Comparative Analysis and Isoform-Specific Therapeutic Vulnerabilities of KRAS Mutations in Non–Small Cell Lung Cancer

This article is featured in Highlights of This Issue, p. 1475

KRAS is one of the most commonly mutated oncogenes in human cancer, with selectively high frequency in tumors of the pancreas, colon, and lung. In non–small cell lung cancer (NSCLC), KRAS mutations are detected in about 30% of patients and are associated with poor prognosis (1, 2). Several strategies have been investigated over the last decades to target KRAS mutation in lung cancer, including the inhibition of downstream effectors such as MEK (3, 4). However, none of these strategies were effective in clinical trials, and the optimal first-line therapy for advanced KRAS-mutant lung adenocarcinoma (LUAD) still consists of PD-(L)1 blockade alone or in combination with platinum-based chemotherapy (5). More recently, alternative strategies targeting KRAS-mutant lung cancer have been investigated in preclinical models and are in various stages of clinical development (1, 6, 7).

To date, the greatest success in targeting KRAS mutations is represented by the development of isoform-specific direct KRAS inhibitors. Recent data from early-phase clinical trials of KRAS G12C direct inhibitors sotorasib and adagrasib have shown encouraging activity in KRAS G12C-mutant NSCLCs, with responses occurring in ∼30%–40% of patients (8–10). Based on the results of the CodeBreaK 100 study and KRYSTAL-1, sotorasib and adagrasib were granted accelerated FDA approval and breakthrough therapy designation (BTD), respectively, for patients with NSCLC who have received prior systemic therapy. Nevertheless, an urgent need remains for better stratification and effective therapeutic strategies to improve outcomes for KRAS-mutant cancer patients.

Although RAS mutations were discovered over 30 years ago (11), the detailed understanding of the biological properties of oncogenic KRAS is still far to be complete. New aspects of KRAS biology have been recently described, including the requirement of dimerization for oncogenic signaling (12), lipid-sorting specificity into the membrane (13), and tissue-specific comutation networks (14).

One intriguing aspect of KRAS biology is the potential impact of distinct KRAS-activating mutations on downstream signaling and drug sensitivity (15, 16). Until very recently, RAS proteins with alterations in codon 12, 13, or 61 have been considered oncogenic equivalents; however, emerging data suggest functional differences for each RAS mutation in colon cancer (17), melanoma (18), and NSCLC, where different mutant KRAS proteins are known to be heterogeneous in terms of clinical outcome (19–20). Recently, studies in isogenic cell lines demonstrated that each individual mutation generates a distinct oncogenic and network response (14, 15, 21–23). In terms of translational impact, the study of differences among specific KRAS mutants by means of more informative experimental approaches will be fundamental to discover new therapeutic strategies for individual KRAS mutations.

Here, we developed an isogenic system where a series of mutant KRAS isoforms commonly found in lung cancer were directly compared to highlight distinct GTP-loading levels, differential drug sensitivity to MEK inhibitors, and unique growth properties in vivo. To gather evidence on whether these differences also translated in different clinical outcomes, we investigated the impact of the different KRAS isoforms on survival in a large cohort of patients with non-squamous NSCLC. Together, these results suggest that genomic coalterations and residue-specific properties of KRAS mutations have important biological and clinical implications.

KRas^lox KRAS^MUT cells were generated as previously described (12). Briefly, KRAS^MUT retroviral constructs were created by point mutagenesis from pBABE HA-tagged KRAS^WT plasmid (a gift from Channing Der, Addgene plasmid #75282). Retroviruses were generated by cotransfection of pBABE plasmids together with pAmpho plasmid into 293T cells using FuGENE HD Transfection Reagent (Promega). The retroviruses were transduced into HRas^−/−; NRas^−/−; KRas^lox/lox MEFs (24) followed by puromycin selection (1 μg/mL). Selected cells were then cultured in the presence of 4-hydroxytamoxifen (4OHT; Sigma, 600 nmol/L) for at least 10 days in order to achieve complete deletion of endogenous KRas alleles. Cells were not cultured longer than 3 months after thawing from frozen stocks and were routinely checked for Mycoplasma as determined by the Mycoplasma Plus PCR Primer Set (Agilent).

Cells (1×10³) were seeded in 96-well plates in 150 μL complete medium. The following day, cells were PBS washed, and 150 μL of medium with different concentrations of FBS was added to the plates. Cells were then incubated in the IncuCyte Zoom for real-time imaging, with three fields imaged per well under 10× magnification every two hours for a total of 60–96 hours. Data were analyzed using the IncuCyte Confluence version 1.5 software, which quantified cell-surface area coverage as confluence values. IncuCyte experiments were performed in triplicate. A single representative growth curve is shown for each condition. For hypoxic conditions, cells (1×10³) were seeded in 96-well plates in 150 μL complete medium. The following day, cells were PBS washed, and 150 μL of medium with different concentrations of FBS was added to the plates that were incubated in 1% O₂ incubator. Cell viability was assessed by CellTiter-Glo 2.0 (Promega) at the indicated time points.

