KRAS^G12C Mutation Is Associated with Increased Risk of Recurrence in Surgically Resected Lung Adenocarcinoma

Abstract

Purpose:

KRAS^G12C is the most common KRAS mutation in primary lung adenocarcinoma (LUAD). Phase I clinical trials have demonstrated encouraging clinical activity of KRAS^G12C inhibitors in the metastatic setting. We investigated disease-free survival (DFS) and tumor genomic features in patients with surgically resected KRAS^G12C-mutant LUAD.

Experimental Design:

Patients who underwent resection of stage I-III LUAD and next-generation sequencing (NGS) were evaluated. Exclusion criteria were receipt of induction therapy, incomplete resection, and low-quality NGS. Mutations were classified as KRAS wild-type (KRAS^wt), G12C (KRAS^G12C), or non-G12C (KRAS^other). DFS was compared between groups using the log-rank test; factors associated with DFS were assessed using Cox regression. Mutual exclusivity and co-occurrence, tumor clonality, and mutational signatures were assessed.

Results:

In total, 604 patients were included: 374 KRAS^wt (62%), 95 KRAS^G12C (16%), and 135 KRAS^other (22%). Three-year DFS was not different between KRAS-mutant and KRAS^wt tumors. However, 3-year DFS was worse in patients with KRAS^G12C than KRAS^other tumors (log-rank p=0.029). KRAS^G12C tumors had more lymphovascular invasion (51% vs. 37%; p=0.032) and higher tumor mutation burden (median [interquartile range], 7.0 [5.3–10.8] vs. 6.1 [3.5–9.7]; p=0.021), compared with KRAS^other tumors. KRAS^G12C mutation was independently associated with worse DFS on multivariable analysis. Our DFS findings were externally validated in an independent The Cancer Genome Atlas cohort.

Conclusions:

KRAS^G12C mutations are associated with worse DFS after complete resection of stage I-III LUAD. These tumors harbor more-aggressive clinicopathologic and genomic features than other KRAS-mutant tumors. We identify a high-risk group for whom KRAS^G12C inhibitors may be investigated to improve survival.

Introduction

The KRAS gene encodes an oncoprotein involved in key signaling pathways for tumor growth and differentiation and is the most frequently mutated oncogene in all cancer types (1). Somatic mutations in KRAS are found in 25% to 33% of primary lung adenocarcinoma (LUAD) (2,3). In stage IV LUAD, patients with KRAS-mutant primary tumors have demonstrated poor overall survival (4) and less clinical benefit from standard-of-care systemic therapies, compared with KRAS^wt tumors (5).

During the last few decades, the KRAS protein has been viewed as undruggable, owing to the lack of deep pockets for direct small-molecule-inhibitor binding. Consequently, efforts to develop targeted therapies for KRAS-mutant tumors have focused on inhibition of downstream effector proteins in the MAPK pathway, such as BRAF, MEK, and ERK. However, inhibition of these downstream targets is often accompanied by on-target, nontumor toxicities, due to the inhibition of this signaling pathway in normal cells (6,7). As a result, this narrow therapeutic index has precluded the successful clinical development of agents targeting KRAS (5,8).

The majority of KRAS mutations in LUAD are single-base substitutions (SBSs) in codons 12 or 13—the most common being G12C, which occurs in 13% to 16% of LUAD (3,9). The recent discovery of direct KRAS^G12C inhibition by use of a trapping mechanism has led to promising preclinical and early-phase drug development (1,10–12). For example, in two phase I clinical trials of sotorasib and adagrasib (NCT03600883 and NCT03785249), encouraging clinical activity has been observed in patients with non-small cell lung cancer (NSCLC) (13–15).

Given the encouraging early results in the development of KRAS^G12C inhibitors, a greater understanding of the genomic complexity of KRAS^G12C-mutant tumors and the oncologic outcomes of patients with these tumors is needed, especially in the setting of curative-intent surgery for earlier-stage disease (16,17). To address this knowledge gap, we investigated and compared tumor genomic features and disease-free survival (DFS) in patients with surgically resected LUAD harboring mutations in KRAS^G12C, compared to KRAS^other and KRAS^wt tumors. The results of our study offer insight into the potential impact that KRAS^G12C inhibitors may have in patients with early-stage LUAD who undergo surgery as their first treatment.

