Comutations and KRAS^G12C Inhibitor Efficacy in Advanced NSCLC

Activating mutations in the KRAS proto-oncogene are detected in 25% to 30% of nonsquamous non–small cell lung cancer (NSCLC) and most frequently (∼42%) involve a glycine to cysteine substitution at residue 12 (G12C) as a result of a smoking-related G>T transversion (1). Replacement of glycine in codon 12 of KRAS is thought to sterically hinder insertion of the arginine finger (R-finger) of canonical GTPase activating proteins (GAP; such as neurofibromin and p120RasGAP) into the GTPase active site and impairs GAP-stimulated GTP hydrolysis (2), thus shifting the KRAS nucleotide cycling equilibrium toward the active, GTP-bound state. For over 30 years since its initial discovery, KRAS remained an elusive therapeutic target due to (i) picomolar binding affinity for its guanine nucleotide substrates coupled with a high intracellular concentration of GTP, thus precluding the development of competitive inhibitors; (ii) a featureless protein surface devoid of deep pockets suitable for docking of small-molecule inhibitors; (iii) on-target toxicity from wild-type KRAS inhibition or concurrent targeting of the downstream effector RAF/MEK/ERK and PI3K/AKT/MTOR pathways; (iv) paradoxical increase in RAS signaling with downstream pathway inhibitors due to release of negative feedback; and (v) redundant prenylation pathways that control KRAS plasma membrane localization (3). The groundbreaking identification of compounds and subsequent development of covalent allosteric inhibitors that bind irreversibly to cysteine 12 and occupy a cryptic induced pocket in the switch II region of GDP-bound KRAS, trapping the oncoprotein in its inactive conformation, has enabled effective inhibition of KRAS^G12C (4, 5). Sotorasib (formerly AMG510), the first-in-class KRAS^G12C inhibitor (KRAS^G12Ci), and adagrasib (formerly MRTX849) both yielded robust single-agent clinical activity in previously treated patients with advanced KRAS^G12C-mutant NSCLC, producing objective response rates (ORR) of 37% to 43% in single-arm registrational phase II studies (6, 7). Based on these results, both sotorasib and adagrasib received FDA accelerated approval for previously treated patients with advanced KRAS^G12C-mutant NSCLC; furthermore, sotorasib improved progression-free survival (PFS) and ORR compared with docetaxel in the randomized phase III CodeBreaK 200 trial (8). Several additional KRAS^G12C inhibitors are undergoing clinical development, with initial reports indicating comparable single-agent activity (9–12).

Despite promising ORR, KRAS^G12Ci produce a median PFS of approximately 6 to 7 months (6, 7), which is inferior to what has been reported for targeted therapies in other oncogene-addicted NSCLC subsets (e.g., EGFR mutations or ALK rearrangements; refs. 13, 14). For individual patients, clinical outcomes with KRAS^G12Ci vary widely from long-term durable responses and prolonged survival—with a 2-year overall survival (OS) rate of 32.5% reported in CodeBreaK 100—to early disease progression seen in ∼5% to 16% of treated patients (6, 7, 15). De novo as well as adaptive and acquired resistance collectively curtail the efficacy of KRAS^G12Ci monotherapy (7, 15–20), and support the need for improved patient selection for sotorasib or adagrasib monotherapy and for combination regimens directed at treatment intensification. However, molecular or clinical determinants of distinct clinical outcomes with KRAS^G12Ci are hitherto poorly defined, and validated markers for patient stratification prior to treatment initiation are lacking. Co-occurring genomic alterations in key tumor suppressor genes underpin the molecular diversity of KRAS-mutant NSCLC and impact both tumor cell–intrinsic and nontumor cell–autonomous cancer hallmarks, including shaping its immune contexture (21, 22). Critically, comutations can impact responses to standard-of-care systemic therapies, including both chemotherapy and immunotherapy (22–26). Here, we systematically dissected the impact of genomic and clinical features on outcomes with KRAS^G12Ci in the largest cohort to date of NSCLC patients treated with sotorasib or adagrasib, encompassing 424 patients from 21 centers in the United States and Europe. We demonstrate that prevalent coalterations in KEAP1, SMARCA4, and CDKN2A are associated with inferior clinical outcomes with KRAS^G12Ci therapy and collectively define a subgroup of patients with poor prognosis. In addition, we identify less prevalent candidate genomic modifiers of KRAS^G12Ci efficacy and propose a framework for patient stratification with implications for treatment selection and clinical trial development for KRAS^G12C-mutant NSCLC.

In order to comprehensively interrogate the impact of baseline clinicogenomic parameters on clinical outcomes with KRAS^G12Ci, we assembled the largest cohort to date of patients with KRAS^G12C-mutant NSCLC who were treated with sotorasib or adagrasib, encompassing 424 unique evaluable patients across 21 centers in the United States and Europe (Supplementary Table S1). The study cohort was established by merging two independently collected retrospective cohorts [cohort A (N = 330) and cohort B (N = 94)], which were also analyzed separately to provide additional validation of key findings (Supplementary Tables S2 and S3; see Methods for detailed study eligibility criteria). In the overall cohort, the median age was 68 years, patients were predominantly current or former smokers (96.9%), and most patients had Eastern Cooperative Oncology Group performance status (ECOG PS) 0 to 1 (82.1%). Adenocarcinoma was the most common histology (92.7%). All patients had metastatic disease at the start of KRAS^G12Ci therapy, and 35.2% had a history of brain metastases (26.2% previously treated, 9.0% untreated). The majority of patients received treatment with sotorasib (83.3%). Most patients received prior treatment with PD-1/PD-L1 inhibitors and platinum-based chemotherapy (75.9%). This cohort was overall representative of the general population of patients with KRAS^G12C-mutant NSCLC (6, 7). Most patients had genomic profiling performed on tumor tissue (62.3%), 18.2% had genomic profiling results from the liquid biopsy, and 13.7% had both tumor and liquid biopsy profiling; 5.8% of patients had confirmed KRAS^G12C status from analysis of tumor DNA but did not undergo next-generation sequencing–based profiling. Patient characteristics for the overall study cohort are summarized in Supplementary Table S1. In the overall cohort, ORR was 34.0% [95% confidence interval (CI), 29.4–38.8], median PFS was 5.2 months (95% CI, 4.7–5.6), and median OS was 10.7 months (95% CI, 8.8–12.6; Fig. 1A). The estimated 12-month PFS and OS rates were 22.2% and 46.3%, respectively, whereas the estimated 24-month PFS and OS rates were 6.4% and 23.3%, respectively. We observed similar results when analyzing the individual cohorts separately (Supplementary Fig. S1A and S1B). ECOG PS of 1 or 2 was associated with shorter PFS and OS compared with PS 0, and patients with a history of brain metastases had worse PFS and OS with KRAS^G12Ci therapy compared with those without prior history of brain metastasis (Fig. 1B). No difference in PFS and OS was observed depending on the KRAS^G12Ci used (Fig. 1B). When the analysis was limited to previously treated patients with ECOG PS 0 to 1 and either absent or treated and stable brain metastases at start of KRAS^G12Ci therapy (comparable with the patient population enrolled in the registrational CodeBreaK 100 and KRYSTAL-1 clinical trials; refs. 6, 7), the ORR was 35.0% (95% CI, 29.1–41.1), the median PFS was 5.5 months (95% CI, 4.9–6.0) and the median OS was 11.4 months (95% CI, 8.8–14.1; Supplementary Fig. S2A). Patients with untreated brain metastases had similar survival compared with those with previously treated brain metastases [PFS: 5.0 vs. 4.3 months, log-rank P = 0.964, multivariable (MV) hazard ratio (HR) 0.95 (95% CI, 0.82–1.44); OS: 8.8 vs. 7.8 months, log-rank P = 0.741, MV HR 1.13 (95% CI, 0.68–1.88); Supplementary Fig. S2B]. Tumor cell PD-L1 expression and exposure to immune-checkpoint inhibitors in prior line(s) of therapy were not associated with PFS or OS (Fig. 1B; Supplementary Fig. S2C and S2D).

