Molecular dynamics simulation-guided drug sensitivity prediction for lung cancer with rare EGFR mutations

Significance

A variety of rare mutations account for 10–20% of EGFR mutations in nonsmall cell lung cancer. However, due to high diversity, proper medication for patients with such mutations is impossible in daily clinic. To appropriately treat lung cancer patients harboring such rare EGFR mutations, a robust prediction model to predict sensitivities of rare EGFR mutants to existing drugs is strongly needed. Using molecular dynamics simulation-based model, we successfully predicted diverse sensitivities of EGFR exon 20 insertion mutants to existing inhibitors. The findings suggest the usefulness of in silico simulation to overcome mutation diversity at a clinically relevant level. The present in silico model will help in selecting effective drugs for these patients.

Abstract

Next generation sequencing (NGS)-based tumor profiling identified an overwhelming number of uncharacterized somatic mutations, also known as variants of unknown significance (VUS). The therapeutic significance of EGFR mutations outside mutational hotspots, consisting of >50 types, in nonsmall cell lung carcinoma (NSCLC) is largely unknown. In fact, our pan-nation screening of NSCLC without hotspot EGFR mutations (n = 3,779) revealed that the majority (>90%) of cases with rare EGFR mutations, accounting for 5.5% of the cohort subjects, did not receive EGFR-tyrosine kinase inhibitors (TKIs) as a first-line treatment. To tackle this problem, we applied a molecular dynamics simulation-based model to predict the sensitivity of rare EGFR mutants to EGFR-TKIs. The model successfully predicted the diverse in vitro and in vivo sensitivities of exon 20 insertion mutants, including a singleton, to osimertinib, a third-generation EGFR-TKI (R² = 0.72, P = 0.0037). Additionally, our model showed a higher consistency with experimentally obtained sensitivity data than other prediction approaches, indicating its robustness in analyzing complex cancer mutations. Thus, the in silico prediction model will be a powerful tool in precision medicine for NSCLC patients carrying rare EGFR mutations in the clinical setting. Here, we propose an insight to overcome mutation diversity in lung cancer.

Recent genome-scale characterization of cancers, including nonsmall cell lung cancer (NSCLC), revealed an extreme diversity of somatic gene mutations (1, 2). In the era of next generation sequencing (NGS) technologies, an overwhelming number of novel, rare, and uncharacterized somatic mutations, classified as variants of unknown significance (VUS), have been identified (3). For the majority of NSCLC patients with rare mutations in oncogenes (i.e., VUS), appropriate precision medicine approaches are not applicable, and therefore, their prognosis remains poor (4). Thus, diversity of gene mutations producing VUS is an emerging problem in oncology.

Lung cancer with epidermal growth factor receptor gene (EGFR) mutations accounts for 10–20% and 40–50% of NSCLC cases in Caucasians and East Asians, respectively (5). In-frame deletions around the LREA motif of exon 19 and exon 21 L858R point mutation, which are classic hotspot EGFR mutations, account for ∼80–90% of EGFR mutations detected in NSCLC (6), while G719X (3% of EGFR mutations) and L861Q (2% of EGFR mutations) are other relatively rare hotspot mutations (5, 7). All these mutations occur in the EGFR tyrosine kinase domain and promote the active conformation of EGFR protein, thereby constitutively activating corresponding oncogenic pathways (8–10). Multiple EGFR tyrosine kinase inhibitors (EGFR-TKIs) have been approved and used in routine cancer clinics to therapeutically inhibit hyperactive EGFR signaling (11–16) based on the fact that a positive relationship between the presence of these EGFR mutations and sensitivity to EGFR-TKIs has been well-established (17–19). In contrast, other EGFR mutations occurring outside hotspots in the kinase domain are VUS, which are largely uncharacterized due to their high diversity. EGFR exon 20 insertion mutations, consisting of >50 types and accounting for 4–10% of all EGFR mutations, are representatives of such VUS (7, 20, 21). Based on several reports that EGFR exon 20 insertion mutants are resistant to EGFR-TKIs (7, 12, 22–24), NSCLC patients with these mutations are not administered EGFR-TKIs as the first-line treatment. However, we previously revealed that an EGFR exon 20 insertion mutant, A763_Y764insFQEA, is sensitive to the first- and second-generation EGFR-TKIs (23). Therefore, it is possible that a fraction of patients with EGFR exon 20 insertion mutations might benefit from therapy of some EGFR-TKIs. However, the high diversity of these mutations as well as the presence of many singleton mutations prevents the comprehensive characterization of the presently known mutants. Furthermore, the number of novel EGFR mutations is increasing owing to the use of NGS-based tests in lung cancer clinics. Thus, a rapid and robust method to accurately predict the sensitivity of EGFR rare mutants to existing TKIs in the clinical setting is necessary to tackle the problem that NSCLC patients with rare EGFR mutations often lose the chance of being treated with appropriate EGFR-TKIs.

