Development of Lung Adenocarcinomas with Exclusive Dependence on Oncogene Fusions

Lung adenocarcinoma (LADC) is the most frequent histologic type of lung cancer and its incidence is rising in Asian and Western countries. Oncogenic fusions of the protein tyrosine kinase genes ALK, RET, and ROS1, identified by us and others, are believed to drive the development of a subset (3%–4%, 1%–2%, and 1%–2%, respectively) of LADCs (1–4). These fusion-positive LADCs often, but not always, show mucinous–cribriform patterns (2, 5–7). In addition, fusion-positive LADCs tend to occur in young and non/light-smoking individuals (2, 4, 8–10), and show a high therapeutic response to tyrosine kinase inhibitors (TKI) that suppress the kinase activity of the fusion products (11–13). These results indicate that fusion-positive LADCs are a distinct LADC molecular entity. Despite recent large-scale genome sequencing studies in LADCs (14–16), the genetic profile of fusion-positive LADCs remains unknown due to the rarity of these tumors. Better genetic characterization of fusion-positive LADCs is required to improve therapeutic strategies. If other genetic abnormalities are detected, agents targeting these defects could be used in combination with TKIs to improve efficacy and outcome (11, 17, 18).

LADCs carrying ALK, ROS1, and RET fusions have already been shown to lack activating mutations in other oncogenes, such as EGFR, KRAS, BRAF, and HER2/ERBB2 (2–4, 14–16); however, the mutational status of other genes frequently mutated in lung and other cancers, including those identified in the cancer gene census (CGC; ref. 19) or those identified as significantly mutated genes (SMG) in 12 common cancers (20), is unknown. These gene sets include tumor-suppressor genes, such as TP53, CDKN2A, KEAP1, and STK11/LKB1, and chromatin remodeling/modifying genes, such as ARID1A and SMARCA4, which are the targets of genetic loss-of-function aberrations in cancer cells (20). Notably, recent studies suggested that these deleterious aberrations are therapeutically targetable; drugs restoring the function of mutant p53 proteins are being developed (21, 22), and synthetic lethality-based therapies have been considered by us and others to treat cancers with TP53, LKB1, ARID1A, and SMARCA4 deficiencies (23–28).

We performed the comparative genetic aberration profiling of oncogenic fusion-positive and -negative LADCs. Two hundred cases of snap-frozen surgical LADC tissues were subjected to whole-exome sequencing using a next-generation sequencer and to copy-number variation analysis using a DNA chip. The selected cases were enriched in oncogenic ALK, RET, or ROS1 fusions (n = 31, 15.5%) and included 96 cases (48.0%) of activating mutations in EGFR, KRAS, HER2, BRAF, and HRAS oncogenes and 73 cases (35.5%) without any such aberrations. The study revealed that fusion-positive LADCs have a unique genetic profile that includes fewer genetic aberrations than other LADCs.

A total of 200 LADC cases (the study cohort) were selected from 608 consecutive cases (the NCC cohort; NCC Biobank) who underwent surgical resection between 1997 and 2008 at the National Cancer Center Hospital, Tokyo, and for whom snap-frozen cancerous and noncancerous lung tissues were available (Fig. 1A and B). All of the 608 cases were screened for EGFR, KRAS, BRAF, and HER2 hot spot mutations by the HRM method, and for EML4– and KIF5B–ALK, KIF5B– and CCDC6–RET, and CD74–, EZR– and SLC34A2–ROS1 fusions by RT-PCR, as described previously (Supplementary Table S2; refs. 4, 7). In addition, a case with a novel type of RET fusion, KIAA1468-RET, which was detected by whole RNA sequencing, was included in this cohort.

Driver fusion study subjects (n = 31), that is, those with ALK (n = 11), RET (n = 11), or ROS1 (n = 9) fusions were selected from all 50 fusion-positive cases in the NCC original cohort, that is, those with ALK fusions (n = 23), RET (n = 13), or ROS1 (n = 14) fusions based on the criterion that sufficient amounts of genomic DNA for whole-exome sequencing were available (Fig. 1A). EGFR-positive (n = 72), other driver mutation (n = 23), and pan-negative cases (n = 74), with sufficient amounts of genomic DNA, were randomly selected from the NCC original cohort together with EGFR-positive (n = 282), other driver mutation (n = 79), and pan-negative cases (n = 197) to obtain a EGFR-positive case:pan-negative case ratio of approximately 1:1 and to make the total number of samples 200. The exome sequencing analysis revealed an activating HRAS mutation (Q61L), so this case was classified as driver mutation. Thus, the study cohort included 73 pan-negative cases and an HRAS mutation–positive case. The selection resulted in a cohort that was more enriched in driver fusion cases than the original cohort (Fig. 1B). The study subjects were diagnosed according to the seventh TNM classification of malignant tumors (29, 30). The study was approved by the Institutional Review Boards of the NCC.