Materials and Methods

Study Population and Data Collection

Following institutional review board approval at Memorial Sloan Kettering Cancer Center, patients who underwent complete (R0) surgical resection for pathologic stage I-III LUAD from February 2010 to December 2018 and had targeted next-generation sequencing (NGS; MSK-IMPACT) (18) performed on the primary tumor were identified. All patients provided written informed consent to participate in the institutional review board–approved protocol, and all studies were conducted in accordance with the ethical guidelines of the Declaration of Helsinki. Exclusion criteria were as follows: receipt of induction therapy, microscopic (R1) or macroscopic (R2) residual disease, and low-quality NGS (see CONSORT diagram, Supplementary Figure 1). Tumors were classified according to primary tumor KRAS mutation status as either wild-type (KRAS^wt), KRAS(G12C) mutation (KRAS^G12C), or other KRAS mutation (KRAS^other).

Prospectively collected demographic, imaging, staging (American Joint Committee on Cancer 8^th edition), pathologic, genomic, recurrence, and follow-up data were reviewed. Predominant invasive LUAD histologic subtypes were classified as either lepidic, acinar, papillary, micropapillary, or solid (19). Follow-up was performed in accordance with National Comprehensive Cancer Network guidelines (20). Metachronous lesions were distinguished from recurrences using Martini and Melamed criteria (21), with confirmation of clonal relatedness using genomic data, as previously reported by our group (22).

MSK-IMPACT Sequencing

Tumor genomic profiling was performed and analyzed as previously described (18). The sequencing breadth of the IMPACT panel has increased over time, resulting in 8, 190, and 406 patients sequenced with 341-, 410-, and 468-gene panels, respectively. Tumor mutation burden (TMB) was defined as the total number of nonsynonymous single-nucleotide insertion or deletion mutations per megabase and was normalized by panel size by dividing the total number of mutations by the length of the coding region captured by each panel (0.98, 1.06, and 1.22 Mb in the 341-, 410-, and 468-gene panels, respectively). Fraction of genome altered (FGA) was computed from the output of the FACETS (Fraction and Allele-specific Copy number Estimates from Tumor Sequencing) algorithm, which provides accurate, purity- and ploidy-corrected, integer DNA copy number calls from sequenced samples (23). FGA is defined as the fraction of the genome that differs from the major integer copy number, which represents the integer total copy number spanning the largest portion of the genome.

Statistical and Genomic Data Analysis

The primary endpoint—DFS—was defined as the time from surgery to first recurrence or death from any cause; patients were otherwise censored at the time of last follow-up. DFS was estimated using the Kaplan-Meier method and compared between groups using the log-rank test. Median follow-up was estimated using the reverse Kaplan-Meier method. Patterns of recurrence were assessed using the cumulative incidence of locoregional recurrence (CIR-LR) and distant recurrence (CIR-DR) in the KRAS^G12C and KRAS^other groups. Patients with both locoregional and distant recurrence were included in the distant recurrence analysis. Clinicopathologic and genomic variables (including specific genes, pathways, TMB, and FGA) were compared between groups using Fisher’s exact test for categorical factors and the Wilcoxon rank-sum test for continuous factors. A total of two Cox regression models were used to quantify the association between clinicopathologic or genomic factors and DFS—(1) within the overall cohort and (2) within only the KRAS^G12C group—using hazard ratios (HRs) and 95% confidence intervals (CIs). Both regression analyses used identical lists of variables, with the exception of the KRAS-mutation group, which was included in only the Cox regression for the overall cohort. Multivariable models for all factors were adjusted for pathologic stage. To quantify the associations between specific genes and DFS, an additional univariable analysis was performed using all genes with an alteration frequency >8% after false-discovery rate (FDR) correction.

Analysis of specific somatic alterations was performed using OncoKB (24) to remove variants of unknown significance. For the analysis of co-occurrence and mutual exclusivity, we assessed all genes known to be drivers in LUAD (24). In total, 121 genes were identified at the intersection of the a priori pathway templates and the MSK-IMPACT panel. A pathway was considered to be altered in a tumor if at least one gene within the corresponding pathway template was altered. Mutual exclusivity and co-occurrence of alterations in genes and oncogenic cell signaling pathways were assessed using Fisher’s exact test, and p values were adjusted to correct for multiple comparisons using FDR correction.