Recently, computational structural modeling and molecular dynamics (MD) simulations have helped us clarify the activation mechanism of EGFR at the atomic level (25–27). In addition, predictions of sensitivity of EGFR mutants to EGFR tyrosine kinase inhibitors were performed for several EGFR mutations using binding free energy calculated with MD simulation (28, 29) and fitness scores calculated by molecular docking simulation (30). However, there is still room for discussion on the prediction accuracy and robustness of these models. Also, whether these methods can be applied to predict the sensitivity of various rare EGFR mutants to existing TKIs at a clinically relevant level remains elusive.

We have previously developed the supercomputer-based binding free energy calculation model utilizing MD simulation (31, 32) and applied our model to secondary ALK and RET mutants, which appeared during therapy using TKIs (33, 34). Based on our previous work, we hypothesized that our supercomputer-based model would allow us to predict the sensitivity of rare EGFR mutants to EGFR-TKIs at a clinically relevant level. To this end, we performed an interdisciplinary study, where computer science, cancer biology, and clinical oncology approaches were applied.

Results

High Diversity of Rare EGFR Mutations in NSCLC.

To obtain clinically relevant information regarding individual EGFR mutations and to assess their diversity, we led the Lung Cancer Genomic Screening Project for Individualized Medicine in Japan (LC-SCRUM-Japan), a prospective nationwide lung cancer clinical and genomic characterization network, in which 217 institutions in Japan participated as of May 2017 (35). In this project, NSCLC cases in which the major EGFR mutations (exon 19 deletions, L858R, G719X, or L861Q) were not detected by the routine clinical testing underwent NGS evaluation for possible somatic alterations using a panel of cancer-related genes. Within LC-SCRUM-Japan, 3,779 NSCLC patients were enrolled from February 2013 to March 2017. Of these patients, 201 from October 2013 to June 2014 (first cohort) and 1,963 from March 2015 to March 2017 (second cohort) were subjected to NGS. The NGS study revealed that major EGFR mutations (exon 19 deletions, L858R, G719X, or L861Q) were detected in 2.7% (53/1,963) of such patients in the second cohort (Fig. 1A), indicating false negative results of routine clinical tests. Rare EGFR mutations were detected in 108 (5.5%) patients. The frequency of rare EGFR mutations was higher than those of ROS1 fusions (3.6%) and RET fusions (2.9%). In addition, the frequency of rare EGFR mutations comprised approximately one-third of KRAS mutation frequency (15.3%). Rare EGFR mutations were detected in both nonsquamous NSCLC and squamous cell carcinoma (SI Appendix, Fig. S1). These data indicate that rare EGFR mutations account for a significant proportion of NSCLC cases.

Fig. 1.

Frequency and distribution of variants of unknown significance in EGFR. (A) Pie chart showing the frequency of genetic alterations of indicated genes. (B) Distribution of variants of unknown significance in the EGFR gene. (C) Oncopanel illustrating the genetic alterations in NSCLC patients with variants of unknown significance in EGFR.

