l e t t e r s

Genomic sequencing of meningiomas identifies oncogenic SMO and AKT1 mutations

Priscilla K Brastianos1–4,11, Peleg M Horowitz3–6,11, Sandro Santagata3,7, Robert T Jones1,8, Aaron McKenna4, Gad Getz4, Keith L Ligon3,7, Emanuele Palescandolo8, Paul Van Hummelen1,8, Matthew D Ducar1,8, Alina Raza1,8, Ashwini Sunkavalli1,8, Laura E MacConaill1,8, Anat O Stemmer-Rachamimov3,9, David N Louis3,9,10, William C Hahn1,3,4,8, Ian F Dunn3,4,6 & Rameen Beroukhim1,3–5,8

Meningiomas are the most common primary nervous system 76% and 32%, respectively1,3,4. Although there are recent reports of tumor. The tumor suppressor NF2 is disrupted in approximately a stepwise progression of a subset of grade I meningiomas to higher half of all meningiomas1, but the complete spectrum of genetic grades5,6, these secondary grade II and III meningiomas may differ changes remains undefined. We performed whole-genome or fundamentally from spontaneously arising grade II and III tumors. whole-exome sequencing on 17 meningiomas and focused Radiation is frequently used as an adjunct to surgery; however, there sequencing on an additional 48 tumors to identify and validate are no effective chemotherapeutic options when surgery and radiation somatic genetic alterations. Most meningiomas had simple fail to offer durable long-term disease control7. genomes, with fewer mutations, rearrangements and copy- We sequenced DNA from a discovery set of grade I meningiomas number alterations than reported in other tumors in adults. using whole-genome (n = 11) and whole-exome (n = 6) techniques However, several meningiomas harbored more complex patterns to identify somatic copy-number alterations (SCNAs), rearrange- of copy-number changes and rearrangements, including one ments, mutations and insertions and/or deletions (indels) through- tumor with chromothripsis. We confirmed focal NF2 inactivation out the genome. We then sequenced 645 known cancer-associated in 43% of tumors and found alterations in epigenetic modifiers , including genes altered in the discovery set, in a validation in an additional 8% of tumors. A subset of meningiomas lacking set of 30 additional grade I tumors. We also sequenced these genes NF2 alterations harbored recurrent oncogenic mutations in in an extrapolation set of 18 grade II or III meningiomas to evaluate

© 2013 Nature America, Inc. All rights reserved. America, Inc. © 2013 Nature AKT1 (p.Glu17Lys) and SMO (p.Trp535Leu) and exhibited whether findings in grade I tumors extended to tumors of higher immunohistochemical evidence of activation of these pathways. grade (Online Methods and Supplementary Tables 1 and 2). These mutations were present in therapeutically challenging Meningiomas in the discovery set had a small number of somatic tumors of the skull base and higher grade. These results begin to genetic events compared to other tumor types8–14. We sequenced to npg define the spectrum of genetic alterations in meningiomas and high depth (median coverage of 57× for the genome, 181× for the identify potential therapeutic targets. exome and 154× for the validation and extrapolation) and used large- insert libraries in whole-genome samples (500 bp and 800 bp) to opti- Meningiomas, tumors that arise from the arachnoidal cap cells of the mize our detection of rearrangements and other events. Nevertheless, leptomeninges, constitute approximately one-third of primary central we found that the median meningioma had SCNAs affecting only nervous system (CNS) tumors1. Most meningiomas (80%) are World 3.3% of the genome, one rearrangement and eight nonsynonymous Health Organization (WHO) grade I and are treated by surgical resec- mutations. Among mutations, cytosine-to-thymidine transitions tion. However, resection of some meningiomas, particularly at the skull within a CpG context predominated, consistent with spontaneous base, is associated with high morbidity. Moreover, 18% of these tumors deamination (Supplementary Fig. 1a)15. The rates of alteration we recur within 5 years, and patients with grade I tumors have significantly observed in meningiomas were significantly lower (often by a factor reduced long-term survival that is related to both tumor recurrence and of ten or more) than previously determined rates for other tumors stroke risk2. Recurrence rates for grade II and III meningiomas can be that have been sequenced to lower depth and with smaller insert sizes as high as 40% and 80%, with 5-year overall survival of approximately (Fig. 1a–c). These findings may reflect differences in mitotic index,

1Department of Medical Oncology, Dana-Farber Cancer Institute, Boston, Massachusetts, USA. 2Department of Hematology/Oncology, Massachusetts General Hospital, Boston, Massachusetts, USA. 3Harvard Medical School, Boston, Massachusetts, USA. 4Broad Institute of MIT and Harvard, Boston, Massachusetts, USA. 5Department of Cancer Biology, Dana-Farber Cancer Institute, Boston, Massachusetts, USA. 6Department of Neurosurgery, Brigham and Women’s Hospital, Boston, Massachusetts, USA. 7Department of Pathology, Brigham and Women’s Hospital, Boston, Massachusetts, USA. 8Center for Cancer Genome Discovery, Dana-Farber Cancer Institute, Boston, Massachusetts, USA. 9Pathology Service, Massachusetts General Hospital, Boston, Massachusetts, USA. 10Cancer Center, Massachusetts General Hospital, Boston, Massachusetts, USA. 11These authors contributed equally to this work. Correspondence should be addressed to R.B. ([email protected]), I.F.D. ([email protected]) or W.C.H. ([email protected]).