Differences in primary tumor clonality were investigated between groups using the cancer cell fraction, as calculated by FACETS. Clonality assessment was able to be performed on 72 patients in the KRAS^G12C group and 95 patients in the KRAS^other group. Primary tumor clonality was defined as a cancer cell fraction >0.8, as in prior reports (25,26). Variant allele frequency is the fraction of sequence reads that contain a specific DNA variant, divided by the overall coverage at that locus.

Mutational signatures were computed using the most recent version of the SBS signatures defined in the Catalogue of Somatic Mutations in Cancer database for somatic mutation signature analysis (27). Tumors with ≥13.8 mutations/Mb were evaluated for the KRAS^G12C (n=67) and KRAS^other (n=83) groups, in accordance with the previously published threshold (28). Tumors were considered to have a detectable signature if the mean signature value was >0.1, as previously described (27).

Analyses were conducted using Stata 15.0 (StataCorp, College Station, TX) and R 3.5.1 (R Core Team, Vienna, Austria). Statistical tests were two-sided, and p<0.05 was considered statistically significant.

External Validation

External validation of the relationship between KRAS^G12C mutation and DFS was performed using The Cancer Genome Atlas (TCGA) LUAD data set (N=476) (20). Patients with pathologic stage I-III LUAD who did not receive induction therapy were included. Tumors were classified according to KRAS-mutation status. DFS was estimated using the Kaplan-Meier method and was compared between groups using the log-rank test.

Results

Clinicopathologic Characteristics

A total of 604 patients met the inclusion criteria (Table 1). KRAS mutation status was as follows: 374 KRAS^wt (62%), 95 KRAS^G12C (16%), and 135 KRAS^other (22%). The median age at resection was 68 years (interquartile range [IQR], 62–74), and two-thirds of patients were women (n=402 [67%]). Most patients (77%) had a history of smoking, with a median of 27 pack-years (IQR, 15–45). Patients with KRAS-mutant LUAD were more commonly smokers, compared with KRAS^wt LUAD. Sublobar resection was performed in 214 patients (35%). A majority of patients had pathologic stage I LUAD (n=447 [74%]); 95 patients (16%) had stage II LUAD, and 62 patients (10%) had stage III LUAD. Of note, patients with KRAS^G12C and KRAS^other tumors were more likely to have high (≥50%) programmed death-ligand 1 (PD-L1) expression (n=35/95 [37%] and n=47/135 [35%], respectively), compared with patients with KRAS^wt tumors (n=94/374 [25%]) (p=0.013). Adjuvant therapy was administered to 116 patients (19%) and was not statistically significantly different between groups. Regimens included chemotherapy only (n=95 [82%]), chemoradiation (n=18 [16%]), radiotherapy only (n=2 [1.7%]), and immunotherapy (n=1 [0.8%]). Median follow-up was 2.51 years (IQR, 1.74–3.30).

Table 1.

Comparison of clinicopathologic characteristics between the KRAS^wt, KRAS^G12C, and KRAS^other groups

