Supplemental figures Supplemental figure 1: Tumor sample selection. A total of 98 thymic tumor specimens were stored in Memorial Sloan-Kettering Cancer Center tumor banks during the study period. 64 cases corresponded to previously untreated tumors, which were resected upfront after diagnosis. Adjuvant treatment was delivered in 7 patients (radiotherapy in 4 cases, cyclophosphamide- doxorubicin-vincristine (CAV) chemotherapy in 3 cases). 34 tumors were resected after induction treatment, consisting of chemotherapy in 16 patients (cyclophosphamide-doxorubicin- cisplatin (CAP) in 11 cases, cisplatin-etoposide (PE) in 3 cases, cisplatin-etoposide-ifosfamide (VIP) in 1 case, and cisplatin-docetaxel in 1 case), in radiotherapy (45 Gy) in 1 patient, and in sequential chemoradiation (CAP followed by a 45 Gy-radiotherapy) in 1 patient. Among these 34 patients, 6 received adjuvant radiotherapy. 1 Supplemental Figure 2: Amino acid alignments of KIT H697 in the human protein and related orthologs, using (A) the Homologene database (exons 14 and 15), and (B) the UCSC Genome Browser database (exon 14). Residue H697 is highlighted with red boxes. Both alignments indicate that residue H697 is highly conserved. 2 Supplemental Figure 3: Direct comparison of the genomic profiles of thymic squamous cell carcinomas (n=7) and lung primary squamous cell carcinomas (n=6). (A) Unsupervised clustering analysis. Gains are indicated in red, and losses in green, by genomic position along the 22 chromosomes. (B) Genomic profiles and recurrent copy number alterations in thymic carcinomas and lung squamous cell carcinomas. Gains are indicated in red, and losses in blue. 3 Supplemental Methods Mutational profiling The exonic regions of interest (NCBI Human Genome Build 36.1) were broken into amplicons of 500 bp or less, and specific primers were designed using Primer 3 (on the World Wide Web for general users and for biologist programmers (see Supplemental Table 2) [1]. M13 tails were added to facilitate Sanger sequencing. PCR reactions were carried out in 384 well plates, in a Duncan DT-24 water bath thermal cycler, with 10 ng of DNA as template, using a touchdown PCR protocol with HotStart Taq (Kapa Biosystems, Cape Town, South Africa). The touchdown PCR method consisted of : 1 cycle of 95°C for 5 min; 3 cycles of 95°C for 30 sec, 64°C for 30 sec, 72°C for 60 sec; 3 cycles of 95°C for 30 sec, 62°C for 30 sec, 72°C for 60 sec; 3 cycles of 95°C for 30 sec, 60°C for 30 sec, 72°C for 60 sec; 37 cycles of 95°C for 30 sec, 58°C for 30 sec, 72°C for 60 sec; 1 cycle of 70°C for 5 min. Templates were purified using AMPure (Agencourt Biosciences, Beverly, MA). The purified PCR reactions were split into two, and sequenced bidirectionally with M13 forward and reverse primers and Big Dye Terminator Kit v.3.1 (Applied Biosystems, Foster City, CA), at Agencourt Biosciences. Dye terminators were removed using the CleanSEQ kit (Agencourt Biosciences), and sequence reactions were run on ABI PRISM 3730xl sequencing apparatus (Applied Biosystems, Foster City, CA). 1 Mutation detection Raw mass-spectrometry data were processed using the SpectroTYPER software (Sequenom, Sequenom, San Diego, CA). DNA sequencing chromatograms were analyzed using an institutional mutation detection pipeline. Briefly, bi-directional reads and mapping tables (to link read names to sample identifiers, gene names, read direction, and amplicon) were subjected to a QC filter which excludes reads that have an average phred score of < 10 for bases 100-200. Passing reads were assembled against the reference sequences for each gene, containing all coding and UTR exons including 5Kb upstream and downstream of the gene, using command line Consed 16.0 (PMID: 9521923) [2]. Assemblies were passed on to Polyphred 6.02b (PMID: 9207020) which generated a list of putative candidate mutations, and to Polyscan 3.0 (PMID: 17416743) which generated a second list of putative mutations. The lists were merged together into a combined report, and the putative mutation calls were normalized to ‘+’ genomic coordinates and annotated using the Genomic Mutation Consequence Calculator (PMID: 17599934). The resulting list of annotated putative mutations was loaded into a Postgres database along with select assembly details for each mutation call (assembly position, coverage, and methods supporting mutation call). To reduce the number of false positives generated by the mutation detection software packages, only point mutations which are supported by at least one bi-directional read pair and at least one sample mutation called by Polyphred were considered, and only the putative mutations which are annotated as having non-synonymous coding effects, occur within 11 bp of an exon boundary, or have a conservation score > 