Purpose:

The majority of broad-panel tumor genomic profiling has used a gene-centric approach, although much of that data is unused in clinical decision making. We hypothesized that a pathway-centric approach using next-generation sequencing (NGS), combined with conventional clinicopathologic features, may better predict disease-free survival (DFS) in early stage lung adenocarcinoma.

Experimental Design:

Utilizing our prospectively maintained database, we analyzed 492 patients with primary, untreated, completely surgically resected lung adenocarcinoma. Ten canonical pathways were analyzed using broad-panel NGS. The correlations of DFS and number (and type) of pathway (NPA) were analyzed using the Kaplan–Meier method and log-rank test. Associations between altered pathways and clinicopathologic variables, as well as identification of actionable therapeutic strategies were explored.

Results:

Median NPA for the cohort was two (range, 0–5). Smoking status, solid morphologic appearance on preoperative CT, maximal standardized uptake value, pathologic tumor size, aggressive histologic subtype, lymphovascular invasion, visceral pleural invasion, and positive lymph nodes were significantly associated with NPA (P < 0.05). Of 543 actionable genetic alterations identified, 455 (84%) were within the RTK/RAS pathway. A total of 86 tumors had actionable therapeutic genomic alterations in >1 pathway. On multivariable analysis, higher NPA was significantly associated with worse DFS (HR, 1.31; P = 0.014).

Conclusions:

NPA and specific pathway alterations are associated with clinicopathologic features in patients with surgically resected lung adenocarcinoma. Cell cycle, Hippo, TGFβ, and p53 pathway alterations are associated with poor DFS. Finally, NPA is an independent risk factor for poor DFS in our cohort.

See related commentary by Blakely, p. 7269

Translational Relevance

Broad-panel tumor genomic profiling typically uses a gene-centric approach for analysis and to guide clinical decision making, although most of the genomic data remain unused. We hypothesized that a pathway-centric approach using next-generation sequencing (NGS) may offer additional information to identify patients at high risk of recurrence following resection of lung adenocarcinoma. We interrogated 492 lung adenocarcinoma specimens and identified number of pathway alterations (NPA) to be independently associated with disease-free survival. Moreover, NPA and specific pathway perturbations were associated with clinicopathologic features of tumors at high risk of recurrence. Finally, we identified specific pathway alteration cooccurrences that implicate possible functional synergies for targeting both pathways in patients at high risk of recurrence. Collectively, these observations highlight the prognostic importance and potential discovery of therapeutic vulnerabilities when a tumor genomic pathway-centric analysis is utilized using broad-panel NGS.

Broad-panel genomic sequencing is increasingly used to select targeted therapies and precision medicine strategies for patients with lung cancer (1). However, a recent study found that <5% of patients who underwent broad-panel genomic sequencing benefited from it, compared with patients who received only EGFR and/or ALK testing (2). This raises the possibility that much of the information included in genomic sequencing panels is currently not well-utilized. Heretofore, the majority of tumor genomic analyses have focused on gene-centric approaches that detect driver alterations and identify functionally irrelevant passenger events, particularly in solid tumors with higher mutational burdens (3). Cancer cell biology and its resulting clinical phenotypes are driven by different pathways, many of which have upstream oncogenic alterations that lead to similar downstream alterations and phenotypes within the same pathway (3, 4). Therefore, a pathway-centric approach has been proposed to better identify functional alterations in selected oncogenic pathways, to explore cooccurrence and mutual exclusivity between pathways, and to identify relevant pathway alteration(s) that may be therapeutically exploited. We recently examined The Cancer Genome Atlas (TCGA) and comprehensively identified 10 canonical pathways covering 89% of 9,125 different solid tumor types, which now provides a curated and standardized pipeline to perform pathway-centric next-generation sequencing (NGS) studies (5).

Lung adenocarcinoma is a genomically well-annotated malignancy with a high background mutational burden. The majority of lung adenocarcinoma genomic analyses have focused on metastatic disease (6), with the 2014 TCGA analysis the only study to examine genomic changes in earlier stage lung adenocarcinoma (7). While surgical resection is the preferred treatment for early stage lung adenocarcinoma, there remains a relatively high recurrence rate, even in pathologic node-negative disease (8–12). Therefore, a significant knowledge gap exists in identifying patients at risk for recurrence following complete surgical resection of lung adenocarcinoma. To address this, we performed a comprehensive analysis to investigate associations between tumor genomic pathway alterations and selected clinicopathologic features and disease-free survival (DFS) in patients with lung adenocarcinoma who have undergone complete resection.

Patient cohort

This study was approved by the Institutional Review Board at Memorial Sloan Kettering. Using our prospectively maintained database, we identified patients who underwent complete surgical resection (R0) for lung adenocarcinoma and had targeted NGS (MSK-IMPACT) performed on their primary tumor between February 2008 and January 2018. Exclusion criteria included induction therapy, microscopic or macroscopic residual disease (R1/R2 resection), low-quality NGS, stage IV disease, and mixed tumor type (Supplementary Fig. S1).

Clinical characteristics, preoperative CT, PET images, and pathologic reports [8th edition American Joint Committee on Cancer (AJCC) Cancer Staging Manual] were reviewed. Follow-up was performed in accordance with NCCN guidelines (13). Recurrences were distinguished from metachronous tumors using Martini and Melamed criteria, with confirmation from pathologic and genomic relatedness when available (14).

MSK-IMPACT sequencing

Sequencing for MSK-IMPACT was performed as described previously (15). Patient clinicopathologic data were matched with genomic data and visualized using the cBioPortal for Cancer Genomics (16, 17). Tumor DNA and corresponding patient-matched blood DNA were extracted. All exons and selected introns were sequenced using the MSK-IMPACT panel to identify somatic alterations, copy number alterations, and mutations. Median sequencing coverage was 764X (range, 164–1,424). Selected panel sizes have been used over time (341-, 410-, and 468-gene panels for 14, 228, and 250 patients, respectively). Tumor mutational burden (TMB) was defined as the total number of nonsynonymous single-nucleotide or insertion/deletion mutations divided by the number of Mbs in the coding region captured by each panel (0.98, 1.06, and 1.22 Mb in the 341-, 410-, and 468-gene panels, respectively; ref. 18). We have previously shown that TMB calculations using this NGS panel are strongly associated with the TMB assessed by whole-exome sequencing (18). The fraction of genome altered (FGA) was defined as the fraction of log2 copy number variation (gain or loss) >0.2 divided by the size of the genome whose copy number was profiled. FGA was corrected for tumor purity, ploidy, and clonal heterogeneity using the FACET method (19).

