Somatic mutation of CDKN1B in small intestine ... - Nature

0 downloads 0 Views 415KB Size Report
The diagnosed incidence of small intestine neuroendocrine tumors (SI-NETs) is increasing, and the underlying genomic mechanisms have not yet been defined.
letters

npg

© 2013 Nature America, Inc. All rights reserved.

Somatic mutation of CDKN1B in small intestine neuroendocrine tumors Joshua M Francis1,2,18, Adam Kiezun1,18, Alex H Ramos1,2,17,18, Stefano Serra3, Chandra Sekhar Pedamallu1,2, Zhi Rong Qian2, Michaela S Banck4,5, Rahul Kanwar4, Amit A Kulkarni4, Anna Karpathakis6,7, Veronica Manzo2, Tanupriya Contractor8, Juliet Philips2, Elizabeth Nickerson1, Nam Pho1, Susanne M Hooshmand2, Lauren K Brais2, Michael S Lawrence1, Trevor Pugh1, Aaron McKenna1, Andrey Sivachenko1, Kristian Cibulskis1, Scott L Carter1, Akinyemi I Ojesina1,2, Samuel Freeman2, Robert T Jones9, Douglas Voet1, Gordon Saksena1, Daniel Auclair1, Robert Onofrio1, Erica Shefler1, Carrie Sougnez1, Jonna Grimsby1, Lisa Green1, Niall Lennon1, Tim Meyer6,7, Martyn Caplin7, Daniel C Chung10,11, Andreas S Beutler4,5, Shuji Ogino2,12,13, Christina Thirlwell6,7, Ramesh Shivdasani2, Sylvia L Asa3,14, Chris R Harris8,15,16, Gad Getz1, Matthew Kulke2 & Matthew Meyerson1,2,9 The diagnosed incidence of small intestine neuroendocrine tumors (SI-NETs) is increasing, and the underlying genomic mechanisms have not yet been defined. Using exome- and genome-sequence analysis of SI-NETs, we identified recurrent somatic mutations and deletions in CDKN1B, the cyclindependent kinase inhibitor gene, which encodes p27. We observed frameshift mutations of CDKN1B in 14 of 180   SI-NETs, and we detected hemizygous deletions encompassing CDKN1B in 7 out of 50 SI-NETs, nominating p27 as a tumor suppressor and implicating cell cycle dysregulation in the etiology of SI-NETs. NETs are rare neoplasms (~1 per 100,000 in the United States) that are thought to arise from endocrine precursor cells and occur most commonly in the lung, pancreas and small intestine1. Well-differentiated NETs are typically more indolent than other epithelial malignancies but can nevertheless metastasize1. Both germline and somatic mutations of MEN1, the multiple endocrine neoplasia type 1 gene, are common in lung and pancreatic NETs1. Pancreatic NETs are also characterized by recurrent somatic mutations in DAXX, ATRX, PTEN and TSC2 (ref. 2). In SI-NETs, by contrast, evidence for focal events that are indicative of driver alterations has remained inconclusive; hemizygous loss of chromosome 18 is the most frequent known genomic event, followed by arm-level gains of chromosomes 4, 5, 14 and 20 (refs. 3–5). Recently, a whole-exome sequencing study of 48 SI-NETs examining somatic

