ARTICLES Distinct patterns of somatic genome alterations in lung adenocarcinomas and squamous cell carcinomas Joshua D Campbell1,2, Anton Alexandrov3,4, Jaegil Kim1, Jeremiah Wala1,2, Alice H Berger1,2, Chandra Sekhar Pedamallu1,2, Sachet A Shukla1,2, Guangwu Guo1,2, Angela N Brooks1,2, Bradley A Murray1,2, Marcin Imielinski1,2,5, Xin Hu6, Shiyun Ling6, Rehan Akbani6, Mara Rosenberg1, Carrie Cibulskis1, Aruna Ramachandran1,2, Eric A Collisson7, David J Kwiatkowski1,8, Michael S Lawrence1, John N Weinstein6, Roel G W Verhaak6, Catherine J Wu1,2, Peter S Hammerman1,2, Andrew D Cherniack1,2, Gad Getz1,9, Cancer Genome Atlas Research Network10, Maxim N Artyomov3, Robert Schreiber3, Ramaswamy Govindan11, Matthew Meyerson1,2,12 To compare lung adenocarcinoma (ADC) and lung squamous cell carcinoma (SqCC) and to identify new drivers of lung carcinogenesis, we examined the exome sequences and copy number profiles of 660 lung ADC and 484 lung SqCC tumor–normal pairs. Recurrent alterations in lung SqCCs were more similar to those of other squamous carcinomas than to alterations in lung ADCs. New significantly mutated genes included PPP3CA, DOT1L, and FTSJD1 in lung ADC, RASA1 in lung SqCC, and KLF5, EP300, and CREBBP in both tumor types. New amplification peaks encompassed MIR21 in lung ADC, MIR205 in lung SqCC, and MAPK1 in both. Lung ADCs lacking receptor tyrosine kinase–Ras–Raf pathway alterations had mutations in SOS1, VAV1, RASA1, and ARHGAP35. Regarding neoantigens, 47% of the lung ADC and 53% of the lung SqCC tumors had at least five predicted neoepitopes. Although targeted therapies for lung ADC and SqCC are largely distinct, immunotherapies may aid in treatment for both subtypes. Lung cancer remains the leading cause of death from cancer around genes can be challenging because of the large number of passenger the world1. An estimated 221,000 new cases and 158,000 deaths mutations that can accumulate from prolonged exposure to tobacco Nature America, Inc. All rights reserved. Inc. Nature America, from lung cancer occurred in the United States alone in 2015 (ref. 2). carcinogens and from inherent mutagenic processes such as aberrant 6 The two major histological classes are non-small-cell lung can- activity of APOBEC cytidine deaminases9. Profiling larger numbers cer (NSCLC) and small-cell lung cancer (SCLC). NSCLCs mostly of samples within a tumor type and combining samples across tumor © 201 comprise lung ADCs and lung SqCCs. These two NSCLC subtypes types can help overcome this problem, by providing the additional have both unique and shared clinical and histopathological charac- statistical power necessary to distinguish important genes mutated teristics. For example, whereas smoking is the major risk factor for at a lower frequency than other genes with passenger mutations10. In both subtypes, approximately 10–15% of lung ADCs are observed addition, a comprehensive comparison of recurrently altered genes in never-smokers3. Molecularly targeted therapies directed against found in lung ADC and lung SqCC has not been performed. Such receptor tyrosine kinases (RTKs) lead to dramatic responses in analyses may yield insights into the similarities and differences in subsets of patients with lung ADCs harboring activating genomic carcinogenesis between the diseases and elucidate the degree to which alterations in the corresponding kinase genes, including EGFR, ALK, common or distinct targeted and immunological therapeutic strate- and ROS1 (ref. 4). Other targeted therapies under current investi- gies can be used to treat each cancer type. gation are directed against activating alterations in the MET, RET, NTRK1, NTRK2, ERBB2, and BRAF kinases4,5. RESULTS Recent efforts have focused on comprehensively characterizing the Comparison of somatically altered genes changes found in the genome, epigenome, transcriptome, and pro- To compare the somatic profiles of lung ADC and lung SqCC and teome in lung ADCs and SqCCs to discover new cancer driver genes to identify new genetic alterations, we studied 660 lung ADC–nor- that may be clinically actionable6–8. Identifying new cancer-related mal paired exome sequences (including 274 previously unpublished 1Cancer Program, Eli and Edythe L. Broad Institute of Harvard and MIT, Cambridge, Massachusetts, USA. 2Department of Medical Oncology, Dana-Farber Cancer Institute, Boston, Massachusetts, USA. 3Department of Pathology and Immunology, Washington University, St. Louis, Missouri, USA. 4Computer Technologies Laboratory, ITMO University, St. Petersburg, Russia. 5Molecular Pathology Unit, Massachusetts General Hospital, Charlestown, Massachusetts, USA. 6Department of Bioinformatics and Computational Biology, University of Texas MD Anderson Cancer Center, Houston, Texas, USA. 7Department of Medicine, University of California, San Francisco, San Francisco, California, USA. 8Department of Medicine, Brigham and Women’s Hospital, Boston, Massachusetts, USA. 9Department of Pathology, Massachusetts General Hospital, Boston, Massachusetts, USA. 10A full list of members and affiliations appears at the end of the paper. 