Data from a large-scale pharmacogenomics study, the Cancer Genome Project (CGP, data set version 2020), were accessible from http://www.cancerrxgene.org. A total of 310 lung, pancreas, and colon cancer cell lines and their drug-sensitivity values to MEK/ERK inhibitors in the CGP data set were used for analysis. In these data, the natural logarithm of the IC₅₀ value represents the drug-sensitivity value. Cell lines were grouped according to KRAS mutational subtypes, and pairwise comparisons were performed using the Mann–Whitney U test.

Cells (1 × 10³) were seeded in 96-well plates. The following day, cells were treated with drugs using a ten-point dose titration scheme. After 72 hours, cell viability was assessed using the colorimetric MTS assay. Absolute inhibitory concentration values were calculated using four-parameter logistic curve fitting. All experimental points were a result of three to six replicates, and all experiments were repeated at least three times. The data were graphically displayed using GraphPad Prism 5 for Windows (GraphPad Software). Each point (mean ± standard deviation) represents growth of treated cells compared with untreated cells. The curves were fitted using a nonlinear regression model with a sigmoidal dose response. For synergy distribution assays, cells were seeded at 1,500 cells per well in a 384-well plate (Corning). The following day, cells were treated with selumetinib and SHP099, varying concentration from 0.25 to 15 μmol/L and from 0.5 to 15 μmol/L, respectively. After 72 hours, cell viability was measured by CellTiter-Glo 2D (Promega). Four replicates of each mutant were analyzed implementing LOEWE model (Combenefit software); experiments were repeated three times.

Cells were grown in 0.1% FBS for 24 hours, stimulated with EFG (Thermo Fisher Cat#PHG0311) 50 ng/mL for 5 minutes, and Ras-GTP levels were assessed by Active Ras Detection Kit (Cell Signaling, cat. #8821) using Raf-RBD fused to GST to bind active (GTP-bound) Ras. Protein lysates (500 μg) were incubated with 30 μL glutathione resin and GST protein binding domains for one hour at 4°C to capture active small GTPases according to the manufacturer's protocol. After washing, the bound GTPase was recovered by eluting the GST-fusion protein from the glutathione resin. The purified GTPase was detected by Western blot using mouse monoclonal anti-KRAS (F234) (Santa Cruz Biotech, cat. #sc-30).

After trypsinization, cells from the plastic culture were resuspended in ice-cold Matrigel (Corning) and incubated on prewarmed 6-cm plate to solidify. Organoids were cultured in Renaissance Essential Tumor Medium (Cellaria) with B-27 supplement (Thermo Fisher Scientific) and passaged over three times before performing any experiments. For the growth assay, organoids in Matrigel were dissociated into single cells using TrypLE Express (Invitrogen) at 37°C. One thousand cells were plated into each well of 384-well ultra-low attachment microplates (Corning) with the media and 10% Matrigel. On the next day, cells were treated with dose titration of trametinib or selumetinib for three days (from 1 nmol/L to 10 μmol/L). Each dose had more than three replicates. The viability assay was tested using CellTiter-Glo 3D (Promega) according to the manufacturer's protocol. The statistical significance was assessed using ANOVA and Tukey posttest for multiple comparison.

Crl:NU-Foxn1^nu mice (females, 8 weeks old) were purchased from Charles River. For lung colonization assays, KRas^lox KRAS^MUT cells (1×10⁶) were injected in the tail vein as single-cell suspension in 200 μL of sterile PBS. Mice were sacrificed synchronously at a single time point of one month, and lungs were collected in formalin. For drug treatment assays, KRas^lox KRAS^MUT cells (2×10⁶) were injected subcutaneously in a 1:1 mix of serum-free DMEM and Matrigel (phenol red–free; BD Biosciences) in both flanks of recipient mice. Once a palpable tumor formed, mice were randomly assigned to either selumetinib, sotorasib, or vehicle treatment, and measurements were taken daily using calipers. Selumetinib and sotorasib, resuspended in 2% hydroxypropyl methyl cellulose (HPMC), 1% Tween 80, were administered daily by oral gavage at a dose of 50 mg/kg. For IVIS longitudinal monitoring, KRas^lox KRAS^MUT cells were transduced with pFUGW-Pol2-ffLuc2-eGFP (Addgene plasmid #71394) and FACS-sorted for GFP expression. KRas^lox KRAS^MUT-luciferase cells (2 × 10⁶) were injected in the tail vein of 8-week-old NCr nude mice (Taconic Biosciences) as single-cell suspension in 200 μL of sterile PBS and were monitored using an IVIS imaging system (PerkinElmer). All care and treatment of experimental animals were in strict accordance with Good Animal Practice as defined by the US Office of Laboratory Animal Welfare and approved by the Dana-Farber Cancer Institute Institutional Animal Care and Use Committee and by the Italian Health Minister (authorization n° 1227/2020-PR).