Received 10 August 2012; accepted 19 December 2012; published online 20 January 2013; doi:10.1038/ng.2526

Nature Genetics ADVANCE ONLINE PUBLICATION  l e t t e r s

Figure 1 Landscape of somatic alterations a NS NS ** ** d NF2 Focally altered Not focally altered in the grade I meningioma genome. 0.8 (a–c) Fraction of the genome altered by 1 SCNAs (a), number of somatic intra- and 0.6 2 interchromosomal rearrangements (b) and 3 number of nonsynonymous mutations (c) 0.4 4 in meningiomas and other tumor types. MEN, meningioma; PR, prostate cancer; 5 0.2 6 affected by SCNAs CRC, colorectal carcinoma; BCL, diffuse large Fraction of genome 7 B-cell lymphoma; MEL, melanoma; MM, multiple 8 myeloma; HN, head and neck cancers. 0 MEN PR CRC BCL MEL 9 Blue bars represent the medians. **P < 0.01; 10 ***P < 0.0001; NS, not significant. (d) Somatic 11 b ** ** *** 12 genetic alterations in individual grade I meningio­ 250 13 mas of the discovery cohort. Top, copy-number 14 15 profiles (red, gain; blue, loss). Middle, number 200 16 17 of somatic rearrangements per tumor. Samples 18 19 150 20 with labels starting with MENex are exome- 21 22 sequenced samples in which no rearrangements 100 X were detected. Bottom, total nonsynonymous Number of and synonymous mutations in exons. rearrangements 50 40 30 Interchromosomal 0 Intrachromosomal MEN MEL CRC PR 20

prior treatments or carcinogen exposure Number of 10 rearrangements between these tumor types. c 4 10 0 In contrast, rates of genomic disruption were ** *** *** *** *** *** 25 Synonymous Nonsynonymous higher in the grade II and III meningiomas 103 20 in the extrapolation set than for their grade 15 2 I counterparts (Supplementary Fig. 2). The 10 10

mean rate of nonsynonymous mutation within Number of 5 Number of 1 the 645 sequenced genes was nearly twice as 10 mutations in exons 0

high in higher-grade tumors (3.0 versus 1.6; nonsynonymous mutations P < 0.02), and the proportion of the genome 100

MEN PR MM HN BCL CRC MEL MEN0008 MEN0009 MEN0018 MEN0017 MEN0016 MEN0019 MEN0022 MEN0020 MEN0015 MEN0011 MEN0021 affected by SCNAs was markedly higher MENex002 MENex003 MENex006 MENex004 MENex005 MENex001 (median of 12.8% versus 0.3%; P < 0.001). Among SCNAs, loss of chromosome 22 (containing NF2) was We identified a total of 110 candidate somatic rearrangements in sam- the most frequent genetic alteration in the discovery cohort, ples that underwent whole-genome analysis, of which 93 were confirmed occurring in all ten tumors with focal NF2 alteration (defined to have reads spanning the putative fusion site (Supplementary Fig. 4 as mutation, indel or rearrangement within the NF2 locus) and and Supplementary Table 3). Most of these rearrangements were con- © 2013 Nature America, Inc. All rights reserved. America, Inc. © 2013 Nature one without (Fig. 1d and Supplementary Fig. 3). The association centrated in three tumors (Fig. 2), one of which had chromothripsis—the between NF2 mutation and chromosome 22 loss extended to the simultaneous rearrangement of one or more chromosome, estimated validation and extrapolation sets (P < 0.0001) and confirms previ- to occur in 2–3% of aggressive malignancies19,20. One of the genes dis- npg ous reports using other methods16. In the grade I tumors, we also rupted by chromothripsis was the putative tumor suppressor NEGR1, found recurrent significant losses on 1p, 7p, 14p and 19 and gains which was also rearranged in a second tumor. NEGR1 has a role in CNS on 5 and 20 (q < 0.1), some of which have been development and is recurrently deleted in neuroblastoma21, although its described previously17. Higher-grade tumors of the extrapolation role in tumorigenesis remains to be elucidated. A third sample harbored set had additional recurrent losses on 10q and 14q, previously disruptions in the known and putative tumor suppressors NF1, NF2 and reported in atypical and anaplastic meningiomas1. We did not CDK14. A copy-neutral 12.5-Mb inversion caused reciprocal fusion of observe previously described losses of FPB41L3 (also known as NF2 (before exon 2) and TCF20, representing a new mechanism of NF2 DAL-1), a related to NF2 on 18p (ref. 18). inactivation in meningiomas. We also identified other translocations