Pathway alteration and therapeutic actionability identification

We evaluated 10 canonical signaling pathways using the templates provided in the signaling pathways manuscript from the TCGA PanCancer Atlas project (5). The pathways analyzed were (i) cell cycle, (ii) Hippo, (iii) Myc, (iv) Notch, (v) oxidative stress response/Nrf2, (vi) PI3K, (vii) receptor tyrosine kinase (RTK)/RAS/MAPK, (viii) TGFβ, (ix) p53, and (x) β-catenin/Wnt. In total, 109 genes were identified at the intersection of the a priori pathway templates (5) and the MSK-IMPACT panel (Supplementary Table S1). A tumor was considered “altered” in the specific pathway when ≥1 gene relative to control in the corresponding pathway template was altered. The status of specific pathways was determined to be either altered or wild-type for each patient. Number of pathway alterations (NPA) was calculated as the total number of altered pathways out of the 10 identified pathways for each patient.

Therapeutic actionability information was annotated using OncoKB on June 5, 2018 (20). Each potentially actionable genomic alteration was stratified into 1 of 6 levels. Detailed methods used for stratification of therapeutic actionable genetic events are included in the Supplementary Materials and Methods. CDK4 and MDM2 genomic data from the TCGA PanCancer Atlas lung adenocarcinoma cohort (6) were accessed using cBioPortal.

Statistical analysis

Associations between clinicopathologic and genomic characteristics, specific pathway alterations, and NPA were analyzed using Fisher exact test or Cochran–Armitage test for categorical variables and Wilcoxon rank-sum test or Spearman rank correlation test for continuous variables. When comparisons were repeatedly performed across 10 pathways, P values were adjusted using the FDR method when needed.

The primary outcome of interest was DFS, defined as the duration between surgery and recurrence or death without recurrence. Patients were censored at the last follow-up. Median follow-up was calculated on the basis of the reverse Kaplan–Meier approach (21). DFS was estimated using the Kaplan–Meier approach and compared between clinicopathologic characteristics and pathway alterations using log-rank tests, stratified by pathologic stage where appropriate. For the DFS analysis, NPA was considered a linear factor, as there was inadequate evidence to reject the linearity assumption (Wald F-Statistic χ2 = 0.09; P = 0.8) using restricted cubic splines.

The primary objective was to quantify the prognostic value of NPA. The relationships between NPA and clinically relevant factors were quantified using Cox proportional hazards models, stratified by pathologic stage [except in the cases of stage, tumor size, lymph node status, and visceral pleural invasion (VPI)]. A multivariable model was constructed starting with all factors with P < 0.1 in the univariable analyses. Multiple imputations were conducted to address missing data (details in Supplementary Data). To avoid loss of information by using the simple summary of NPA, we compared an NPA-only model with a model that included all 10 pathways as individual variables in a multivariable penalized Cox model. Performance of each model was quantified as discrimination (C-index) or calibration (details in Supplementary Data).

Pathway or gene mutual exclusivity or cooccurrence was analyzed using Fisher exact test, and P values were adjusted (FDR method). All analyses were two-sided, and P < 0.05 was considered statistically significant. Statistical analyses were conducted using Stata 13.1 (Stata) and R 3.5.1.

In total, 492 patients met the inclusion criteria, of whom 356 (72%) had stage I disease. Median follow-up was 19 months [95% confidence interval (CI), 18.4–20.1]. Alterations in common lung adenocarcinoma driver genes included 40% KRAS, 26% EGFR, 5% BRAF, 1.6% ALK, 0.8% ROS1, and 0.6% MET (Fig. 1A). Alterations in these driver genes were found to be evenly distributed across all NPA groups. The alteration frequencies of the 109 genes utilized for pathway analysis are included in Supplementary Table S2. Median NPA was 2 (range, 0–5; Fig. 1A; Table 1). The most and least frequently altered pathways were RTK/RAS [n = 415 patients (84%)] and TGFβ [n = 13 patients (3%)]. Of the 164 patients with only one pathway alteration, 141 (86%) had the RTK/RAS pathway altered. Comparatively, 90% of patients (160/178) with NPA = 2, 91% (86/94) with NPA = 3, 85% (23/27) with NPA = 4, and 83% (5/6) with NPA = 5 had the RTK/RAS pathway altered. The RTK/RAS pathway was divided into two pathways as a function of NPA to better visualize the alteration frequencies of individual genes in the RTK and RAS/RAF pathways (Supplementary Fig. S2). Median normalized TMB was 4.7 (range, 0–164.2), and median FGA was 4.3% (range, 0%–49.8%).

Figure 1.

Association between pathway alterations and genomic features. A, Oncoprint of all patients. B, Violin plot of TMB versus NPAs. C, Violin plot of FGA versus NPA. D, Scatterplot of TMB versus FGA. Bar plots of TMB (E) and FGA (F) versus pathway alteration status [altered and wild-type (WT)]. SE displayed as error bar. *, logarithmic scale.

Figure 1.

Association between pathway alterations and genomic features. A, Oncoprint of all patients. B, Violin plot of TMB versus NPAs. C, Violin plot of FGA versus NPA. D, Scatterplot of TMB versus FGA. Bar plots of TMB (E) and FGA (F) versus pathway alteration status [altered and wild-type (WT)]. SE displayed as error bar. *, logarithmic scale.

Close modal
Table 1.