Clinicopathologic and genomic data from patients with NSCLC whose tumors underwent targeted next-generation sequencing at the Dana-Farber Cancer Institute who had provided written informed consent to institutional review board–approved protocols DF/HCC#02-180, #11-104, #13–364, and/or #17-000 were included in this study. This study was conducted in agreement with the Declaration of Helsinki.

Tumor mutational burden (TMB), defined as the number of somatic, coding, base substitution, and indel mutations per megabase (Mb) of genome examined, was calculated from the DFCI OncoPanel next-generation sequencing platform. PD-L1 expression on tissue was assessed by IHC using an anti–PD-L1 rabbit monoclonal antibody (clone E1L3N, Cell Signaling Technology).

Overall survival (OS) was defined as the time from the date of initial diagnosis to death. Patients who were still alive at the data cutoff were censored at the date of last contact. OS was compared between KRAS-mutated and KRAS wild-type cases, as well as across KRAS isoforms.

Fisher exact test P values and conditional odds ratios were used to assess co-occurrence and mutual exclusivity for genes with at least 2% frequency in the relevant groups. Positive odds ratios represented tendency to co-occurrence and negative odds ratios represented tendency to mutual exclusivity. Multiple comparison correction was performed using the Benjamini–Hochberg procedure using the qvalue package in R.

Gene-expression data were downloaded from the Firehose website (TCGA Firehose Legacy version), whereas somatic mutation data were downloaded from the cBioPortal website (cbioportal.org). The RSEM V2 values were used to represent gene expression and genes with counts fewer than 10 were filtered out. Gene-expression profiles were analyzed according to KRAS mutation status. Median expression within each group was used to estimate expression fold change (FC) to minimize the possible impact of outlier samples. Gene differential expression analyses across KRAS allele subgroups were conducted using R package DESeq2. P values were corrected for multiple hypothesis testing via false discovery rate (FDR) adjustment. FC threshold of an absolute value greater than 1.5 and FDR-adjusted P value threshold less than 0.1 were utilized to identify differentially expressed genes. Pathway enrichment analyses were conducted separately for upregulated and downregulated genes using R package Reactome.

We summarized continuous and categorical variables using percentages and medians. To test for differences between continuous variables we used the Wilcoxon-rank sum test and Kruskal–Wallis, when appropriate. The Fisher exact test was used to compare associations between categorical variables. Estimate event–time distributions were examined using the Kaplan–Meier methodology, and the log-rank test was used to test for differences in event-time distributions. Hazard ratios were estimated using the Cox regression models, as previously described (25). To investigate MEK/ERK inhibitor sensitivities for different KRAS-mutant isoforms across cancers, we grouped cancer cell lines based on their KRAS mutation status into the different groups. Mann–Whitney U test was used to compare drug sensitivities between two groups. All P values were two-sided with confidence intervals set at the 95% level. P < 0.05 were defined as significant. Multiple comparison correction was performed using the Benjamini–Hochberg procedure using the qvalue package in R. All statistical analyses were performed using R version 3.6.3.

The institutional clinicopathologic and genomic data that support the findings of our study are available upon reasonable request from the corresponding author. RNA-seq data used in this study are publicly available from the TCGA (accession number: phs000178). Genomic data from the AACR project GENIE (Genomics Evidence Neoplasia Information Exchange) are publicly available (phs001337).

Among KRAS-mutant lung cancer, codon 12 mutations predominate, accounting for nearly 90% of all KRAS mutations, followed by mutations in codons 13 and 61 (14). One question that arises from somatic genetic analysis is whether different KRAS mutations determine the clinical aspects of a given cancer. To investigate the role of individual KRAS isoforms on the biology of mutant KRAS avoiding context-dependent differences associated with different genetic backgrounds, we developed an isogenic system generated from HRas^−/−; NRas^−/−; KRas^lox/lox mouse embryonic fibroblasts (MEF; ref. 24). In these cells, KRas^lox/lox allele excision is controlled by a resident 4OHT-dependent CRE^ERT2 recombinase. We transduced HRas^−/−; NRas^−/−; KRas^lox/lox MEFs with retroviruses encoding for different human HA-tagged KRAS mutants, including the most frequent mutant isoforms detected in human lung adenocarcinoma (G12C, G12D, G12V, G12A, G13D, and Q61H; ref. 12).

In all cell lines, the presence of the respective KRAS mutation was confirmed by Sanger sequencing (data not shown). Cells were treated with 4OHT to abolish the expression of endogenous wild-type KRas, thus allowing the characterization of specific oncogenic KRAS mutations in isogenic cell lines without the interference of H- or NRas isoforms or the endogenous wild-type KRas (herein referred to KRas^lox KRAS^MUT system; Fig. 1A). Interestingly, all KRAS mutants showed increased expression upon deletion of the endogenous wild-type KRas allele. Therefore, we utilized only 4OHT-treated cells in order to eliminate any bias imposed by the wild-type allele. In the full-serum condition, phosphorylation of ERK, MEK, SRC, and AKT was slightly increased in all KRAS mutants compared with wild-type control cells (Fig. 1B). In contrast, STAT3 and EGFR showed marked differences between mutants in terms of both phosphorylation and expression levels (Fig. 1B). Remarkably, KRAS^GTP levels upon stimulation with EGF showed profound variability among mutants, with KRAS G12D and G12V being already saturated in untreated condition in contrast with the other mutants that are still responsive to mitogenic stimulation (Fig. 1C).