The distribution of rare mutations throughout the EGFR gene sequence is illustrated in Fig. 1B. Most mutations were found in the region encoding the tyrosine kinase domain, in exons 18–21, while some mutations outside the tyrosine kinase domain, particularly in exons 6, 7, 8, 12, 15, and 17, were also detected. The most frequent rare EGFR mutations were EGFR exon 20 insertion mutations (Fig. 1B and SI Appendix, Table S1). Of the 113 NSCLC cases with rare EGFR mutations, including the five cases in the first cohort and the 108 cases in the second cohort, 52 (46.0%) harbored EGFR exon 20 insertion mutations, indicating that such mutations comprised about a half of rare EGFR mutations. Of the identified 73 types of rare EGFR mutations, 68 types (93.1%) were found in only one or two cases. These data indicate a high diversity of EGFR mutations in NSCLC.

Low Chance of EGFR-TKI Therapy for Rare EGFR Mutation Cases.

Clinical data were available for 53, 47, and 61 NSCLC patients with EGFR major mutations (exon 19 deletions, L858R, G719X, or L861Q), exon 20 insertion mutations, and other rare mutations, respectively (SI Appendix, Table S2). Notably, the characteristics of NSCLC patients with exon 20 insertion mutations were similar to those of patients with major mutations. However, the characteristics of NSCLC patients with other rare mutations were slightly different by including more male and heavy smoking patients. The majority of rare EGFR mutations were mutually exclusive with other driver oncogene mutations, although coexistence with ERBB2 mutation or amplification, PIK3CA mutation, KRAS mutation, and MET amplification was observed in a small subset of patients (Fig. 1C). Of the 113 patients with rare EGFR mutations, 82 patients, including 33 cases with exon 20 insertion mutations, were available for information on chemotherapy after gene testing (Table 1 and SI Appendix, Table S3). Of those 82 patients, 77 (93.9%) received cytotoxic chemotherapy as first-line treatment, while only five patients (6.1%) were treated with EGFR-TKIs as first-line treatment. Thirty-two (97.0%) patients with EGFR exon 20 insertion mutations were treated with cytotoxic chemotherapy as first-line treatment.

Table 1.

Treatment response of NSCLC harboring rare EGFR mutations

Chemotherapy	Mutations	Median line: 2 (range 0–8 except Exon 20 ins range 0–6)
		First	Second	Third	Fourth and more	Total response rate, % (n)
Cytotoxic agent	All rare mutations	77	42	27	22	—
	Exon 20 ins	32	19	11	6	—
	Other rares	45	23	16	16	—
EGFR-TKI
Afatinib	All rare mutations	3	4	3	7	17.6 (3/17)
	Exon 20 ins	1	1	1	3	16.7 (1/6)
	Other rares	2	3	2	4	18.2 (2/11)
Erlotinib	All rare mutations	1	1	0	1	33.3 (1/3)
	Exon 20 ins	0	1	0	0	0.0 (0/1)
	Other rares	1	0	0	1	50.0 (1/2)
Gefitinib	All rare mutations	1	2	0	0	0.0 (0/3)
	Exon 20 ins	0	0	0	0	—
	Other rares	1	2	0	0	0.0 (0/3)
ICI
Nivolumab	All rare mutations	0	8	8	13	3.4 (2/29)
	Exon 20 ins	0	4	5	5	0.0 (0/14)
	Other rares	0	4	3	8	13.3 (2/15)
Total number	All rare mutations	82	57	38	43	—
	Exon 20 ins	33	25	17	14	—
	Other rares	49	32	21	29	—

n = 82, 33, and 49 for All rare mutations, Exon 20 ins, and Other rares, respectively. ICI, immune checkpoint inhibitor.

The overall response rate to EGFR-TKIs in patients with rare EGFR mutations was only 17.4% (4/23; Table 1). Notably, one of the four cases, who responded to afatinib, carried NSCLC with an exon 20 insertion mutation, A767_V769dupASV (SI Appendix, Table S3), and the result was consistent with our previous study (36). This result validates that while the overall response to first- and second-generation EGFR-TKIs for NSCLC cases with rare EGFR mutations is low, there is a responsive subgroup among them.