single nucleotide variants (SSNVs) identified mutations in several cancer genes, although none were recurrently altered5. To identify genomic alterations driving tumorigenesis in SI-NETs, we profiled 55 tumors from 50 individuals using a combination of whole-exome and whole-genome sequencing (Fig. 1a and Supplementary Tables 1–5). Mutation analysis of the exome sequencing data with the MuTect algorithm6,7 identified a total of 1,230 genes with somatic mutations, 90% of which (1,113/1,230) were mutated in only a single individual. We observed a relatively low non-silent SSNV rate of 0.77 per Mb (range, 0.13–2.51 per Mb) (Fig. 1a, Supplementary Fig. 1a and Supplementary Table 6). Of the 1,230 genes that were mutated in our study, 21 were also found to be mutated in the previous SI-NET study and another 17 were mutated in a pancreatic NET study, including the cancer census genes ATRX and COL1A1, each in a single individual (Supplementary Fig. 1b–d)2,5,8. The lack of substantial overlap in recurrently altered genes suggests that many of the mutations are passengers. There are potentially therapeutically targetable mutations in several genes, including SRC, FYN, KDR and IDH1 (encoding p.Arg132His); however, each was only present within a single individual (Supplementary Table 6). We identified significantly mutated genes by measuring ­nucleotideand sample-specific mutation rates in the SI-NET sequence data, computing an expected gene-specific mutation frequency for the SI-NETs on the basis of the size and nucleotide composition of each gene and then comparing the actual mutation frequency for each

1Broad

Institute, Cambridge, Massachusetts, USA. 2Department of Medical Oncology, Dana-Farber Cancer Institute, Boston, Massachusetts, USA. 3Department of Pathology, University Health Network, Toronto, Ontario, Canada. 4Division of Medical Oncology, Mayo Clinic, Rochester, Minnesota, USA. 5Mayo Clinic Cancer Center, Mayo Clinic, Rochester, Minnesota, USA. 6University College London Cancer Institute, London, UK. 7Royal Free Hospital NET Unit, London, UK. 8Raymond and Beverly Sackler Foundation, New Brunswick, New Jersey, USA. 9Center for Cancer Genome Discovery, Dana-Farber Cancer Institute, Boston, Massachusetts, USA. 10Gastrointestinal Unit, Massachusetts General Hospital, Boston, Massachusetts, USA. 11Cancer Center, Massachusetts General Hospital, Boston, Massachusetts, USA. 12Department of Pathology, Brigham and Women’s Hospital and Harvard Medical School, Boston, Massachusetts, USA. 13Department of Epidemiology, Harvard School of Public Health, Boston, Massachusetts, USA. 14Ontario Cancer Institute, University of Toronto, Ontario, Canada. 15Cancer Institute of New Jersey, University of Medicine and Dentistry of New Jersey, New Brunswick, New Jersey, USA. 16Department of Pediatrics, University of Medicine and Dentistry of New Jersey, New Brunswick, New Jersey, USA. 17Present address: H3 Biomedicine, Cambridge, Massachusetts, USA. 18These authors contributed equally to this work. Correspondence should be addressed to M.K. ([email protected]) or M.M. ([email protected]). Received 22 March; accepted 10 October; published online 3 November 2013; doi:10.1038/ng.2821

Nature Genetics  VOLUME 45 | NUMBER 12 | DECEMBER 2013

1483

letters a

3.0 2.5 2.0 1.5 1.0 0.5 0 CDKN1B Mutations per Mb

Synonomous

Nonsynonymous

Frameshift

SCNAs

CDKN1B Chr4 Chr5 Chr14 Chr20

Amplification Deletion

Chr18 Primary Small intestine Lymph node

Tumor type Site

npg

c Frameshift deletion

Frameshift insertion

Samples

(Deletion)

D195_TP (∆chr12:12.9–48.1 Mb) A7_TM (∆chr12:9.6–31.2 Mb) H6_TM (∆chr12:12.8–31.2 Mb) B8_TM (∆chr12:8.6–22.9 Mb) H58_TM (∆chr12:10.3–16.3 Mb) D188_TP (∆chr12:12.8–26.3 Mb) A14_TM (∆chr12:12.8–14.6 Mb)

Q197fs

P191fs

T187fs

D176fs

L118fs N124fs H129fs D136fs

Q65fs

R58fs

P94fs P95fs

100

Whole exome and whole genome Whole exome Whole genome

CDKN1B locus deletions Chr12:12,870,302–12,875,305 CDKN1B

CDKN1B mutations

0

H32_TP H27_TM B14_TP H10_TM H53_TP B15_TP T3_TP A9_TM

D192_TM B11_TP H61_TP B5_TP T9_TP B4_TP H54_TP B2_TM T8_TM H55_TP A6_TM H15_TP B17_TP T10_TM H56_TM H57_TP H7_TP H26_TP H52_TP B1_TM H5_TP B13_TM D183_TM D187_TP A8_TM T7_TP T6_TM A4_TP D2810_TP A11_TP D184_TP

b

D37fs

© 2013 Nature America, Inc. All rights reserved.