To dissect the impact of the tumor comutational landscape on clinical outcomes with KRAS^G12Ci, we first classified patients into subgroups with durable clinical benefit (PFS ≥6 months; N = 131) or early progression (PFS ≤3 months; N = 124; total N = 255; ref. 18). Patients censored with less than 3 months of follow-up were excluded from this analysis. We then performed an unbiased enrichment analysis of the most prevalent coalterations (detected in at least 5% of patients) in the overall study cohort (see Methods for additional details). We found that comutations in three tumor suppressor genes were significantly enriched in the early progression subgroup: KEAP1 [Fisher exact test P < 0.001, false discovery rate (FDR) q = 0.004], SMARCA4 (Fisher exact test P = 0.001, FDR q = 0.010), and CDKN2A (Fisher exact test P = 0.006, FDR q = 0.034; Fig. 2A). Patients bearing KEAP1 comutated tumors (KEAP1^MUT) exhibited significantly shorter PFS [2.8 vs. 5.4 months, log-rank P < 0.001, MV HR 2.26 (95% CI, 1.60–3.19)] and OS [6.3 vs. 11.1 months, log-rank P < 0.001, MV HR 2.03 (95% CI, 1.38–2.99)] compared with those harboring KEAP1 wild-type (KEAP1^WT) NSCLC (Fig. 2B). SMARCA4 comutations were associated with markedly worse PFS and OS compared with SMARCA4 wild-type [SMARCA4^MUT vs. SMARCA4^WT PFS: 1.6 vs. 5.4 months, log-rank P < 0.001, MV HR 3.04 (95% CI, 1.80–5.15); OS: 4.9 vs. 11.8 months, log-rank P < 0.001, MV HR 3.07 (95% CI, 1.69–5.60); Fig. 2C]. Coalterations in CDKN2A were also associated with worse PFS and OS upon treatment with KRAS^G12Ci compared with CDKN2A wild-type [CDKN2A^MUT vs. CDKN2A^WT PFS: 3.4 vs. 5.3 months, log-rank P < 0.001, MV HR 1.98 (95% CI, 1.32–2.97); OS: 6.4 vs. 10.7 months, log-rank P = 0.009, MV HR 1.66 (95% CI, 1.03–2.68); Fig. 2D]. Similar findings were observed when cohorts A and B were analyzed separately (Supplementary Fig. S3A–S3C) and when limiting the analysis only to patients who received prior immune-checkpoint inhibitor therapy (Supplementary Fig. S4A–S4C). KEAP1 comutations were associated with numerically lower ORR compared with KEAP1^WT, whereas there was no significant difference in ORR between patients with SMARCA4^MUT versus SMARCA4^WT and CDKN2A^MUT versus CDKN2A^WT NSCLC (Fig. 2B–D).

STK11 was the fourth most enriched somatically mutated gene in patients with early progression with KRAS^G12Ci (Fisher exact test P = 0.019, FDR q = 0.082; Fig. 2A). Patients with STK11^MUT NSCLC had shorter PFS compared with patients who harbored STK11^WT tumors [4.4 vs. 5.5 months, log-rank P = 0.010, MV HR 1.32 (95% CI, 1.00–1.73)]. No significant difference was observed between patients bearing STK11^MUT and STK11^WT tumors for OS [9.8 vs. 10.5 months, log-rank P = 0.167, MV HR 1.18 (95% CI, 0.85–1.64)] or ORR (31.5% vs. 34.3%, Fisher exact P = 0.616; Fig. 3A). Because STK11 and KEAP1 mutations frequently overlap in KRAS-mutant NSCLC (21), we sought to deconvolute their individual impact by comparing outcomes with KRAS^G12Ci in three distinct genomi­cally defined subgroups: (i) KRAS^G12C/KEAP1^WT/STK11^WT; (ii) KRAS^G12C/KEAP1^WT/STK11^MUT; and (iii) KRAS^G12C/KEAP1^MUT/STK11^{MUT or WT}. The KRAS^G12C/KEAP1^MUT/STK11^{MUT or WT} subgroup exhibited significantly shorter PFS and OS compared with the KRAS^G12C/KEAP1^WT/STK11^WT subgroup [PFS: 2.8 vs. 5.3 months, log-rank P < 0.001, MV HR 2.30 (95% CI, 1.60–3.30); OS: 6.3 vs. 10.7 months, log-rank P < 0.001, MV HR 2.13 (95% CI, 1.41–3.20)], and numerically lower ORR (22.0% vs. 34.9%, Fisher exact test P = 0.114). The KRAS^G12C/KEAP1^WT/STK11^MUT and KRAS^G12C/KEAP1^WT/STK11^WT subgroups had similar PFS [KRAS^G12C/KEAP1^WT/STK11^MUT vs. KRAS^G12C/KEAP1^WT/STK11^WT PFS: 5.6 vs. 5.3 months, MV HR 1.03 (95% CI, 0.74–1.46)], OS [12.3 vs. 10.7 months, MV HR 1.05 (95% CI 0.69–1.61)], and ORR (40.6% vs. 34.9%; Fig. 3B). Similar results were observed when cohorts A and B were analyzed separately (Supplementary Fig. S5A and S5B). Further deconvolution of patients with KEAP1^MUT NSCLC based on STK11 mutation status yielded similar findings; each of the KRAS^G12C/KEAP1^MUT/STK11^MUT and KRAS^G12C/KEAP1^MUT/STK11^WT subgroups exhibited worse PFS and OS when compared with KRAS^G12C/KEAP1^WT/STK11^MUT or KRAS^G12C/KEAP1^WT/STK11^WT subgroups (Supplementary Fig. S5C). We therefore conclude that STK11 comutations without concurrent KEAP1 mutations may not significantly influence outcomes with KRAS^G12Ci monotherapy. This finding was also upheld when the analysis was limited to KEAP1^WT, SMARCA4^WT, and CDKN2A^WT (KSC^WT) tumors (Supplementary Fig. S5D).

TP53 was the most frequently comutated gene in the overall cohort (45.7%), but TP53 mutations were not associated with clinical outcomes with KRAS^G12Ci (Supplementary Fig. S6A). This was further validated when cohorts A and B were analyzed separately (Supplementary Fig. S6B and S6C).

Next, we interrogated our patient cohort to determine less prevalent functionally related comutations that are enriched in patients with early progression or durable clinical benefit. We focused this analysis on an expanded set of genes that were comutated in at least three patients. Due to the large-size dominant effects of KEAP1, SMARCA4, and CDKN2A (KSC^MUT), this analysis was limited to patients whose tumors were KSC^WT (N = 128). CHEK2 and ATRX comutations were enriched in patients with durable clinical benefit with KRAS^G12Ci [odds ratio (OR) ≤ −2.0], whereas tumors harboring (i) KRAS amplification, (ii) comutations in TSC1, TSC2, MTOR, or PTEN, encoding components of the PI3K/AKT/MTOR pathway, and (iii) comutations in some additional driver oncogenes (such as ALK, ROS1, and NTRK3) were enriched in patients with early progression (OR ≥2.0; Fig. 4A).