Genome copy-number and allelic status were assessed in all 200 study cases by Illumina OMNI 2.5M array analysis using both cancerous and noncancerous lung DNA. Tumor cell content in each tumor sample and copy numbers for each gene were deduced using the Global Parameter Hidden Markov Model method (31).

Exome sequencing was conducted from 2.5 μg of cancerous or noncancerous DNA isolated from snap-frozen tissues. Exome capture was performed using the Agilent SureSelect Human All Exon 50-Mb, V4 or V5 according to the manufacturer's instructions. Exome sequencing was performed on the Illumina HiSeq 2000 platform using 75 bp paired-end reads (Illumina). Basic alignment and sequence quality control were conducted using the Picard and Firehose pipelines. The reads were aligned against the reference human genome from UCSC human genome 19 (Hg19) using the Burrows Wheeler Aligner Multi-Vision software package. Because duplicate reads were generated during the PCR amplification process, paired-end reads that aligned to the same genomic positions were removed using SAMtools.

Somatic SNVs were called by the MuTect program, which applies a Bayesian classifier to allow the detection of somatic mutations with a low allele frequency (32). Somatic InDel mutations were called by the GATK Somatic Indel Detector (33). SNV and InDel detection was supported by visual examination using the Integrative Genomics Viewer software (34).

Mutations in all coding exons of the following 28 genes were examined by targeted genome capture and massively parallel sequencing using an Illumina HiSeq 2000 system and the Haloplex Custom Enrichment Kit (Agilent Technologies): 10 representative cancer census genes (19), AKT1, APC, CTNNB1, KEAP1, MAP2K1, MET, NRAS, PIK3CA, STK11 and TP53, and 18 SWI/SNF chromatin remodeling genes (35) whose mutations were detected in one or more tumors by exome sequencing, ACTL6B, ARID1A, ARID1B, ARID2, BPTF, DPF1, EP400, HLTF, PBRM1, RAD54L2, SHPRH, SMARCA2, SMARCA4, SMARCAD1, SMARCAL1, SMARCB1, SMARCC1, and SRCAP. Average read depths were approximately 1,000.

A list of somatic mutations from the CGC was downloaded from the most recently released COSMIC V70 (36).

Validation analysis was performed using The Cancer Genome Atlas (TCGA) LADC study data (16). The TCGA cohort of 230 cases was selected from 678 patients with previously untreated LADC based on the tumor percentage, availability of clinical data, and availability of sufficient amounts of nucleic acid (37). Driver gene fusions and mutations were evaluated in all cases; thus, the 230 cases were subgrouped according to driver gene type: 9 driver fusion cases (3.9%) with ALK (n = 3), RET (n = 2), and ROS1 (n = 2) fusions, 121 driver mutation cases (52.6%) with hot spot mutations in EGFR (n = 26), KRAS (n = 74), HER2 (n = 4), BRAF (n = 16), and HRAS (n = 1), and 100 pan-negative cases (43.5%; Supplementary Fig. S4A). Among the cancer-related and SWI/SNF chromatin remodeling genes investigated in Fig. 2, information on CDKN2A, RBM10, RB1, NF1, KEAP1, MET, MGA, U2AF1, PIK3CA, STK11, TP53, SMARCA4, and ARID1A was available, and therefore was used in the analysis.

Statistical analyses of differences in genetic alterations, clinical and pathologic factors between the driver aberration groups, or smoking status were assessed by using the Kruskal–Wallis test, two-sided Mann–Whitney test, two-sided Fisher exact test or χ² test, and two-sided Spearman r test in GraphPad Prism 5 software (GraphPad Software). Multivariate regression analysis, including the number of nonsynonymous mutations per Mb and clinicopathologic factors (age, gender, smoking status, pathologic stage, and tumor differentiation), was conducted using JMP 10 software (SAS Institute). A P value of <0.05 was considered significant.