H62_TP A17_TP A12_TM A16_TP D188_TP A7_TM H6_TM D195_TP H58_TM A14_TM B8_TM

Analysis

Metastasis Liver Soft tissue

198 [20 genes]

Figure 1  Mutational analysis of 31 primary and 19 metastatic SI-NETs. (a) Top, somatic mutation rate per Mb of the covered target sequence in the 50 cases. Bottom, the recurrent somatic mutation of CDKN1B and prominent somatic copy number alterations (SCNAs) found in each tumor. The sites and type of sequencing performed are indicated by the colored boxes. TP, primary tumor; TM, metastatic tumor. (b) Schematic representation of the frameshift mutations identified in CDKN1B. (c) Schematic of the hemizygous deletions identified that target CDKN1B. The minimally deleted region contains 20 genes (Supplementary Table 8).

gene to the calculated expected number9. This analysis of the 50 cases identified statistically significant mutations in only one gene, the cell cycle regulator CDKN1B (P = 6.5 × 10−10). In total, we found small insertions and deletions within CDKN1B in 10% (5/50) of the cases (Fig. 1a and Supplementary Table 7), which lead to frameshift mutations (Fig. 1b). We validated these mutations by independent PCR and sequencing. Furthermore, copy number analysis identified hemizygous deletions encompassing CDKN1B in seven cases (Fig. 1c and Supplementary Table 8). Four out of these seven SI-NETs harboring CDKN1B deletions retained both copies of chromosome 18 compared to 8 out of 35 SI-NETs without CDKN1B deletions (P = 0.048, two-tailed Fisher’s exact test). The region encompassing CDKN1B, at 12p13, is frequently hemizygously deleted in ovarian, prostate and non–small cell lung cancers and multiple hematological malignancies10–16. To confirm the incidence of CDKN1B mutations in SI-NETs, we analyzed two independent cohorts: 48 SI-NETs reported by Banck et al.5 and an extension set of 81 SI-NETs sequenced to a mean 800fold coverage at CDKN1B. We detected two previously unreported somatic deletions within CDKN1B in the Banck et al.5 cohort that were not previously analyzed for insertions or deletions (indels) 5, both of which result in frameshift mutations. Analysis of the extension cohort identified seven small indels within CDKN1B that lead to frameshifts; the extension set did not have paired germline DNA, so we cannot exclude the possibility that some of these inactivating alterations are germline. Overall we detected heterozygous frameshift mutations in CDKN1B in 8% (14/180) of the individuals analyzed. 1484

The presence of heterozygous inactivating mutations in CDKN1B is consistent with the possibility that CDKN1B acts as a haploinsufficient tumor-suppressor gene in SI-NETs. One possible explanation is that some expression of p27 is necessary for cell proliferation, as has been described in certain oncogenic models17,18, thus making biallelic deletion disfavored. Several cancer genes that are recurrently mutated, including FBXW7, PTEN, DICER1 and CREBBP, were recently reported to be haploinsufficient tumor suppressors in mouse genetic models of cancer19–22. The increased susceptibility to tumors after DNA damage that is observed in Cdkn1b heterozygous knockout mouse models along with elevated cellular proliferation23–26 is consistent with the hypothesis that CDKN1B is haploinsufficient for tumor suppression. We found hemizygous loss of chromosome 18 (log2(copy number/2) < −0.1) in ~78% (43/55) of SI-NETs, but this loss was associated with only a slight increase in the mutation rate genome wide (Supplementary Figs. 2 and 3). We found two genes, including BCL10, which is mutated in colorectal cancer27, to be altered exclusively in the 12 cases with diploid chromosome 18 (Supplementary Fig. 3d). Because of the high frequency of hemizygous deletion of chromosome 18, we examined the cohort for somatic mutations to growth-inhibitory or ‘STOP’28 genes within the three frequently deleted regions of this chromosome. Although we observed no somatic mutations, it is possible that hemizygous loss of these genes may contribute to SI-NET tumorigenesis through altered gene dosage (Supplementary Table 9). In addition, comparison with genes that are mutated in small cell lung cancer29,30, a tumor type that shares neuroendocrine characteristics with SI-NETs, showed 199 genes with mutations in common in both VOLUME 45 | NUMBER 12 | DECEMBER 2013  Nature Genetics