Next, we aimed at characterizing the proliferation kinetics of KRas^lox KRAS^MUT cells both in vitro and in vivo. We tested growth properties of different KRAS mutants by IncuCyte in full-serum and starving conditions. All mutants displayed a strong proliferative advantage over wild-type KRAS controls in full-serum; nevertheless, no significant differences were observed among mutants (Fig. 1D). Under starving conditions, however, a clear stratification was evident, with G12D, and G12V to a lesser extent, showing the best fitness over the other KRAS mutants (Fig. 1D). Of note, G12D and G12V are the mutant isoforms displaying higher GTP-bound fraction in limited serum conditions (Fig. 1C). Similar results were obtained in hypoxic conditions, with G12D mutant cells showing improved fitness compared with the other KRAS mutants upon concomitant starvation (Supplementary Fig. S1A). In 3D conditions, all KRAS-mutant isoforms showed comparable growth kinetics with the exception of Q61H, which was significantly less proliferative (Fig. 1D). The advantage of G12D versus the other mutants was also maintained in tumor growth assays in vivo. We injected KRas^lox KRAS^MUT cells expressing different mutants into the tail vein of nude mice and euthanized the animals after one month in order to check for colonization property into the lungs. At histologic evaluation, G12D-mutant cells displayed tumor nodule formation into the lungs, whereas all the other mutants did not (Fig. 1E). Using the same mutant cells expressing luciferase gene, we noninvasively monitored tumor growth in the mice. Consistent with Fig. 1E and in vitro results, G12D and G12V mutants were established in the mouse lung much quicker than other mutants (Supplementary Fig. S1B and S1C). These results confirmed an in vivo advantage for KRAS^G12D mutants (26), although further assessment at longer time points should be performed in order to detect delayed tumor onset in mutants other than G12D.

Due to the prominent role of the MAPK pathway in KRAS signaling, MEK inhibitors were thought to be able to achieve disease control in KRAS-mutant tumors. However, phase III clinical trials demonstrated no overall benefit compared with chemotherapy (4). Retrospective analysis in different subpopulations of KRAS codon mutations showed that patients with KRAS G12C or G12V receiving selumetinib + docetaxel had improved OS compared with other KRAS mutations, suggesting that the type of KRAS mutations may affect MEK inhibitor sensitivity in LUAD (20). We next surveyed the CGP pharmacogenomic data set (27) to identify correlations between sensitivity to MEK inhibitors and KRAS-mutant isoforms in a broad panel of lung, pancreas, and colon cancer cell lines. We could not observe any obvious trend for increased sensitivity for specific isoforms other than G12R, likely due to the limited representation of small number of cell lines (Supplementary Fig. S2). Considering that cell line heterogeneity poses challenging limits to the characterization of KRAS isoform as a single variable to predict MEK inhibitor sensitivity, we tested a panel of MAPK inhibitors in our isogenic KRas^lox KRAS^MUT system. GSK1120212 (trametinib) showed similar antiproliferative effects in all KRAS mutants, whereas selumetinib and the other MAPK inhibitors exhibited up to 10-fold IC₅₀ variability in different KRAS isoforms, being G12C and Q61H among the most sensitive (Fig. 2A; Supplementary Fig. S3; refs. 28, 29). The same trend was observed in 3D culture (Fig. 2B). Western blot assessment of the MAPK pathway confirmed dose-dependent reduction of pERK in selumetinib-treated KRas^lox KRAS^MUT cells, with the exception of G12A and G13D mutants (Fig. 2C). To validate these findings into an in vivo setting, we implanted KRas^lox KRAS^MUT cells into nude mice and followed tumor growth in the presence or absence of selumetinib treatment (50 mg/kg daily). In agreement with our findings in 2D and 3D conditions, we could stratify the outcome as tumor regression (G12C and Q61H), stable disease (G12D, G12V, and G13D), and no response (G12A) based on tumor volume FC on treatment (Fig. 2D).

Considering that preclinical in vivo studies demonstrated that in KRAS-mutant LUAD, combined SHP2/MEK inhibition is therapeutically effective (30, 31), we tested the same combination in KRas^lox KRAS^MUT cells. Interestingly, we observed that combined SHP2/MEK blockade was more effective in G12C, G12D, and G12V mutants compared with the other mutants, in agreement with previous findings (refs. 31, 32; Supplementary Fig. S4).