Y Y Y X X X 22 22 22 Figure 2 Somatic rearrangements disrupt 1 1 1 21 21 21 20 20 20 tumor suppressors in several meningiomas. 2 2 2 19 19 19 Circos plots show the SCNAs (inner ring 18 18 NEGR1 18 NF2 3 3 3 heat map) and intra- and interchromosomal 17 17 17 rearrangements (green and purple arcs, 16 16 16 NEGR1 4 4 4 respectively) in three meningiomas for 15 15 15 NF1 which the whole genomes were sequenced. 14 5 14 5 14 5 Left, chromothripsis of chromosome 1 (enlarged in inset) in an NF2-mutated 13 6 13 6 13 CDK14 6 sample (MEN0017) disrupts the putative 12 7 12 7 12 7

11 8 11 8 11 8 tumor suppressor NEGR1. Middle, a 9 9 9 10 10 10 second NF2-mutated sample (MEN0018) harbors 29 rearrangements, including disruption of NEGR1. Right, an inversion on chromosome 22 disrupts NF2, and other rearrangements affect NF1 and CDK14 (MEN0009). Data from the remaining tumors are shown in Supplementary Figure 4.

 aDVANCE ONLINE PUBLICATION Nature Genetics l e t t e r s

Figure 3 Significant and selected cancer-related Discovery set Validation set Extrapolation set Grade somatic mutations, indels and translocations in Grade I Grade I Grade II III FDR q value NF2 meningiomas. Mutation subtypes are denoted by SMO color. If multiple mutations were found in a gene AKT1 in a single sample, only one is shown. Discovery MTOR CBL set tumors are shown in the same order as in STK11 Figure 1d. The significance of mutations in each NOTCH2

gene is shown to the right by false discovery rate Cell APC

signaling CDKN2C (FDR) q value. The full list of mutated genes is NF1 given in Supplementary Table 4. NEGR1 KDM5C KDM6A SMARCB1 modifiers Epigenetic TP53 in samples that were not subjected to whole- CHEK2 CD300C Other genome analysis, including a truncating trans- genes RGPD3 location in CHEK2 (Supplementary Fig. 4i and Synonymous Splice site Frame shift Translocation 0 2 4 6 8 10 Supplementary Table 3). The loss of CHEK2 Missense Nonsense In-frame indel Not covered –log10(q value) may lead to impaired DNA repair, as has been described in meningioma cell lines22. The most frequently mutated gene was NF2, which exhibited a Among genes mutated in our cohort and previously associated with total of nine nonsense mutations, nine splice-site mutations and nine cancer, three (KDM5C, KDM6A and SMARCB1) are epigenetic modi- frameshift indels in addition to the above-mentioned translocation fiers. Mutations involving these genes affected 8% of the cohort. Both (Supplementary Fig. 5). The majority of these events overlapped KDM5C (also known as JARID1C) and KDM6A are histone demethy- with previously reported events, and our rates of NF2 inactivation lases23; SMARCB1 (also known as SNF5 and INI1) is a member of were consistent with published rates1. The nonsynonymous muta- the SWI/SNF chromatin-remodeling complex. SMARCB1 is located tion rates of tumors with and without focal NF2 alterations did not 6 Mb away from NF2, and a four-hit model of biallelic inactivation differ significantly, nor did the spectra of their mutation subtypes of both genes has been described in familial schwannomas24. The (Supplementary Fig. 1b–d). mutation we identified (encoding a p.Arg374Gln alteration) is near We also identified nonsynonymous mutations in 190 other genes a mutational hotspot (p.Arg377His) described in meningiomas25, (Supplementary Tables 4 and 5). We determined the significance and germline mutations in SMARCB1, including ones encoding a of mutation rates in individual genes relative to genome-wide back- p.Arg374Gln alteration, have been reported in the congenital disorder ground rates of mutation. Six genes reached statistical significance Coffin-Siris syndrome26. (Fig. 3, Supplementary Fig. 5 and Supplementary Table 4). Four We observed mutations of SMO, a member of the Hedgehog of these (NF2, KDM5C, SMO and AKT1) have previously been signaling pathway, in three tumors (5%). None of these tumors implicated in cancer, and the other two (RGPD3 and CD300C) had focal alteration of NF2. Two samples harbored a mutation in have not. Isolated or rare mutations that did not reach signifi- SMO encoding a p.Trp535Leu alteration, a known oncogenic muta- cance across the entire cohort were observed in 23 genes with tion in basal-cell carcinomas27, and a third harbored a mutation known involvement in cancer, including TP53, APC, CBL, STK11 encoding a p.Leu412Phe alteration, previously reported in desmo- © 2013 Nature America, Inc. All rights reserved. America, Inc. © 2013 Nature and NOTCH2. plastic medulloblastoma28. The allelic fractions of these mutations (45%, 55% and 47%, respectively) were among the highest in these tumors, sug- npg a AKT1 mutated SMO mutated gesting clonality. Moreover, dysregulation NF2 mutated of the Hedgehog pathway has previously Others been described in meningiomas29,30. Gorlin syndrome (nevoid basal-cell carcinoma syndrome) is caused by a germline PTCH1

Meningothelial NF2 mutated Transitional Fibroblastic Figure 4 Associations between mutations Angiomatous in Hedgehog and AKT-mTOR pathways Atypical and histological findings. (a) Samples with SMO mutated Anaplastic AKT1-MTOR mutated mutations in SMO and AKT1-MTOR are predominantly of the meningothelial subtype Positive control SMO-WT meningioma SMO-mutated meningioma Atypical meningioma b (P = 0.005 and 0.009, respectively). NF2-mutated samples are predominantly fibroblastic and transitional (P = 0.013). GAB1 Samples underwent hematoxylin and eosin staining (left). The distribution of samples within different histological subtypes is shown Positive control AKT1-WT meningioma AKT1-mutated meningioma SMO-mutated meningioma by pie chart (right). (b) Immunohistochemistry indicates activation of the Hedgehog (GAB1) and AKT-mTOR (STMN1) pathways in tumors harboring and mutations,

STMN1 SMO AKT1 respectively (P = 0.0008 and 3 × 10−6). Scale bars, 50 µm. WT, wild type.