letters Figure 2  Somatic mutation and copy number discordance between primary and metastatic tumors. (a) Venn diagram depicting the concordance and discordance in somatic mutation calls in primary or metastatic lesions analyzed in five cases. (b) Copy number profiles for concordant and discordant primary and metastatic tumors.

a

Primary

7 Germline SNP concordance:

18

D184 Metastasis

14

18

99.7%

B14 Metastasis

99.9%

Primary

14

31

B17 Metastasis

15

A17 Metastasis

Primary

29

11

99.7%

17

19

99.2%

Chromosome 2

4 3

6 5

8 7

10 9

12 11

14 13

16 15

17

18 20 22 Y 19 21 X

Discordant metastasis

studies; however, this overlap may be due to the high overall mutation rate in small cell 22 12 lung cancer rather than a shared mechanism of tumorigenesis (Supplementary Fig. 4 and Germline SNP 99.7% Supplementary Tables 10 and 11). concordance: To survey for genomic rearrangements, we performed whole-genome sequencing on b 24 tumor-normal sample pairs. The number 1 of somatic rearrangements detected by A16_TP A16_TM paired-end and split-read mapping31,32 D184_TP ranged from 0 to 45 per case, with a median of D184_TM B14_TP 7 (Supplementary Fig. 5 and Supplementary B14_TM Table 12). Of those rearrangements, 20% B17_TP B17_TM (33/163) involved genes or promoter A17_TP A17_TM regions, leading to five potential fusion proteins and two in-frame deletions; Deletion however, none of these rearrangements was recurrent (Supplementary Table 13). The concordance of SSNVs identified through whole-genome and whole-exome sequencing was ~95% on average when sufficient coverage was available6. Tumor heterogeneity within epithelial tumors can be exceptionally complex, and cells shed from the primary tumor can form distant metastases33. Approximately 25% of SI-NETs are multifocal tumors at the time of resection, and 10–15% of neuroendocrine metastases are diagnosed as being of unknown primary origin34,35. When we compared exomic mutations and copy number data for paired primary and metastatic tumors, we observed no overlap in SSNVs or somatic copy number alterations between the primary tumor and metastasis in two out of five primary-metastatic-normal trios (Fig. 2 and Supplementary Table 14). We confirmed that germline SNPs were concordant in the trios to exclude sample mix up. In one particular case (A16), the primary tumor contained a CDKN1B frameshift mutation and the metastasis did not, a phenomenon that has also been reported for PIK3CA and EGFR in breast and non–small cell lung cancer, respectively36,37. It is hypothesized that the metastases in these two cases may have been derived from an undiagnosed independent primary lesion, a subclonal population that was not detected by sequencing or a clone that was shed from the primary tumor early in the progression and before the acquisition of major genomic events. In contrast, Banck et al.5 assessed the overlap of 35 gene mutations in paired primary and metastasis samples and observed an 83% concordance in five cases. Given the small number of primary-metastaticnormal SI-NET trios (five cases each) in our two cohorts, the difference between these data sets is consistent with statistical fluctuation. A Fisher’s exact test comparing a case series with two primarymetastatic discordances and three concordances to a case series with five primary-metastatic somatic mutation concordances and no discordances yielded P = 0.22, which is consistent with the null hypothesis that the two data sets are identical. This suggests that SI-NETs are observed in a population in which a subset of patients may harbor multifocal tumors and another subset may harbor unifocal tumors.