The study cohort of 200 LADC cases included 31 cases (15.5%) with ALK, RET, or ROS1 fusions, 96 cases (48.0%) with hot spot mutations in EGFR, KRAS, HER2, BRAF, or HRAS, and 73 cases (36.5%) without any of these driver gene aberrations (Table 1). The study cohort samples were selected from the original National Cancer Center (NCC) cohort samples (n = 608) to enrich fusion-positive cases (Fig. 1A and B; Supplementary Table S1). The driver gene aberrations detected were mutually exclusive, as predicted (14, 16, 38). There was no case of NTRK1 fusion, a recently reported oncogene fusion (39), in this cohort. The 31, 96, and 73 cases were classified into three groups designated the “driver fusion,” “driver mutation,” and “pan-negative” groups, respectively (Fig. 1A and B).

Patient characteristics are summarized in Table 1. The driver fusion cases included a significantly higher frequency of young, female and never-smoker cases, and the pan-negative cases included a significantly higher frequency of older, male, and heavy smokers (P = 0.024 by Kruskal–Wallis test, 0.001 by χ² test, and <0.0001 by χ² test, respectively) than the other two groups, representing the characteristics of the original NCC cohort (Supplementary Table S2). In addition, the pan-negative cases showed a significant predominance of poor differentiation (P = 0.0007 by χ² test). These features are consistent with previous reports (2, 4–6, 8–10), indicating the authenticity and suitability of our study cohort to establish the genetic profile of LADCs driven by specific aberrations.

Genomic DNA from cancerous and noncancerous lung tissues was subjected to whole-exome sequencing and SNP chip analyses. The average sequencing depth of the driver fusion, driver mutation, and pan-negative groups was similar; median: 106 (range, 84–218), 98 (82–216), and 104 (82–145), respectively (P ≥ 0.05 by Kruskal–Wallis test; Supplementary Fig. S1A). The tumor contents deduced from genome-wide copy-number and allelic imbalance data obtained by SNP chip analysis were also similar (P ≥ 0.05 by Kruskal–Wallis test; Supplementary Fig. S1B). Therefore, having ruled out differential sensitivity between the groups, the sample set was found suitable to compare the genetic aberration profiles of the three groups. In addition, the SMGs deduced by the MutSigCV analysis were consistent with recent large-scale sequencing studies (14–16), supporting the authenticity of the present sample set (Supplementary Table S3).

Nonsynonymous mutations (missense, nonsense, indel, and splicing site mutations) based on driver genes are depicted in Fig. 2, focusing on CGC genes, SMGs, and SWI/SNF chromatin remodeling complex genes (20, 40). The driver fusion cases appeared to harbor fewer mutated genes. In fact, the median number of nonsynonymous mutations per Mb was lowest in driver fusion cases (0.37; range, 0–1.5) compared with driver mutation cases (0.65; 0–6.0) and pan-negative cases (0.87; 0–24.8; P < 0.0001 by Kruskal–Wallis test; Fig. 3A and Supplementary Fig. S2A). The median number of mutated CGC genes was smaller in driver fusion cases (1.0; range, 0–4) than in driver mutation cases (3.0; range, 1–12) and pan-negative cases (3.0; range, 0–46; P < 0.0001 by Kruskal–Wallis test). The median number of mutated SMGs was also smaller in driver fusion cases (1.0; range, 0–4) than in driver mutation cases (3.0; range, 1–7) and pan-negative cases (2.0; range, 0–25; P < 0.0001 by Kruskal–Wallis test). A positive correlation was observed between the number of nonsynonymous mutations in all genes and that in CGC genes or SMGs (P < 0.0001 by Spearman test; Supplementary Fig. S2B). Driver fusion was associated with fewer mutations regardless of age, gender, smoking status, pathologic stage, or tumor differentiation (P < 0.0001 by multivariate regression model analysis; Supplementary Table S4). Thus, driver fusion-positive LADCs seem to develop from the accumulation of a significantly smaller number of gene mutations than other types of LADCs.

Mutations in several genes known to contribute to lung carcinogenesis were examined. The TP53 gene is a representative tumor-suppressor gene, included both in the CGC genes and SMGs, whose mutation occurs during the progression of LADCs following EGFR and KRAS mutations (41). TP53 was the most frequently mutated gene in our study cohort (Table 2) and in the U.S. cases (20), with a mutation frequency of 41 of 96 (42.7%) in the driver mutation cases and 29 of 73 (39.7%) in the pan-negative cases. Notably, TP53 mutations were significantly less frequent in the driver fusion group (5/31 or 16.1%, P = 0.026 by χ² test) than in the other two groups (Fig. 3B and Supplementary Fig. S3A). Truncating mutations in SWI/SNF chromatin remodeling complex genes, which are frequently observed in a variety of cancers (40), were not detected either (Figs. 2 and 3B and Supplementary Fig. S3B). The SMGs included 20 cellular process genes, most of which were less frequently mutated in fusion-positive cases than in mutation-positive and pan-negative cases (Supplementary Table S5). Two WNT signaling genes, APC and CTNNB1/β-catenin, and a PI(3)K signaling gene, PIK3CA, which activates PI(3)K signaling, are representative signaling genes known to be mutated in a small subset of LADCs (14, 16). Notably, these three genes were mutated irrespective of driver gene status (Fig. 3B, Table 2; Supplementary Table S5 and Supplementary Fig. S3C and S3D).