Calculation of Binding Energy of EGFR-TKIs for EGFR with Rare Mutations.

The above findings regarding EGFR mutation diversity and the efficacy of EGFR-TKIs for a subset of mutants prompted us to apply our in silico drug sensitivity prediction model (31, 32). We have previously indicated that a differential sensitivity of two representative rare EGFR mutants, exon 20 insertion mutants A763_Y764insFQEA and A767_V769dupASV, to two EGFR-TKIs, afatinib and osimertinib, revealed that A763_Y764insFQEA was more sensitive than A767_V769dupASV (36). We generated Ba/F3 cells harboring these EGFR mutations and evaluated the sensitivity by MTS cell proliferation assay. Here, we verified that the IC₅₀ values of afatinib and osimertinib for the EGFR mutant harboring A763_Y764insFQEA were 10 times lower than those for the A767_V769dupASV mutant (11 vs. 171 nM and 33 vs. 321 nM, respectively). First, to evaluate whether our supercomputer-based binding free energy (ΔG_bind) calculation model could predict the difference in sensitivity conferred by these mutations, we calculated ΔG_bind values for the binding of afatinib to these EGFR mutants. In this model, the structures of EGFR molecules harboring rare mutations were built by homology modeling, and their binding affinities for EGFR-TKIs were evaluated using massively parallel computation of absolute binding free energy with a well-equilibrated system (the MPCAFEE method) (31, 32, 37). The modeled structure with bound afatinib is shown in SI Appendix, Fig. S2A. The calculated ΔG_bind values were −23.3 and −18.5 for A763_Y764insFQEA and A767_V769dupASV, respectively (SI Appendix, Fig. S2B), indicating a more stable binding of afatinib to A763_Y764insFQEA than to A767_V769dupASV. Although a limited efficacy of afatinib for EGFR exon 20 insertion mutation-positive NSCLC patients was reported (38), the potential efficacy of osimertinib for these patients was reported (36, 39, 40), and several human clinical trials for evaluating the efficacy of osimertinib for EGFR exon 20 insertion mutations are ongoing. Therefore, we applied the model to osimertinib and nine recurrent or novel EGFR exon 20 insertion mutations found in our LC-SCRUM-Japan cohort including A763_Y764insFQEA, Y764_V765insHH, A767_S768insSVD, A767_V769dupASV, V769_D770insDNP, D770_N771insNPG, D770_N771insNPH, N771_P772insPGD, and P772_H773insHV. The modeled structures with bound osimertinib are shown in Fig. 2A and SI Appendix, Figs. S3 and S4. The calculated ΔG_bind values are shown in SI Appendix, Table S4. Among the variable ΔG_bind values, that of A763_Y764insFQEA was the lowest, −15.4 (kcal/mol), indicating the sensitivity of this mutation to osimertinib. In contrast, although the amino acid sequence was similar between D770_N771insNPG and D770_N771insNPH, the calculated ΔG_bind values were distinctive [−13.2 (kcal/mol) for D770_N771insNPG and −10.0 (kcal/mol) for D770_N771insNPH]. These data may demonstrate structural complexity of osimertinib binding to mutant EGFR proteins. Finally, we found that the ΔG_bind value for N771_P772insPGD, a mutation found in this cohort, was the lowest [−14.3 (kcal/mol)], except for that of A763_Y764insFQEA, among the mutations. These data indicate the potential sensitivity of this mutation to osimertinib.

Fig. 2.

Calculation of binding energy values for EGFR exon 20 insertion mutations. (A) Structures of mutated EGFR kinase domains for EGFR A763_Y764insFQEA, A767_V769dupASV, D770_N771insNPG, and N771_P772insPGD were modeled using wild-type EGFR data (Protein Data Bank ID code 4ZAU). The structures were extracted from the trajectories of 50-ns molecular dynamics simulations that were used for binding affinity calculations. The structures in gray and in colors are wild-type and mutated EGFRs, respectively. Mutated amino acids in EGFR structures are indicated in red. Osimertinib molecules are shown as sticks (white, carbon; blue, nitrogen; red, oxygen). (B) Plot of ΔG_bind values against negative log-transformed IC₅₀ values. Each mutated EGFR is indicated by a dot. Dashed line represents a linear fit with squared correlation coefficient R².