Patient characteristics (N = 492)

CharacteristicNo. (%)
Age at surgery, years, median (range) 69 (34–87) 
Sex  
 Male 326 (66) 
 Female 166 (34) 
Smoking status  
 Ever 380 (77) 
 Never 112 (23) 
Pack-yearsa 20 (0–120) 
Tumor morphologic appearance, CT criteria  
 Pure ground glass opacity 72 (15) 
 Mixed ground glass opacity/subsolid 90 (18) 
 Solid 330 (67) 
Tumor maximal standardized uptake value, median (range) 3.4 (0–31) 
Procedure type  
 Lobectomy 340 (69) 
 Segmentectomy 47 (9) 
 Wedge 102 (21) 
 Pneumonectomy 3 (1) 
Pathologic tumor size, cm, median (range) 1.8 (0.08–19.5) 
Predominant histologic subtype  
 Lepidic, AIS, MIA 73 (15) 
 Acinar 263 (53) 
 Papillary 37 (8) 
 Micropapillary 32 (7) 
 Solid 60 (12) 
 Other 27 (5) 
Lymphovascular invasion  
 Positive 302 (62a
 Negative 185 (38a
Visceral pleural invasion  
 Positive 77 (16a
 Negative 414 (84a
Lymph node status  
 Positive 83 (17a
 Negative 394 (83a
Pathologic stage  
 I 356 (72) 
 II 80 (16) 
 III 56 (11) 
Pathway alteration  
 Cell cycle 77 (16) 
 Hippo 14 (3) 
 Myc 36 (7) 
 Notch 19 (4) 
 Nrf2 27 (5) 
 PI3K 119 (24) 
 TGFβ 13 (3) 
 RTK/RAS 415 (84) 
 p53 201 (41) 
 Wnt 19 (4) 
Number of pathway alterations  
 0 23 (5) 
 1 164 (33) 
 2 178 (36) 
 3 94 (19) 
 4 27 (6) 
 5 6 (1) 
TMB, median (range) (IQR) 4.7 (0–164.2) (2.5–8.2) 
FGA, median (range) (IQR) 0.25 (0–1) (0.1–0.5) 
CharacteristicNo. (%)
Age at surgery, years, median (range) 69 (34–87) 
Sex  
 Male 326 (66) 
 Female 166 (34) 
Smoking status  
 Ever 380 (77) 
 Never 112 (23) 
Pack-yearsa 20 (0–120) 
Tumor morphologic appearance, CT criteria  
 Pure ground glass opacity 72 (15) 
 Mixed ground glass opacity/subsolid 90 (18) 
 Solid 330 (67) 
Tumor maximal standardized uptake value, median (range) 3.4 (0–31) 
Procedure type  
 Lobectomy 340 (69) 
 Segmentectomy 47 (9) 
 Wedge 102 (21) 
 Pneumonectomy 3 (1) 
Pathologic tumor size, cm, median (range) 1.8 (0.08–19.5) 
Predominant histologic subtype  
 Lepidic, AIS, MIA 73 (15) 
 Acinar 263 (53) 
 Papillary 37 (8) 
 Micropapillary 32 (7) 
 Solid 60 (12) 
 Other 27 (5) 
Lymphovascular invasion  
 Positive 302 (62a
 Negative 185 (38a
Visceral pleural invasion  
 Positive 77 (16a
 Negative 414 (84a
Lymph node status  
 Positive 83 (17a
 Negative 394 (83a
Pathologic stage  
 I 356 (72) 
 II 80 (16) 
 III 56 (11) 
Pathway alteration  
 Cell cycle 77 (16) 
 Hippo 14 (3) 
 Myc 36 (7) 
 Notch 19 (4) 
 Nrf2 27 (5) 
 PI3K 119 (24) 
 TGFβ 13 (3) 
 RTK/RAS 415 (84) 
 p53 201 (41) 
 Wnt 19 (4) 
Number of pathway alterations  
 0 23 (5) 
 1 164 (33) 
 2 178 (36) 
 3 94 (19) 
 4 27 (6) 
 5 6 (1) 
TMB, median (range) (IQR) 4.7 (0–164.2) (2.5–8.2) 
FGA, median (range) (IQR) 0.25 (0–1) (0.1–0.5) 

Abbreviations: AIS, adenocarcinoma in situ; IQR, interquartile range; MIA, minimally invasive adenocarcinoma.

aDenominator excluding patients with missing data.

Pathway alterations and genomic features

TMB and FGA are essential genomic features reported to be associated with survival and recurrence in patients with cancer (22–25). We investigated the correlation of TMB and FGA with pathway alterations and found, as NPA increased, TMB (ρ = 0.50; P < 0.001; Fig. 1B) and FGA (ρ = 0.31; P < 0.001; Fig. 1C) increased correspondingly. However, the correlation between TMB and FGA was significant, but the magnitude of correlation (ρ = 0.12; P = 0.006) is considered negligible on the basis of the interpretation (26) by Hinkle and colleagues (Fig. 1D).

We next examined the association between TMB, FGA, and individual pathway alterations (Supplementary Table S3). Tumors with cell-cycle, Hippo, Myc, Notch, Nrf2, PI3K, p53, or Wnt pathway alterations had significantly higher TMB than tumors without these pathway alterations (Fig. 1E). Similarly, tumors with cell-cycle and p53 pathway alterations had significantly higher FGA than tumors without these pathway alterations (Fig. 1F).

Pathway alterations and clinicopathologic features

We then investigated associations between NPA and selected poor-risk clinicopathologic factors (Fig. 2A–H). As NPA increased, the rate of ever-smokers also significantly increased (Cochran–Armitage test, P < 0.001). Similar analyses showed increasing proportion of solid tumor morphologic appearance on preoperative CT, PET tumor maximal standardized uptake value (SUVmax; ρ = 0.28; P < 0.001), pathologic tumor size (ρ = 0.20; P < 0.001), aggressive histologic subtype (micropapillary or solid; Cochran–Armitage test, P < 0.001), lymphovascular invasion (LVI; Cochran–Armitage test, P < 0.001), VPI (Cochran–Armitage test, P = 0.04), and lymph node status (Cochran–Armitage test, P = 0.001) were all associated with increasing NPA.

Figure 2.

NPAs and clinicopathologic features and DFS. Association between NPA and smoking status (A), morphologic appearance on CT (B), pathologic tumor size (C), tumor maximal standardized uptake value (SUVmax; D), predominate histologic subtype (E), lymphovascular invasion (F), visceral pleural invasion (G), and lymph node status (H). I, Staging and outcome distribution among different NPA groups. DOD, dead of disease; NED, no evidence of disease. J, Association between DFS and NPA.