All 31 driver fusion cases and five driver mutation cases were subjected to deep resequencing of 28 genes with a mean depth of 1,000, using the same DNA samples. The 28 genes included 10 cancer-related genes and 18 SWI/SNF chromatin remodeling genes. All 19 nonsynonymous mutations detected by exome sequencing in these 36 cases were confirmed. In addition, an ARID1A missense mutation, which was not detected because of low depth by exome sequencing (i.e., no mutant reads among 27 reads), was detected in a driver fusion case. This resequencing verified that LADCs with oncogenic fusions develop through a pathway that involves only a small number of gene mutations.

Similar findings were also observed in a LADC cohort from the United States consisting of 230 cases (16). This cohort included 9 fusion-positive cases consisting of three ALK fusions, two RET fusions, and four ROS1 fusions. Mutation distributions according to driver gene status are summarized in Supplementary Table S6. In this cohort, driver fusion cases carried fewer nonsynonymous mutations (P = 0.046 by Kruskal–Wallis test) and cancer-related gene mutations, including TP53 (P = 0.003 by χ² test) mutations, than other cases (Supplementary Fig. S4A–A4C).

Genomic copy-number gains in nine oncogenes recently defined as amplified in histologic types of lung cancer (15) were examined to assess whether genomic aberrations other than mutations contribute to the development of fusion-positive LADCs (Supplementary Table S7). According to DNA chip analysis, these aberrations were infrequent regardless of the driver fusion.

The mutational characteristics of pan-negative cases were also investigated. These cases were associated with higher proportion of individuals of male gender, smokers, and individuals with poorly differentiated tumors (Table 1), and had the highest median number of nonsynonymous mutations per Mb among the three groups (Fig. 3A and Supplementary Fig. S2A). Furthermore, they had more frequent mutations in CGC genes (P < 0.0001 by Kruskal–Wallis test; Fig. 3A) than the other cases, and most of the SMGs were more highly mutated in these cases than in the other groups (Supplementary Table S5). Among 127 SMGs, TSHZ3, SETBP1, EPHA3, and NAV3 were preferentially mutated in pan-negative cases (Table 2). In addition, truncating mutations in SWI/SNF chromatin remodeling genes, such as ARID1A and SMARCA4/BRG1, which have been reported to be mutated in LADC (14, 16), and PBRM, another SWI/SNF gene frequently mutated in renal cell carcinoma (ccRCC; ref. 42), were significantly predominant in pan-negative cases (P < 0.0001 by Kruskal–Wallis test; Fig. 3B and Supplementary Fig. S3B; and Supplementary Table S5). On the other hand, KEAP1, NF1, and RIT1 mutations, which were previously shown to occur frequently in pan-negative tumors in the TCGA cohort (16), were not frequent in our cases (Supplementary Table S6). Therefore, there might be a difference in the carcinogenic pathways of pan-negative cases between Asians and Europeans/Americans.

Some pan-negative cases did not show mutations in cancer-related genes (Fig. 2). A small number of mutations were associated with cases with low tumor content, suggesting that the failure to detect mutations in some pan-negative cases was due to the low mutation detection power in tumors with low purity (Supplementary Fig. S5). To further address this point, we subjected 29 pan-negative cases, in which few or no cancer-related gene mutations were detected, to deep resequencing, in the same way as we did for fusion-positive cases. In addition to the nonsynonymous mutations already detected by exome sequencing, we additionally detected several mutations, including those with a lower mutation allele frequency than expected from tumor content (Supplementary Table S8). Therefore, some of the pan-negative tumors most likely had intratumor heterogeneity that hampered the detection of mutations by exome sequencing.