Characteristic	Total Cohort (n=604)	KRASwt (n=374)	KRASG12C (n=95)	KRASother (n=135)
Age at resection, years	68 (62–74)	68 (62–74)	69 (64–75)	67 (61–73)
Sex
Male	202 (33)	130 (35)	34 (36)	38 (28)
Female	402 (67)	244 (65)	61 (64)	97 (72)
Smoking status
Never	138 (23)	125 (33)	4 (4.2)	9 (6.7)
Ever	466 (77)	249 (67)	91 (96)	126 (93)
Pack-years (n=465)	27 (15–45)	10 (0–30)	27 (16–43)	30 (14–46)
FEV1 (n=592)	94 (83–106)	95 (84–108)	90 (77–104)	94 (83–105)
DLCO (n=586)	84 (69–97)	87 (72–99)	79 (64–92)	80 (66–95)
SUVmax (n=537)	3.7 (1.9–7.2)	3.4 (1.8–6.7)	4.1 (2.3–8.1)	4.0 (2.1–7.5)
Operative approach
VATS	521 (86)	324 (87)	79 (83)	118 (87)
Open	83 (14)	50 (13)	16 (17)	17 (13)
Operative procedure
Lobectomy or pneumonectomy	390 (65)	245 (66)	61 (64)	84 (62)
Sublobar	214 (35)	129 (34)	34 (36)	51 (38)
Pathologic tumor size, cm	1.8 (1.2–2.8)	1.8 (1.2–2.8)	1.8 (1.3–2.7)	2.0 (1.4–3.1)
LVI (n=599)	248 (41)	150 (41)	49 (52)	49 (36)
VPI	109 (18)	70 (19)	20 (21)	19 (14)
STAS (n=544)	337 (62)	183 (55)	68 (75)	86 (70)
Histologic subtype
Lepidic	88 (15)	59 (16)	10 (11)	19 (14)
Acinar	368 (61)	218 (58)	64 (67)	86 (64)
Papillary	43 (7.1)	28 (7.5)	4 (4.2)	11 (8.1)
Micropapillary	37 (6.1)	25 (6.7)	5 (5.3)	7 (5.2)
Solid	68 (11)	44 (12)	12 (13)	12 (8.9)
pN status (n=603)
N0	506 (84)	311 (83)	75 (79)	120 (89)
N1 or N2	97 (16)	62 (17)	20 (21)	15 (11)
pStage
I	447 (74)	278 (74)	66 (69)	103 (76)
II	95 (16)	61 (16)	16 (17)	18 (13)
III	62 (10)	35 (9.4)	13 (14)	14 (10)
PD-L1 status
None (<1%)	343 (57)	233 (62)	45 (47)	65 (48)
Low (1–49%)	85 (14)	47 (13)	15 (16)	23 (17)
High (≥50%)	176 (29)	94 (25)	35 (37)	47 (35)
Adjuvant therapy	116 (19)	68 (18)	21 (22)	27 (20)

Data are no. (%) or median (interquartile range). DLCO, diffusing capacity of the lungs for carbon monoxide; FEV1, forced expiratory volume in 1 second; LVI, lymphovascular invasion; PD-L1, programmed death-ligand 1; STAS, spread through air spaces; SUVmax; maximum standardized uptake value; VPI, visceral pleural invasion; wt, wild-type.

KRAS Mutation Status, Recurrence Patterns, and Survival

The specific types of KRAS alterations and associated frequencies are listed in Supplementary Table 1. The KRAS^other group predominantly comprised patients with G12D (n=33/135 [24%]), G12V (n=32/135 [24%]), and G12A (n=23/135 [17%]) mutations. Three-year DFS for the overall cohort was 73.6% (95% CI, 69.5%−78.0%). DFS was not statistically significant different between patients with any KRAS mutation (3-year DFS, 75.3% [95% CI, 69.2%−82.0%]) and patients with KRAS^wt LUAD (3-year DFS, 72.5% [95% CI, 67.1%−78.3%]), with a hazard ratio (HR) of 0.95 (95% CI, 0.68–1.34) (p=0.785; Figure 1A). However, among patients with KRAS-mutant tumors (n=230), DFS was substantially worse for patients with KRAS^G12C mutations (3-year DFS, 68.7% [95% CI, 58.9%−80.3%]) than for those with KRAS^other mutations (3-year DFS, 80.0% [95% CI, 72.4%−88.3%]), with an HR of 1.82 (95% CI, 1.06–3.15) (p=0.029; Figure 1B).

Figure 1. Association between KRAS mutation status and disease-free survival in the study cohort (N=604).

Three-year disease-free survival for (A) all KRAS-mutant (KRAS^mut) tumors versus KRAS wild-type (KRAS^wt) tumors and for (B) KRAS^G12C tumors versus all other KRAS-mutant tumors (KRAS^other).

The CIR-LR was higher in the KRAS^G12C group (3-year CIR-LR, 8.0% [95% CI, 2.8%−16.8%]) than in the KRAS^other group (3-year CIR-LR, 2.3% [95% CI, 0.6%−6.1%) but did not reach statistical significance (Gray’s p=0.119; Supplementary Figure 2A). The subtle observation of higher CIR-LR appeared to occur past the 2-year mark after surgery. Similarly, the CIR-DR was higher in the KRAS^G12C group (3-year CIR-DR, 20.8% [95% CI, 12.6%−30.4%]) than in the KRAS^other group (3-year CIR-DR, 13.0% [95% CI, 7.2%−20.7%]) but again did not reach statistical significance (Gray’s p=0.070; Supplementary Figure 2B). The observed divergence, however, appeared to occur earlier in the follow-up period. Overall, at 3 years after surgery, the CIR-DR was more than double the CIR-LR across both groups.