Figure 2.

NPAs and clinicopathologic features and DFS. Association between NPA and smoking status (A), morphologic appearance on CT (B), pathologic tumor size (C), tumor maximal standardized uptake value (SUVmax; D), predominate histologic subtype (E), lymphovascular invasion (F), visceral pleural invasion (G), and lymph node status (H). I, Staging and outcome distribution among different NPA groups. DOD, dead of disease; NED, no evidence of disease. J, Association between DFS and NPA.

Close modal

Finally, we investigated the association between clinicopathologic features and individual pathway alterations (Supplementary Table S3, Supplementary Fig. S3). Nrf2 and PI3K pathway alterations were associated with smoking status. Cell-cycle, Hippo, Nrf2, PI3K, p53, and Wnt pathway alterations were associated with solid tumor morphologic appearance on CT scan. Cell-cycle, p53, and Wnt pathway alterations were associated with higher tumor SUVmax. Cell-cycle, p53, and Wnt pathways were associated with pathologic tumor size. P53 pathway alteration was associated with aggressive histologic subtype and LVI.

Pathway alterations and DFS

A DFS event occurred in 79 individuals, with 70 (89%) experiencing recurrence of their disease. The recurrence rate was 33% for patients with ≥4 NPA and 9% for those with 0 NPA (Fig. 2I). The median DFS for patients with ≥4 NPA was 25.6 months (95% CI, 9.7–41.6); patients with 0 NPA had a median DFS of 89.4 months (95% CI, 10.7–168.1; Fig. 2J). We next examined pathologic stage–specific DFS on the basis of NPA and observed that increasing NPA was associated with worse DFS in stages I, II, and III lung adenocarcinoma (Supplementary Fig. S4). Altered pathways were associated with worse 2-year DFS (95% CI) versus wild-type pathways for cell cycle [70.5% (59.0%–84.2%) vs. 82.4% (77.6%–87.5%); log rank P = 0.03], Hippo [38.9% (14.8%–100.0%) vs. 81.6% (77.1%–86.4%); log rank P = 0.046], TGFβ [21.2% (4.0%–100.0%) vs. 82.2% (77.8%–86.8%); log rank P < 0.001], and p53 [70.9% (63.3%–79.4%) vs. 87.0% (81.8%–92.7%); log rank P < 0.001; Fig. 3]. On univariable analysis, when stratified by stage, increase in NPA was associated with worse DFS [HR, 1.38 (95% CI, 1.12–1.69); P = 0.002; Table 2]. Other clinicopathologic factors, including solid tumor morphologic appearance, tumor SUVmax, pathologic tumor size, pathologic stage, LVI, predominant histologic subtype, VPI, lymph node status, and FGA, were significantly associated with DFS in univariable models (Table 2). However, subsequent multivariable analysis revealed only NPA [HR, 1.31 (95% CI, 1.06–1.62); P = 0.01], solid tumor morphologic appearance, tumor SUVmax, and pathologic stage were independently associated with DFS (Table 2).

Figure 3.

Associations of individual pathway alterations and DFS.

Figure 3.

Associations of individual pathway alterations and DFS.

Close modal
Table 2.

Univariable and multivariable Cox proportional hazards model for DFSa

Univariable ModelMultivariable Model
VariableHR (95% CI)PHR (95% CI)P
Age at surgery, years 1.01 (0.99–1.04) 0.40   
Female vs. male 1.28 (0.81–2.04) 0.29   
Smoking status: ever 0.89 (0.53–1.49) 0.66   
Solid morphologic appearance on preoperative CT 5.93 (2.11–16.68) 0.001 4.39 (1.54–12.49) 0.006 
Tumor SUVmax 1.09 (1.05–1.14) <0.001 1.07 (1.03–1.12) 0.001 
Pathologic stage (8th edition AJCC)     
 II vs. I 3.33 (1.88–5.89) <0.001 1.59 (0.86–2.95) 0.14 
 III vs. I 7.92 (4.71–13.31) <0.001 3.92 (2.25–6.81) <0.001 
Pathologic tumor size, cm 1.20 (1.13–1.28) <0.001   
Lymphovascular invasion 2.51 (1.47–4.28) 0.001   
Predominate histologic subtypeb 1.41 (0.86–2.33) 0.18   
Visceral pleural invasion 2.76 (1.69–4.50) <0.001   
Lymph node status: positive 4.90 (3.13–7.66) <0.001   
Procedure type     
 Sublobar vs. lobectomy 1.96 (0.93–4.12) 0.08   
TMB 1.01 (0.99–1.03) 0.26   
FGA 2.13 (0.94–4.84) 0.07   
Number of pathways altered 1.38 (1.12–1.69) 0.002 1.31 (1.06–1.62) 0.01 
Univariable ModelMultivariable Model
VariableHR (95% CI)PHR (95% CI)P
Age at surgery, years 1.01 (0.99–1.04) 0.40   
Female vs. male 1.28 (0.81–2.04) 0.29   
Smoking status: ever 0.89 (0.53–1.49) 0.66   
Solid morphologic appearance on preoperative CT 5.93 (2.11–16.68) 0.001 4.39 (1.54–12.49) 0.006 
Tumor SUVmax 1.09 (1.05–1.14) <0.001 1.07 (1.03–1.12) 0.001 
Pathologic stage (8th edition AJCC)     
 II vs. I 3.33 (1.88–5.89) <0.001 1.59 (0.86–2.95) 0.14 
 III vs. I 7.92 (4.71–13.31) <0.001 3.92 (2.25–6.81) <0.001 
Pathologic tumor size, cm 1.20 (1.13–1.28) <0.001   
Lymphovascular invasion 2.51 (1.47–4.28) 0.001   
Predominate histologic subtypeb 1.41 (0.86–2.33) 0.18   
Visceral pleural invasion 2.76 (1.69–4.50) <0.001   
Lymph node status: positive 4.90 (3.13–7.66) <0.001   
Procedure type     
 Sublobar vs. lobectomy 1.96 (0.93–4.12) 0.08   
TMB 1.01 (0.99–1.03) 0.26   
FGA 2.13 (0.94–4.84) 0.07   
Number of pathways altered 1.38 (1.12–1.69) 0.002 1.31 (1.06–1.62) 0.01 

aAll models were stratified by pathologic stage except for stage, tumor size, lymph node status, and visceral pleural invasion.

bTumor histologic subtypes collapsed as micropapillary + solid versus other subtypes.