Primary

Concordant metastasis

© 2013 Nature America, Inc. All rights reserved.

Primary

npg

A16 Metastasis

Nature Genetics  VOLUME 45 | NUMBER 12 | DECEMBER 2013

Neutral

Amplification

The discordance between primary tumors and metastasis, along with the multifocal nature of SI-NETs, highlights a challenge in identifying underlying driving events in these tumors. Somatic mutations targeting the cell cycle regulatory gene CDKN1B were the most frequent gene-specific events in SI-NETs. CDKN1B encodes a cyclin-dependent kinase inhibitor that binds to and inhibits CDK2 and CDK4 (refs. 38,39). Mouse models of Cdkn1b haploinsufficiency have larger body and organ size and enhanced sensitivity to mitogenic stimulation owing to greater Cdk2 activity23–25. In contrast to CDKN2A and related genes encoding inhibitors of CDK4–CDK6, somatic mutations in CDKN1B were recently reported at a low frequency in breast and prostate cancers16,40,41. CDKN1B is also known as MEN4 and is mutated in the germline of families with a phenotype of MEN-1 syndrome without an identifiable mutation in MEN1 (ref. 42). Furthermore, menin, the product of MEN1, associates with promoter regions to mediate the expression of CDKN1B and CDKN2C through epigenetic regulation43,44. In summary, we present a comprehensive genomic analysis of somatic variants and whole genomes of SI-NETs. SI-NETs are dominated by large, arm-level copy number gains and losses, but we found strikingly few recurrent somatic gene alterations. The discovery of recurrent CDKN1B mutations raises the possibility that the p21p27-p57 family proteins may be haploinsufficient tumor suppressors and suggests a focus on cell cycle regulation in understanding the pathogenesis of SI-NETs. Methods Methods and any associated references are available in the online version of the paper. Accession codes. Whole-exome sequencing of 55 SI-NETs and wholegenome sequencing of 24 SI-NETs in this study have been deposited in the dbGaP repository under accession code phs000579.v1.p1. 1485

letters Note: Any Supplementary Information and Source Data files are available in the online version of the paper. Acknowledgments This work was supported by grants from the Caring for Carcinoid Foundation (S.L.A. and M.M.), the Raymond and Beverly Sackler Foundation for the Arts and Sciences (C.T., A. Karpathakis, S.L.A. and M.M.) and Cancer Research UK (C.T. and A. Karpathakis).

npg

© 2013 Nature America, Inc. All rights reserved.

