited a limited ROS increase when treated with MEKi (Supplementary Fig. S3L).

It has been observed that KRAS inhibitors and oxidative stress can induce cell cycle arrest, and previous studies established that NRF2 promotes cell cycle transition (58–61). Therefore, we examined whether KEAP1 knockout and NRF2 accumulation could mitigate cell cycle arrest induced by KRAS inhibition. We performed cell cycle analyses on KEAP1^KO and control (eGFP^KO) PANC-1 and Pa14C cells treated with G12Di for three days. Control Pa14C cells showed a pronounced G₁ arrest upon treatment with G12Di, while KEAP1^KO Pa14C cells only exhibited modest increase in G1 population under G12Di treatment (Supplementary Fig. S3M). Similarly, in PANC-1, G12Di caused an increase in G₁ population, while negligible changes in cell cycle distribution were observed in KEAP1^KO PANC-1. These findings suggest that a subset of KEAP1^KO PDAC cells are refractory to cell cycle arrest induced by KRAS inhibition.

Distinct Mechanisms of Resistance Due to Loss of KEAP1 Versus Activation of KRAS-MYC or YAP/TAZ-TEAD

MEK-ERK, MYC, and YAP/TAZ-TEAD activation has been reported to limit the effectiveness of KRAS inhibition (12,17,20,22). To assess the degree of overlap between genes regulated by these mechanisms of resistance with KEAP1-NRF2, we compared the KEAP1^KO UP transcriptome to transcriptomes associated with KRAS, MYC and YAP/TAZ-TEAD. For these analyses, we utilized data from gene sets established by acute 24 hour siRNA knockdown of KRAS (KRAS UP) (45) or MYC (MYC UP) (62) in a panel of eight KRAS-mutant PDAC cell lines, or pharmacologic inhibition with 48 hour treatment of an NF2-null human mesothelioma cell line with the pan-TEAD inhibitor GNE-7883 (TEAD UP; GSE273175) (63). We performed the network analysis on genes that were significantly upregulated from each dataset and weighted based on their weighted log₂FC (Supplementary Fig. S4A).

We found very significant overlap with the KRAS-, MYC-, and TEAD-dependent gene networks (607 genes), while the KEAP1^KO-dependent transcriptome was distinct from the other three gene sets, with only 4–13% overlap (Fig. 4A). Moreover, analysis of gene essentiality scores in pancreatic and lung cancers from the Dependency Map (DepMap) portal shows that the KRAS-, MYC-, and TEAD-dependent genes generally exhibited higher essentiality scores, whereas the KEAP1^KO-dependent UP transcriptome was comprised primarily of non-essential genes. Collectively, these findings suggest that the KEAP1-NRF2 signaling network promotes drug resistance utilizing a distinct molecular mechanism from those driven by KRAS, MYC, or TEAD.

Figure 4.

KEAP1-loss transcriptome is distinct from KRAS-, TEAD-, and MYC-dependent transcriptome and associated with the resistance to KRAS inhibitors. A, The network analysis on genes upregulated upon KEAP1 knockout (KEAP1^KO UP), as well as genes dependent on KRAS (KRAS UP), MYC transcription factor (MYC UP), or TEAD transcription factor (TEAD UP). Each node denotes a gene, and the size and color intensity for each node represent the medium DepMap essentiality scores across PDAC and LUAD cohorts. The edges are color coded and connect each node to their associated regulatory programs (KRAS, TEAD, MYC, or KEAP1^KO). The width of each edge represents the magnitude of weighted log₂FC from each dataset. B, The heatmap for the expression changes (weighted log₂FC) of genes within the PKD signature. The frequency (ChIP-Seq frequency, green) of genes that were found to be NRF2 transcriptional targets in the ENCODE database were plotted underneath the plot. C, The gene expression scores of PKD, ROMERO, SINGH, CHORLEY, and IBRAHIM signatures were determined using Singscore on mRNA expressions collected from KRAS^G12C-mutant patients before treated with adagrasib. The p values between patients who responded (sensitive, red) or were resistant to adagrasib (resistant, blue) were determined using Wilcoxon test. D, The gene expression scores of PKD, ROMERO, SINGH, CHORLEY, and IBRAHIM signatures were determined using Singscore on mRNA expressions collected from PDX models treated with sotorasib. The p values between initially sensitive tumors (sensitive, red) or relapsed tumors (resistant, blue) were determined using Wilcoxon test.

Establishment of a PDAC KEAP1-Deficiency (PKD) Gene Signature

With KEAP1 loss associated with RASi resistance and deregulated gene transcription, we next determined a KEAP1-deficiency gene signature that may have predictive value for response to RASi. To define a focused gene set, we utilized the top 200 significantly upregulated and protein-coding genes upon KEAP1^KO as the PDAC KEAP1-deficiency (PKD) gene signature (Fig. 4B; Supplementary Table 3). We first examined the proportion of direct NRF2 transcriptional targets within the PKD signature by cross-referencing chromatin immunoprecipitation sequencing (ChIP-seq) data from the ENCODE database (64). We found 100 genes (50%) within the PKD signature were detected as direct NRF2 gene targets (Fig. 4B). Moreover, HOMER motif analysis identified NFE2L2, BACH1, and NRF2 as top enriched transcription factor binding motifs in the PKD signature, where BACH1 has been reported to act as a repressor to AREs as a negative feedback mechanism of NRF2 activation (Supplementary Fig. S4B) (65). In summary, the PKD signature captures a significant portion of direct NRF2 transcriptional targets as well as secondary targets where their genetic expressions were induced upon KEAP1^KO.

To elucidate the pathways regulated by genes within the PKD signature, ORA was carried out to identify significantly enriched pathways using Reactome, KEGG, and GO gene sets (Supplementary Fig. S4C). The analysis revealed a robust enrichment associated with oxidative stress response pathways. Given our previous finding that KRAS inhibition elevates ROS levels in PDAC cells (Supplementary Fig. S3L), these ORA results underscore the important role of oxidative stress relief of the PKD genes.