Given the very recent FDA approval of the selective G12C inhibitor sotorasib (CodeBreaK 100, NCT03600883), we aimed at assessing the efficacy of G12C inhibition in our isoform-specific KRAS^G12C system, both in vitro and in vivo. As expected, drug–response curves with sotorasib demonstrated a clear selectivity for G12C-mutant cells compared with the other KRAS-mutant isoforms (Fig. 3A). To determine how G12C inhibition compared with MEK inhibition in this model, we next implanted KRas^lox KRAS^G12C cells into nude mice and randomized these mice to receive either vehicle, selumetinib (50 mg/kg daily), sotorasib (50 mg/kg daily), or selumetinib + sotorasib. Interestingly, selumetinib exerted better therapeutic response in comparison with sotorasib, possibly likely due to the delay in implementing feedback mechanisms for MEK inhibitors as compared with selective G12C inhibitors as single agent. Importantly, combined treatment with selumetinib + sotorasib resulted in complete growth abrogation and disease control in mice treated with this combination (Fig. 3B–E). This preliminary finding further supports the hypothesis that combined KRAS G12C and MEK inhibition may improve clinical outcomes in patients with KRAS-mutant NSCLC over G12C inhibition alone. A phase Ib clinical trial of sotorasib in combination with other anticancer agents, including MEK inhibitors is currently ongoing (NCT04185883).

In order to translate our findings based on genetically defined cell lines to a clinically relevant context, a total of 3,560 patients with NSCLCs who underwent successful comprehensive tumor genomic profiling at the Dana-Farber Cancer Institute were identified. Of these, 3,126 (87.8%) were non-squamous NSCLC and were included in the final analysis (Supplementary Fig. S5A). The median age of the 3,126 patients with non-squamous NSCLC was 66 (range, 18–99), 61.2% were female, 18.6% were current smokers, 58.4% were former smokers, and 18.6% were never smokers. KRAS mutation (KRAS^MUT) occurred in 1,131 cases (36.2%), whereas the remaining 1,995 cases (63.8%) had a KRAS wild-type (KRAS^WT) genotype (Supplementary Table S1). The most common KRAS mutations identified occurred at codon 12 (990/1131, 87.5%), of which the KRAS G12C was the most frequent (476/990, 47.2%; Supplementary Fig. S5B).

We first investigated differences in clinicopathologic and genomic factors according to KRAS mutation status. Compared with KRAS^WT cases, patients with KRAS^MUT tumors were more likely to be older (P < 0.001), females (P < 0.0001), current/former smokers (P < 0.0001), and to have adenocarcinoma histology (P < 0.01), consistent with previous reports (33, 34). Baseline clinicopathologic features of patients with KRAS^WT versus KRAS^MUT are shown in Supplementary Table S2. We next focused on comutations in preselected genes of interest that define major subsets of NSCLC, including TP53, STK11, and KEAP1. Of note, although mutations in TP53 were more likely to occur in KRAS^WT tumors (P < 0.0001), STK11 and KEAP1 were significantly enriched in KRAS-mutated tumors (P < 0.0001; Supplementary Fig. S5C). We also examined whether PD-L1 expression and TMB distributions differed between KRAS^MUT and KRAS^WT NSq-NSCLCs. We noted that both median PD-L1 expression (10 vs. 5%, P < 0.001) and median TMB (9.8 vs. 8.4 mut/Mb, P < 0.0001) were significantly higher in KRAS^MUT versus KRAS^WT tumors (Supplementary Fig. S5D).

Next, we examined the impact of KRAS mutation on survival among patients with NSq-NSCLC. Although we found no significant impact of KRAS mutation on OS in all comers (HR: 1.07, P = 0.21), as well as among patients with stage I–III NSq-NSCLC (HR: 1.08, P = 0.39), we observed a significant deleterious impact of KRAS mutation among patients with stage IV tumors (16.2 vs. 25.5 months, HR:1.37, P < 0.0001; Fig. 4A–C).

Although we identified a detrimental impact of KRAS mutation among patients with advanced NSq-NSCLC, this difference likely reflects the use of highly effective targeted therapies among patients with KRAS wild-type tumors harboring EGFR mutations and ALK fusions.

Because we have shown that different KRAS mutation have different in vitro and in vivo growth properties, and differential responses to MEK inhibitors, we lastly investigated whether different KRAS mutation affected survival in patients with NSq-NSCLC. The clinicopathologic characteristics of patients according to KRAS alleles are shown in Supplementary Table S3. Consistent with previous reports (35), we found a significantly higher proportion of never smokers among patients with KRAS G12D mutation (21.8%) as opposed to the other KRAS variants (<10%). Although there was no difference in terms of enrichment in TP53 mutation across the KRAS variants, the rates of concurrent STK11 and KEAP1 mutations were highest among KRAS codon 13 mutations (Supplementary Fig. S5C). Pairwise comparisons between KRAS alleles in terms of TP53, STK11, and KEAP1 comutation patterns are shown in Supplementary Fig. S6A. When we examined PD-L1 expression and TMB distributions in this across the different KRAS alleles, we found that the median TMB differed significantly across the seven groups, with KRAS G12D-mutated tumors having the lowest number of non-synonymous mutations/megabase (Supplementary Fig. S5D). By contrast, no difference in PD-L1 expression was observed among these groups (Supplementary Fig. S5D). Pairwise comparisons between KRAS alleles in terms of PD-L1 and TMB distributions are shown in Supplementary Fig. S6B.