AUTHOR CONTRIBUTIONS J.M.F., A. Kiezun, A.H.R., R.S., S.L.A., C.R.H., M.K. and M.M. conceived and designed the experiments. J.M.F., A. Kiezun, A.H.R., S.S., C.S.P., Z.R.Q., A. Karpathakis, S.O., C.T., R.S., S.L.A., C.R.H., G.G., M.K. and M.M. analyzed the data. C.S.P., Z.R.Q., M.S.B., R.K., A.A.K., A. Karpathakis, V.M., T.C., J.P., N.P., S.M.H., L.K.B., M.S.L., T.P., A.M., A.S., K.C., S.L.C., A.I.O., S.F., R.T.J., D.V., G.S., T.M., M.C., D.C.C., A.S.B., S.O., C.T., R.S., S.L.A., C.R.H., G.G., M.K., J.G., L.G. and N.L. contributed reagents, materials and tools. E.N., D.A., R.O., E.S. and C.S. provided project management support. J.M.F., A. Kiezun, A.H.R., S.L.A., M.K. and M.M. wrote the manuscript with contributions from all other authors. COMPETING FINANCIAL INTERESTS The authors declare competing financial interests: details are available in the online version of the paper. Reprints and permissions information is available online at http://www.nature.com/ reprints/index.html. 1. Vinik, A.I. et al. NANETS consensus guidelines for the diagnosis of neuroendocrine tumor. Pancreas 39, 713–734 (2010). 2. Jiao, Y. et al. DAXX/ATRX, MEN1, and mTOR pathway genes are frequently altered in pancreatic neuroendocrine tumors. Science 331, 1199–1203 (2011). 3. Kulke, M.H. et al. High-resolution analysis of genetic alterations in small bowel carcinoid tumors reveals areas of recurrent amplification and loss. Genes Chromosom. Cancer 47, 591–603 (2008). 4. Cunningham, J.L. et al. Common pathogenetic mechanism involving human chromosome 18 in familial and sporadic ileal carcinoid tumors. Genes Chromosom. Cancer 50, 82–94 (2011). 5. Banck, M.S. et al. The genomic landscape of small intestine neuroendocrine tumors. J. Clin. Invest. 123, 2502–2508 (2013). 6. Cibulskis, K. et al. Sensitive detection of somatic point mutations in impure and heterogeneous cancer samples. Nat. Biotechnol. 31, 213–219 (2013). 7. Lawrence, M.S. et al. Mutational heterogeneity in cancer and the search for new cancer-associated genes. Nature 499, 214–218 (2013). 8. Futreal, P.A. et al. A census of human cancer genes. Nat. Rev. Cancer 4, 177–183 (2004). 9. Getz, G. et al. Comment on “The consensus coding sequences of human breast and colorectal cancers”. Science 317, 1500 (2007). 10. Hatta, Y., Takeuchi, S., Yokota, J. & Koeffler, H.P. Ovarian cancer has frequent loss of heterozygosity at chromosome 12p12.3–13.1 (region of TEL and Kip1 loci) and chromosome 12q23-ter: evidence for two new tumour-suppressor genes. Br. J. Cancer 75, 1256–1262 (1997). 11. Kibel, A.S., Freije, D., Isaacs, W.B. & Bova, G.S. Deletion mapping at 12p12–13 in metastatic prostate cancer. Genes Chromosom. Cancer 25, 270–276 (1999). 12. Takeuchi, S. et al. Frequent loss of heterozygosity in region of the KIP1 locus in non-small cell lung cancer: evidence for a new tumor suppressor gene on the short arm of chromosome 12. Cancer Res. 56, 738–740 (1996). 13. Pietenpol, J.A. et al. Assignment of the human p27Kip1 gene to 12p13 and its analysis in leukemias. Cancer Res. 55, 1206–1210 (1995). 14. Hetet, G. et al. Recurrent molecular deletion of the 12p13 region, centromeric to ETV6/TEL, in T-cell prolymphocytic leukemia. Hematol. J. 1, 42–47 (2000). 15. Le Toriellec, E. et al. Haploinsufficiency of CDKN1B contributes to leukemogenesis in T-cell prolymphocytic leukemia. Blood 111, 2321–2328 (2008). 16. Barbieri, C.E. et al. Exome sequencing identifies recurrent SPOP, FOXA1 and MED12 mutations in prostate cancer. Nat. Genet. 44, 685–689 (2012).