Beyond oxidative stress control, the PKD signature consists of genes potentially involved in additional mechanisms promoting resistance to KRAS inhibitors. ORA revealed that pathways related to chemical stress or toxic substance response were highly enriched. These include ABCC3 and SLC47A1, which encode membrane transporters that facilitate drug efflux, diminishing inhibitor efficacy (66–68). Aldo-keto reductase genes, such as AKR1B1 and AKR1C4, were identified, and they have been shown to confer chemoresistance and resistance to small molecule inhibitors via NADPH-dependent carbonyl reduction to external chemicals (69–72). Autophagy related genes (TXNRD1 and SQSTM1) were further notable components, and we and others previously reported that increased autophagy led to resistance to ERK inhibition in PDAC cells (73,74). The SRC tyrosine kinase oncogene was also identified in the PKD signature and is frequently overexpressed in PDAC and has been widely reported to cause drug resistance (75–77).

Finally, we examined the clinical relevance of the PKD signature in LUAD patients harboring mutant KEAP1. Utilizing RNA-seq data from the LUAD cohort in TCGA database, we categorized patients based on their KEAP1 mutational status and calculated the enrichment scores of genes within the PKD signature. The KEAP1 mutant tumors exhibited significantly elevated expressions of the PKD signature relative to those carrying wild type KEAP1 (Supplementary Fig. S4D). These data further confirm that the PKD gene set is a robust indicator of aberrant KEAP1 signaling in not only PDAC but also LUAD.

Increased Expression of Genes within the PKD Signature is Associated with Resistance to KRAS Inhibition.

We next determined if the PKD signature is associated with resistance to KRAS inhibitors. We examined gene expressions within the PKD, ROMERO, SINGH, CHORLEY, and IBRAHIM signatures in the RNA expression data from patients or patient-derived xenografts (PDX) treated with KRAS inhibitors. The transcriptome of KRAS^G12C-mutant NSCLC or CRC patients before treatment with adagrasib (45) showed that the two patients who responded to the drug had the lowest PKD expression scores, and those were significantly separated from the non-responder group (Fig. 4C). In contrast, expression scores of ROMERO, SINGH, CHORLEY, and IBRAHIM signatures did not effectively differentiate the response status.

A similar analysis was performed on PDX models treated with sotorasib (GSE204753) (78), and we found the significantly lower expression scores of PKD, ROMERO, and SINGH signatures in the sensitive samples, distinguishing them from those which were resistant (Fig. 4D). The CHORLEY signature showed limited distinction while IBRAHIM expression scores were indistinguishable between the two groups.

In summary, analyses across patient cohorts and PDX models under RASi treatments demonstrated that the PKD gene signature was strongly associated with resistance to KRAS inhibitors and poor clinical outcomes. We conclude that, although other previously established gene sets encompass a broader coverage of KEAP1-NRF2-related genes, the PKD signature can more effectively distinguish RASi resistant cases from sensitive samples.

KEAP1^KO PDAC cells Exhibit Enhanced Dependency on Glutamine Metabolism

We have previously demonstrated that KEAP1^KO lung cancer cells become more sensitive to the glutaminase inhibitor CB-839 (GLSi) (43). Glutaminase facilitates the conversion of glutamine to glutamate, and it is critical for glutathione synthesis and metabolic reprogramming in cancer cells (79,80). In fact, we and others have reported that cells with KEAP1 inactivation or NRF2 activation can reduce oxidative stress by increasing glutathione levels (26,30,43,81). In the present study, we found that the expressions of genes regulating glutamine and glutathione metabolism were enriched in the cancer cells with KEAP1 loss, such as PGD, G6PD, GSR, GCLC, and GCLM. This finding led us to investigate whether KEAP1 loss in PDAC cells is associated with increased dependency on glutamine metabolism.

We evaluated cell proliferation upon treating control and KEAP1^KO PDAC cells with GLSi. We first applied a glutaminase activity assay using isolated mitochondria and confirmed the inhibitory effect of GLSi on glutaminase activity (Supplementary Fig. S5A). The proliferation assay performed on SW1990, Pa16C, and Pa14C PDAC showed enhanced sensitivity to glutaminase inhibition upon KEAP1 knockout, while the sensitivity of GLSi changed only marginally in PANC-1 cells (Fig. 5A). We then examined glutamate secretion and glutamine consumption in KEAP1^KO SW1990 and Pa14C cell lines. We found that glutamate secretion and glutamine uptake significantly increased in the KEAP1^KO cells (Supplementary Fig. S5B and S5C), confirming elevated glutamine dependency.

Figure 5.

Concurrent inhibition of glutaminase and KRAS showed enhanced anti-tumor activity in PDAC. A, PDAC cell lines with stable knockout of KEAP1 (KEAP1–1, red; and KEAP1–2, green), or eGFP (eGFP, blue), were treated with GLSi (0 – 500 nM) for 5 days, and cell viability was determined by the addition of Calcein-AM assay. Normalized cell counts compared to the control conditions, which were normalized to 1.0, were plotted, and error bars denoted standard error of at least three independent biological replicates. Concentrations of GLSi that inhibit 50% viability were defined as GI₅₀ and are shown in the corresponding colors underneath each plot. B, PDAC cell lines with stable knockout of KEAP1 (KEAP1–1, red; and KEAP1–2, green), or eGFP (eGFP, blue), were treated with GLSi (100 nM) and either G12Di (top panel, 0 – 300 nM) or RASi (bottom panel, 0 – 300 nM) for 5 days, and cell viability was determined by the addition of Calcein-AM assay. Normalized cell counts compared to the control conditions, which were normalized to 1.0, were plotted, and error bars denoted standard error of at least three independent biological replicates. C, The GI₅₀ (nM) of PDAC cell panel treated with G12Di, and without (dark blue) or with (light blue) 100 nM GLSi. D, The GI₅₀ (nM) of PDAC cell panel treated with RASi. and without (dark blue) or with (light blue) 100 nM GLSi. E, The GI₅₀ (nM) of LUAD cell panel treated with RASi, and without (dark blue) or with (light blue) 100 nM GLSi.