We lastly analyzed the impact of KRAS alleles on survival in patients with non-squamous NSCLC. We found no significant difference across KRAS mutation subtype in all comers, as well as among those with stage I–III or stage IV disease (Fig. 4D–F).

Although we did not find differences in survival in patients with NSCLC with different KRAS alleles, we identified several differences in clinicopathologic and genomic factors among NSCLC harboring different KRAS alleles in the DFCI cohort. Therefore, we hypothesized that one of the reasons by which our findings in isogenic models did not translate into different clinical outcomes could be the biological heterogeneity of KRAS-mutant LUADs. To further dissect this heterogeneity in patients with KRAS-mutant LUADs, we interrogated a large publicly available cohort of 12,931 LUADs (GENIE v.9.0) that underwent tumor genomic profiling to comprehensively determine differences in comutation patterns across KRAS variants. After QC removing duplicate samples and gene annotation, 10,663 unique samples were included in the final analysis. KRAS mutation was detected in 3,843 of cases (36%). In this cohort, the most common genomic alterations included TP53 (33%), STK11 (16%), ATM (11%), KEAP1 (11%), and RBM10 (7%; Fig. 5A). When compared with KRAS wild-type cases, however, KRAS-mutant LUADs were found to be significantly enriched in STK11, KEAP1, ATM, RBM10 mutations, whereas TP53, EGFR, BRAF, and ERBB2 were less likely to co-occur with any KRAS mutation (Supplementary Fig. S7A and S7B).

We next focused on the individual KRAS variants and compared the comutation patterns of each KRAS variant with KRAS wild-type LUADs as reference. We identified that all KRAS variants tended to be mutually exclusive with EGFR mutation and were more likely to lack concurrent TP53 mutations (Fig. 5B). Of note, among LUADs with codon 12 mutations, STK11 and ATM mutations was significantly enriched among tumors harboring KRAS G12C, G12A, and G12V mutations, whereas such enrichment was not detected in LUADs with KRAS G12D (Fig. 5B). Additionally, we also found that KEAP1 mutation was significantly enriched among KRAS G12C and KRAS G13X LUADs. Importantly, the highest frequency of concurrent STK11 and KEAP1 mutation was observed in tumors with KRAS codon 13 mutations (Supplementary Fig. S7C). Pairwise comparisons in concurrent and mutually exclusive mutations between each KRAS variant and the remaining KRAS alleles are reported in Supplementary Fig. S8A and S8B.

Having shown that the KRAS-mutant LUADs are highly heterogeneous in terms of concurrent genomic alterations, we lastly examined whether tumors with different KRAS mutations also had different gene-expression profiles. To address this question, we performed a gene ontology analysis of differentially expressed genes in each KRAS variant versus KRAS wild-type tumors in the TCGA LUAD cohort. We found that several pathways were uniquely upregulated and downregulated in specific KRAS isoform (Fig. 5C and D). Interestingly, KRAS G12D-mutant LUADs showed a significant downregulation of pathways involved in chromosome maintenance, DNA double break repair, and cellular senescence (Fig. 5C), whereas KRAS G13X LUADs had a significant downregulation of pathways involved in innate and adaptative immunity, including PD-1, CD28, INFγ, and IL2 signaling, which is consistent with the high rate of concurrent STK11 and/or KEAP1 (36, 37) mutations we identified in this group of LUADs. A full list of differentially regulated pathways in all KRAS isoforms is reported in Supplementary Table S4.

Activating mutations in the KRAS oncogene are known to trigger and sustain the development of tumors with different histology, being particularly frequent in lung, colon, and pancreatic cancers (1). The comparison between predicted and observed frequency of different mutant KRAS isoforms suggests the existence of tissue-specific mechanisms of biological selection (14); however, the connection between KRAS-mutant isoforms and therapeutic vulnerabilities has not been fully addressed yet. Very recently, the development of two specific inhibitors of the KRAS-G12C allele, sotorasib and adagrasib, which were granted accelerated FDA approval and BTD, respectively, in patients with previously treated advanced NSCLC (10, 38–41) highlighted the need to identify new isoform-specific therapeutic strategies. Here we demonstrate that KRAS isoforms are characterized by different oncogenic properties and therapeutic sensitivities in isogenic models, and that KRAS-mutant LUADs are highly heterogeneous in terms of clinicopathologic, genomic, and transcriptomic features.