1486

17. Muraoka, R.S. et al. ErbB2/Neu-induced, cyclin D1–dependent transformation is accelerated in p27-haploinsufficient mammary epithelial cells but impaired in p27-null cells. Mol. Cell Biol. 22, 2204–2219 (2002). 18. Gao, H. et al. A critical role for p27kip1 gene dosage in a mouse model of prostate carcinogenesis. Proc. Natl. Acad. Sci. USA 101, 17204–17209 (2004). 19. Sancho, R. et al. F-box and WD repeat domain–containing 7 regulates intestinal cell lineage commitment and is a haploinsufficient tumor suppressor. Gastroenterology 139, 929–941 (2010). 20. Ying, H. et al. PTEN is a major tumor suppressor in pancreatic ductal adenocarcinoma and regulates an NF-κB–cytokine network. Cancer Discov. 1, 158–169 (2011). 21. Kumar, M.S. et al. Dicer1 functions as a haploinsufficient tumor suppressor. Genes Dev. 23, 2700–2704 (2009). 22. Zimmer, S.N. et al. Crebbp haploinsufficiency in mice alters the bone marrow microenvironment, leading to loss of stem cells and excessive myelopoiesis. Blood 118, 69–79 (2011). 23. Kiyokawa, H. et al. Enhanced growth of mice lacking the cyclin-dependent kinase inhibitor function of p27(Kip1). Cell 85, 721–732 (1996). 24. Nakayama, K. et al. Mice lacking p27(Kip1) display increased body size, multiple organ hyperplasia, retinal dysplasia, and pituitary tumors. Cell 85, 707–720 (1996). 25. Fero, M.L. et al. A syndrome of multiorgan hyperplasia with features of gigantism, tumorigenesis, and female sterility in p27(Kip1)-deficient mice. Cell 85, 733–744 (1996). 26. Fero, M.L., Randel, E., Gurley, K.E., Roberts, J.M. & Kemp, C.J. The murine gene p27Kip1 is haplo-insufficient for tumour suppression. Nature 396, 177–180 (1998). 27. The Cancer Genome Atlas Network. Comprehensive molecular characterization of human colon and rectal cancer. Nature 487, 330–337 (2012). 28. Solimini, N.L. et al. Recurrent hemizygous deletions in cancers may optimize proliferative potential. Science 337, 104–109 (2012). 29. Rudin, C.M. et al. Comprehensive genomic analysis identifies SOX2 as a frequently amplified gene in small-cell lung cancer. Nat. Genet. 44, 1111–1116 (2012). 30. Peifer, M. et al. Integrative genome analyses identify key somatic driver mutations of small-cell lung cancer. Nat. Genet. 44, 1104–1110 (2012). 31. Banerji, S. et al. Sequence analysis of mutations and translocations across breast cancer subtypes. Nature 486, 405–409 (2012). 32. Imielinski, M. et al. Mapping the hallmarks of lung adenocarcinoma with massively parallel sequencing. Cell 150, 1107–1120 (2012). 33. Gerlinger, M. et al. Intratumor heterogeneity and branched evolution revealed by multiregion sequencing. N. Engl. J. Med. 366, 883–892 (2012). 34. Habbe, N., Fendrich, V., Heverhagen, A., Ramaswamy, A. & Bartsch, D.K. Outcome of surgery for ileojejunal neuroendocrine tumors. Surg. Today 43, 1168–1174 (2013). 35. Polish, A., Vergo, M.T. & Agulnik, M. Management of neuroendocrine tumors of unknown origin. J. Natl. Compr. Canc. Netw. 9, 1397–1402 (2011). 36. Curtit, E. et al. Discordances in estrogen receptor status, progesterone receptor status, and HER2 status between primary breast cancer and metastasis. Oncologist 18, 667–674 (2013). 37. Park, S. et al. Discordance of molecular biomarkers associated with epidermal growth factor receptor pathway between primary tumors and lymph node metastasis in non-small cell lung cancer. J. Thorac. Oncol. 4, 809–815 (2009). 38. Polyak, K. et al. Cloning of p27Kip1, a cyclin-dependent kinase inhibitor and a potential mediator of extracellular antimitogenic signals. Cell 78, 59–66 (1994). 39. Toyoshima, H. & Hunter, T. p27, a novel inhibitor of G1 cyclin-Cdk protein kinase activity, is related to p21. Cell 78, 67–74 (1994). 40. The Cancer Genome Atlas. Comprehensive molecular portraits of human breast tumours. Nature 490, 61–70 (2012). 41. Stephens, P.J. et al. The landscape of cancer genes and mutational processes in breast cancer. Nature 486, 400–404 (2012). 42. Pellegata, N.S. et al. Germ-line mutations in p27Kip1 cause a multiple endocrine neoplasia syndrome in rats and humans. Proc. Natl. Acad. Sci. USA 103, 15558–15563 (2006). 43. Milne, T.A. et al. Menin and MLL cooperatively regulate expression of cyclindependent kinase inhibitors. Proc. Natl. Acad. Sci. USA 102, 749–754 (2005). 44. Karnik, S.K. et al. Menin regulates pancreatic islet growth by promoting histone methylation and expression of genes encoding p27Kip1 and p18INK4c. Proc. Natl. Acad. Sci. USA 102, 14659–14664 (2005).