We further examined the association of comutations in the identified candidate genes and clinical outcomes with KRAS^G12Ci in the evaluable population for each individual gene. Patients whose tumors harbored comutations in CHEK2 had longer PFS compared with those whose tumors were CHEK2^WT, and median OS was not reached in patients harboring CHEK2^MUT NSCLC (Supplementary Fig. S7A). CHEK2 is a tumor suppressor gene that encodes a serine/threonine kinase involved in signal transduction in the cellular response to DNA double-strand breaks (DSB; ref. 27). We then further explored the impact of somatic genomic alterations in a group of well-validated DNA damage response (DDR) genes: BRCA1/2, ATM, ATR, CHEK1/2, PALB2, RAD50/51/51B/51C/51D. Alterations in this group of DDR genes were present in 32.1% of patients. Patients whose tumors harbored DDR gene comutations had higher ORR (52.2% vs. 27.7%, Fisher exact test P = 0.001; Fig. 4B) and significantly longer PFS with KRAS^G12Ci compared with patients whose tumors were DDR gene wild-type [5.9 vs. 4.6 months, log-rank 0.030, HR 0.68 (95% CI, 0.48–0.97)], although there was no statistically significant difference in OS between patients harboring DDR gene comutated and wild-type tumors [13.0 vs. 8.4 months, log-rank P = 0.075, HR 0.69 (95% CI, 0.46–1.04); Fig. 4C]. Somatic mutations in ATRX were also enriched in patients with durable clinical benefit with KRAS^G12Ci therapy (Fig. 4A) and were associated with longer PFS and OS with KRAS^G12Ci therapy when compared with patients bearing ATRX wild-type tumors (Supplementary Fig. S7B). ATRX encodes an ATP-dependent chromatin remodeling protein, a member of the SWI/SNF family, that interacts with the histone chaperone DAXX to deposit the variant histone H3.3 at sites of nucleosome turnover (28). The ATRX/DAXX complex has been implicated in transcriptional regulation and control of DNA replication, recombination, and repair (28, 29). Patients whose tumors harbored somatic mutations in the ATRX/DAXX genes had longer PFS and OS with sotorasib and adagrasib compared with patients whose tumors were ATRX/DAXX wild-type (Fig. 4D).

Patients with NSCLC harboring additional (beyond the qualifying KRAS^G12C mutation) coalterations in RAS genes (KRAS/NRAS/HRAS, including both somatic mutations and/or gene amplifications) prior to starting KRAS^G12Ci therapy exhibited worse PFS and OS compared with those bearing tumors without additional RAS gene alterations in the mutation-evaluable population (Fig. 4E) as well as in the KSC^WT population (Supplementary Fig. S8A). Presence of comutations in a group of functionally related PI3K/AKT/MTOR pathway genes (including AKT1, PIK3CA, MTOR, TSC1/2, and PTEN) was not associated with survival in the overall mutation-evaluable population (Supplementary Fig. S8B). This may be attributable to the large-size dominant effects of co-occurring KEAP1, SMARCA4, and CDKN2A mutations on clinical outcomes with KRAS^G12Ci therapy (Fig. 2B–D). Therefore, we tested the association of PI3K/AKT/MTOR pathway genes with survival in the KSC^WT population and observed that patients whose tumors harbored PI3K/AKT/MTOR pathway comutations had significantly shorter PFS with sotorasib and ada­grasib compared with patients harboring PI3K/AKT/MTOR pathway gene wild-type tumors (Fig. 4F). We also found that patients whose tumors harbored missense mutations in ROS1, ALK, and NTRK1–3 oncogenes—assessed together—had shorter PFS and OS with KRAS^G12Ci therapy compared with patients whose tumors were ROS1, ALK, and NTRK1–3 wild-type (Supplementary Fig. S8C and S8D). Additional comutated genes that were enriched in patients with early progression included LRP1B, KDM5C, FAT1, NOTCH2, NFE2L2, FLT1, and RAD50 (Fig. 4A). However, none of these genes were associated with survival with KRAS^G12Ci therapy [LRP1B^MUT vs. LRP1B^WT: 3.0 vs. 5.1 months, log-rank P = 0.585, HR 1.24 (95% CI, 0.57–2.67); KDM5C^MUT vs. KDM5C^WT: 2.2 vs. 5.3 months, log-rank P = 0.143, HR 1.83 (95% CI, 0.80–4.19); FAT1^MUT vs. FAT1^WT: 4.1 vs. 4.7 months, log-rank P = 0.263, HR 1.66 (95% CI, 0.68–4.09); NOTCH2^MUT vs. NOTCH2^WT: 1.9 vs. 4.8 months, log-rank P = 0.427, HR 1.39 (95% CI, 0.61–3.16); NFE2L2^MUT vs. NFE2L2^WT: 5.5 vs. 4.7 months, log-rank P = 0.701, HR 1.17 (95% CI, 0.52–2.65); FLT1^MUT vs. FLT1^WT: 2.8 vs. 4.7 months, log-rank P = 0.187, HR 1.93 (95% CI, 0.71–5.24); and RAD50^MUT vs. RAD50^WT: 2.7 vs. 4.7 months, log-rank P = 0.832, HR 1.13 (95% CI, 0.36–3.59)]. These findings collectively suggest that baseline coalterations in RAS genes and PI3K/AKT/MTOR pathway genes may exert a deleterious effect on clinical outcomes with KRAS^G12Ci. It remains unclear if individual missense mutations in oncogenic drivers such as ALK, ROS1, and NTRK1–3 are functional and expressed in the absence of corresponding gene rearrangements. Meanwhile, somatic mutations in genes involved in DDR and chromatin remodeling/epigenetic regulation may have favorable impact on treatment outcomes with KRAS^G12Ci. Due to the exploratory nature of this analysis, these findings warrant validation in subsequent preclinical and clinical studies.

Next, we aimed to determine the prevalence and overlap of enriched comutations in patients with either early progression or durable clinical benefit with KRAS^G12Ci in order to further explore their clinical relevance and interrelationships. For this purpose, we focused our analysis on patients whose tumors underwent comprehensive NGS profiling (≥400 covered genes). As expected, KEAP1, SMARCA4, and CDKN2A coalterations were prevalent in patients with early disease progression (Fig. 5A). In KSC^WT tumors, additional alterations in RAS genes (KRAS, NRAS, and HRAS), mutations in PI3K/AKT/MTOR pathway genes (AKT1, PIK3CA, MTOR, TSC1/2, and PTEN), and somatic mutations in select driver oncogenes (ALK, ROS1, and NTRK1/2/3) were identified in 11.1% (3/27), 18.5% (5/27), and 33.3% (9/27) of patients with early disease progression, respectively (Fig. 5B). Comutations in the individual pathway genes PTEN, TSC1, TSC2, and MTOR were identified in 2%, 4%, 2%, and 4% of patients with early progression, respectively, and were absent in patients with durable clinical benefit (Fig. 5A). Meanwhile, comutations in CHEK2, PALB2, and ATRX were present in 8%, 2%, and 6% of patients with durable clinical benefit and were lacking in those with early progression (Fig. 5A). Comutations in ATRX/DAXX were present in 10%, and comutations in a group of well-validated DDR genes—BRCA1/2, ATM, ATR, CHEK1/2, PALB2, RAD50/51/51B/51C/51D—were present in 40% of patients with durable clinical benefit (Fig. 5C).