Nature Genetics ADVANCE ONLINE PUBLICATION  l e t t e r s

mutation (upstream of SMO) and is characterized by predisposition are driven by Hedgehog pathway mutations39. Likewise, inhibitors to multiple cancers, including desmoplastic medulloblastomas and of the PI3K-AKT-mTOR pathway have shown promise in preclinical meningiomas31,32. Although we did not observe somatic mutations and clinical trials in multiple cancer types40. The paucity of addi- in PTCH1, two samples harbored germline alterations that were also tional genetic events in meningiomas harboring mutations of the present in the tumors (encoding p.Asp436Asn and p.[Glu44Gly]+ PI3K and Hedgehog signaling pathways suggests that patients with [Pro1282Leu] alterations). However, these individuals did not exhibit these meningiomas may benefit from such targeted therapies already manifestations of Gorlin syndrome, such that the functional relevance in development or use. of these specific events is unclear. Six samples had mutations of the phosphatidylinositol 3- URLs. Picard, http://picard.sourceforge.net; hg19 reference genome (PI3K)-AKT-mTOR pathway. None of these had mutations of NF2 sequence, ftp://ftp-trace.ncbi.nih.gov/1000genomes/ftp/technical/ or SMO (P = 0.03). Five samples harbored identical AKT1 muta- reference/; Burrows-Wheeler Aligner (BWA), http://bio-bwa.source tions (p.Glu17Lys) known to be oncogenic in breast, colorectal and forge.net/bwa.shtml; SAMtools, http://samtools.sourceforge.net; lung cancers33. The glutamic acid–to-lysine substitution results in Genome Analysis Toolkit (GATK), http://www.broadinstitute.org/ constitutive AKT1 activation, which stimulates downstream mTOR gatk; algorithms for variant and germline calling, https://conflu- signaling. The sixth sample had a new MTOR mutation (encoding a ence.broadinstitute.org/display/CGATools; Circos, http://circos.ca/; p.Asp1279Val alteration). UniProt, http://www.uniprot.org/; Integrative Genomics Viewer, The mutations in the Hedgehog and PI3K pathways were also http://www.broadinstitute.org/igv/. associated with particular histopathological subtypes (Fig. 4). Meningiomas with mutated SMO and AKT1-MTOR were predomi- Methods nantly of the meningothelial subtype (P = 0.009 and 0.005, respec- Methods and any associated references are available in the online tively, Fisher’s exact test). This is in contrast to the grade I NF2-altered version of the paper. tumors, which were predominantly fibroblastic and/or transitional (P = 0.013), consistent with previous reports34. We did not find statis- Accession codes. Data, including sequence data and analyses, tically significant correlations with other clinically relevant variables, are available for download from the database of Genotypes and including age, tumor location or previous resection. However, one Phenotypes (dbGaP) under accession phs000552.v1.p1. AKT1- and two SMO-mutated tumors were resected from the skull base, a region that offers particular challenges to resection and higher Note: Supplementary information is available in the online version of the paper. 35 rates of recurrence (Supplementary Table 5). Acknowledgments To validate these findings, we also genotyped an additional 46 grade The authors would like to thank M. Lawrence, N. Stransky, M. Imielinski, I and 49 grade II or III meningiomas for AKT1 (p.Glu17Lys) and SMO K. Cibulskis, S. Carter, C. Stewart, C.-Z. Zhang, Y. Drier, S. Schumacher and (p.Trp535Leu and p.Leu412Phe) mutations using a mass spectrometry– B. Tabak for their assistance with genomic analyses; P. Wen, O. Al-Mefty, T. Batchelor, B. Reisler, M. Johnson, W. Richards and J. Kim for their assistance based method (Sequenom hME). We found one SMO mutation with tissue acquisition; and A. Pisarek-Horowitz, S.A. Horowitz, G. Bergthold (p.Leu412Phe) and three additional AKT1 mutations (p.Glu17Lys) and C.H. Brastianos for critical review of the manuscript. This work was supported (Supplementary Table 6). One of the latter was in a grade III tumor, by the Brain Science Foundation (I.F.D., P.K.B., R.B. and S.S.), the Pediatric Low- and the remaining three were in skull base tumors, of which two were Grade Astrocytoma Foundation (R.B. and W.C.H.), the American Brain Tumor Association (P.K.B.), a Conquer Cancer Foundation Young Investigator Award of meningothelial histology.