0.699 (http://genome.ucsc.edu/cgi- bin/hgTrackUi?hgsid=108554407&g=multiz17way) were included in the final candidate 2 list. Indels called by any method were manually reviewed and included in the candidate list if found to hit an exon. All data were also manually reviewed to identify putative mutations. Synonymous variants and those with entries in the NCBI SNP database were excluded. All putative mutations were confirmed by a second PCR and sequencing reaction, in parallel with amplification and sequencing of matched normal tissue DNA. Genomic profiling data interpretation Significant transitions in copy number were identified from the calculated fluorescence ratios of the scanned images, using the circular binary segmentation algorithm [3]. Gains and losses for a subset of analyses were defined using sample- specific thresholds. Specifically, median absolute deviation (MAD) of the difference between the observed values and the segmentation means was used to estimate the within-sample variation. The sample-specific threshold to call gains and losses is then defined as 1.5 MAD above or below the sample median of the segmented values. Regions of high level amplification or deletion (amplitude higher than 3 MAD) were declared to be recurrent if present in at least two samples. For comparison with profiles from pulmonary primary squamous cell carcinomas, we used genomic profiles generated from a different Memorial Sloan-Kettering Cancer Center Institutional Review Board- approved study. Expression profiling data interpretation The robust multichip average (RMA) method was used to estimate expression of probe sets [4]. For unsupervised clustering, we applied a hierarchical clustering algorithm 3 on the most variable 1429 genes, with the Pearson correlation coefficient as the measure of similarity and average linkage as the method to join clusters. Immunohistochemistry A tissue microarray (TMA) was constructed using formalin-fixed, paraffin-embedded previously untreated tumor tissue specimens, which were reviewed by a single pathologist (M.F.Z.). Triplicate 0.6 mm biopsy cores were taken from selected areas containing a high ratio of tumor epithelial cells on parallel hematoxylin and eosin (H&E)- stained sections and arrayed onto master paraffin blocks using a manual tissue arrayer (Beecher Instruments, Sun Prairie, WI). For previously treated tumors, immunohistochemistry was done on individual paraffin-embedded tumor blocks. Five- micrometer sections from the tissue microarray or the tumor blocks were deparaffinized in xylene and dehydrated in graded alcohols. Ba/F3-KIT mutant transformant assays The KITV560del and KITH697Y mutations were generated using the QuikChange II XL site-directed Mutagenesis Kit (Stratagene, Agilent Technologies, Santa Clara, CA) and pMSCV-wtKIT-IRES-GFP, a retroviral expression vector (kindly provided by Dr. Gary Gilliland, Harvard Medical School) containing a wild-type human KIT cDNA (GNNK- isoform). Ba/F3 cells were co-transfected with the KITV560del or the KITH697Y cDNA (25 μg) and a hygromycin-resistance gene (25:1 ratio) by electroporation as previously described [5]. 4 Ba/F3 cells were maintained with 10 ng/ml recombinant murine IL-3 and 200 μg hygromycin for 10 days, and then selected for IL-3-independent growth. The IL3- independent BaF3 KITV560del and KITH697Y cells reached optimal growth with 20 ng/ml human stem cell factor (hSCF). KIT cell surface expression was confirmed by flow cytometry using PE-conjugated anti-CD117 (BD Biosciences, San Jose, CA) antibody. Total KIT expression was confirmed by immunoblots using an anti-KIT (Oncogene, San Diego, CA) antibody (data not shown). IL-3-independent clones were verified by direct sequencing to harbor mutant cDNAs. For drug sensitivity assays, Ba/F3 KITV560del and KITH697Y cells were starved from hSCF for 4 hours. 15 minutes prior to drug administration, 20ng/ml of hSCF was added. References 1. Krawetz S, Misener S (eds): Bioinformatics Methods and Protocols: Methods in Molecular Biology. Humana Press, Totowa, NJ, 2000;pp 365-6. 2. Gordon D, Abajian C, Green P. Consed: A Graphical Tool for Sequence Finishing. Genome Res 1998;8:195-202. 3. Olshen AB, Venkatraman ES, Lucito R, Wigler M. Circular binary segmentation for the analysis of array-based DNA copy number data. Biostatistics 2004;5:557-72. 4. Irizarry RA, Hobbs B, Collin F, et al. Exploration, normalization, and summaries of high density oligonucleotide array probe level data. Biostatistics 2003;4:249-64. 5 5. Guo T, Agaram NP, Wong GC, et al. Sorafenib inhibits the imatinib-resistant KITT670I gatekeeper mutation in gastrointestinal stromal tumor. Clin Cancer Res 2007;13:4874-81. 6 Supplemental Tables Supplemental Table 1: List of mutations screened using mass spectrometry-based