This study compared gene aberrations based on driver gene status in a cohort enriched in LADCs with oncogenic ALK, RET, and ROS1 fusions. The fusion-positive cases showed significantly fewer mutations than the other cases in all genes and in known cancer-related genes represented by the CGC genes and SMGs. The lower mutation rate was independent of age, gender, smoking status, pathologic stage, and tumor differentiation, and was validated in a LADC cohort consisting of 230 selected U.S. patients (Supplementary Fig. S4; ref. 16). The rate of nonsynonymous mutation in driver fusion LADCs was similar to that in ovarian, breast, brain, kidney, and hematopoietic tumors, for which mutation rates are low (20, 43, 44). Notably, fusion-positive LADCs had a lower frequency of C>A transversion, which is predominant in tumors of ever-smokers and is a signature of cigarette smoke exposure, than others (Supplementary Fig S6A; refs. 20, 43, 44). Thus, LADCs with oncogene fusions develop through a distinct pathway that includes fewer gene aberrations than other LADCs. Interestingly, similar differential genomic profiles were observed in PAX fusion-positive and -negative rhabdomyosarcomas (45). This might be due to the strong ability of oncogene fusions to drive carcinogenesis; therefore, tumor cells with such fusions might not need many other genetic alterations to progress. The fact that LADC patients with oncogene fusions are younger than those without fusions in this and other populations (Table 1) might also reflect the oncogenic robustness of these fusions (2, 8–10, 38).

By contrast, pan-negative LADCs appear to develop from a large number of gene mutations, including mutations in TP53 and other cancer-related genes. In addition, pan-negative cases carry truncating mutations in several SWI/SNF chromatin remodeling genes more frequently than other cases. Notably, truncating mutations in PBRM1 were detected only in pan-negative cases. Truncating mutations in PBRM1 and other SWI/SNF genes are frequent in ccRCC (42); therefore, some pan-negative LADC and ccRCC might develop through a common carcinogenic pathway. Smoking is linked to high numbers of mutations (Supplementary Fig. S6B and S6C), suggesting that the large numbers of gene mutations caused by exposure to tobacco carcinogens can lead to tumor development without any known driver oncogene alteration. In addition, a subset of pan-negative tumors were found to have high intratumor heterogeneity that hampered mutation detection by the exome sequencing method. Thus, the mutation profiles of pan-negative tumors require further investigation by more sensitive methods, such as deep whole-exome sequencing, to understand how these tumors develop.

The present study has implications for therapeutic approaches to fusion-positive LADCs. Personalized therapy targeting fusion products with TKIs has become the first-line therapeutic method in advanced and/or recurrent tumors (12, 46, 47). This study indicates that targeting fusion products is the best approach as only a small number of mutations have occurred in other genes. Notably, driver fusion cases (and also driver mutation cases) lack SWI/SNF chromatin remodeling gene aberrations. Recent studies demonstrated that deficiencies in SMARCA4 and ARID1A make cancer cells treatable by inhibiting the activity of their paralogs, SMARCA2 and ARID1B, based on synthetic lethality (23, 24, 28); however, LADC patients with driver fusions will not benefit from such therapies because their tumors have a low frequency of aberrations in SMARCA4 and ARID1A genes. Notably, a small subset of fusion-positive LADCs carry gene mutations in genes, such as PIK3CA, APC, and CTNNB1, that could affect signal transduction, although how prevalent or specific these mutations are to fusion-positive LADCs remains unclear due to a small number of study subjects. This issue should be further investigated to unravel the carcinogenic pathways that lead to fusion-positive LADC and find efficient therapeutic targets for this type of cancer.

No potential conflicts of interest were disclosed.

Conception and design: M. Saito, J. Yokota, T. Kohno

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Shimada, K. Shiraishi, H. Sakamoto, K. Tsuta, H. Ichikawa, S.-i. Watanabe, T. Yoshida

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. Saito, K. Shiraishi, H. Totsuka, S. Chiku, M. Kato, T. Kohno

Writing, review, and/or revision of the manuscript: M. Saito, T. Kohno

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y. Shimada, H. Sakamoto, K. Tsuta, T. Yoshida

Study supervision: J. Yokota

The authors thank Sumiko Ohnami, Yoko Odaka, and Misuzu Okuyama for technical assistance.

This work was supported by the Advanced Research for Medical Products Mining Program of the National Institute of Biomedical Innovation (NIBIO); grants-in-aid from the Japan Agency for Medical Research and Development (AMED) for the Health, and Labor Sciences Research Expenses for Commission (the Practical Research for Innovative Cancer Control: H26-practical-general-007); a grant-in-aid from the Japanese Society for the Promotion of Science (JSPS) and Scientific Research (B: 26293200 and C: 15K10275), and the National Cancer Center Research and Development Fund (26A-1).

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.