We next compared the simple pathway analysis model (simple-PA model; includes only NPA) with the complex pathway analysis model (complex-PA model; includes the 10 pathways; Supplementary Table S4) in terms of performance by quantifying their discrimination and calibration. The cross-validated C-indexes were 0.658 (95% CI, 0.576–0.732) and 0.657 (95% CI, 0.530–0.750) for the simple-PA and complex-PA models, respectively. Bias-corrected calibration curves confirmed the simple-PA model was well-calibrated, whereas the complex-PA model tended to underestimate DFS (Supplementary Fig. S4).

Mutual exclusivity, cooccurrence, and therapeutic actionabilities

The mutual exclusivity of a pathway or gene may reflect functional redundancy or synthetic lethality (27, 28), whereas cooccurrence(s) may reflect functional synergies important in resistance to targeted therapies (29). Of the 10 pathways investigated, two pairs, p53–cell cycle (OR, 5.78; P < 0.001) and PI3K-Nrf2 (OR, 7.17; P < 0.001), cooccurred with statistical significance (Fig. 4A). We found no mutual exclusivity among pathways in our cohort. Within the p53–cell cycle and PI3K-Nrf2 pathways, alterations in three pairs of genes, CDK4-MDM2 (log2 OR > 3; P < 0.001), RB1-TP53 (log2 OR = 2.81; P = 0.03), and STK11-KEAP1 (log2 OR > 3; P < 0.001), cooccurred with statistical significance (Fig. 4B).

Figure 4.

Mutual exclusivity, cooccurrence, and therapeutic actionability. A, Cooccurrence and mutual exclusivity between pathways. *, P value is significant (P < 0.001, adjusted using the FDR method). Color scale shows the significance of cooccurrence or mutual exclusivity (−log P value). B, Cooccurrence and mutual exclusivity of single genes within PI3K-Nrf2 pathways and p53–cell-cycle pathways. *, P value is significant (CDK4-MDM2, P < 0.001; RB1-TP53, P = 0.03; and STK11-KEAP1, P < 0.001). C, Oncoprint of the CDK4-MDM2–altered patients in this study and the TCGA pan-cancer atlas cohort. D, Actionable genetic alterations within 10 pathways annotated using OncoKB and mean/median NPAs for each pathway. E, Stacked bar plot of percentage of NPA (x-axis) in patients with specific pathway (y-axis) altered. F, Venn diagram showing patients who have actionable alterations in the four indicated pathways. Intersections of the circles show the number of patients who have >1 actionable pathway.

Figure 4.

Mutual exclusivity, cooccurrence, and therapeutic actionability. A, Cooccurrence and mutual exclusivity between pathways. *, P value is significant (P < 0.001, adjusted using the FDR method). Color scale shows the significance of cooccurrence or mutual exclusivity (−log P value). B, Cooccurrence and mutual exclusivity of single genes within PI3K-Nrf2 pathways and p53–cell-cycle pathways. *, P value is significant (CDK4-MDM2, P < 0.001; RB1-TP53, P = 0.03; and STK11-KEAP1, P < 0.001). C, Oncoprint of the CDK4-MDM2–altered patients in this study and the TCGA pan-cancer atlas cohort. D, Actionable genetic alterations within 10 pathways annotated using OncoKB and mean/median NPAs for each pathway. E, Stacked bar plot of percentage of NPA (x-axis) in patients with specific pathway (y-axis) altered. F, Venn diagram showing patients who have actionable alterations in the four indicated pathways. Intersections of the circles show the number of patients who have >1 actionable pathway.

Close modal

Cooccurrences of CDK4 (cell-cycle pathway) and MDM2 (p53 pathway) amplification were significant (n = 26, cell cycle; n = 30, p53; n = 16, both; Fig. 4C). This pair of genes also has corresponding level 2B (palbociclib and abemaciclib for CDK amplification) and level 4 (DS-3032b and RG7112 for MDM2 amplification) actionable drugs. This observation suggests patients with tumors with cooccurent CDK4 and MDM2 amplifications may benefit from combined therapies that target both pathways. Supporting our observation, analysis of the TCGA lung adenocarcinoma cohort (5) reveals these genes have a significant cooccurrence (log2 OR > 3; P < 0.001; Fig. 4C).

Finally, using OncoKB, we investigated the therapeutic actionability of selected genomic alterations. We identified 543 actionable genomic alteration events across 437 tumors (Fig. 4D). A list of tumors with targetable alterations at the pathway and gene level with associated therapies and levels of evidence as curated using OncoKB can be found in Supplementary Table S5. Of these 543 actionable alterations, 455 (84%) were within the RTK/RAS pathway, including all level 1 and level 2A actionable alterations. Interestingly, tumors with RTK/RAS pathway alterations had the lowest mean NPA (n = 2) and the highest proportion of tumors with 1 NPA (Fig. 4D and E). Of the 437 tumors with actionable alterations, 351 had actionable therapeutic alterations in only one pathway; the other 86 had actionable therapeutic alterations in >1 pathway (Fig. 4F).

To our knowledge, this is the first study to comprehensively investigate the association between selected pathway alterations and detailed clinicopathologic features and DFS in patients with completely resected lung adenocarcinoma using broad-panel NGS. Ninety-five percent of our cohort had at least one pathway alteration, and our analysis used 109 genes, of which 57 were altered in <2% of patients. This “long tail” distribution (30–32) of altered genes may complicate the identification of the functionally relevant driver genes from the passenger genes. However, as we show, aggregating these genes into biologically relevant molecular pathways permits the identification of associations with relevant clinicopathologic features and DFS.