VOLUME 45 | NUMBER 12 | DECEMBER 2013  Nature Genetics

ONLINE METHODS

Whole-exome sequencing. The generation of sequencing data was performed using a protocol that has been detailed previously31,32,45. In brief, exonic regions were captured using the Agilent V2 capture probe set and sequenced by 76-bp paired-end reads using Illumina HiSeq2000 instruments. A median of 9.15 Gb of unique sequence data were generated for each sample, 91% of which were aligned to the target exome using the Burrows-Wheeler Aligner (BWA)46, resulting in a median coverage of each base of 140× (Supplementary Table 3). Whole-genome sequencing. Genomic DNA from 23 tumor-normal pairs was sent to Axeq Technologies for whole-genome library construction and sequencing. Genomic DNA was sheared to an average size of 400 bp, and libraries were prepared following the Illumina TruSeq protocol. Each library was sequenced on an Illumina HiSeq2000 to generate 101-bp paired-end reads. Each whole-genome library was sequenced to generate 108 Gb of raw data with a mean depth of 30× alignable coverage (Supplementary Table 4). Variant calling. Downstream sequencing analysis was performed as described6,32. Before mutation and indel calling, sequencing reads were locally realigned to improve the detection of indels and decrease the number of false-positive SNVs caused by misaligned reads, particularly at the 3′ end47. For mutation detection, >14 reads in the tumors and >8 reads in the normal samples were necessary to call candidate somatic base substitutions, and indels were detected using MuTect6. Germline mutations were detected using the UnifiedGenotyper47. All somatic mutations were manually reviewed and visually confirmed with the Integrated Genomics Viewer (http:// www.broadinstitute.org/igv/). Extension cohort and validation sequencing. Approximately 100–200 ng of DNA was used as the template for a multiplex PCR reaction tiling CDKN1B

using the Fluidigm Access Array microfluidic device. PCR products were barcoded and pooled in sets of 48 samples and subjected to Illumina sequencing on MiSeq to a mean coverage of 800×. Reads were aligned to hg19 using BWA and analyzed as previously described48. CDKN1B indel calling in the Banck et al.5 cohort. The whole-exome sequencing files from the current study and the study by Banck et al.5 were run through the Strelka algorithm 49 to identify indels. The Strelka algorithm identified 100% of the CDKN1B indels identified with the MuTect algorithms6. Copy number profiling. DNA from 61 samples was hybridized to Affymetrix SNP6.0 arrays using standard approaches. Raw data were preprocessed, normalized and segmented as previously described in detail32. Segmented copy number data were visualized using the Integrated Genomics Viewer (http://www.broadinstitute.org/igv). Segmented copy number profiles were generated from whole-exome sequencing using the SegSeq algorithm50. To analyze segmented copy number profiles, the GISTIC 2.0 algorithm was used to identify arm-level and focal regions of copy number alterations as described51. A length threshold of 80% of a chromosome arm was used to distinguish between arm-level and focal events. To remove false-positive segments resulting from hypersegmentation, we further filtered segments using an amplitude threshold at a copy difference of 0.1. The frequencies of samples with chromosome arms displaying gains (relative copy number >2.1) and losses (relative copy number