Clinicopathologic Factors Associated with DFS

On univariable analysis (N=604), the following factors were associated with DFS: diffusing capacity of the lungs for carbon monoxide (DLCO), primary tumor maximum standardized uptake value, open (thoracotomy) resection, pathologic tumor size, lymphovascular invasion (LVI), visceral pleural invasion (VPI), spread through air spaces (STAS), micropapillary or solid histologic subtype, pathologic node positivity, pathologic stage II or III LUAD, KRAS^G12C mutation, TMB, and FGA (p<0.1; Supplementary Table 2). On multivariable analysis, after adjustment for pathologic stage, KRAS^G12C mutation was independently associated with worse DFS (HR, 1.84 [95% CI, 1.01–3.36]; p=0.046) (Table 2). In addition, DLCO (HR, 0.99 [95% CI, 0.98–1.00]; p=0.031), LVI (HR, 2.36 [95% CI, 1.43–3.91]; p=0.001), VPI (HR, 1.66 [95% CI, 1.10–2.51]; p=0.015), STAS (HR, 1.81 [95% CI, 1.02–3.23]; p=0.044), and pathologic stage II or III (vs. stage I; HR, 1.80 [95% CI, 1.18–2.76]; p=0.007) were also independently associated with worse DFS.

Table 2.

Multivariable analysis for disease-free survival for the total cohort and the KRAS^G12C group

Group/Variablea	HRa	95% CI	p
Total cohort (N=604)
Primary tumor SUVmax	1.03	1.00–1.07	0.065
DLCO	0.99	0.98–1.00	0.031
LVI	2.36	1.43–3.91	0.001
VPI	1.66	1.10–2.51	0.015
STAS	1.81	1.02–3.23	0.044
pStage
I	Ref	—	—
II or III	1.80	1.18–2.76	0.007
KRAS mutation status
Non-G12C	Ref	—	—
Wild-type	1.45	0.85–2.47	0.2
G12C	1.84	1.01–3.36	0.046
KRASG12C group (n=95)
LVI	9.57	2.20–41.54	0.003
VPI	2.25	1.03–4.94	0.042
Histologic subtype
Lepidic, acinar, or papillary	Ref	—	—
Micropapillary or solid	2.15	0.95–4.87	0.067

CI, confidence interval; DLCO, diffusing capacity of the lungs for carbon monoxide; HR, hazard ratio; LVI, lymphovascular invasion; STAS, spread through air spaces; SUVmax, maximum standardized uptake value; VPI, visceral pleural invasion

^aMultivariable models for all factors were adjusted for pathologic stage.

Factors prognostic for DFS were then investigated within the KRAS^G12C group (n=95). On univariable analysis, age at resection, pack-years, DLCO, primary tumor maximum standardized uptake value, primary tumor size, LVI, VPI, STAS, micropapillary or solid subtype, pathologic node positivity, pathologic stage II or III, and TMB were associated with DFS (p<0.1; Supplementary Table 3). On multivariable analysis, primary tumor LVI (HR, 9.57 [95% CI, 2.20–41.54]; p=0.003) and VPI (HR, 2.25 [95% CI, 1.03–4.94]; p=0.042) were independently associated with DFS (Table 2).

Clinicopathologic and Genomic Differences Between Patients with KRAS^G12C and KRAS^other Tumors

Clinicopathologic and genomic factors were then compared between patients with KRAS^G12C and KRAS^other tumors (Figure 2). LVI (51% vs. 37%; p=0.032) and pathologic lymph node metastasis (21% vs. 12%; p=0.059) were more common in the KRAS^G12C group. KRAS^G12C tumors were also found to have higher TMB (median [IQR], 7.0 [5.3–10.8] vs. 6.1 [3.5–9.7]; p=0.021) and FGA (x100; median [IQR], 3.8 [0.4–8.9] vs. 1.5 [0.2–7.3]; p=0.053).

Figure 2. Clinicopathologic and genomic features of KRAS-mutant tumors.