It is unknown which pathway alteration(s) initiate the progressive evolution of normal tissue along the continuum toward a neoplastic state. Our results revealed that, among patients with 1 NPA, 86% had an alteration in the RTK/RAS pathway. Of the 10 oncogenic pathways studied, the RTK/RAS pathway comprised the greatest number of clinically relevant genes (n = 38), including three common driver genes in lung adenocarcinoma (KRAS, EGFR, and BRAF). Activation of the RTK/RAS pathway leads to cell proliferation and growth (33), which are fundamental for tumor cell development (34, 35). This may explain the high frequency of RTK/RAS pathway alteration in our study cohort. Similar findings were reported in the lung adenocarcinoma TCGA cohort (5). Furthermore, the proportion of patients with alterations in the RTK/RAS pathway remained stable across all NPA groups. This suggests RTK/RAS pathway activation may be an early event in lung adenocarcinoma. However, this finding warrants further validation from dynamic tumor genomic profiling and multisite tumor sequencing.

Increases in NPA were significantly associated with increases in TMB (ρ = 0.50; P < 0.001) and FGA (ρ = 0.31; P < 0.001). This suggests that, as more mitogenic signaling pathway alterations occur, there may also be more gene replication errors (such as gene mutations or copy number alterations). However, although significant, the magnitude of correlation (Spearman correlation coefficient = 0.12) between TMB and FGA is considered negligible on the basis of the interpretation (26) by Hinkle and colleagues. Neither was shown to be an independent predictor of DFS on multivariable analysis. In a pathway-centric analysis, some pathways, such as cell cycle, are more copy number driven, whereas other pathways, such as TGFβ, are more mutation driven (5). This suggests TMB and FGA can each play a unique role in assessing therapeutic response and clinical outcomes in patients with lung adenocarcinoma. For example, higher TMB is associated with potentially better efficacy of immunotherapy (18), whereas higher FGA may be associated with worse outcomes following immunotherapy (36). Finally, high TMB has been associated with conflicting prognoses in lung cancer (23, 25), whereas increasing FGA has been associated with worse outcomes in prostate, breast, and colorectal cancers (24, 37, 38).

Increases in NPA were significantly associated with tumor SUVmax, solid morphologic appearance on preoperative CT, pathologic tumor size, aggressive histologic subtype, LVI, VPI, and pathologic lymph node status. All these clinicopathologic features have been previously associated with poor outcomes following complete resection of lung adenocarcinoma (11, 39, 40). Finally, we found that increasing NPA was independently associated with worse DFS.

Analysis of mutual exclusivity and cooccurrence revealed interplay within and across pathways. When these findings were combined with therapeutic actionability from a pathway-centric perspective, we found that 84% of actionable alterations were located within the RTK/RAS pathway, and tumors with RTK/RAS pathway alterations had the lowest mean NPA. Because activation of alternative pathways is an important mechanism of acquired resistance to targeted therapy (41), our results suggest the success of drugs in targeting tumors with an isolated RTK/RAS pathway alteration may be secondary to the intrinsically low number of alternative pathways. As MDM2 and CDK4 are found in a similar location on chromosome 12q13-15, the amplification of these two genes often occurs together. Using OncoKB to investigate therapeutic actionability, we found that both genes are actionable (level 2B and level 4). This implies the combination of targeted therapies may be effective in patients with alterations in both pathways, which is supported by preclinical studies (42).

Limitations of our study include using DNA sequencing data alone, whereas previous pathway analyses were curated using RNA-seq data (5). Therefore, if a gene is epigenetically silenced through promoter hypermethylation, it would be missed in our analysis. Moreover, the lack of RNA-seq or reverse phase protein array data precludes validation of somatic alterations that can affect the transcriptomic and protein level(s). We also did not evaluate tumor clonality. When two pathways are coaltered, we are unable to distinguish between clonal and subclonal driver mutations. Our ability to assess spatial heterogeneity of pathway alterations over time is limited, as we obtained a single sample per patient. More longitudinal data (i.e., multisampling over time) would allow investigation of the temporal ordering of pathway alterations. We analyzed primarily mitogenic signaling pathways; pathways of other oncogenic processes, such as DNA repair, ubiquitination, and metabolic pathways, were not examined. Most of our cohort is early stage; thus, our results may not be applicable to patients with more advanced stage disease.

Oncogenic pathway analysis using a 10-pathway template characterized by 109 genes from broad-panel NGS revealed important associations with clinicopathologic features. NPA is an independent risk factor for DFS following complete resection of lung adenocarcinoma and is associated with potential therapeutic actionability. Integration of pathway alteration annotation into broad-panel NGS may help stratify patients' risk of recurrence and aid in selection of optimized induction or adjuvant therapeutic strategies.

V.W. Rusch reports receiving commercial research grants from Genentech and Genelux, is an unpaid consultant/advisory board member for Bristol-Myers Squibb, and reports receiving other remuneration from NIH. N. Rekhtman reports receiving speakers bureau honoraria from Grand Rounds. D.B. Solit is an employee/paid consultant for Pfizer, Loxo Oncology, Lilly Oncology, Illumina, and Vivideon Therapeutics. M.F. Berger is an employee/paid consultant for Roche. D.R. Jones is an employee/paid consultant for Diffusion Pharmaceuticals, Merck, and AstraZeneca. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J. Zhou, F. Sanchez-Vega, G.D. Jones, M. Bott, F. Yang, D.R. Jones

Development of methodology: J. Zhou, G.D. Jones, M. Bott, M. Ladanyi, D.B. Solit

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J. Zhou, W.S. Brandt, G.D. Jones, S. Yan, P.S. Adusumilli, J. Huang, J.M. Isbell, M. Ladanyi, D.B. Solit, D.R. Jones

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Zhou, F. Sanchez-Vega, R. Caso, K.S. Tan, W.S. Brandt, G.D. Jones, P.S. Adusumilli, M. Bott, S. Sihag, V.W. Rusch, W.K. Chatila, N. Rekhtman, M. Ladanyi, D.B. Solit, M.F. Berger, N. Schultz, D.R. Jones

Writing, review, and/or revision of the manuscript: J. Zhou, F. Sanchez-Vega, R. Caso, K.S. Tan, W.S. Brandt, G.D. Jones, P.S. Adusumilli, M. Bott, J. Huang, J.M. Isbell, D. Molena, V.W. Rusch, W.K. Chatila, N. Rekhtman, D.B. Solit, N. Schultz, D.R. Jones

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J. Zhou, P.S. Adusumilli, F. Yang, D.B. Solit, D.R. Jones

Study supervision: F. Yang, D.B. Solit, D.R. Jones

We gratefully acknowledge the members of the Molecular Diagnostics Service in the Department of Pathology. David B. Sewell, of the MSK Department of Surgery, provided superb editorial assistance. This work was supported by the NCI (grant nos. R01CA217169 to D.R. Jones, R01CA240472 to D.R. Jones, and R01CA236615 to P.S. Adusumilli); the Department of Defense (grant no. LC160212 to P.S. Adusumilli); and the NIH (grant no. T32CA009501 to W.S. Brandt). In addition, this work was funded in part by the Marie-Josée and Henry R. Kravis Center for Molecular Oncology and by the NCI/NIH (Cancer Center Support Grant P30CA008748).