Through an unbiased approach, we identified genes that when comutated were associated with early progression with KRAS^G12Ci therapy (Fig. 2A). Prevalent alterations in KEAP1, SMARCA4, and CDKN2A (collectively identified in 32.0% of KRAS^G12C-mutant NSCLC in our overall cohort) were the most enriched in this group and captured 49.3% of patients with early disease progression with KRAS^G12Ci (Fig. 6A). Figure 6A and Supplementary Fig. S9A show the overlap between KEAP1, SMARCA4, and CDKN2A comutations. The KSC^MUT subgroup exhibited numerically lower ORR compared with the KSC^WT subgroup (25.3% vs. 38.1%, Fisher exact test P = 0.065; Fig. 6B). Despite approximately a quarter of patients achieving an early response, PFS and OS were significantly curtailed in the KSC^MUT subgroup compared with the KSC^WT subgroup [PFS: 2.8 vs. 5.9 months, log-rank P < 0.001, MV HR 2.51 (95% CI, 1.79–3.52); OS: 6.9 vs. 13.0 months, log-rank P < 0.001, MV HR 2.05 (95% CI, 1.38–3.02); Fig. 6C]. Furthermore, the KSC^MUT subgroup had markedly inferior 6- and 12-month PFS and OS compared with the KSC^WT subgroup (estimated 6- and 12-month PFS rate: 15.7% vs. 49.5%, and 3.3% vs. 28.5%, respectively; estimated 6- and 12-month OS rate: 54.7% vs. 75.2%, and 27.0% vs. 54.6%, respectively; Fig. 6C). We also observed an incrementally detrimental effect based on the comutation overlap of the KSC genes. Patients whose tumors harbored two or more comutations in any of the KSC genes exhibited significantly worse PFS and OS compared with patients with KSC^WT NSCLC and with those with tumors bearing a single altered KSC gene upon treatment with KRAS^G12Ci (Supplementary Fig. S9B and S9C). Importantly, KEAP1, SMARCA4, and CDKN2A comutations were each independently associated with shorter PFS with KRAS^G12Ci in a multivariable model that also incorporated key clinical characteristics (Supplementary Table S4). KEAP1 and SMARCA4 were also independently associated with shorter OS (Supplementary Table S5). Thus, comutations in KEAP1, SMARCA4, and CDKN2A are robust independent determinants of KRAS^G12Ci efficacy that consistently segregate patients with advanced KRAS^G12C-mutant NSCLC into groups with markedly dissimilar clinical outcomes.

In this study, we identified genomic modifiers of KRAS^G12Ci efficacy in advanced NSCLC through an unbiased clinicoge­nomic analysis of the largest cohort to date of patients treated with sotorasib or adagrasib. Prevalent coalterations in KEAP1, SMARCA4, and CDKN2A were each associated with early disease progression and poor clinical outcomes with KRAS^G12Ci monotherapy—independently of key clinical covariates—and collectively define subgroups of KRAS^G12C-mutant NSCLC patients with markedly divergent therapeutic response trajectories and overall prognosis. Furthermore, in an exploratory analysis, we identified less frequent baseline somatic alterations in RAS genes and PI3K/AKT/MTOR pathway genes as candidate mediators of inferior clinical outcomes with KRAS^G12Ci, whereas grouped alterations in DDR genes and components of the ATRX/DAXX chromatin remodeling complex were associated with prolonged clinical benefit. These findings shed light on the molecular underpinnings of KRAS^G12Ci clinical response heterogeneity in NSCLC and suggest a framework for patient stratification as well as for personalization of KRAS^G12Ci-anchored combination therapeutic strategies (Supplementary Fig. S10).

Examined individually, coalterations in KEAP1, SMARCA4, and CDKN2A were consistently associated with significantly shorter PFS and OS with KRAS^G12Ci in two independently established cohorts of patients with advanced KRAS^G12C-mutant NSCLC, as well as in the overall merged cohort. In contrast, their impact on ORR was more heterogeneous and did not reach statistical significance, although a trend toward lower ORR was observed for KEAP1 mutations. KEAP1 comutations were associated with numerically lower ORR with both sotorasib and adagrasib in the phase II component of the CodeBreaK 100 and KRYSTAL-1 clinical trials, respectively, but in both cases the CIs overlapped (6, 7); surprisingly, higher ORR was reported with adagrasib in patients with CDKN2A^MUT compared with CDKN2A^WT NSCLC in KRYSTAL-1 (7). Biologically, this discrepancy may underlie the emergence of adaptive—rather than primary—resistance, which can develop expeditiously in response to KRAS^G12Ci (16) and manifest as rapid disease progression after initial radiologic response. Therefore, assessment of the impact of comutations based on ORR alone may underestimate or fail to adequately capture their effect on the efficacy of KRAS^G12Ci monotherapy. When assessed together, KSC alterations were identified in 32.0% of patients in the overall cohort and accounted for approximately half (49.3%) of patients who exhibited early disease progression (PFS ≤3 months) with sotorasib or adagrasib. The median PFS in patients with KSC^MUT NSCLC was 2.8 months (compared with 5.9 months in KSC^WT), and the estimated 12-month PFS rate was 3.3% (compared with 28.5% in KSC^WT). Thus, comutations in key tumor suppressor genes delineate subsets of KRAS^G12C-mutant NSCLC with strikingly dissimilar clinical outcomes with KRAS^G12Ci.

A limitation of the current study is that it does not allow for the separation of predictive from prognostic effects of individual genomic alterations. However, both KEAP1 and CDKN2A loss were previously identified as drivers of improved cellular fitness under adagrasib selection in CRISPR/Cas9-based in vitro and in vivo knockout screens, thus supporting a causal—albeit context-dependent—role in mediating KRAS^G12Ci insensitivity (5). KEAP1 encodes an adapter protein that engenders substrate specificity for the CUL3/RBX E3 ubiquitin ligase complex and is critical for the ubiquitylation and proteasomal degradation of NRF2 (encoded by the NFE2L2 gene), a master regulator of cellular antioxidant, anti-inflammatory, and cytoprotective signals (30). Importantly, NRF2 is involved in the transcriptional control of genes encoding efflux transporters as well as several genes involved in xenobiotic detoxification (30). Inactivating KEAP1 somatic mutations have been associated with poor prognosis and inferior clinical outcomes with radiotherapy or chemoradiation (31, 32), platinum-doublet chemotherapy (22, 24, 33), PD-1 axis inhibitor monotherapy (24, 33, 34), and chemoimmunotherapy (25, 26) in NSCLC, particularly in the context of KRAS-mutant tumors (22). Furthermore, KEAP1 depletion promoted resistance to multiple targeted therapies against components of the RTK/RAS/MAPK pathway in NSCLC cell lines by decreasing drug-induced generation of ROS and increasing glutathione synthesis (35). It should be noted that although NRF2 nuclear accumulation is considered the dominant molecular event downstream of KEAP1 loss in terms of carcinogenesis and therapeutic response, several NRF2-independent effects of KEAP1 inactivation have also been recognized (36). Inactivation of CDKN2A alone or in combination with the genetically and functionally related CDKN2B gene as a result of somatic mutation or biallelic deletion (frequently involving both genes as a result of an arm-level event in 9p21) can ostensibly promote KRAS^G12Ci resistance by decoupling cell-cycle progression from signaling downstream of KRAS^G12C. In this context, it is plausible that less prevalent alterations in other components of the cell-cycle machinery may also influence individual responses to KRAS^G12Ci as a result of dysregulated cell-cycle control. Finally, deleterious somatic mutations in SMARCA4 encoding BRG1, one of two possible and mutually exclusive ATP-dependent core catalytic subunits of mammalian SWI/SNF ATP-dependent chromatin remodeling complexes, were previously linked with dedifferentiated histology and an atypical club cell lung cancer cell of origin in genetically engineered mouse models (37). In addition, SMARCA4 somatic mutations portend poor prognosis in patients with both early-stage and advanced NSCLC—particularly among those that harbor KRAS-mutant tumors—although reports of their impact on immune-checkpoint inhibitor efficacy have been conflicting (38–40). The mechanisms by which SMARCA4 loss may modulate response to KRAS^G12Ci are currently unknown, but previously reported pleiotropic functions in the regulation of cellular differentiation, DNA replication, and repair as well as cell-cycle progression are likely to be involved (41). The SMARCA4 genomic locus resides on the short arm of chromosome 19 (19p13.2), in topological proximity to KEAP1 and STK11, thus increasing susceptibility to codeletion events that contribute to the frequent co-occurrence of alterations in the three genes.