Next, we investigated the mechanism underlying the enhanced glutamine dependency caused by KEAP1 depletion. Glutamate secretion is mediated by the anti-transporter SLC7A11, which exchanges intracellular glutamate for extracellular cystine, another precursor for glutathione synthesis (43,82,83). We showed previously that SLC7A11 overexpression is associated with increased sensitivity to glutaminase inhibition (43), and in the present study we determined that SLC7A11 mRNA and protein expression were upregulated upon KEAP1 depletion (Fig. 2A and 3A). To determine whether KEAP1^KO-associated glutamine addiction resulted from SLC7A11 overexpression in PDAC cells, we knocked down SLC7A11 with two distinct siRNAs (siSLC7A11–1 and siSLC7A11–2), and with non-specific siRNA as controls (siNS), in control (eGFP^KO) or KEAP1^KO SW1990 and PA14C cells. The reduction of SLC7A11 expression (Supplementary Fig. S5D) resulted in decreased sensitivity to GLSi in KEAP^KO cells (Supplementary Fig. S5E) (43). Collectively, these findings support a mechanism where KEAP1-depleted PDAC cells become dependent on glutaminase activity primarily through upregulation of SLC7A11 expression.

We next determined if combining GLSi with a KRAS inhibitor can more effectively inhibit cancer cell proliferation. With the addition of GLSi, we found that the GI₅₀ of G12Di or RASi became notably lower in KEAP1^KO cells, and this combination also enhanced the cytotoxic activity of KRAS inhibitors in control cells (Fig. 5B; Supplementary Fig. S5F and S5G). We also evaluated the sensitivity of G12Di or RASi without or with GLSi in a panel of primary PDAC and NSCLC cell lines. We found that concurrent treatment with GLSi generally augmented the activity of G12Di and RASi across PDAC and LUAD cell lines, with Pa01C, HPAF-II, HPAC, and A549 cells presented the strongest enhancement (Fig. 5C–5E; Supplementary Fig. S5H–L). These observations support the benefit of co-targeting glutamine metabolism and KRAS signaling in suppressing KRAS-mutant PDAC and LUAD growth.

Glutamine Antagonist DRP-104 Enhances RAS inhibitor Anti-tumor Activity

It has been shown that GLSi exhibited limited efficacy in vivo, which may result from cancer cells compensating for glutamate depletion following prolonged GLSi treatment (84). Thus, we hypothesized that an alternative approach targeting global glutamine metabolism may be necessary to achieve a pronounced anti-tumor effect in vivo. Recently, DRP-104, a prodrug of a potent glutamine antagonist 6-diazo-5-oxo-l-norleucine (DON), was developed to globally target glutamine-utilizing enzyme and found selective in malignant tumors (85). We and others have shown that DRP-104 can effectively and sustainably suppress KEAP1 mutant LUAD and PDAC tumor growth with tolerable toxicity (86,87). Moreover, the drug is currently under phase 2 clinical evaluation in KEAP1 or NFE2L2 altered NSCLC patients (NCT07249372) and a phase 1/2 clinical trial assessing the combination with the anti-PD-L1 inhibitor durvalumab in fibrolamellar hepatocellular carcinoma (NCT06027086). Therefore, we hypothesized that DRP-104 treatment could sensitize tumors to KRAS inhibition.

We first tested the combination of RASi and DRP-104 in two WT and four KRAS-mutant PDAC patient-derived organoids (Supplementary Fig. S6A). Among these organoid cultures, PANFR0185 harbors WT NRF2 amplification (copy number >5) and PANFR0165 contains the NRF2^D29H mutation which impairs interaction with KEAP1(88). RASi demonstrated potent inhibition in organoid growth in all KRAS mutant but not WT organoids (Fig. 6A and 6B; Supplementary Fig. S6B). All six organoids were sensitive to DRP-104, with PANFR0185 showing especially high sensitivity to DRP-104 (Supplementary Fig. S6C). Treatment with both drugs enhanced the growth suppression of these organoids compared with monotherapy. These observations support the therapeutic potential of co-targeting KRAS and glutamine metabolism in the setting of NRF2 activation.

Figure 6.

The combination of glutamine antagonist DRP-104 with RAS inhibitors effectively suppresses KRAS-mutant PDAC and LUAD tumor growth. A, Representative images of PANFR0185 organoids cultured in medium containing DMSO, RASi (24.7 nM), DRP-104 (400 nM), and the combination of both drugs (RASi + DRP-104). The scale bars denote 50 μm. B, PDAC patient derived organoids were treated with indicated concentrations of RASi and DRP-104 for 5 days, and cell viability was determined by 3D CellTiter Glo. Normalized cell viability compared to the control conditions, which were normalized to 1.0, were plotted, and error bars denoted standard error of at least three independent biological replicates. C, PANFR0185 organoids were subcutaneously injected into mice to induce PDAC patient-derived xenografts (PDX). Mice were randomly assigned to groups treated with vehicle, 3 mg/kg daraxonrasib, 3 mg/kg DRP-104, or the combination of both drugs, and tumor sizes were monitored for 23 days. D, KPK lung cancer cells were subcutaneously injected into mice to induce allografts. Mice were randomly assigned to groups treated with vehicle, 10 mg/kg RASi, 3 mg/kg DRP-104, or the combination of both drugs, and tumor sizes were monitored for 14 days. E, KP lung cancer cells expressing Keap1^R470C were subcutaneously implanted in mice to induce xenograft. Mice were treated with vehicle, 10 mg/kg G12Di, 3 mg/kg DRP-104, and the combination of the two drugs, and the tumor sizes were monitored for 24 days or until tumors reached the sacrificing endpoint.