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.

1.
Korf
BR
,
Rehm
HL
. 
New approaches to molecular diagnosis
.
JAMA
2013
;
309
:
1511
21
.
2.
Presley
CJ
,
Tang
D
,
Soulos
PR
,
Chiang
A
,
Longtine
J
,
Adelson
K
, et al
Association of broad-based genomic sequencing with survival among patients with advanced non–small cell lung cancer in the community oncology setting
.
JAMA
2018
;
320
:
469
77
.
3.
Castellano
E
,
Santos
E
. 
Functional specificity of ras isoforms: so similar but so different
.
Genes Cancer
2011
;
2
:
216
31
.
4.
Yuen
ST
,
Davies
H
,
Chan
TL
,
Ho
JW
,
Bignell
GR
,
Cox
C
, et al
Similarity of the phenotypic patterns associated with BRAF and KRAS mutations in colorectal neoplasia
.
Cancer Res
2002
;
62
:
6451
5
.
5.
Sanchez-Vega
F
,
Mina
M
,
Armenia
J
,
Chatila
WK
,
Luna
A
,
La
KC
, et al
Oncogenic signaling pathways in The Cancer Genome Atlas
.
Cell
2018
;
173
:
321
37
.
6.
Jordan
EJ
,
Kim
HR
,
Arcila
ME
,
Barron
DA
,
Chakravarty
D
,
Gao
J
, et al
Prospective comprehensive molecular characterization of lung adenocarcinomas for efficient patient matching to approved and emerging therapies
.
Cancer Discov
2017
;
7
:
596
609
.
7.
Cancer Genome Atlas Research Network
. 
Comprehensive molecular profiling of lung adenocarcinoma
.
Nature
2014
;
511
:
543
50
.
8.
Hung
JJ
,
Hsu
WH
,
Hsieh
CC
,
Huang
BS
,
Huang
MH
,
Liu
JS
, et al
Post-recurrence survival in completely resected stage I non-small cell lung cancer with local recurrence
.
Thorax
2009
;
64
:
192
6
.
9.
Kelsey
CR
,
Marks
LB
,
Hollis
D
,
Hubbs
JL
,
Ready
NE
,
D'Amico
TA
, et al
Local recurrence after surgery for early stage lung cancer: an 11-year experience with 975 patients
.
Cancer
2009
;
115
:
5218
27
.
10.
Saynak
M
,
Veeramachaneni
NK
,
Hubbs
JL
,
Nam
J
,
Qaqish
BF
,
Bailey
JE
, et al
Local failure after complete resection of N0–1 non-small cell lung cancer
.
Lung Cancer
2011
;
71
:
156
65
.
11.
Taylor
MD
,
Nagji
AS
,
Bhamidipati
CM
,
Theodosakis
N
,
Kozower
BD
,
Lau
CL
, et al
Tumor recurrence after complete resection for non-small cell lung cancer
.
Ann Thorac Surg
2012
;
93
:
1813
20
.
12.
Brandt
WS
,
Bouabdallah
I
,
Tan
KS
,
Park
BJ
,
Adusumilli
PS
,
Molena
D
, et al
Factors associated with distant recurrence following R0 lobectomy for pN0 lung adenocarcinoma
.
J Thorac Cardiovasc Surg
2018
;
155
:
1212
24
.
13.
Ding
L
,
Bailey
MH
,
Porta-Pardo
E
,
Thorsson
V
,
Colaprico
A
,
Bertrand
D
, et al
Perspective on oncogenic processes at the end of the beginning of cancer genomics
.
Cell
2018
;
173
:
305
20
.
14.
Martini
N
,
Melamed
MR
. 
Multiple primary lung cancers
.
J Thorac Cardiovasc Surg
1975
;
70
:
606
12
.
15.
Cheng
DT
,
Mitchell
TN
,
Zehir
A
,
Shah
RH
,
Benayed
R
,
Syed
A
, et al
Memorial sloan kettering-integrated mutation profiling of actionable cancer targets (MSK-IMPACT): a hybridization capture-based next-generation sequencing clinical assay for solid tumor molecular oncology
.
J Mol Diagn
2015
;
17
:
251
64
.
16.
Cerami
E
,
Gao
J
,
Dogrusoz
U
,
Gross
BE
,
Sumer
SO
,
Aksoy
BA
, et al
The cBio Cancer Genomics Portal: an open platform for exploring multidimensional cancer genomics data
.
Cancer Discov
2012
;
2
:
401
4
.
17.
Gao
J
,
Aksoy
BA
,
Dogrusoz
U
,
Dresdner
G
,
Gross
B
,
Sumer
SO
, et al
Integrative analysis of complex cancer genomics and clinical profiles using the cBioPortal
.
Sci Signal
2013
;
6
:
pl1
.
18.
Rizvi
H
,
Sanchez-Vega
F
,
La
K
,
Chatila
W
,
Jonsson
P
,
Halpenny
D
, et al
Molecular determinants of response to anti-programmed cell death (PD)-1 and anti-programmed death-ligand 1 (PD-L1) blockade in patients with non-small-cell lung cancer profiled with targeted next-generation sequencing
.
J Clin Oncol
2018
;
36
:
633
41
.
19.
Shen
R
,
Seshan
VE
. 
FACETS: allele-specific copy number and clonal heterogeneity analysis tool for high-throughput DNA sequencing
.
Nucleic Acids Res
2016
;
44
:
e131
.
20.
Chakravarty
D
,
Gao
J
,
Phillips
SM
,
Kundra
R
,
Zhang
H
,
Wang
J
, et al
OncoKB: a precision oncology knowledge base
.
JCO Precis Oncol
2017
;
2017
.
21.
Schemper
M
,
Smith
TL
. 