Comutations in STK11, when present in the absence of concurrent alterations in KEAP1 (or KEAP1/SMARCA4/CDKN2A), did not affect ORR, PFS, or OS with KRAS^G12Ci. This finding has implications for clinical trial design and interpretation, because STK11 alterations are drivers of poor clinical outcomes with first-line PD-(L)1 inhibitor–encompassing chemoimmunotherapy regimens in advanced NSCLC (25, 26) and constitute an eligibility criterion for clinical trials evaluating KRAS^G12Ci in previously untreated patients (NCT04933695 and NCT03785249). Furthermore, these results argue against a purely prognostic role for STK11 somatic mutations in NSCLC.

In order to identify additional, less prevalent candidate mediators of diverse therapeutic outcomes with KRAS^G12Ci, we adopted a pathway-level approach by initially surveying individual somatically mutated genes that were enriched in either the durable benefit (PFS ≥6 months) or early progression (PFS ≤3 months) subgroups and subsequently assessing their combined impact on KRAS^G12Ci clinical outcomes. This analysis revealed an association of mutations in genes implicated in DNA damage response and repair with improved clinical outcomes with KRAS^G12Ci. Recurrent mutations in two distinct groups of genes were enriched in the durable benefit group including: (i) DDR pathway genes, such as CHEK2, and (ii) ATRX and DAXX. The ATRX/DAXX complex has been implicated in the maintenance of genomic integrity through diverse effects in DNA repair, replication, methylation, gene expression, and telomere homeostasis; accordingly, ATRX- or DAXX-deficient tumors exhibit DNA repair defects and display genomic instability (28, 29, 42, 43). Therefore, convergence on impaired DDR and genome maintenance pathways may underpin the increased KRAS^G12Ci sensitivity of several low penetrance comutations. Notably, enrichment for DDR gene mutations in patients with durable clinical benefit was not uniform across individual genes and was not observed for ATM or RAD50; acquisition of secondary genomic alterations in this heavily chemotherapy-pretreated patient cohort may account for this discordant observation. Due to the exploratory nature of this analysis, these findings require further evaluation and validation in future studies.

Baseline coalterations in RAS genes, including high-level focal KRAS amplifications and coexisting oncogenic somatic mutations in KRAS/HRAS/NRAS, were enriched in patients with early progression and were associated with worse PFS and OS with KRAS^G12Ci. These results are aligned with prior preclinical and clinical work demonstrating that de novo and acquired RAS alterations are associated with and lead to resistance to the single-agent KRAS^G12Ci adagrasib and sotorasib (17, 18). Co-occurring alterations in components of the PI3K/AKT/MTOR pathway were also associated with inferior PFS with KRAS^G12Ci in KSC^WT tumors; mutations in these genes can promote KRAS^G12Ci insensitivity by establishing bypass signaling tracts, in agreement with direct effects in preclinical models (5). Finally, somatic mutations in some oncogenic kinase genes, including ROS1, ALK, and NTRK1/2/3, were also associated with inferior PFS and OS with KRAS^G12Ci in KSC^WT tumors. Gradual expansion of subclonal mutations under the selective pressure imposed by KRAS^G12Ci therapy may explain their more modest impact on clinical outcomes.

Taken together, our data establish comutations in KEAP1, SMARCA4, and CDKN2A as major independent determinants of inferior clinical outcomes with KRAS^G12Ci monotherapy in advanced NSCLC. Additional granularity and accuracy in forecasting individual clinical response trajectories and patient stratification into distinct prognostic groups will likely be achieved by incorporation of less prevalent genomic as well as baseline and on-treatment transcriptomic and proteomic biomarkers (Supplementary Fig. S10). For example, the expression of RGS3, a noncanonical, mutant KRAS–inclusive GAP correlated with in vivo KRAS^G12Ci sensitivity in a panel of NSCLC patient-derived xenograft models (44). Lineage- or cell state–specific as well as nontumor cell–intrinsic effects may also contribute to future integrated KRAS^G12Ci efficacy predictive models. Finally, beyond patient stratification and individual clinical response prediction, our results are relevant for prioritization and precise tailoring of KRAS^G12Ci-based combination therapeutic strategies—including currently ongoing and planned combinations with CDK4/6, mTOR, DNA repair, SHP2, EGFR, and MEK/ERK inhibitors—to the comutation status of individual tumors in order to maximize therapeutic benefit.

Electronic medical record review was performed for two independently collected patient cohorts from 21 academic institutions in the United States and Europe. Cohort A includes MD Anderson Cancer Center, Cleveland Clinic, University of Chicago, Yale University, University of Cologne, University of Heidelberg, Columbia University Medical Center, Gustave Roussy, Henry Dunant Hospital Center, Johns Hopkins, Ohio State University, Instituto Nazionale Tumori Regina Elena-Rome, Stanford University, University of Torino–Orbassano, University of California Davis, University of California Los Angeles, University of California San Diego, University of California San Francisco, and Moffitt Cancer Center. Cohort B includes Dana-Farber Cancer Institute and Massachusetts General Hospital. Patients with stage IV KRAS^G12C-mutant NSCLC who received treatment with the single-agent KRAS^G12Ci sotorasib or adagrasib were alive for ≥14 days after the start of treatment, had ECOG PS ≤2, and had genomic profiling results available from tumor or blood prior to starting KRAS^G12Ci were eligible. Patients with acquired KRAS mutation in the context of other oncogene-addicted NSCLC (e.g., EGFR and ALK) were excluded. Patients were treated between November 2018 and October 2022, and the dataset was locked on October 1, 2022, for the outcome analysis. Patient information was collected through chart review. Cohorts A and B were analyzed separately and in combination (overall study cohort) for scientific rigor and transparency to provide further validation of key findings. The number of prior lines of therapy was defined as lines of systemic therapy received for metastatic disease. Tumor cell PD-L1 expression was determined with the Dako 22C3 (61.8%), E1L3N (23.6%), Ventana SP263 (12%), QR1 (1.2%), Ventana SP142 (0.6%), and IHC411 (0.6%) assays. The study was Institutional Review Board approved at participating centers and included a waiver of patient informed consent. This study was conducted in accordance with ethical guidelines including the Declaration of Helsinki and the U.S. Common Rule.

Patients must have had genomic profiling results from tumor and/or plasma prior to starting KRAS^G12Ci to be included in the analysis. Tests performed through commercially approved assays or in a Clinical Laboratory Improvement Amendments–certified laboratory were allowed (see Supplementary Table S6 for included assays). When available, we integrated results from tumor and plasma profiling for the analysis. Test results for each individual patient were curated and annotated for pathogenic somatic nonsynonymous variants. Variants reported as germline were excluded. To be classified as pathogenic, a variant must meet at least one of four criteria: (i) be defined as pathogenic per Catalogue of Somatic Mutations in Cancer (COSMIC; RRID:SCR_002260) entry; (ii) be defined as pathogenic on the ClinVar database (RRID:SCR_006169); (iii) have a PolyPhen (Polymorphism Phenotyping; RRID:SCR_013189) score ≥0.95 (45); or (iv) have a SIFT (Sorting Intolerant From Tolerant; RRID:SCR_012813) score ≤0.05 (46). Biallelic (homozygous) copy-number losses for tumor suppressor genes, amplifications for oncogenes, and gene rearrangements—where reported—were considered relevant alterations and were included in the analysis.