Sensitivity of EGFR with Rare Mutations to EGFR-TKIs and Its Correlation with Calculated Binding Free Energy Values.

To support our hypothesis that the ΔG_bind values of rare EGFR mutations predict sensitivity to osimertinib, we generated a Ba/F3 library of rare EGFR mutations that were found in the LC-SCRUM-Japan cohort. EGFR transgenes bearing individual mutations were transduced into Ba/F3 cells, mouse pro-B cells (23). To examine the sensitivity of cells transduced with EGFR incorporating rare mutations to EGFR-TKIs, we performed MTS proliferation assays. We used the following EGFR-TKIs: erlotinib (first generation), afatinib (second generation), rociletinib, and osimertinib (third generation). The calculated IC₅₀ values are shown in SI Appendix, Table S5. Interestingly, the sensitivity data of D770_N771insNPG and D770_N771insNPH were distinct (76 nM for D770_N771insNPG and 427 nM for D770_N771insNPH), as predicted by binding energy calculations. In addition, of the nine mutations examined in this study, EGFR A763_Y764insFQEA exhibited the lowest binding energy, indicating the most stable binding of osimertinib. The IC₅₀ value of osimertinib in cells that expressed A763_Y764insFQEA (33 nM) was the lowest, which matched the prediction of binding energy calculation. Moreover, the highest IC₅₀ value of osimertinib in cells that expressed D770_N771insNPH (427 nM) was also predicted by binding energy calculation. Remarkably, the ΔG_bind values calculated with our model showed a statistically significant correlation (R²: 0.7232, P = 0.0037) with experimentally observed IC₅₀ values (Fig. 2B).

As mentioned above, to predict the sensitivity of EGFR mutants to TKIs, some approaches employing the molecular docking method or MD simulation have been reported (28–30). First, to compare the usefulness of the molecular docking method with our model, we calculated the fitness (docking) score of TKIs to EGFR mutants using rDock (30), which is one of the most widely used molecular docking programs for structure-based virtual screening. The calculated docking scores did not show a correlation with the experimentally observed IC₅₀ values (R²: 0.0054, P = 0.8508; SI Appendix, Fig. S5). The calculated docking scores using rDock are shown in SI Appendix, Table S6. These data indicate that the molecular docking method is not useful for sensitivity prediction of EGFR exon 20 insertion mutants.

Next, to compare the usefulness of other MD simulation-based methods with our model, we calculated binding free energies using MM-PBSA (Molecular Mechanics Poisson-Boltzmann Surface Area) and MM-GBSA (Molecular Mechanics Generalized Born Surface Area) with the MMPBSA.py module (41) in Amber Tools, which were employed in previous studies to investigate the sensitivity of EGFR mutants to TKIs (28, 29). Although the calculated binding free energies by MM-GBSA and MM-PBSA showed statistically significant correlations (R²: 0.5733, P = 0.018 for MM-GBSA and R²: 0.5744, P = 0.018 for MM-PBSA; SI Appendix, Fig. S6), the correlations were lower than those of our model. The calculated binding free energy values are shown in SI Appendix, Table S7. These results indicate that the sensitivity of cells expressing EGFR exon 20 insertion mutations to EGFR-TKIs can be predicted more accurately using our in silico prediction method based on MD simulation. In addition, to evaluate the robustness of our model, we calculated the binding energies or rDock for nonexon 20 insertion mutations to osimertinib. We selected representative single nucleotide variations (SNVs) and combinations of SNVs (SNV combination) mutations. The calculated values are shown in SI Appendix, Tables S8 and S9. Interestingly, our model demonstrated higher correlations with the experimentally observed IC₅₀ values (R²: 0.8392, P = 0.0288 for SNV and R²: 0.8768, P = 0.0191 for SNV combinations) (SI Appendix, Fig. S7) than other models. These data indicate the robustness of our model for several types of EGFR mutations. In addition, to predict the therapeutic window of each EGFR-TKI, we have proposed that the selectivity index (SI), i.e., the ratio of log-transformed IC₅₀ values in Ba/F3 cells transduced with mutated and wild-type EGFR, should be used (36). SI values for EGFR-TKIs that inhibited the growth of cells with EGFR exon 20 insertions and other rare EGFR mutations are shown in SI Appendix, Figs. S8 and S9. These data indicate that osimertinib would affect a wide spectrum of lung cancers with rare EGFR mutations. However, sensitivity to osimertinib varied among the cells with rare EGFR mutations as there was a >100-fold difference between the lowest and the highest osimertinib IC₅₀ values. These data also indicate the variation in sensitivity of rare EGFR mutations, including EGFR exon 20 insertion mutations to osimertinib.