Comparison of clinicopathologic variables, genomic factors, and specific genes between the KRAS^G12C and other KRAS mutation (KRAS^other) groups. All genes with an alteration frequency >8% are shown. *p<0.05 for comparison between groups using Fisher’s exact test for categorical factors and the Wilcoxon rank-sum test for continuous factors. Acn, acinar; FGA, fraction of genome altered; Lep, lepidic; LVI, lymphovascular invasion; MiP, micropapillary; Pap, papillary; Sol, solid; STAS, spread through air spaces; TMB, tumor mutation burden; VPI, visceral pleural invasion.

The ten most commonly altered genes were compared between groups (Figure 2). Although no differences reached statistical significance, KRAS^G12C tumors had more STK11 and NF1 mutations than KRAS^other tumors. Conversely, KRAS^other tumors were nearly twice as likely to have a truncating mutation in RBM10, an RNA-binding protein and splicing regulator (29).

Genes Associated with DFS

To determine whether the differences in alteration frequencies described above were also prognostic for DFS, an additional univariable analysis was performed using the same list of genes (Supplementary Table 4). In the overall cohort, two genes (TP53: HR, 1.65 [95% CI, 1.18–2.31]; Q=0.024; and RBM10: HR, 0.43 [95% CI, 0.22–0.84]; Q=0.024) were associated with DFS after FDR correction, a finding that was previously reported by our group (30). In the KRAS^G12C group, no genes were statistically significantly associated with DFS after FDR correction.

Mutual Exclusivity and Co-occurrence Patterns

Mutual exclusivity and co-occurrence patterns of individual genes and oncogenic pathways were then explored between groups. In the overall cohort, STK11 and KEAP1 were significantly co-occurrent, as previously described (31), whereas any KRAS mutation was predictably mutually exclusive with other RAS pathway genes (e.g., EGFR, BRAF, MET) (FDR-p<0.05; Supplementary Figure 3A). Interestingly, no genes were found to be statistically significantly co-occurrent with KRAS^G12C mutation or KRAS^other mutation. Among KRAS^G12C tumors, TP53 and NF1 mutations were observed to be significantly co-occurrent (p=0.03; Supplementary Figure 3B). Finally, among KRAS^other tumors, STK11 and KEAP1 were again found to statistically significantly co-occur (p=0.007; Supplementary Figure 3C).

Clonality Patterns and Somatic Mutation Signatures of KRAS^G12C and KRAS^other Tumors

Next, differences in primary tumor clonality were explored between the KRAS^G12C and KRAS^other groups. KRAS mutation was found to be a clonal event in 90% of KRAS^G12C tumors (n=65/72) and 91% (n=86/95) of KRAS^other tumors (Supplementary Figure 4). This confirms that KRAS mutations are an early, truncal alteration in the evolution of LUAD tumors.

Detectable somatic mutation signatures were then investigated between the KRAS^G12C and KRAS^other groups (Figure 3). Overall, the smoking signature (SBS4) was most commonly detectable in both groups (33% vs. 42%, respectively; p=0.24). Interestingly, KRAS^G12C tumors had a statistically significantly higher prevalence of the SBS2 signature (attributed to APOBEC activity(27)), compared with KRAS^other tumors (16% vs. 6%; p=0.04). Conversely, KRAS^other tumors had higher rates of SBS16 (17% vs. 6%; p=0.041) and SBS18 (10% vs. 2%; p=0.043), compared with KRAS^G12C tumors.

Figure 3. Analysis of somatic mutational signatures in KRAS-mutant tumors.

Comparison of the relative frequencies of 8 detectible mutational signatures between the KRAS^G12C and other KRAS mutation (KRAS^other) groups, and proposed etiologies for these signatures.

External Validation of DFS Association

To externally validate our findings, we queried a TCGA data set for all patients with pathologic stage I-III LUAD who did not undergo induction therapy; a total of 476 patients were identified. KRAS mutation status was as follows: 323 KRAS^wt (68%), 63 KRAS^G12C (13%), and 90 KRAS^other (19%) (Supplementary Table 5). No difference was noted in 3-year DFS between patients with any KRAS mutation and patients with KRAS^wt LUAD (HR, 1.18 [95% CI, 0.89–1.56]; p=0.244; Figure 4A). However, among patients with KRAS-mutant tumors (n=153), patients with KRAS^G12C mutation were again found to have significantly worse 3-year DFS than patients with KRAS^other mutation (31.3% versus 54.2%, respectively; HR, 1.88 [95% CI, 1.18–3.00]; p=0.007; Figure 4B), confirming the relationship between KRAS^G12C mutation and DFS in our study.