Studies have been inconsistent in reporting whether KRAS isoforms differ in terms of oncogenic properties, preferred signaling pathways, genomics, and clinical outcomes in patients with LUAD (1). Previously, NSCLC patient samples and cell lines harboring KRAS G12A have been shown to primarily rely on phosphatidylinositol 3-kinase (PI3K) and mitogen-activated protein/extracellular signal-regulated kinase kinase (MEK) signaling, whereas those with mutant KRAS G12C or mutant KRAS G12V are more likely to activate Ral signaling and decrease growth factor-dependent Akt activation (19). Similarly, in KRAS-overexpressing clones from human cell lines harboring the three most common KRAS mutations in NSCLC (G12C, G12V, and G12D), the expression of specific KRAS isoforms induced a different sensitivity pattern to cytotoxic agents, such as cisplatin, pemetrexed, and taxanes (42). Nonetheless, subsequent retrospective analysis yielded highly variable results in terms of clinical outcomes to either chemotherapy or PD-(L)1 inhibition in patients with advanced NSCLC, as well as among those with resected tumors (19, 42, 43). However, data on KRAS comutation pattern have been underreported in these studies; therefore, dissecting the biological differences among KRAS isoforms and their impact on clinical outcomes in patients with LUAD remains challenging.

To study the oncogenic properties of KRAS isoforms in a genetic-controlled system with the same genomic background, we generated isogenic models to dissect potential isoform-dependent therapeutic vulnerabilities. It has already been demonstrated that specific KRAS isoforms exhibit diverse biochemical and biological features in vitro and in vivo (19, 21, 25, 44, 45). However, the therapeutic impact of different KRAS alleles on treatment outcome has had limited investigated so far (29, 46, 47). In our KRas^lox KRAS^MUT platform expressing the most frequent KRAS-mutant alleles (G12C, G12D, G12V, G12A, G13D, and Q61H), we observed that G12D-mutant cells show a significant growth advantage over the other mutants both in vitro and in vivo and are characterized, together with G12V, by the highest levels of GTP-bound fraction even in the absence of upstream EGF stimulation.

To determine whether these differences translated in differential clinical outcomes, we analyzed a large cohort of patients with LUAD. Although we did not identify significant differences in survival in patients with either early-stage or advanced LUADs according to the different KRAS variants, we found that each KRAS isoform was associated with distinct clinicopathologic, genomic, and transcriptomic features, which may have affected our ability to identify differences in outcomes. These data are in agreement with current knowledge confirming no significant differences in survival in LUAD patients according to specific KRAS-mutant alleles. Still, KRAS isoform specificity could be used as a stratification criterion for better tailored therapies based on exclusive vulnerabilities. For instance, KRAS G12D-mutated LUADs were more likely to develop in patients with no history of tobacco use. Consistently, tumors with KRAS G12D mutation also had the lowest TMB among KRAS isoforms, were less likely to harbor concurrent mutations in STK11 or KEAP1, and had a marked downregulation of pathways involved in cellular senescence and chromosome maintenance. In contrast, the highest rate of concurrent mutations in STK11 and KEAP1 was identified among KRAS G13X-mutated tumors. Concurrent mutations in STK11 and KEAP1 define unique subsets of KRAS-mutant LUADs, with distinct immunophenotype and transcriptomic profile, and are generally characterized by significantly worse survival with both chemotherapy and PD-(L)1 blockade (48, 49). Consistently, we identified that G13X LUADs had a significant downregulation of pathways involved in innate and adaptative immunity, such as PD-1 signaling, TCR signaling, and INFγ signaling, among others. It should be highlighted that gene-expression profiling was performed in NSCLC samples from the TCGA, rather than in our isogenic models. Therefore, these results should be interpreted in the context of the different genomic backgrounds which characterize each KRAS allele. Overall, our results support the hypothesis that in addition to the allele-specific and -intrinsic differences that we identified in our isogenic, each KRAS allele has a different spectrum of comutations and transcriptomic profiles in patient samples, which may not be dependent uniquely from KRAS alleles, but still have great biological relevance. Together, these results may have direct and immediate implications for the design and interpretation of panKRAS or multi/KRAS-targeted therapies, as each individual allele has unique oncogenic properties and features.

To further dissect this question, we also investigated differential response to targeted therapy in KRas^lox KRAS^MUT cells. As a proof of concept we chose MEK inhibitors, which showed promising results in a phase II clinical trial but failed to provide any significant benefit in addition to chemotherapy in KRAS-mutant LUAD in a phase III trial (4). Interestingly, retrospective analysis of the phase II trial showed a trend for KRAS G12C and G12V patients toward greater improvement in OS, progression-free survival, and overall response rate compared with other KRAS mutations (20). Indeed, we observed a consistently increased sensitivity of G12C and Q61H mutants with a panel of MEK inhibitors in vitro, which was maintained in a preclinical setting in vivo where G12C and Q61H implants showed greater tumor volume reduction compared with other mutants. This effect may partly rely on increased affinity to CRAF for some isoforms (29, 50). Also, we validated the convenience of our preclinical tool in surveying combination therapies that can be easily tested in high-throughput screening, such as G12C/MEK inhibitors or SHP2/MEK inhibitors (30, 31).