Biological Confirmation of the Sensitivity of EGFR N771_P772insPGD to Osimertinib.

In this study, we have shown several previously unreported EGFR mutations. Of these EGFR mutations, we analyzed the EGFR exon 20 insertion mutation, N771_P772insPGD, found in a 45-y-old NSCLC patient diagnosed with lung adenocarcinoma in 2016. The positron emission tomography-computed tomography scan of the patient and the N771_P772insPGD sequences are shown in Fig. 3 A and B, respectively. Both in silico model and experimentally obtained IC₅₀ values predicted the sensitivity of this mutation to osimertinib. To validate the prediction that N771_P772insPGD is sensitive to osimertinib in vivo, we generated a patient-derived xenograft (PDX) model from pleural effusion of this patient. The histology of the PDX tumor in the mouse was also confirmed as adenocarcinoma (SI Appendix, Fig. S10). Consistent with the in silico prediction results, once-daily administration of 25 mg/kg (a dose that approximates the clinically approved 80-mg dose; ref. 39) or 50 mg/kg osimertinib induced a significant regression of the tumor (Fig. 3C and SI Appendix, Fig. S11). These data indicate that osimertinib could be effective in NSCLC cases with this mutation. These data support the applicability of the in silico prediction model for predicting the sensitivity of NSCLC with rare EGFR mutations to EGFR-TKIs.

Fig. 3.

Biological confirmation of sensitivity of N771_P772insPGD to osimertinib. (A) A positron emission tomography-computed tomography scan showing 18F-fluorodeoxyglucose accumulation in the right lung filed of a NSCLC patient. (B) Results of Sanger sequencing of EGFR from patient-derived xenograft tumor DNA showing three amino acid (PGD) insertion in N771_P772. (C) Effect of osimertinib (25 or 50 mg/kg) on the size of patient-derived xenograft tumors in nonobese diabetic/severe combined immunodeficiency mice. Each group n = 5. Data are presented as the mean ± SD. **P < 0.01 for the combination of osmertinib (25 mg/kg or 50 mg/kg) versus control (Upper). Pictures of the tumors (Lower).

Discussion

In this study, we clarified the diversity and driver roles of rare EGFR mutations in a large prospective Japanese NSCLC cohort and revealed the limited efficacy of precision medicine approaches for NSCLC patients with rare EGFR mutations. Among the 73 types of rare EGFR mutations detected, 68 (93.1%) were found in only one or two patients; the frequency of each rare EGFR mutation was less than 0.1%. In addition, sensitivity to EGFR-TKIs was quite diverse even if the mutation sequences were similar, e.g., for EGFR with D770_N771insNPG and D770_N771insNPH mutations. These data point to the diversity of rare EGFR mutations in their structural interaction with EGFR-TKIs. In vitro and/or in vivo experimental evaluation of drug sensitivity is effective in the evaluation of a few EGFR exon 20 mutants as shown by recent studies (42, 43). However, considering the extremely low frequency of each mutation and the continual appearance of novel mutations, such methods are not realistic in the clinical setting.