A note on quantifying follow-up in studies of failure time
.
Control Clin Trials
1996
;
17
:
343
6
.
22.
Ciriello
G
,
Gatza Michael
L
,
Beck Andrew
H
,
Wilkerson Matthew
D
,
Rhie Suhn
K
,
Pastore
A
, et al
Comprehensive molecular portraits of invasive lobular breast cancer
.
Cell
2015
;
163
:
506
19
.
23.
Devarakonda
S
,
Rotolo
F
,
Tsao
M-S
,
Lanc
I
,
Brambilla
E
,
Masood
A
, et al
Tumor mutation burden as a biomarker in resected non–small-cell lung cancer
.
J Clin Oncol
2018
;
36
:
2995
3006
.
24.
Hieronymus
H
,
Schultz
N
,
Gopalan
A
,
Carver
BS
,
Chang
MT
,
Xiao
Y
, et al
Copy number alteration burden predicts prostate cancer relapse
.
Proc Natl Acad Sci U S A
2014
;
111
:
11139
44
.
25.
Owada-Ozaki
Y
,
Muto
S
,
Takagi
H
,
Inoue
T
,
Watanabe
Y
,
Fukuhara
M
, et al
Prognostic impact of tumor mutation burden in patients with completely resected non-small cell lung cancer: brief report
.
J Thorac Oncol
2018
;
13
:
1217
21
.
26.
Hinkle
DE
,
Wiersma
W
,
Jurs
SG
.
Applied statistics for the behavioral sciences
. 5th ed. Boston: Houghton Mifflin; 
2003
.
27.
Etemadmoghadam
D
,
Weir
BA
,
Au-Yeung
G
,
Alsop
K
,
Mitchell
G
,
George
J
, et al
Synthetic lethality between CCNE1 amplification and loss of BRCA1
.
Proc Natl Acad Sci U S A
2013
;
110
:
19489
94
.
28.
Mina
M
,
Raynaud
F
,
Tavernari
D
,
Battistello
E
,
Sungalee
S
,
Saghafinia
S
, et al
Conditional selection of genomic alterations dictates cancer evolution and oncogenic dependencies
.
Cancer Cell
2017
;
32
:
155
68
.
29.
Nissan
MH
,
Pratilas
CA
,
Jones
AM
,
Ramirez
R
,
Won
H
,
Liu
C
, et al
Loss of NF1 in cutaneous melanoma is associated with RAS activation and MEK dependence
.
Cancer Res
2014
;
74
:
2340
50
.
30.
Armenia
J
,
Wankowicz
SAM
,
Liu
D
,
Gao
J
,
Kundra
R
,
Reznik
E
, et al
The long tail of oncogenic drivers in prostate cancer
.
Nature Genet
2018
;
50
:
645
51
.
31.
Leiserson
MDM
,
Vandin
F
,
Wu
HT
,
Dobson
JR
,
Eldridge
JV
,
Thomas
JL
, et al
Pan-cancer network analysis identifies combinations of rare somatic mutations across pathways and protein complexes
.
Nature Genet
2015
;
47
:
106
14
.
32.
Garraway
LA
,
Lander
ES
. 
Lessons from the cancer genome
.
Cell
2013
;
153
:
17
37
.
33.
Imperial
R
,
Toor
OM
,
Hussain
A
,
Subramanian
J
,
Masood
A
. 
Comprehensive pancancer genomic analysis reveals (RTK)-RAS-RAF-MEK as a key dysregulated pathway in cancer: its clinical implications
.
Semin Cancer Biol
2019
;
54
:
14
28
.
34.
Hanahan
D
,
Weinberg
RA
. 
Hallmarks of cancer: the next generation
.
Cell
2011
;
144
:
646
74
.
35.
Feitelson
MA
,
Arzumanyan
A
,
Kulathinal
RJ
,
Blain
SW
,
Holcombe
RF
,
Mahajna
J
, et al
Sustained proliferation in cancer: mechanisms and novel therapeutic targets
.
Semin Cancer Biol
2015
;
35
:
S25
S54
.
36.
Davoli
T
,
Uno
H
,
Wooten
EC
,
Elledge
SJ
. 
Tumor aneuploidy correlates with markers of immune evasion and with reduced response to immunotherapy
.
Science
2017
;
355
:
pii
:
eaaf8399
.
37.
Bao
L
,
Qian
Z
,
Lyng
MB
,
Wang
L
,
Yu
Y
,
Wang
T
, et al
Coexisting genomic aberrations associated with lymph node metastasis in breast cancer
.
J Clin Invest
2018
;
128
:
2310
24
.
38.
Zhang
L
,
Feizi
N
,
Chi
C
,
Hu
P
. 
Association analysis of somatic copy number alteration burden with breast cancer survival
.
Front Genet
2018
;
9
:421.
39.
Isaka
M
,
Kojima
H
,
Takahashi
S
,
Omae
K
,
Ohde
Y
. 
Risk factors for local recurrence after lobectomy and lymph node dissection in patients with non-small cell lung cancer: implications for adjuvant therapy
.
Lung Cancer
2018
;
115
:
28
33
.
40.
Uramoto
H
,
Tanaka
F
. 
Prediction of recurrence after complete resection in patients with NSCLC
.
Anticancer Res
2012
;
32
:
3953
60
.
41.
Nagano
T
,
Tachihara
M
,
Nishimura
Y
. 
Mechanism of resistance to epidermal growth factor receptor-tyrosine kinase inhibitors and a potential treatment strategy
.
Cells
2018
;
7
:
pii
:
E212
.
42.
Laroche-Clary
A
,
Chaire
V
,
Algeo
MP
,
Derieppe
MA
,
Loarer
FL
,
Italiano
A
. 
Combined targeting of MDM2 and CDK4 is synergistic in dedifferentiated liposarcomas
.
J Hematol Oncol
2017
;
10
:
123
.