Figure 4. Association between KRAS-mutation status and disease-free survival in The Cancer Genome Atlas cohort (N=476).

Three-year disease-free survival for (A) all KRAS-mutant (KRAS^mut) tumors versus KRAS wild-type (KRAS^wt) tumors and for (B) KRAS^G12C tumors versus all other KRAS-mutant tumors (KRAS^other).

Discussion

The recent development of KRAS^G12C inhibitors and their promising early results in phase I clinical trials have necessitated the genomic characterization and examination of long-term oncologic outcomes in patients with surgically resected KRAS^G12C-mutant tumors. We have shown that, compared with KRAS^other mutations, KRAS^G12C primary tumor mutations portended worse DFS in both our institutional data set as well as an external TCGA data set. KRAS^G12C mutation was also independently associated with worse DFS in our cohort. KRAS^G12C tumors appeared to contain a greater proportion of aggressive pathologic features (LVI, positive lymph nodes) and genomic perturbations (higher TMB and FGA) than KRAS^other tumors. Through extensive genomic characterization, we discovered that no common oncogenes or tumor suppressors were co-occurrent or mutually exclusive with KRAS^G12C tumors. However, in the overall cohort, STK11 and KEAP1 were significantly co-occurrent, a finding supported by studies in the metastatic setting (32). Finally, we have shown that the vast majority of both KRAS^G12C and KRAS^other tumors harbor clonal populations of KRAS-mutant cells, confirming that acquisition of this alteration is an early event in the mutagenesis of these tumors.

KRAS somatic mutations have been shown to be associated with decreased survival in prior studies, with 5-year overall survival ranging from 22% to 30% in patients with KRAS-mutant NSCLC (4,32). Although this association with poor prognosis is well documented for advanced disease, the influence of KRAS mutation on survival in patients with early-stage disease remains poorly characterized. In a smaller series (N=179), Nadal and colleagues found overall KRAS mutation was associated with worse DFS (log-rank p=0.006) and overall survival (log-rank p=0.046) (17). However, this cohort included all stages of LUAD, with 20% of KRAS-mutant tumors (n=21) being stage III or IV in this study. When the analysis was repeated for only patients with stage I disease (n=121), the survival difference substantially diminished (log-rank p=0.049). In the two larger, early-stage cohorts in our study (study cohort, N=604; TCGA external validation cohort, N=476), 3-year DFS was not statistically significantly different between patients with KRAS-mutant and KRAS^wt tumors, indicating there was no prognostic significance for the overall KRAS mutation population.

Targeting KRAS^G12C—the most prevalent of the KRAS alterations, present in up to half of KRAS-mutant tumors (33)—has led to encouraging tumor responses in phase I clinical trials (13,34). In the Nadal study, the 2-year DFS for patients with KRAS^G12C tumors was 42.9%, compared with approximately 65.0% for patients with other KRAS-mutant tumors. However, the KRAS^G12C group comprised only 35 patients, with an unreported stage distribution. In the present study, 3-year DFS for patients with KRAS^G12C tumors was also substantially worse than that for patients with KRAS^other tumors in our institutional cohort (68.7% vs 80.0%, respectively). Additionally, analysis of the TCGA database yielded similar results, with a 3-year DFS of 31.3% versus 54.2%, respectively, in this external validation cohort. Interestingly, the observed survival differences do not appear to be linked to comutation patterns, as no genes were significantly co-occurrent with KRAS^G12C alteration on genomic analysis, and likewise no other genes were prognostic for DFS within this group. One possible explanation for this survival detriment may involve the lack of immunogenicity in KRAS^G12C tumors. A recent study by Aredo and colleagues found that these tumors were significantly associated with low (1%−49%) PD-L1 expression (35). In the present study, KRAS^G12C-mutant tumors with high PD-L1 expression (≥50%)—although these were in the minority—were significantly more common in the KRAS^G12C group than in the KRAS^wt group (37% vs. 23%; p=0.013). On the basis of preclinical data, KRAS^G12C inhibitors enhance antitumor immunity, which may be helpful for eradicating micrometastatic disease (9). The ability to target this alteration in a newly defined high-risk population—whether in the adjuvant or neoadjuvant setting—has major therapeutic implications.