Finally, we extended the analyses of this combination in vivo in KEAP1 or NRF2 altered models. First, we utilized the PANFR0185 (KRAS^G12R/NRF2^Amp) organoid culture to induce xenograft tumors in mice, and we evaluated the anti-tumor effect of concurrent treatment with G12Di, clinical candidate analog of RMC-7977 (RASi), the RAS(ON) multi-inhibitor daraxonrasib (RMC-6236) and DRP-104. The dosages and schedules were selected based on previously established protocols demonstrating potent anti-tumor activity and high therapeutic selectivity (86,89). Monotherapy with either DRP-104 or daraxonrasib exhibited limited anti-tumor activity. In contrast, the combination of DRP-104 and daraxonrasib significantly suppressed the tumor growth (Fig. 6C). Consistent with these findings, combining RASi with DRP-104 exhibited potent anti-tumor effect in mice bearing lung KPK subcutaneous tumors (Fig. 6D). Importantly, both experiments demonstrated limited toxicity profiles for either monotherapy or the combination, as there was no significant weight loss observed across different conditions (Supplementary Fig. S6D and S6E).

We then examined the combination of G12Di and DRP-104 in lung tumors induced by subcutaneous implantation of the Kras^G12D/p53^Δ/Δ/Keap1^R470C mouse LUAD cells (44). Consistent with the RASi studies, the treatments were tolerable and did not cause significant weight loss in mice (Supplementary Fig. S6F). Although treating mice with G12Di or DRP-104 alone was able to reduce the rate of tumor growth, the combination caused near-complete suppression of tumor growth (Fig. 6E). Collectively, these results support the potential of combination therapy targeting KRAS and glutamine metabolism in inhibiting KRAS-mutant, KEAP1-mutant or NRF2 activated pancreatic and lung cancers.

DISCUSSION

KEAP1 mutations have been associated with poor patient responses to KRAS^G12C inhibitors, the molecular mechanisms underlying this resistance remain unclear (16,17,23,39). NRF2 overexpression caused by KEAP1 loss-of-function has been linked to reduced sensitivity to KRAS^G12C inhibitors in NSCLC patients (39). While KEAP1 mutation is not commonly found in PDAC, the increased expression of NRF2 at advanced stages of PDAC highlights the importance of potential NRF2-mediated resistance (29,31). Here, we elucidated the mechanism of resistance to KRAS inhibition mediated by the KEAP1-NRF2 pathway to identify pharmacologic approaches to enhance the efficacy of KRAS-targeted therapy.

We determined that KEAP1 depletion conferred NRF2-dependent resistance to KRAS inhibitors and established a KEAP1-dependent transcriptional program in PDAC. Our KEAP1^KO UP/DN signature differed significantly from established KEAP1/NRF2 gene sets, in part because of genetic (KRAS mutation) and cell type differences. This heterogeneity between the signatures is likely due to the diverse approaches to activate NRF2 (small molecule inducers, KEAP1 silencing, NRF2 overexpression), the methodologies (RNA-seq, ChIP-seq, or literature curation), different cancer or cell line models (osteosarcoma, lymphoblastoid, lung cancer patients, HEK293T, and PDAC), and distinct gene-selection criteria in generating each signature. There is a core of nine NRF2-regulated genes (GCLC, GCLM, ME1, PIR, GSR, SLC7A11, NQO1, TXNRD1, and SRXN1) that are shared across nearly all of these gene sets. Overall, our KEAP1^KO UP/DN signature, along with others, reflects the transcriptomic alterations under different contexts, and the convergence on core NRF2-regulated targets among them implicates that fundamental transcriptional program of NRF2 was generally captured.

The resistance to KRAS inhibitors has been linked to activation of KRAS-MYC and YAP/TAZ-TEAD signaling pathways (12,17,20,22). Here, we compared the upregulated transcriptome associated with KEAP1-deficient, KRAS-, MYC-, and TEAD-dependent programs (45,62,63). The KRAS-, MYC-, and TEAD-dependent, upregulated transcriptome are interconnected and regulate many essential genes promoting tumor progression, based on DepMap gene essentiality scores. However, the KEAP1-deficient transcriptome was found distinct from the other three and contains primarily non-essential genes. These findings are consistent with the nature of KEAP1-NRF2 pathway, which activates cytoprotective genes primarily under stress conditions (24). The dysregulation of KRAS, MYC, and TEAD has been established as drivers to cancer progression, agreeing with these highly essential genes detected under these axes (3,90,91). Notably, the unique profile of KEAP1^KO transcriptome highlights the heterogeneity of resistance mechanisms to KRAS inhibition, implicating the importance of tailored therapeutics to address patients with tumor relapsed after the KRAS inhibitor treatment under different genetic contexts.

Moreover, we determined a 200-gene PDAC KEAP1-deficient (PKD) signature comprising commonly and significantly upregulated genes upon KEAP1 loss. By cross referencing the ChIP-seq results from the ENCODE database as well as HOMER motif analysis, genes that are directly regulated by NRF2 consist of 50% of the signature. Collectively, these findings further verify the hypothesis that NRF2 is the major driver of the resistance to KRAS inhibition promoted by KEAP1 knockout. Importantly, the PKD signature stands out to be highly correlated to the poor drug response to KRAS inhibitors in both KRAS-mutant PDAC and lung cancers. Future integration of more extensive clinical sequencing datasets will be critical to validate our findings and to establish the PKD signature as a predictive biomarker in the responses of KRAS inhibitors in patients.

Our RNA-seq analysis also revealed an increased dependence of KEAP1-deficient cells on glutamine metabolism, driven in part by NRF2-induced upregulation of SLC7A11. We exploited the anti-tumor effects of combining the clinical candidate, glutamine antagonist DRP-104 with inhibitors targeting RAS, and we demonstrated that this combination strongly suppressed proliferation in PDAC patient-derived organoids, especially those with NRF2 gain-of-function alterations. In addition, this combination also potently inhibited KEAP1/NRF2-altered PDAC or LUAD tumor growth in vivo with no overt toxicity. These results support the potential of future clinical evaluation of combining DRP-104 with a RAS inhibitor in targeting KEAP1-mutant lung cancer or NRF2-overexpressing PDAC.