To overcome the problem of such mutation diversity, we, here, applied a supercomputer-based in silico prediction model that can be used to promptly estimate sensitivity to existing TKIs, obviating the need for time-consuming “wet” experiments. In this model, ΔG_bind values that are highly correlated with sensitivity to drugs can be obtained in approximately 1 wk. Although several reports have proposed the availability of the in silico approach, by employing MD simulation, for predicting drug sensitivity of several EGFR mutations (28, 29), our model demonstrated the highest prediction accuracy based on comparisons with experimentally obtained sensitivity values. These results indicate the usefulness and robustness of our model and show the potential to overcome mutation diversity in cancer. Here, in this study, we evaluated our model based on experimentally observed values for binding affinity. Since osimertinib is one of the targeted covalent inhibitors, further discussion on both the binding affinity and the rate of the subsequent bond-forming reaction will be needed. Of course, the clinical utility of our method should be evaluated in humans since it has been evaluated only in the PDX model. In this study, we indicated the efficacy of osimertinib for several EGFR exon 20 mutants. Thus, we have launched a clinical trial to evaluate the efficacy of osimertinib for cases with EGFR exon 20 insertion mutations [University Hospital Medical Information Network (UMIN) 000031929]. Prospective in silico and in vivo studies of these patients will prove the utility of our prediction method. Recent large-scale genomic characterization programs as well as rapid technological advancements have enabled the application of NGS in the clinical setting to identify numerous mutations in a variety of genes in lung and other cancers. Nonhot spot EGFR mutations have been discovered in small fractions of several cancers other than NSCLC. Application of our method might improve the prognosis of cancer patients by guiding drug development or promoting drug repositioning based on information on VUS in cancers.

Methods

Patients.

LC-SCRUM-Japan is a prospective, nationwide clinical and genomic screen of lung cancer (UMIN ID: UMIN000010234). In this study, a total of 3,779 NSCLC patients were enrolled from February 2013 to March 2017. Cases with NSCLC stage II or more advanced stage [tumor, nodes, metastasis (TNM) classification version 7], which were confirmed to have no major EGFR mutations (exon 19 deletions, L858R, G719X, or L861Q) by local hospitals, were included. The methods of EGFR mutation detection performed in local hospitals included PNA-LNA PCR clamp, Scorpion-ARMS, Cyclerve PCR, PCR-invader, or Cobas EGFR mutation assay v.2. All patients provided written informed consent for the entry to the LC-SCRUM-Japan study.

DNA Extraction and Next Generation Sequencing.

In this study, DNA samples were extracted from fresh frozen specimens or pleural effusion. From October 2013 to June 2014 (first cohort), DNA samples from 201 cases were analyzed by a targeted NGS assay, the Ion Ampliseq Cancer Hotspot Panel (Thermo Fisher Scientific). From March 2015 to March 2017 (second cohort), DNA samples from 1,963 cases were analyzed by another targeted NGS assay, the Oncomine Comprehensive Assay (OCA v.1) (Thermo Fisher Scientific). Sequencing of the paired normal tissues or blood was not performed in this cohort. We selected potential somatic mutations those are registered as “confirmed somatic” in the COSMIC (Catalogue Of Somatic Mutations In Cancer) database. The mutations confirmed as somatic mutations by the COSMIC database are indicated with * in SI Appendix, Table S1. The present study was approved by the institutional Ethics Committee of all 217 institutions (SI Appendix, Table S10) that participated in the LC-SCRUM-Japan cohort. All patients provided written informed consent for the molecular analysis of their samples. All analyses were done at SRL, Inc.

Statistical Analysis.

IBM SPSS Statistics 24 was used for data management and statistical analyses. For descriptive analysis, quantitative variables are expressed as the median and range. Categorical variables are expressed as the number of cases and percentage. Two-sided Student’s t tests were used for pairwise comparisons. The Pearson’s correlation test was performed to calculate R and P values. All statistical analyses were conducted with a significance level of α = 0.05 (P < 0.05). All P values are two-sided.

Other methods are described in SI Appendix.