11Department of Medicine, Washington University School of Medicine, St. Louis, Missouri, USA. 12Department of Pathology, Harvard Medical School, Boston, Massachusetts, USA. Correspondence should be addressed to M.M. ([email protected]) or R.G. ([email protected]). Received 4 September 2015; accepted 12 April 2016; published online 9 May 2016; doi:10.1038/ng.3564 NATURE GEnETICS ADVANCE ONLINE PUBLICATION 1 ARTICLES a Significantly mutated genes b Focal amplifications c Focal deletions TP53 NKX2−1 CDKN2A KRAS –64 –128 KEAP1 NF1 10 10 EGFR RB1 TERT –8 BRAF ARID1A –32 MYC –64 10 ERBB2 STK11 10 MCL1 10 SMARCA4 CDK4 MDM2 PTPRD SETD2 –32 RBM10 –16 KRAS 10 MGA 10 CCND1 MET EGFR –16 B2M alue alue alue –4 ATM CDKN2A MECOM–TERC 10 v –8 CCNE1 v v 10 U2AF1 WWOX LRP1B q 10 q PDE4D q RIT1 ERBB2 –8 WHSC1L1–FGFR1 10 10–4 SMAD4 RB1 ARID2 –4 SMAD4 MET 18q11.2 10 Xp22.2 –2 PDGFRA–KIT–KDR 21q21.1 CSMD1 10 CTNNB1 NFE2L2 10–2 Lung ADC –2 Lung ADC Lung ADC APC 10 4q22.1 PIK3CA PTEN RAF1 –1 6q22.31 MLL2 –1 ZMYND11 NRAS 10 CDK6 REL–BCL11A 10 MAP2K1 19p13.2 FOXP1 10–1 HRAS KDM6A NOTCH1 FAT1 MYCL1 BCL2L1 NFE2L2 SOX2 KDM6A NF1 FAT1 PTEN NS NS NS NS 10–1 10–2 10–4 10–8 NS 10–1 10–2 10–4 10–8 10–1610–3210–6410–128 NS 10–1 10–2 10–4 10–810–1610–3210–6410–128 Lung SqCC q value Lung SqCC q value Lung SqCC q value Figure 1 Distinct somatic alterations in lung ADC and lung SqCC. (a) The MutSig2CV algorithm10 was used to identify significantly mutated genes across 660 lung ADCs and 484 lung SqCCs. Genes with q values <0.1 were considered to be significantly mutated. The q value for each gene in the lung ADC cohort is plotted against the respective q value in the lung SqCC cohort. The majority of significantly mutated genes were unique to either tumor type. (b,c) The GISTIC 2.0 algorithm was used to identify significantly recurrent copy number gains and losses. The q values for amplifications (b) and deletions (c) in the lung ADC cohort are plotted against the respective q values in the lung SqCC cohort. Peaks with q values <0.25 were considered to be significant. Deletions located within putative fragile sites are highlighted with green labels. Only points corresponding to genes with a previously characterized role in lung cancer are labeled. NS, not significant. cases and 227 previously described cases from The Cancer Genome Mutational signatures in lung cancer Atlas (TCGA)6 together with 159 cases from the cohort in Imielinski Various carcinogenic and cancer-related processes contribute to the et al.8) and 484 lung SqCC–normal paired exome sequences (includ- mutational patterns observed in tumors13,14. Previous large-scale ing 308 previously unpublished cases and 176 previously described studies of lung cancer genomes have identified signatures associated cases from TCGA7; Supplementary Tables 1–4). Similarly to with non-smoking and smoking cases6,8,15; here we extend these previous studies6,7, we observed median somatic mutation rates findings through the improved statistical power of our larger sample of 8.7 mutations/Mb and 9.7 mutations/Mb for lung ADCs and set. Using non-negative matrix factorization (NMF)13,16 (Online SqCCs, respectively. After excluding genes with lower median Methods), we identified six mutational signatures in this cohort, many expression (log2 (FPKM) <6.16 for lung ADCs and <6.27 for lung of which are strongly correlated with previously defined signatures SqCCs; Online Methods and Supplementary Fig. 1), we identi- in the Catalogue of Somatic Mutations in Cancer (COSMIC) data- fied 38 genes as significantly mutated in lung ADC and 20 genes base13,17 (Supplementary Figs. 3–5 and Supplementary Table 7). as significantly mutated in lung SqCC using MutSig2CV10 (q value These included a UV-related signature of C>T changes at TCC or Nature America, Inc. All rights reserved. Inc. Nature America, < 0.1; Supplementary Tables 5 and 6). Only six genes—TP53, RB1, CCC sites (COSMIC signature 7, abbreviated SI7), a smoking-related 6 ARID1A, CDKN2A, PIK3CA, and NF1—were significantly mutated signature of C>A transversions (SI4), a mismatch-repair (MMR) in both tumor types, and, of these, TP53, CDKN2A, and PIK3CA signature of C>T changes at GCG sites (SI15/SI6), two APOBEC- © 201 had a significantly higher mutation frequency in lung SqCC tumors related signatures of C>G or C>T changes at TCT or TCA sites (P < 0.01, Fisher’s exact test; Fig. 1a). Likewise, only 11 of 42 focal (SI13 and SI2), and a final signature with a moderate correlation to amplification peaks were identified as altered in both tumor types COSMIC signature 5 (SI5) with putative ‘molecular clock’ properties18 (Fig. 1b), and 13 of 50 focal deletion peaks were altered in both (Supplementary Fig. 5). In addition to identifying mutational signa- tumor types (Fig.