© 2013 Nature America, Inc. All rights reserved. America, Inc. © 2013 Nature (P.K.B.), US National Institutes of Health (NIH) grants K08 CA122833 (R.B.), Mutations in SMO, AKT1 and MTOR also correlated with the pres- K08 NS064168 (S.S.) and K12 CA090354-11 (P.K.B.), and US Department of ence of markers of pathway activation. GAB1 immunoreactivity is Defense grant W81XWH-12-1-0136 (P.M.H.). used in medulloblastomas to characterize tumors with Hedgehog AUTHOR CONTRIBUTIONS npg pathway activation36. Of the 65 meningiomas, 7 exhibited strong P.K.B., P.M.H., I.F.D., R.B. and W.C.H. conceived the study, designed the immunoreactivity for GAB1, including the 3 tumors harboring SMO experiments, analyzed the data and wrote the manuscript. P.M.H. performed the mutations (P = 0.0008). The other four GAB1-positive tumors were bioinformatics analyses, and A.M., G.G., R.B. and M.D.D. provided analytical grade II meningiomas. Likewise, STMN1 expression is a marker of advice. S.S., A.O.S.-R. and D.N.L. reviewed the histopathology, and S.S. performed PI3K-AKT pathway activation37. Ten meningiomas exhibited strong the immunohistochemistry. K.L.L. managed the tissue repository. R.T.J., E.P., P.V.H., A.R., A.S. and L.E.M. provided technical support and performed STMN1 expression, including all six AKT1- and MTOR-mutated sequencing. W.C.H., I.F.D. and R.B. supervised the study. All authors discussed the −6 meningiomas (P = 3 × 10 ). Notably, all 3 SMO-mutated menin- results and implications and commented on the manuscript. giomas also exhibited strong (n = 2) or moderate (n = 1) STMN1 staining. This is consistent with studies that have suggested that the COMPETING FINANCIAL INTERESTS 38 The authors declare competing financial interests: details are available in the online Hedgehog and PI3K-AKT-mTOR pathways may interact . version of the paper. We observed recurrent mutations in signaling pathways and epi- genetic modifiers, but our data also highlight the heterogeneity of Published online at http://www.nature.com/doifinder/10.1038/ng.2526. mutations in these tumors. Among the 17 tumors we comprehen- Reprints and permissions information is available online at http://www.nature.com/ reprints/index.html. sively characterized, 3 (18%) did not have mutations in any of the significantly mutated genes. Genomic characterization of additional 1. Choy, W. et al. The molecular genetics and tumor pathogenesis of meningiomas tumors, particularly those of higher grade, is likely to reveal additional and the future directions of meningioma treatments. Neurosurg. Focus 30, oncogenic mechanisms. E6 (2011). 2. van Alkemade, H. et al. Impaired survival and long-term neurological problems in These observations also have the potential to guide new thera- benign meningioma. Neuro-oncol. 14, 658–666 (2012). peutic strategies. We observed SMO and AKT1 mutations in tumors 3. Durand, A. et al. WHO grade II and III meningiomas: a study of prognostic factors. that pose special therapeutic challenges, including one high-grade J. Neurooncol. 95, 367–375 (2009). 4. Perry, A., Scheithauer, B.W., Stafford, S.L., Lohse, C.M. & Wollan, P.C. “Malignancy” and six skull-base tumors. Inhibitors of SMO have generated high in meningiomas: a clinicopathologic study of 116 patients, with grading implications. response rates in patients with basal-cell carcinoma, many of which Cancer 85, 2046–2056 (1999).