To determine genomic modifiers of clinical outcomes with KRAS^G12Ci, we first classified patients into two subgroups—durable clinical benefit (PFS ≥6 months) or early progression (PFS ≤3 months) with sotorasib and adagrasib—following similar methodology as previously reported (18). Patients censored with less than 3 months of follow-up were excluded. We then performed an unbiased enrichment analysis of the most prevalent coalterations (detected in at least 5% of patients) in the overall study cohort. If a given patient underwent profiling (tumor or plasma) with an NGS panel that did not cover a specific gene, then that patient was removed from the analysis of that specific gene. Differences between durable clinical benefit and early progression subgroups were assessed with the Fisher exact test adjusted for multiple comparisons using FDR (Benjamini–Hochberg procedure). Significance was established at P ≤ 0.05 and FDR q ≤ 0.10.

To identify less prevalent comutations that might be associated with clinical outcomes upon treatment with KRAS^G12Ci, we performed an exploratory analysis focusing on (i) KSC^WT tumors and (ii) genes with comutations present in at least three patients. Genes of interest were selected based on Log₂ OR ≥2.0 or ≤ −2.0 for early progression (patients with PFS ≤3 months) relative to durable clinical benefit (patients with PFS ≥6 months).

For the PFS analysis, patients who were alive and had no evidence of progression at the time of dataset lock or who were lost to follow-up were censored at the time of the last radiologic tumor assessment. For the OS analysis, patients who were alive or lost to follow-up at the time of dataset lock were censored at the time of the last documented patient contact. The Kaplan–Meier method was used to estimate PFS and OS, and differences were assessed by log-rank test. HRs and corresponding CIs were estimated with the use of a stratified Cox proportional-hazards model adjusting for clinical variables [age, history of brain metastasis, prior lines of therapy for metastatic disease (0 vs. ≥1), PS (0–1 vs. 2)]. Univariate analysis was performed for the exploratory analysis of less prevalent candidate genes identified through the unbiased enrichment analysis and for gene groups established by biological significance. Best response was determined through investigator-assessed RECIST v1.1 without central review. Patients who died ≥14 days after the start of KRAS^G12Ci, but prior to the first restaging scan, were considered to have progressive disease. Differences in categorical variables were assessed by two-sided Fisher exact test. Significance was established at P ≤ 0.05. Statistical analysis was performed on IBM SPSS Statistics (RRID:SCR_002865), R (RRID:SCR_001905), Microsoft Excel (RRID:SCR_016137), and SAS 9.4 (RRID:SCR_008567).

The individual patient data generated in this study are governed by all participating institutions. To preserve patient confidentiality, to protect patient-related information, and to remain compliant with each participating institution's regulatory requirements, aggregate and/or summary deidentified data may be made available upon reasonable academic request to the corresponding author.