In conclusion, our study comprehensively characterized the KRAS inhibitor-resistant transcriptome induced by KEAP1 depletion, and we derived a PDAC KEAP1-deficient (PKD) gene signature from the RNA-seq transcriptomic analysis. The transcriptome is not only highly driven by NRF2 activation but also strongly associated with resistance to KRAS inhibitors. This mode of resistance is largely distinct from established mechanisms driven by KRAS, MYC, or TEAD. Building upon the transcriptomic profiling, we identified a drug combination co-targeting KRAS and glutamine metabolism and showed that this treatment gave rise to potent anti-tumor activities. These findings highlight mechanistic insight into KEAP1-NRF2-mediated resistance and provide a rationale for future clinical evaluation of combinations involving KRAS inhibitors and the glutamine antagonist in PDAC and NSCLC.

METHODS

Cell Line, Spheroid, and Organoid Cultures

RASless MEFs stably expressing exogenous KRAS or BRAF were provided by the National Cancer Institute (NCI) RAS Initiative and established as described previously (36). Pa01C (RRID: CVCL_E303), Pa14C (RRID: CVCL_1638), and Pa16C (RRID: CVCL_1639) were provided by Dr. Anirban Maitra (MD Anderson Cancer Center). HEK293T (RRID: CVCL_0063), PANC-1 (RRID: CVCL_0480), SW1990 (RRID: CVCL_1723), AsPC-1 (RRID: CVCL_0152), HPAC (RRID: CVCL_3517), HPAF-II (RRID: CVCL_0313), SU.86.86 (RRID: CVCL_3881), NCI-H441 (RRID: CVCL_1561), A549 (CVCL_0023), NCI-H358 (RRID: CVCL_1559), and NCI-H2030 (RRID: CVCL_1517) were acquired from American Type Culture Collection (ATCC). KP LUAD cell lines were obtained from Dr. Tyler Jacks (Massachusetts Institute of Technology). Further generation of control KP cells, Keap1-KO (KPK), Keap1^WT-, and Keap1^R470C-expressing KPK cell lines were described previously (44,86). Cells were cultured in either DMEM (Thermo Fisher, #11995065) or RPMI-1640 (Thermo Fisher, #11875093) supplemented with 10% fetal bovine serum (FBS, Sigma #2442) and PennStrep (Thermo Fisher, #15140163). All cells were cultured in a humidified incubator under 37°C and 5% CO₂.

To establish RAS inhibitor-resistant cells, PDAC cell lines were initially cultured in appropriate medium containing either RASi or G12Di at their GI₅₀ concentrations. Once the cultures reached 90% confluency, they were then passaged at 2.0 × 10⁶ cells per dish and maintained in growth medium with inhibitor concentrations increased 50% from the previous passage. Cells would be continuously propagated until their sensitivity of the inhibitors reach 100 X of their original GI₅₀ concentrations. Cells were then cultured in the medium with the final drug dosages for future experiments.

Spheroids were generated by seeding 3,000 – 10,000 cells in a well of a round-bottom, low adherent plate in appropriate medium. The plates were then subjected to 300 × g centrifugation for 3 minutes. The plates were incubated in a humidified incubator under 37°C and 5% CO₂ for two days to allow spheroid formation. The spheroids were then subjected to future experiments.

Patient-derived organoids were obtained from Dr. David Tuevson (Cold Spring Harbor Laboratory, hM1A, hT105) and the Dana-Farber Cancer Institute (PANFR0165, PANFR0185, PANFR0088, PANFR0344). The organoid culture protocol was described previously (92). In brief, the organoids were maintained in a humidified incubator under 37°C and 5% CO₂. The organoids were cultured in reduced growth factor Matrigel domes with compete human feeding medium (Advanced DMEM/F12 (Thermo Fisher, #12634028)–based WRN conditioned medium (L-WRN)(ATCC, CRL-3276), 100 ng/mL hFGF10 (Thermo Fisher # 100–26), 10 mM nicotinamide (Sigma #72340), 0.01 μM hGastrin I (TOCRIS #3006), 50 ng/mL hEGF (Thermo Fisher # AF-100–15), 500 nM A83–01 (TOCRIS #2939), 1× B27 supplement (Thermo Fisher, #A1486701), 10 mM HEPES (Thermo Fisher #15630130), and 0.01 μM GlutaMAX (all Thermo Fisher, #35050079). All the patient-derived materials were obtained from patients with written consents. All the patient studies were conducted in accordance with the Declaration of Helsinki and approved by the Institutional Review Board.

CRISPR-Cas9 Knockout

sgRNA (KEAP1–1, CTTGTGGGCCATGAACTGGG; KEAP1–2, TGTGTCCTCCACGTCATGAA; eGFP, GGGCGAGGAGCTGTTCACCG) (30) targeting KEAP1 and eGFP were reported previously and cloned into LentiCRISPRv2 (Addgene #52961, RRID:Addgene_52961) or TLCV2 (Addgene #87360, RRID:Addgene_87360) (37). sgRNA (Keap1,CACCGCCGAAGCCCGTTGGTGAACA, AAACTGTTCACCAACGGGCTTCGGC; Neomycin resistant gene, CCTCGGAACAAGATGGATTGCACGC) for targeting Keap1 or Neomycin resistant gene were cloned into lentiCRISPRv2 puro (Addgene #98290, RRID: Addgene_98280). pLenti-DECKO-Cas9-Blast construct cloned with sgRNA targeting NRF2 (GGAGTAGTTGGCAGATCCAC) were provided from Dr. M. Benjamin Major (Washington University at St. Louis). sgGFP-Cas9-Blast (PXPR007 sgGFP-1, Addgene #74960, RRID: Addgene_74960) construct was acquired from Addgene. Keap1^R470C or Keap1^WT was cloned into a Gibson compatible lentiviral construct as described previously (44).