 aDVANCE ONLINE PUBLICATION Nature Genetics l e t t e r s

5. Goutagny, S. et al. Genomic profiling reveals alternative genetic pathways of 23. Kooistra, S.M. & Helin, K. Molecular mechanisms and potential functions of histone meningioma malignant progression dependent on the underlying NF2 status. Clin. demethylases. Nat. Rev. Mol. Cell Biol. 13, 297–311 (2012). Cancer Res. 16, 4155–4164 (2010). 24. Hadfield, K.D. et al. Molecular characterisation of SMARCB1 and NF2 in familial 6. Krayenbühl, N., Pravdenkova, S. & Al-Mefty, O. De novo versus transformed atypical and sporadic schwannomatosis. J. Med. Genet. 45, 332–339 (2008). and anaplastic meningiomas: comparisons of clinical course, cytogenetics, 25. Schmitz, U. et al. INI1 mutations in meningiomas at a potential hotspot in exon 9. cytokinetics, and outcome. Neurosurgery 61, 495–503, discussion 503–504 Br. J. Cancer 84, 199–201 (2001). (2007). 26. Tsurusaki, Y. et al. Mutations affecting components of the SWI/SNF complex cause 7. Wen, P.Y., Quant, E., Drappatz, J., Beroukhim, R. & Norden, A.D. Medical therapies Coffin-Siris syndrome. Nat. Genet. 44, 376–378 (2012). for meningiomas. J. Neurooncol. 99, 365–378 (2010). 27. Reifenberger, J. et al. Missense mutations in SMOH in sporadic basal cell carcinomas 8. Bass, A.J. et al. Genomic sequencing of colorectal adenocarcinomas identifies a of the skin and primitive neuroectodermal tumors of the central nervous system. recurrent VTI1A-TCF7L2 fusion. Nat. Genet. 43, 964–968 (2011). Cancer Res. 58, 1798–1803 (1998). 9. Berger, M.F. et al. Melanoma genome sequencing reveals frequent PREX2 mutations. 28. Jones, D.T. et al. Dissecting the genomic complexity underlying medulloblastoma. Nature 485, 502–506 (2012). Nature 488, 100–105 (2012). 10. Berger, M.F. et al. The genomic complexity of primary human prostate cancer. 29. Laurendeau, I. et al. profiling of the hedgehog signaling pathway Nature 470, 214–220 (2011). in human meningiomas. Mol. Med. 16, 262–270 (2010). 11. Chapman, M.A. et al. Initial genome sequencing and analysis of multiple myeloma. 30. Aavikko, M. et al. Loss of SUFU function in familial multiple meningioma. Am. J. Nature 471, 467–472 (2011). Hum. Genet. 91, 520–526 (2012). 12. Stransky, N. et al. The mutational landscape of head and neck squamous cell 31. Gorlin, R.J. Nevoid basal-cell carcinoma syndrome. Medicine (Baltimore) 66, carcinoma. Science 333, 1157–1160 (2011). 98–113 (1987). 13. Lohr, J.G. et al. Discovery and prioritization of somatic mutations in diffuse large B-cell 32. Santos, D.C. et al. PTCH1 gene mutations in exon 17 and loss of heterozygosity lymphoma (DLBCL) by whole-exome sequencing. Proc. Natl. Acad. Sci. USA 109, on D9S180 microsatellite in sporadic and inherited human basal cell carcinomas. 3879–3884 (2012). Int. J. Dermatol. 50, 838–843 (2011). 14. Beroukhim, R. et al. The landscape of somatic copy-number alteration across human 33. Bleeker, F.E. et al. AKT1E17K in human solid tumours. 27, 5648–5650 cancers. Nature 463, 899–905 (2010). (2008). 15. Greenman, C. et al. Patterns of somatic mutation in human cancer genomes. 34. Wellenreuther, R. et al. Analysis of the neurofibromatosis 2 gene reveals molecular Nature 446, 153–158 (2007). variants of meningioma. Am. J. Pathol. 146, 827–832 (1995). 16. Mawrin, C. & Perry, A. Pathological classification and molecular genetics of 35. Mathiesen, T., Lindquist, C., Kihlstrom, L. & Karlsson, B. Recurrence of cranial meningiomas. J. Neurooncol. 99, 379–391 (2010). base meningiomas. Neurosurgery 39, 2–7, discussion 8–9 (1996). 17. Ketter, R. et al. Hyperdiploidy defines a distinct cytogenetic entity of meningiomas. 36. Ellison, D.W. et al. Medulloblastoma: clinicopathological correlates of SHH, WNT, and J. Neurooncol. 83, 213–221 (2007). non-SHH/WNT molecular subgroups. Acta Neuropathol. 121, 381–396 (2011). 18. Nunes, F. et al. Inactivation patterns of NF2 and DAL-1/4.1B (EPB41L3) in sporadic 37. Karst, A.M. et al. Stathmin 1, a marker of PI3K pathway activation and regulator meningioma. Cancer Genet. Cytogenet. 162, 135–139 (2005). of microtubule dynamics, is expressed in early pelvic serous carcinomas. Gynecol. 19. Rausch, T. et al. Genome sequencing of pediatric medulloblastoma links catastrophic Oncol. 123, 5–12 (2011). DNA rearrangements with TP53 mutations. Cell 148, 59–71 (2012). 38. Riobó, N.A., Lu, K., Ai, X., Haines, G.M. & Emerson, C.P. Jr. Phosphoinositide 20. Stephens, P.J. et al. Massive genomic rearrangement acquired in a single 3-kinase and Akt are essential for Sonic Hedgehog signaling. Proc. Natl. Acad. Sci. catastrophic event during cancer development. Cell 144, 27–40 (2011). USA 103, 4505–4510 (2006). 21. Takita, J. et al. Aberrations of NEGR1 on 1p31 and MYEOV on 11q13 in 39. Von Hoff, D.D. et al. Inhibition of the hedgehog pathway in advanced basal-cell neuroblastoma. Cancer Sci. 102, 1645–1650 (2011). carcinoma. N. Engl. J. Med. 361, 1164–1172 (2009). 22. Yang, H.W. et al. Alternative splicing of CHEK2 and codeletion with NF2 promote 40. Janku, F. et al. PI3K/AKT/mTOR inhibitors in patients with breast and gynecologic chromosomal instability in meningioma. Neoplasia 14, 20–28 (2012). malignancies harboring PIK3CA mutations. J. Clin. Oncol. 30, 777–782 (2012). © 2013 Nature America, Inc. All rights reserved. America, Inc. © 2013 Nature npg