M.V. Negrao reports grants and other support from Mirati, Genentech, and Novartis, other support from Merck, and grants from AstraZeneca, Pfizer, Alaunos, and Checkmate outside the submitted work. A.J. Cooper reports personal fees from MJH Life Sciences outside the submitted work. J.K. Hicks reports other support from The Jackson Laboratory Maine Cancer Genomics Initiative Genomic Tumor Board and Quest Diagnostics, and grants from OneOme outside the submitted work. M. Aldea reports grants and personal fees from Sandoz, personal fees from Viatris, and other support from Amgen outside the submitted work. M.J. Dennis reports personal fees from MJH Life Sciences, Targeted Healthcare Communications, and OMNI Health Media and Remedy Health Media outside the submitted work. S.C. Scott reports other support from Mirati Therapeutics and personal fees from AstraZeneca, Genentech, Regeneron, and Foundation Medicine outside the submitted work. P. Bironzo reports personal fees from AstraZeneca, BeiGene, Roche, Janssen, and Bristol Myers Squibb, grants from Roche and Pfizer, and other support from Takeda and Daiichi Sankyo outside the submitted work. M. Scheffler reports grants and personal fees from Amgen, and personal fees from Boehringer Ingelheim, Novartis, and Roche during the conduct of the study, as well as grants from Dracen Pharmaceuticals and personal fees from Janssen, Pfizer, Sanofi-Aventis, and Takeda outside the submitted work. P. Christopoulos reports grants and personal fees from Amgen, AstraZeneca, Takeda, Thermo Fisher, Boehringer Ingelheim, Roche, and Novartis, and personal fees from Gilead, Pfizer, Janssen, and Chugai outside the submitted work. A. Stenzinger reports grants and personal fees from Amgen, Bayer, and Bristol Myers Squibb, personal fees from Illumina, Incyte, Takeda, AstraZeneca, Novartis, MSD, and Eli Lilly, and nonfinancial support from Pfizer during the conduct of the study. J.W. Riess reports grants from Merck, Novartis, ArriVent, Revolution Medicines, GSK, Spectrum, and AstraZeneca, other support from AstraZeneca, and personal fees from Mervus, Bristol Myers Squibb, Seattle Genetics, Janssen, BeiGene, Biodesix, Blueprint, Bayer, EMD Serono, Jazz Pharmaceuticals, Novartis, Turning Point, Daiichi Sankyo, Boehringer Ingelheim, Regeneron, and Sanofi outside the submitted work. S.Y. Kim reports personal fees from Amgen outside the submitted work. S.B. Goldberg reports grants and personal fees from AstraZeneca, Boehringer Ingelheim, and Mirati, and personal fees from Bristol Myers Squibb, Genentech, Amgen, Blueprint Medicines, Sanofi Genzyme, Daiichi Sankyo, Regeneron, Takeda, Janssen, Summit Therapeutics, and Merck outside the submitted work. B. Ricciuti reports personal fees from Regeneron outside the submitted work. L. Landi reports personal fees from Pfizer, AstraZeneca, Roche, Takeda, Bayer, Eli Lilly, Sanofi, Amgen, Bristol Myers Squibb, MSD, and Bayer outside the submitted work. M. Nishino reports grants from Canon Medical Systems, Daiichi Sankyo, and AstraZeneca and personal fees from AstraZeneca outside the submitted work. S.R. Digumarthy reports grants from Qure AI, Vuno, Lunit, and GE, and other support from Siemens Healthineers and Elsevier outside the submitted work; provides independent image analysis for hospital-contracted clinical research trial programs for Merck, Pfizer, Bristol Myers Squibb, Novartis, Roche, Polaris, Cascadian, AbbVie, Gradalis, Bayer, Zai Laboratories, Biengen, Resonance, and Analise; and reports research grants from Lunit, GE, Qure AI, and Vuno and honorarium from Siemens. G.R. Blumenschein Jr reports grants and personal fees from Amgen, Daiichi Sankyo, Bristol Myers Squibb, Genentech, Merck, MedImmune, Novartis, Roche, CytomX Therapeutics, Regeneron, and BeiGene, grants from Bayer, Adaptimmune, Immatics, Immunocore, Incyte, Kite Pharma, Macrogenics, AstraZeneca, Xcovery, Tmmunity Therapeutics, Mythic Therapeutics, Takeda, Duality Biologics, Verastem, and Repertoire Immune Medicines, personal fees from Instil Bio, Sanofi, Genzyme, Gilead, Lilly, Janssen, Tyme Oncology, Virogin Biotech, Maverick Therapeutics, Intervenn Biosciences, Onconova Therapeutics, Seagen, Aulos Bioscience, AbbVie, Adicet, Ariad, and Clovis Oncology outside the submitted work, as well as an immediate family member who is employed with Johnson & Johnson/Janssen. J. Zhang reports grants from Merck, grants and personal fees from Johnson & Johnson and Novartis, and personal fees from Bristol Myers Squibb, AstraZeneca, GenePlus, Innovent, and Hengrui outside the submitted work. D.H. Owen reports grants from Bristol Myers Squibb, Merck, Palobiofarma, Genentech, Pfizer, and Onc.AI outside the submitted work. C.M. Blakely reports grants from AstraZeneca, Novartis, Takeda, Spectrum, Puma, and Mirati and personal fees from Janssen outside the submitted work. G. Mountzios reports personal fees from Amgen Greece, Roche Hellas, MSD Greece, Bristol Myers Squibb Greece, AstraZeneca Greece, Novartis Greece, Sanofi Greece, and Takeda during the conduct of the study, and is a principal investigator in sponsored clinical trials from Roche, MSD, Bristol Myers Squibb, Amgen, Novartis, Gilead, Mirati, AstraZeneca, and GSK. C.A. Shu reports personal fees from AstraZeneca, Genentech, Arcus Biosciences, Mirati Therapeutics, Janssen, and Takeda outside the submitted work. C.M. Bestvina reports grants and personal fees from AstraZeneca and Bristol Myers Squibb, and personal fees from CVS, Daiichi Sankyo, EMD Serono, Genentech, Jazz, Johnson & Johnson, Novartis, Novocure, Pfizer, Regeneron, Sanofi, Seattle Genetics, Takeda, and Merck outside the submitted work. M.C. Garassino reports grants and personal fees from AstraZeneca, Merck, and Pfizer, and personal fees from Bayer, Boehringer Ingelheim, Bristol Myers Squibb, AbbVie, Novartis, Roche, Regeneron, Blueprint, Sanofi, Amgen, Mirati, Eli Lilly, Takeda, and Daiichi Sankyo outside the submitted work. K.A. Marrone reports grants and personal fees from Mirati, personal fees from Regeneron, Amgen, Daiichi Sankyo, Janssen, Puma, and AstraZeneca, and grants from Bristol Myers Squibb outside the submitted work. J.E. Gray reports personal fees from AbbVie, Axiom HC Strategies, and Blueprint Medicines, grants from Boehringer Ingelheim, Celgene, Daiichi Sankyo, EMD Serono-Merck KGaA, Gilead Sciences, G1 Therapeutics, Inivata, Janssen Scientific Affairs, Jazz Pharmaceuticals, Loxo Oncology, OncoCyte Biotechnology Company, Sanofi Pharmaceuticals, Takeda Pharmaceuticals, Ludwig Institute of Cancer Research, and Pfizer, and grants and personal fees from AstraZeneca, Bristol Myers Squibb, Genentech, Merck, and Novartis outside the submitted work. S.P. Patel reports nonfinancial support from Amgen, AstraZeneca, Bristol Myers Squibb, Certis, Eli Lilly, Jazz, Genentech, Illumina, Merck, Pfizer, Rakuten, and Tempus, and grants from Amgen, AstraZeneca/Med­Immune, Bristol Myers Squibb, Eli Lilly, Fate Therapeutics, Gilead, Iovance, Merck, Pfizer, Roche/Genentech, and SQZ Biotechnologies during the conduct of the study. H.A. Wakelee reports personal fees from Mirati and grants from AstraZeneca, Genentech/Roche, Merck, Seagen, Xcovery, Helsinn, and Bayer outside the submitted work. J. Wolf reports personal fees from Amgen, AstraZeneca, Bayer, Blueprint, Boehringer Ingelheim, Chugai, Daiichi Sankyo, Lilly, Loxo, Merck, MSD, Nuvalent, Roche, Seattle Genetics, Takeda, and Turning Point, and grants and personal fees from Bristol Myers Squibb, Janssen, Novartis, and Pfizer outside the submitted work. G.V. Scagliotti reports personal fees from AstraZeneca, Eli Lilly, MSD, Pfizer, Roche, Johnson & Johnson, Takeda, Eli Lilly, BeiGene, and AstraZeneca, and grants from Eli Lilly, MSD, and Tesaro outside the submitted work. F. Cappuzzo reports personal fees from Roche, AstraZeneca, Bristol Myers Squibb, Pfizer, Takeda, Lilly, Bayer, Amgen, Sanofi, PharmaMar, Novocure, Mirati, Galecto, OSE, and MSD outside the submitted work. F. Barlesi reports other support from AbbVie, ACEA, Amgen, Astra­Zeneca, Bayer, Bristol Myers Squibb, Boehringer Ingelheim, Eisai, Eli Lilly Oncology, F. Hoffmann-La Roche Ltd, Genentech, Ipsen, Ignyta, Innate Pharma, Loxo, Novartis, MedImmune, Merck, MSD, Pierre Fabre, Pfizer, Sanofi-Aventis, and Takeda outside the submitted work. P.D. Patil reports personal fees from AstraZeneca and Jazz Pharmaceuticals outside the submitted work. L. Drusbosky reports personal fees from Guardant Health during the conduct of the study, as well as personal fees from Guardant Health outside the submitted work. D.L. Gibbons reports grants from the NCI, Rexanna's Foundation for Fighting Lung Cancer, and the Cancer Prevention & Research Institute of Texas during the conduct of the study; grants from AstraZeneca, Ribon Therapeutics, NGM Biotherapeutics, Boehringer Ingelheim, and Astellas, and personal fees from Eli Lilly, 4D Pharma, Onconova, Sanofi, and Menarini Richerche outside the submitted work. F. Meric-Bernstam reports personal fees from AbbVie, Aduro BioTech Inc., Alkermes, DebioPharm, Ecor1 Capital, F. Hoffmann-La Roche Ltd, GT Apeiron, Genentech, Harbinger Health, IBM Watson, Infinity Pharmaceuticals, The Jackson Laboratory, Kolon Life Science, Lengo Therapeutics, Menarini Group, OrigiMed, PACT Pharma, Parexel International, Pfizer, Protai Bio Ltd., Samsung Bioepis, Seattle Genetics, Tallac Therapeutics, Tyra Biosciences, Xencor, Zymeworks, Black Diamond, Biovica, Eisai, Fog­Pharma, Immunomedics, Inflection Biosciences, Karyopharm Thera­peutics, Loxo Oncology, Mersana Therapeutics, OnCusp Therapeutics, Sanofi, Silverback Therapeutics, Spectrum Pharmaceuticals, Zentalis, and Chugai Biopharmaceuticals, grants and personal fees from AstraZeneca, Daiichi Sankyo, eFFECTOR Therapeutics, and Puma Biotechnology, grants from Aileron Therapeutics, Inc., Bayer Healthcare Pharmaceutical, Calithera Biosciences, Curis Inc., CytomX Therapeutics, DebioPharm International, Guardant Health, Klus Pharma, Takeda Pharmaceuticals, Novartis, and Taiho Pharmaceutical, and other support from the European Organisation for Research and Treatment of Cancer and the European Society for Medical Oncology outside the submitted work. J.V. Heymach reports other support from Genentech, Mirati Therapeutics, Eli Lilly, Janssen Pharmaceuticals, BioAlta, Dava Oncology, Regeneron, BerGenBio, Jazz Pharmaceuticals, Curio Science, and Novartis, and grants and other support from Boehringer Ingelheim, Takeda Pharmaceuticals, and AstraZeneca Pharmaceuticals outside the submitted work. D.S. Hong reports other support from Amgen and Mirati, and personal fees from Amgen during the conduct of the study; research (institution)/grant funding (institution) from AbbVie, Adaptimmune, Adlai-Nortye, Amgen, AstraZeneca, Bayer, Biomea, Bristol Myers Squibb, Daiichi Sankyo, Deciphera, Eisai, Eli Lilly, Endeavor, Erasca, F. Hoffmann-La Roche Ltd, Fate Therapeutics, Genentech, Genmab, Immunogenesis, Infinity, Kyowa Kirin, Merck, Mirati, Navier, NCI-Cancer Therapy Evaluation Program, Novartis, Numab, Pfizer, Pyramid Bio, Revolution Medicine, Seagen, STCube, Takeda, TCR2, Turning Point Therapeutics, and VM Oncology; travel, accommodations, and expenses from the American Association for Cancer Research, the American Society of Clinical Oncology, CLCC, Bayer, Genmab, the Society for Immunotherapy of Cancer, and Telperian; consulting, speaker, or advisory roles with 28Bio, AbbVie, Acuta, Adaptimmune, Alkermes, Alpha Insights, Amgen, Affini-T, Astellas, Aumbiosciences, Axiom, Baxter, Bayer, Boxer Capital, BridgeBio, CARSgen, CLCC, COG, COR2ed, Cowen, Ecor1, Erasca, Fate Therapeutics, F. Hoffmann-La Roche Ltd, Genentech, Gennao Bio, Gilead, GLG, Group H, Guidepoint, HCW Precision Oncology, Immunogenesis, InduPro, Janssen, Liberium, MedaCorp, Medscape, Numab, Oncologia Brasil, ORI Capital, Pfizer, Pharma Intelligence, POET Congress, Prime Oncology, Projects in Knowledge, Quanta, RAIN, Ridgeline, Seagen, Stanford, STCube, Takeda, Tavistock, Trieza Therapeutics, Turning Point Therapeutics, WebMD, YingLing Pharma, and Ziopharm; and other ownership interests in Molecular Match (adviser), OncoResponse (founder, adviser), and Telperian (founder, adviser). R.S. Heist reports other support from AbbVie, Claim Therapeutics, Regeneron, Sanofi, Daiichi Sankyo, AstraZeneca, EMD Serono, Novartis, and Lilly, and grants from Novartis, Lilly, Turning Point, Erasca, Daiichi Sankyo, Mirati, Exelixis, AbbVie, Agios, Corvus, and Mythic outside the submitted work. M.M. Awad reports personal fees from Merck, ArcherDx, Mirati, and Gritstone, grants and personal fees from Genentech, Bristol Myers Squibb, and AstraZeneca, and grants from Amgen, Novartis, EMD Serono, and Lilly during the conduct of the study. F. Skoulidis reports grants from the NIH/NCI and Rexanna's Foundation for Fighting Lung Cancer during the conduct of the study, as well as grants, personal fees, and other support from Amgen, personal fees from ­BeiGene, Navire Pharma, VSPO McGill University de Montreal, and RV Mais Promocao Eventos LTDS, grants and personal fees from Novartis, personal fees and other support from AstraZeneca, Guardant Health, BergenBio, the European Society for Medical Oncology, the American Association for Cancer Research, the International Association for the Study of Lung Cancer, the Japan Lung Cancer Society, Tango Therapeutics, Calithera Biosciences, Intellisphere, Medscape, PER, Curio, MI&T, IDEOlogy Health, and MJH Life Sciences, grants from Mirati Therapeutics, Revolution Medicines, Merck, and Pfizer, and other support from BioNTech, Moderna, and Dava Oncology outside the submitted work. No disclosures were reported by the other authors.