Our study has limitations inherent to the artificial generation of our isogenic model and the retrospective assessment of clinical outcomes. However, we believe that our model does isolate the biological contribution of each isoform and allow the exploration of drug sensitivity in the absence of potential confounding factors. This is of particular importance, given that our results indicate that each KRAS isoform is associated with unique comutation patterns and transcriptomic profiles that may affect drug sensitivity independently from KRAS-intrinsic activity. Our results suggest that the use of specific genetically defined in vitro tools together with the development of large patient-derived data sets will facilitate the identification of potentially actionable vulnerabilities for KRAS-mutant NSCLC.

In summary, this study indicates that KRAS isoforms are biologically and clinically heterogeneous, and that such heterogeneity should be considered to optimize drug development strategies for KRAS-mutant NSCLC.

A.J. Redig reports grants from NCI, Conquer Cancer Foundation, A Breath of Hope Foundation, and Harvard Catalyst/KL2 during the conduct of the study. During the time that A.J. Redig was employed by Dana-Farber (ending in September 2017), A.J. Redig had financial consulting relationships that did not fund any of the work included in this paper but would have been present during the time when contributions to this manuscript were conducted; these relationships are now 36 months from publication and are itemized as follows: (i) Boehringer Ingelheim: personal fee; (ii) Medtronic: personal fee, nonfinancial support; (iii) Roche: personal fee, nonfinancial support; (iv) Ariad: personal fee; and (v) AstraZeneca: personal fee, nonfinancial support, grant funding. A.J. Redig also reports employment by Takeda Pharmaceuticals (November 2017–February 2020) and Blackstone Life Sciences (August 2020–present). P.C. Gokhale reports grants from Marengo Therapeutics, Epizyme Inc, Daiichi Sankyo, Foghorn Therapeutics, 28-7 Therapeutics, and Moderna outside the submitted work. M.M. Awad reports grants and personal fees from Genentech, Bristol Myers Squibb, and AstraZeneca; grants from Lilly; and personal fees from Merck, Maverick, Blueprint Medicine, Syndax, Ariad, Nektar, Gritstone, ArcherDx, Mirati, NextCure, Novartis, EMD Serono, NovaRx, and Iovance outside the submitted work. P.A. Jänne reports grants and personal fees from AstraZeneca, Boehringer Ingelheim, Eli Lilly, Daiichi Sankyo, and Takeda Oncology; personal fees from Pfizer, Roche/Genentech, Chugai, SFJ Pharmaceuticals, Voronoi, Biocartis, Novartis, Sanofi Oncology, Mirati Therapeutics, Transcenta, Silicon Therapeutics, Syndax, Nuvalent, Bayer, Esai, Allorion Therapeutics, Accutar Biotech, and AbbVie; and grants from Revolution Medicines and PUMA outside the submitted work. In addition, P.A. Jänne has a patent for EGFR mutations issued and licensed to LabCorp. C. Ambrogio reports other support from Revolution Medicines, Verastem, and Boehringer Ingelheim outside the submitted work. No disclosures were reported by the other authors.

B. Ricciuti: Conceptualization, validation, investigation, methodology, writing–original draft. J. Son: Validation, investigation, methodology, writing–original draft. J.J. Okoro: Validation, investigation, methodology. A. Mira: Validation, investigation, methodology. E. Patrucco: Validation, investigation, methodology. Y. Eum: Validation, investigation, methodology. X. Wang: Validation, investigation, methodology. R. Paranal: Validation, investigation, methodology. H. Wang: Formal analysis. M. Lin: Validation, investigation, methodology. H.M. Haikala: Validation, investigation, methodology. J. Li: Validation, investigation, methodology. Y. Xu: Validation, investigation, methodology. J.V. Alessi: Validation, investigation, methodology. C. Chhoeu: Methodology. A.J. Redig: Formal analysis. J. Köhler: Formal analysis. K.H. Dholakia: Methodology. Y. Chen: Methodology. E. Richard: Validation, investigation, methodology. M.-J. Nokin: Validation, investigation, methodology. D. Santamaria: Formal analysis. P.C. Gokhale: Formal analysis. M.M. Awad: Conceptualization, supervision, writing–original draft. P.A. Jänne: Conceptualization, supervision, writing–original draft. C. Ambrogio: Conceptualization, supervision, funding acquisition, validation, investigation, methodology, writing–original draft.

This work was funded by the Giovanni Armenise–Harvard Foundation, the Lung Cancer Research Foundation, the International Lung Cancer Foundation, the European Research Council under the European Union's Horizon 2020 research and innovation program (grant agreement No. 101001288) and AIRC under IG 2021 ID. 25737 project (to C. Ambrogio). B. Ricciuti's work was supported by the Conquer Cancer Foundation of ASCO YIA, and the IASLC Fellowship Award. C. Ambrogio is supported by the Zanon di Valgiurata family through Justus SS.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.