Nature Genetics ADVANCE ONLINE PUBLICATION  ONLINE METHODS SCNAs and rearrangements. SCNAs were evaluated using SegSeq48, which Sample selection and preparation. This study was reviewed and approved uses a combination of local change-point analysis and subsequent merging by the human subjects institutional review boards (IRBs) of the Dana-Farber of adjacent chromosomal segments with similar copy number. For arm-level Cancer Institute, Brigham and Women’s Hospital and the Broad Institute of significance analysis, pseudo-markers were added every 10,000 bases to seg- Harvard and MIT. Written informed consent was obtained from all partici- mented copy-number data, and these were analyzed by GISTIC 2.0 to deter- pants (Supplementary Note). We chose to initially perform whole-exome mine which events occurred more often than expected by chance14. sequencing in 6 meningiomas as a pilot study to determine the optimal depth Rearrangements and their exact breakpoints were identified using a combi- of whole-genome sequencing needed to provide adequate tumor coverage in nation of dRanger and BreakPointer algorithms8,10. Briefly, dRanger identified spite of stromal contamination. These 6 tumors and 11 additional meningiomas potential rearrangements from read pairs mapping to different chromosomes on which whole-genome sequencing was performed comprised the discovery or to unexpected positions or orientations on the same chromosome, and cohort. To validate clinically actionable mutations identified in the discovery BreakPointer located reads spanning a fusion and maps the exact breakpoint set, we performed focused sequencing on a validation cohort consisting of using modified Smith-Waterman alignment10. Rearrangements were assigned an additional 30 grade I meningiomas. We also assembled an extrapolation a score reflecting the number of tumor reads supporting a breakpoint, the cohort of 15 grade II and 3 grade III meningiomas (Supplementary Table 1) fraction of nearby reads with MAPQ0, the prevalence of other nearby dis- to compare genetic aberrations in low- and high-grade tumors. Our approach cordant pairs and the standard deviation of breakpoint starting positions. of deep, broad sequencing of a small representative cohort followed by valida- To minimize false positive rearrangements, only breakpoints with a score of ≥8 tion in an independent larger group is similar to the approach used by others passing BreakPointer were accepted. The specificity of this type of approach to discover IDH1 (ref. 41) and PIK3CA42 mutations. Three study pathologists has previously been shown to be >90% (ref. 10). Copy-number and rearrange- reviewed the histopathological diagnosis, grade and purity of each tumor. ment results were displayed using the Circos program49. Representative fresh-frozen blocks with estimated purity of ≥90% were selected. DNA was extracted from tissue shavings and buffy coat prepara- Identification of base-pair substitutions and short indels. Mutations were tions of paired blood using standard techniques (Qiagen), then quantified only considered at covered positions (≥8× for normal samples, 14× for tumor using PicoGreen dye (Invitrogen). The identities of tumor-normal pairs were sites). Somatic mutations and short indels were called and filtered using confirmed by mass spectrometry genotyping using a well-established 48-SNP MuTect and IndelLocator10,11. These variations were annotated to genes panel (Sequenom)43. and compared to events in the Catalogue of Somatic Mutations in Cancer (COSMIC) using Oncotator, and spurious calls caused by errors in mapping Sequencing. For whole-genome sequencing (n = 11), DNA was randomly and other previously identified systematic errors were removed using an estab- fragmented, and libraries of two insert sizes were prepared (500 and 800 bp) lished list of known problematic sites11–13. MutSig was used to determine the for paired-end sequencing on an Illumina HiSeq 2000 (refs. 10,11). Target significance of mutated genes, adjusting for cohort- and genome-wide muta- depth (60× for tumors, 30× for normal samples) was achieved in all but one tion rates, local background mutation rate and gene length11–13. We used a sample (MEN0015) that failed 800-bp library preparation; however, adequate cohort-wide background mutation rate instead of the more commonly used coverage of the 500-bp library was achieved. sample-specific mutation rate because the observed low mutation rate in these For whole-exome sequencing (n = 6), 250-bp libraries were prepared by tumors leads to poor sampling of individual mutation subcategories. This Covaris sonication, double size selected (Agencourt AMPure XP beads) and may lead to overly large apparent variations between the background rates of ligated to specific barcoded adaptors (Illumina TruSeq) for multiplexed analy- mutation subtypes in different samples. Mutations are shown in the context of sis. Exome hybrid capture was performed with NimbleGen SeqCap EZ Exome UniProt regions and domains. Post-hoc analyses for significant non-overlap of Library SR v2.2, Agilent SureSelect All Exon v2.0 hybrid capture kit or with NF2, SMO and AKT1 mutations was performed by Fisher’s exact test, as were both approaches. All samples but one achieved >150× coverage in exon regions associations between these mutations and chromosome 22 loss or specific (MENex004; 79× for the tumor, 90× for the normal sample). categorical demographic characteristics. The Mann-Whitney test was used for