M.V. Negrao: Conceptualization, resources, data curation, formal analysis, supervision, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. H.A. Araujo: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. G. Lamberti: Data curation, formal analysis, validation, investigation, methodology, writing–review and editing. A.J. Cooper: Data curation, investigation, writing–review and editing. N.S. Akhave: Data curation, investigation, writing–review and editing. T. Zhou: Data curation, writing–review and editing. L. Delasos: Data curation, writing–review and editing. J.K. Hicks: Data curation, writing–review and editing. M. Aldea: Data curation, writing–review and editing. G. Minuti: Data curation, writing–review and editing. J. Hines: Data curation, writing–review and editing. J.V. Aredo: Data curation, writing–review and editing. M.J. Dennis: Data curation, writing–review and editing. T. Chakrabarti: Data curation, writing–review and editing. S.C. Scott: Data curation, writing–review and editing. P. Bironzo: Data curation, writing–review and editing. M. Scheffler: Data curation, writing–review and editing. P. Christopoulos: Data curation, writing–review and editing. A. Stenzinger: Data curation, writing–review and editing. J.W. Riess: Data curation, writing–review and editing. S.Y. Kim: Data curation, writing–review and editing. S.B. Goldberg: Data curation, writing–review and editing. M. Li: Data curation, writing–review and editing. Q. Wang: Formal analysis, visualization, methodology, writing–review and editing. Y. Qing: Formal analysis, visualization, methodology, writing–review and editing. Y. Ni: Data curation, writing–review and editing. M.T. Do: Methodology, writing–review and editing. R. Lee: Data curation, methodology, writing–review and editing. B. Ricciuti: Data curation, writing–review and editing. J.V. Alessi: Data curation, writing–review and editing. J. Wang: Visualization, methodology, writing–review and editing. B. Resuli: Data curation, writing–review and editing. L. Landi: Data curation, writing–review and editing. S.-C. Tseng: Data curation, writing–review and editing. M. Nishino: Data curation, visualization, writing–review and editing. S.R. Digumarthy: Data curation, visualization, writing–review and editing. W. Rinsurongkawong: Data curation, writing–review and editing. V. Rinsurongkawong: Data curation, writing–review and editing. A.A. Vaporciyan: Resources, project administration, ­writing–review and editing. G.R. Blumenschein Jr: Supervision, writing–review and editing. J. Zhang: Supervision, writing–review and editing. D.H. Owen: Data curation, writing–review and editing. C.M. Blakely: Data curation, writing–review and editing. G. Mountzios: Data curation, writing–review and editing. C.A. Shu: Writing–review and editing. C.M. Bestvina: Data curation, writing–review and editing. M.C. Garassino: Supervision, writing–review and editing. K.A. Marrone: Supervision, writing–review and editing. J.E. Gray: Supervision, writing–review and editing. S.P. Patel: Supervision, writing–review and editing. A.L. Cummings: Data curation, supervision, writing–review and editing. H.A. Wakelee: Supervision, writing–review and editing. J. Wolf: Supervision, writing–review and editing. G.V. Scagliotti: Supervision, writing–review and editing. F. Cappuzzo: Supervision, writing–review and editing. F. Barlesi: Supervision, writing–review and editing. P.D. Patil: Supervision, writing–review and editing. L. Drusbosky: Formal analysis, visualization, methodology, writing–review and editing. D.L. Gibbons: Writing–review and editing. F. Meric-Bernstam: Writing–review and editing. J.J. Lee: Formal analysis, methodology, writing–review and editing. J.V. Heymach: Supervision, writing–review and editing. D.S. Hong: Supervision, writing–review and editing. R.S. Heist: Supervision, writing–review and editing. M.M. Awad: Conceptualization, supervision, investigation, methodology, writing–review and editing. F. Skoulidis: Conceptualization, resources, data curation, formal analysis, supervision, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.

M.V. Negrao acknowledges research funding from Rexanna's Foundation for Fighting Lung Cancer. J. Hines acknowledges research funding from an NIH Clinical Therapeutics Training Grant (T32-GM07019). M.M. Awad was supported in part by the Elva J. and Clayton L. McLaughlin Fund for Lung Cancer Research, Team Stuie, and LUNGSTRONG. Work in F. Skoulidis's laboratory was supported in part by NIH/NCI 1R01 CA262469-01 and the Tammi Hissom Grant from Rexanna's Foundation for Fighting Lung Cancer. This work was supported by the generous philanthropic contributions to The University of Texas MD Anderson Lung Moon Shot Program and the MD Anderson Cancer Center Support Grant P30 CA016672. We acknowledge the MD Anderson GEMINI Team for their contributions to this article.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.