Lentivirus was used to transduce CRISPR-Cas9 system into cells and was generated by transfecting HEK-293T cells with the cloned construct, psPAX2 (Addgene #12260, RRID: Addgene_12260), and pMD2.G (Addgene #12259, RRID: Addgene_12259) via FuGene 6 (Promega #E2691). HEK293T cells were incubated with the transfecting agents for 24 hours, and the medium was replaced with fresh medium supplemented with 10% FBS. After 48 hours, the conditioned medium was collected, filtered using a 0.45 μm filter, and stored at a −80°C freezer for future use.

For virus transduction, cells were cultured in the appropriate medium with the addition of the conditioned medium containing desired lentivirus and 8.0 μg/mL polybrene. The medium was replaced with fresh medium after 24 hours incubation. After an additional day, cells were then selected using 2.0 – 3.0 μg/mL puromycin (Thermo Fisher # A1113803) for 2 days, or 7.5 mg/mL blasticidin (Corning # 30–100-RB) for 7 days. For inducible CRISPR knockdown using TLCV2 construct, knockout was induced by culturing cells with 2.0 – 4.0 μg/mL doxycycline (Sigma #D5207) for at least 5 days. Cells were then subjected to further western blot, proliferation, apoptosis, spheroid, metabolomic, and RNA sequencing analysis.

CRISPR-Cas9 Screen

A single-guide RNA (sgRNA) library targeting 2,240 genes was generated and provided by Dr. Kris Wood (Duke University) (37). PANC-1 cells were transduced with this library at multiplicity of infection (MOI) of 0.2 via lentivirus, followed by puromycin selection (3.0 μg/mL) for 2 days. Cells were harvested immediately after selection (day 0) and after 7 additional days in culture (day 7). Subsequently, cells were split into two groups: one treated with 10 nM G12Di and the other with DMSO. Both groups were continuously cultured under the indicated conditions for 4 weeks before harvested (day 35). Genomic DNA from samples collected at days 0, 7, and 35 was extracted using DNeasy Blood & Tissue Kit (Qiagen # 69504) and subjected to next generation sequencing (NGS). MAGeCK algorithm was used to determine the enrichment of sgRNA for each gene within the library(93).

In Vitro Proliferation Assays

Five hundred – 3,000 cells were seeded in a well of a 96-well plate. Cells were treated with MRTX1133 (Selleckchem #E1051, or Wuxi Network Lab), RMC-7977 (MedChemExpress # HY-156498), AI-1 (Sigma # 492041), CDDO-Me (Sigma # SMB00376), CB-839/GLSi (Selleckchem #S7655), and DRP-104 (Selleckchem #E1203, or Wuxi Network Lab), indicated conditions and incubated for 5 days. Cell viability was measured using Calcein-AM, Hoechst or Cell TiterGlo cell viability assay. For Calcein-AM or Hoechst assay, cell culture medium was aspirated followed by the addition of 50 μL, 0.50 ng/μL Calcein-AM (Thermo Fisher #C34852) or 2 mM Hoechst (Thermo Fisher # H1398) dissolved in phosphate-buffered saline (PBS) to each well of the 96-well plates and incubated in 37°C for 15 minutes. After incubation, the fluorescence images were taken using a plate reader (SpectraMax i3X MiniMax 300 Spectrometer, Molecular Devices; or Cytation 1 Cell Imaging Multi-Mode Reader, BioTek) at 465 nm excitation and 541 nm emission for Calcein AM fluorescence, or at 360 nm excitation and 450 nm emission for Hoechst. The individual cell numbers were counted by fluorescence images.

For CellTiter cell viability assay, CellTiter Luminescent cell viability assay (Promega #G7570) was used for 2D cell culture, and CellTiter-Glo 3D cell viability Assay (Promega #G9683) was performed for 3D spheroid or organoid cultures per manufacturer’s protocol. In brief, a volume of CellTiter reagents equivalent to the existing volume of culture medium in the well was added to the well containing cells or organoids and was mixed by vigorous pipetting for at least 20 times. The plate was then incubated at room temperature for 30 minutes to allow the luminescent signal developed. After the incubation, the solution was transferred to opaque, white plates and luminescent signal was quantified by a plate reader.

NMR Metabolomic Analysis

One × 10⁵ cells were seeded on a 12-well plate and settled overnight. To start the experiment, the medium was aspirated, washed by PBS, and replaced by 750 μL of fresh low-glucose DMEM (Thermo Fisher #11885–084) or RPMI supplemented with 10% FBS for 24 hours. The conditioned medium was collected and filtered using a 10 KDa molecular weight cut off Amicon Ultra centrifugal unit for 30 minutes. Four hundred μL flow-through conditioned medium were then mixed with 200 μL of standard solution containing 6 mM trimethylsilylpropanoic acid (TSP) and 3 mM imidazole prepared in 99% D2O. The NMR spectrums were collected using a Bruker Advance III 700 mHz Spectrometer. The metabolites were determined and quantified using Chenomx.

Western Blot Analyses

Cells were lysed by lysis buffer (20 mM pH 7.4 Tris-HCl, 150 mM NaCl, 2 mM EDTA, and 1% Triton X-100), and the total protein concentrations of lysates were quantified by Bradford Assay. Equal amounts of cell lysates (based on total protein mass) were loaded and resolved on SDS-PAGE gels, transferred to PVDF membranes, and blocked by 5.0% bovine serum albumin (BSA) in TBST buffer (20 mM Tris, 2.7 mM KCl, 140 mM NaCl, and 0.05% Tween-20). The membrane was incubated with the primary antibody with 1000X dilution in TBST in 4°C overnight. Subsequently, the membrane was incubated with the HRP-conjugated secondary antibody with 5000X dilution in TBST in room temperature for 1 hour. The membrane was then washed, incubated with ECL substrate, and imaged using a Bio-Rad ChemiDoc MP imager. The vendors, catalog numbers, and RRIDs of antibodies used in this study are documented in the Supplementary Table 4.

siRNA Knockdown

One × 10⁵ cells were first seeded in a well of six-well plates and incubated in the incubator overnight. siRNAs targeting SLC7A11 (Horizon Discovery, siSLC7A11–1, UUGUAGAGCUUGCGUAGAA, # J-007612–09; siSLC7A11–2, CGGCAAACUUAUUGGGUCU, # J-007612–10) or siRNA with non-specific sequence (Thermo Fisher #4390844) were mixed with OptiMEM (Thermo Fisher #31985088) and lipofectamine RNAiMAX (Thermo Fisher #13778150) and incubated in room temperature for 15 minutes. The siRNA mixture was then introduced to the cells and incubated for 24 hours. Then, medium containing siRNA was removed, and cells were subjected to western blot or proliferation assay for further analysis.