© 2013 Nature America, Inc. All rights reserved. America, Inc. © 2013 Nature We proceeded with sequencing 645 cancer-associated genes in our larger comparisons of mutations and continuous demographic variables. validation and extrapolation sample set (n = 48) because this preselected gene Germline variants were called using the UnifiedGenotyper45,46, filtered set (OncoPanel) included all genes of interest that were also clinically action- against the 1000 Genomes Project data set50 and annotated to genes as able (including AKT1 and SMO) in which mutations were identified in the described. Genes with known syndromic associations to meningiomas npg discovery set. In addition, a wide breadth of other genes in the same families (NF1, NF2, SMARCB1 and PTCH1) were manually reviewed for germ- (such as MTOR in the AKT pathway) and other interesting cancer-related line nonsynonymous mutations. These were visualized in the Integrative genes in other families were included (Supplementary Table 2). Briefly, DNA Genomics Viewer. was sonicated to achieve an average fragment size of 150 bp and was then size selected and barcoded as described. Each multiplexed pool was hybrid- Sequenom validation. We obtained IRB approval for the collection of inde- ized with biotinylated baits (Agilent SureSelect) designed to capture exonic pendent archival paraffin-embedded meningioma samples from the Brigham sequences. Streptavidin-coated beads were used to pull down the complexes, and Women’s Hospital Pathology Department. We identified 95 samples that which were loaded on a HiSeq 2000 for paired-end sequencing. did not overlap with our cohort of 65 tumors, resected between 2005 and 2012. Forty-nine of these were higher-grade meningiomas. Tissue was depar- Sequencing analysis pipeline. For exome- and OncoPanel-sequenced data, affinized and DNA extracted using standard techniques (Qiagen). Variants pooled sample reads were demultiplexed using Picard tools. Read pairs were in SMO (p.Trp535Leu and p.Leu412Phe) and AKT1 (p.Glu17KLys) were aligned to the hg19 reference sequence (Build 37) using the Burrows-Wheeler assayed using Sequenom’s homogeneous Mass-Extend (hME) Genotyping Aligner44 with options −q 5 −l 32 −k 2 −o 1. Data were sorted and duplicates system. Bidirectional probes and primers were designed using Sequenom were marked using SAMtools and Picard. Biases in base quality score assign- MassARRAY Typer software (v 4.0), and hME genotyping was performed ments due to flow cell, lane, dinucleotide context and machine cycle were using 5 ng of unamplified genomic DNA template per assay pool (maximum analyzed and recalibrated, and local realignment around indels was achieved of four assays per pool). using the Genome Analysis Toolkit (GATK)45,46. Variant and germline calling were performed within the Firehose environ- Immunohistochemistry. Tissue sections were deparaffinized and rehy- ment11 using GenePattern47. All algorithms described are available publicly drated using xylene and graded alcohols. Endogenous peroxidase activity (see URLs). All sample pairs passed a quality control pipeline to test for any was blocked with 3% H2O2 in 100% ethanol (1:1; 15 min). Antigen retrieval tumor-normal and interindividual mix-ups by comparing insert-size distribu- was performed with Dako Target Retrieval Solution, pH 6.0, and pressure tions and copy-number profiles, as described previously10. The Mann-Whitney cooker treatment (120 ± 2 °C, 15 ± 5 psi) for 45 s. Sections were incubated test was used to determine the significance of differences between rates of with primary antibodies against GAB1 (Abcam, ab27439; 1:100 dilution for somatic genetic alteration between meningiomas and the other data sets. 45 min), Ki67 (Vector Lab, VP-451; 1:2,500 dilution for 45 min) or Stathmin

Nature Genetics doi:10.1038/ng.2526 (Cell Signaling Technology, 3352; 1:120 dilution at 4 °C for 16 h), followed by 43. Demichelis, F. et al. SNP panel identification assay (SPIA): a genetic-based assay incubation with secondary antibody (Dako Labeled Polymer-HRP anti-rabbit for the identification of cell lines. Nucleic Acids Res. 36, 2446–2456 (2008). 44. Li, H. & Durbin, R. Fast and accurate short read alignment with Burrows-Wheeler ′ IgG; 30 min). Visualization was performed using 3,3 -diaminobenzidine transform. Bioinformatics 25, 1754–1760 (2009). as a chromogen (Envision+System) and Gill hematoxylin counterstaining. 45. DePristo, M.A. et al. A framework for variation discovery and genotyping using Associations between specific mutations and their differences in histopatho- next-generation DNA sequencing data. Nat. Genet. 43, 491–498 (2011). logical subtypes and immunohistochemical staining were determined post 46. McKenna, A. et al. The Genome Analysis Toolkit: a MapReduce framework for analyzing next-generation DNA sequencing data. Genome Res. 20, 1297–1303 (2010). hoc with Fisher’s exact test. 47. Reich, M. et al. GenePattern 2.0. Nat. Genet. 38, 500–501 (2006). 48. Chiang, D.Y. et al. High-resolution mapping of copy-number alterations with massively parallel sequencing. Nat. Methods 6, 99–103 (2009). 41. Parsons, D.W. et al. An integrated genomic analysis of human glioblastoma 49. Krzywinski, M. et al. Circos: an information aesthetic for comparative genomics. multiforme. Science 321, 1807–1812 (2008). Genome Res. 19, 1639–1645 (2009). 42. Samuels, Y. et al. High frequency of mutations of the PIK3CA gene in human 50. 1000 Genomes Project Consortium. A map of variation from cancers. Science 304, 554 (2004). population-scale sequencing. Nature 467, 1061–1073 (2010). © 2013 Nature America, Inc. All rights reserved. America, Inc. © 2013 Nature npg

doi:10.1038/ng.2526 Nature Genetics