Numerous studies have shown that KRAS mutation is an early event in lung tumorigenesis (36–38). In a recent analysis of NSCLC tumor evolution, primary tumors with alterations in any of four genes (TP53, KEAP1, STK11, and EGFR) were shown to have a higher proportion of clonal tumor cell populations, signifying that these genes play a role in tumor initiation (36). In a separate study involving multi-region sequencing, KRAS mutations were present in both minimally invasive adenocarcinoma or adenocarcinoma in situ and paired invasive LUAD, suggesting that KRAS mutation is an early mutagenic event and an indicator of malignant transition (37). Furthermore, in the landmark TRACERx NSCLC tumor evolution study, 88% of samples with KRAS mutations were deemed to be clonal populations (38). In our study, 90% of overall KRAS-mutant tumors, as well as 90% of KRAS^G12C-mutant tumors, were found to harbor clonal cancer cell populations. This knowledge, coupled with the increased risk of recurrence in patients with KRAS^G12C tumors in this study, provides encouraging evidence to support the use of KRAS^G12C inhibitors for preventing or delaying tumor relapse in patients with early-stage LUAD.

Recent large-scale analyses have identified numerous somatic mutational signatures across the spectrum of cancer types (27). Although somatic mutational signature analysis most commonly relies on whole-genome or whole-exome sequencing data, our group has previously shown the feasibility of using MSK-IMPACT (28). In the present study, KRAS^G12C tumors were associated with high activity of the putative APOBEC mutational signature (SBS2), whereas KRAS^other tumors were associated with the putative reactive oxygen species signature (SBS18). Loss-of-function alterations in APOBEC-related genes lead to DNA hypermutation and inaccurate RNA editing and are associated with tumorigenesis as well as drug resistance (19,39,40). Furthermore, a recent analysis of immune-response-related mutational signatures in NSCLC showed that high TMB combined with APOBEC-related mutational signatures was predictive of response to immunotherapy (41), revealing an intriguing new patient population that may derive benefit from these agents.

Limitations of our study include a low number of death events in our early-stage cohort such that overall survival was unable to be explored as an outcome. Another limitation is that our findings require further validation from international data sets such as those from Asia, where the incidence of KRAS mutations in LUAD is lower, which would provide useful information regarding the implications of KRAS^G12C mutation and inhibition in a geographically diverse population. Despite these limitations, this externally validated analysis from the largest data set of patients with resected KRAS^G12C LUAD provides evidence to support the investigation of sotorasib, adagrasib (MRTX849), or other KRAS^G12C inhibitors in the adjuvant setting, with the goal of improving DFS.

Mutations in KRAS^G12C are the most common KRAS mutations in lung cancer and are independently prognostic for poor DFS after complete resection of stage I-III LUAD. We have shown that the vast majority of KRAS^G12C tumors harbor clonal populations of KRAS^G12C-mutant cells, and these tumors appear to harbor more genomic perturbations and aggressive clinicopathologic features than other KRAS-mutant tumors. Additionally, KRAS^G12C mutation was not found to be co-occurrent with actionable alterations in other common LUAD driver genes. Our findings provide evidence supporting the investigation of KRAS^G12C inhibitors in the adjuvant setting in this vulnerable patient population.

Statement of Translational Relevance:

KRAS^G12C is the most prevalent of the KRAS alterations in primary lung adenocarcinoma (LUAD), present in up to half of cases. The association between KRAS somatic mutations and decreased survival is well-documented in metastatic LUAD. However, the influence of KRAS mutation on survival in early-stage disease remains poorly characterized. The recent development of KRAS^G12C inhibitors and their promising early results in phase I clinical trials necessitates the genomic characterization and examination of long-term oncologic outcomes in patients with surgically resected KRAS^G12C-mutant tumors. We have shown that, compared with KRAS^other mutations, KRAS^G12C primary tumor mutations portended worse disease-free survival in both our institutional data set and an external TCGA data set. KRAS^G12C tumors contain a greater proportion of aggressive pathologic features (LVI, positive lymph nodes) and genomic perturbations (higher TMB and FGA) than KRAS^other tumors. We identify a high-risk group for whom KRAS^G12C inhibitors may be investigated to improve survival.