Cell Death Assay

Seven thousand - 10,000 PDAC cells were seeded in a 96-well plate with appropriate medium supplemented with CellTox Green Express (Promega #G8731). Cells were incubated overnight, and the baseline fluorescence signals were read using a plate reader at 485 nm excitation and 535 nm emission. Cells were then treated with indicated concentrations of G12Di. After 3 days, the final fluorescence signals were read. The normalized apoptosis flux is defined as:

$$CellToxFlux=\frac{Finalsignal-Baselinesignal}{BaselineSignal}$$

Oxidative Stress Assay

The oxidative stress assay was determined using CellROX^™ Green (Thermo Fisher # C10444) per manufacturer’s protocol. In brief, 50,000 to 100,000 cells were seeded in petri dishes or six-well plates and allowed to adhere overnight. Cells were then treated with the indicated concentrations of drugs for 72 hours. Following the treatment, cells were stained by 2.0 μM CellROX Green reagent or DMSO for 1 hour. Cells were then washed twice with PBS, harvested, and pelleted by centrifugation at 300 × g for 5 minutes. Cell pellets were resuspended in DMEM and analyzed by flow cytometry using a CytoFLEX S flow cytometer (Beckman Coulter, USA) with 488 nm excitation and 525 nm emission. A minimum of 20,000 cells were recorded to quantify ROS levels.

Cell Cycle Assay

Cells were seeded in petri dishes and allowed to adhere overnight. Cells were then treated with the indicated conditions of drugs for 72 hours. After the treatment, culture media and cells were collected, washed by PBS, and pelleted under 300 × g centrifugation for 5 minutes. Cell pellets were resuspended in 400 μL of ice-cold PBS. To fix the cells, 800 μL of ice-cold 100% ethanol was added and mixed gently, and the samples were incubated at 4°C for two hours to ensure complete fixation. Fixed cells were centrifuged to remove ethanol and incubated with FxCycle^™ PI/RNase Staining Solution (Thermo Fisher # F10797) for 15–30 minutes at room temperature, protected from light. The stained cells were analyzed by a CytoFLEX S flow cytometry with 561 nm excitation and 610 nm emission. A minimum of 20,000 cells were recorded. Cell debris was excluded by SSC-A versus FSC-A gating, and single cells population were identified by FSC-H versus FSC-A gate. A histogram of propidium iodide signal was generated from the singlet cell population, and the G1, S, and G2/M phases of the cell cycle were determined using FCS Express 7 Research (De Novo Software).

Glutaminase Activity Assay

Glutaminase activity was determined by a two-step reaction coupling NAD⁺ reduction to glutamine hydrolysis using the mitochondria fraction of cells, as described previously (94,95). For mitochondrial isolation, cells were cultured in petri dish plates until they reached 90–100% confluency. The cells were washed with PBS, harvested, pelleted by centrifugation (600 × g) for 5 minutes, and resuspended in 1000 μL ice-cold isolation buffer (10 mM pH 7.4 Tris-MOPS, 1 mM EGTA, 10 mM sucrose). The cells were homogenized using a Dounce homogenizer in ice and centrifuged at 600 × g for 10 minutes at 4°C. The supernatant was collected and centrifuged at 7,000 × g for 10 minutes at 4°C, and the resulting pellet containing mitochondria was washed with isolation buffer once and resuspended in isolation buffer.

To determine the glutaminase activity, 10 μg mitochondrial fractions were brought to a final volume of 10 μL by the isolation buffer containing GLSi at the indicated concentrations or DMSO and incubated at 37 °C for 10 minutes. Then, 105 μL of the first reaction mixture (65 mM pH 8.6 Tris-acetate, 0.25 mM EDTA, 75 mM NaCl, 17.4 mM L-glutamine, 130 mM K₂HPO₄) were added to initiate the reaction of glutamine to glutamate at 37°C for 30. The first reaction was quenched by the addition of 10 uL of 3.00 M HCl on ice. Subsequently, 20 μL of the first reaction mixture was then mixed with 200 μL of the second reaction mixture (130 mM pH 9.4 tris-HCl, 0.36 mM ADP, 1.80 mM NAD⁺, 1 unit of glutamate dehydrogenase (Sigma #10197734001)) and incubated at room temperature for 45 minutes to convert glutamate to 2 α-ketoglutarate, coupled with NAD+ to NADH conversion. The absorbance of NADH for the final reaction mixture was measured at 340 nm to determine the glutaminase activity

Clonogenic Growth Assay

Five thousand – 10,000 cells were seeded in a well of a six-well plate and settled overnight. Cells were treated with indicated concentrations of G12Di, and fresh medium with G12Di were replenished every 4–5 days. After 14 days, cells were washed by PBS twice and fixed by 3.7% formaldehyde. The fixed cells were then stained by 0.5% crystal violet for 10 minutes, washed by PBS three times, and imaged using a Typhoon FLA9500 biomolecular imager.

RNA Sequencing

For PDAC studies, 10⁵ cells were seeded in 6-cm petri dishes. The cells were treated with GI₅₀ concentrations of G12Di or DMSO for 3 days. The cells were snap-frozen by liquid nitrogen, and the RNA was extracted using Qiagen RNeasy plus kit (Qiagen # 74134). The samples were frozen in −80°C and submitted to Novogene (USA) for poly A enrichment, directional mRNA library preparation using NEBNext^® Ultra^™ II Directional RNA Library Prep Kit (New England Biolabs #E7760), and next generation sequencing using a Novaseq 6000 sequencer.

For KP lung cancer models, 10³ cells were seeded as spheroids in media containing KI696 and treated with G12Di for 4 days. The spheroids were harvested and snap-frozen, and RNA was extracted using a Purelink RNA mini kit (Invitrogen). mRNA was converted to cDNA by using a SMARTer PCR cDNA synthesis kit (Clonetech), followed by sequencing library preparation via Nextera XT DNA library preparation kit (Illumina). Samples were pooled at equimolar ratios, loaded on SP11 cycle flow cells, and sequenced on Illumina NovaSeq600.

The quality of the NGS was verified by FastQC (v0.12.1). The fastq files were aligned to human reference genome (GRCh38.p14 or GRCm38) by STAR (v2.7.11b). The raw counts of mRNA transcripts were quantified using featureCounts within Samtools (v1.22) package or RSEM (v1.3.3). The following analyses and packages were performed in R environment: differential expression analyses for each paired condition were done using DESeq2 (v1.42.1) or limma (v3.58.1); and the pathway enrichment analyses were conducted by fgsea (v1.28.0), clusterProfiler (v4.10.1), and Singscore (v1.22.0) (96). Motif analyses were defined from −1000 to +1000 nucleotides from the transcription start site (TSS) by HOMER (v5.1) (48).

ChIP-seq Analysis

Optimal IDR thresholded peak files from NRF2/NFE2L2 chip-sequencing experiments were downloaded from ENCODE for HeLa, A549, IMR-90, and Hep G2 cell lines (Acc: ENCFF305KIK, ENCFF418TUX, ENCFF474PPT, ENCFF882YL0). Promoters were defined based on Hg38 Gencode (v30) (97) annotations and summarized to genes using the R package TxDB.Hsapiens.UCSC.hg38.knownGene (v3.21.0). Overlaps between gene promoters and peak files were conducted with the IRanges (v2.38.1) and GenomicRanges (v1.56.2) R packages (98). Gene promoters were prioritized based on frequency of peak occurrence across the four cell line experiments and annotated using BiomaRt (v2.50.3) (99).

Network Analysis

Differential analysis were carried out by limma on RNA-seq data in KEAP1^KO PDAC cells, PDAC cells treated with siRNA targeting KRAS (KRAS UP) (45) or MYC (MYC UP) (62), or cells treated with TEAD inhibitors (GNE-7883, TEAD UP; GSE273175) (63). The log₂FC is normalized by scaled adjusted P values using the following equation to generate weighted log₂FC. Genes with expressions significantly upregulated (adj. p < 0.05, log₂FC > 0) were filtered and subjected to network analysis. The network analysis was conducted using R packages tidygraph (v1.3.1) and igraph (v2.1.4), plotted by ggraph (v2.2.1).

Cancer Gene dependency scores, Mutational Data, and RNA expression

The gene dependency scores, mutational data of cancer cell lines for PDAC and LUAD were acquired from Cancer Dependency Map (DepMap). The data were accessed via R package depmap (v1.16.0). The RNA expressions for TCGA-LUAD cohort were acquired from TCGA database.

Mouse Study

All mouse experiments described in this study were approved by the Institutional Animal Care and Use Committee (IACUC) at New York University (NYU). Mice were housed according to IACUC guidelines in ventilated cages in a pathogen-free animal facility. For all mouse studies, ≥ 4 mice were used for each condition. All transplantation experiments were performed using C57BL/6 J WT (JAX strain 000664) or NOD SCID Gamma (NSG; JAX strain 005557 F) mice. For cell line-derived xenograft experiments, 500,000 cells suspended in 100 μl PBS were injected subcutaneously into the flanks of the experimental mice. For the PDAC patient-derived xenograft experiment, PDAC organoids cultures were homogenized and injected subcutaneously into NSG mice, and after tumor outgrowth, the mice were euthanized, the tumors excised, cut to 25 mm³ pieces and surgically implanted into naive NSG hosts. Tumors were measured with calipers, and volume was calculated based on 0.5 (length × width²). Mice were euthanized when tumors reached the maximum volume of 2,000 mm³ permitted by IACUC at NYU, or sooner if the tumors reached pre-determined endpoints.

For each experiment, once the tumors of a mouse treatment cohort reached 100 – 250 mm³, mice were randomized into respective treatment groups and dosed at the indicated conditions until experimental endpoint. G12Di was dosed at 10 mg/kg or vehicle (50 mM pH 5.0 sodium citrate + 10% captisol) by intraperitoneal injection (daily). Daraxonrasib (RMC-6236) (MedChemExpress # HY-148439) was dosed at 3 mg/kg or vehicle (10% DMSO, 20% PEG400, 10% Solutol HS 15, in water) by oral gavage (daily). RMC-7977/RASi was dosed at 10 mg/kg (same vehicle as daraxonrasib) by oral gavage (daily). DRP-104 was dosed at 3 mg/kg or vehicle (10% Tween 80, 10% ethanol, in 0.9% saline) by subcutaneous injection (five days on, two days off schedule).

Statistical Analyses

Computational analyses were conducted using R (v4.3.2) and RStudio (v2024.9.0.375). All experiments are reported with at least three independent samples. Quantitative results are reported as mean ± standard error of the mean (SEM). Statistical analyses are performed using GraphPad Prism software, and significance is evaluated using Student’s t-tests, Wilcoxon test, or ANOVA. Significance thresholds are defined as follows: *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001.

SIGNIFICANCE:

KEAP1 loss is associated with reduced response to KRAS inhibitor therapy. We demonstrate that KEAP1 loss-associated resistance can be overcome by pharmacologic inhibition of the KEAP1 loss-induced glutamine dependency, establishing a combination to enhance RAS inhibitor clinical efficacy.

Data Availability

Additional data generated in this study are available from the corresponding author upon request.