Mutational heterogeneity in non-serous ovarian cancers.

Tumors were chosen by interrogating Moffitt tissue bank for cases that fulfilled the following criteria: a) non-serous histology; b) Stage I; and c) available fresh-frozen tissue; d) available matched normal blood or tissue. Whole exome sequencing was performed on 9 tumor/normal matched pairs, and one tumor only referred to as the ES (exome sequencing) cohort (Table  1 ). No case with clear cell tumor fulfilled the selection criteria. A median of 184 million reads and 17.5GB of exome sequencing data per sample was obtained. We annotated 32,965 somatic mutations within 100 bases of the exome target region (median = 1,299; range 829–15,767 per tumor, Supplementary Table  3 ) for an overall rate of mutations of 12.1 per Mb. The overview of the study is depicted in a REMARK (Reporting recommendations for tumor marker prognostic studies)-style diagram (Supplementary Figure  1 ). Table 1 Samples used in exome sequencing. Case Histology Grade Stage Age at diagnosis Tissue of normal DNA Main findings OV1 Endometrioid Adenocarcinoma FIGO 2 1C 48 Uterus Glandular and sparse solid patterns. Focal squamous differentiation. OV2 Endometrioid Adenocarcinoma FIGO 2 1C 63 Blood Glandular pattern. OV3 Endometrioid Adenocarcinoma FIGO 1 1A 28 Blood Glandular and villoglandular patterns. Adenofibroma component. OV4 Endometrioid Adenocarcinoma FIGO 2 1C 49 Blood Glandular and sparse solid patterns. OV5 Endometrioid Adenocarcinoma FIGO 1 1C 53 Blood Glandular pattern. Focal squamous differentiation. OV6 Endometrioid Adenocarcinoma FIGO 1 1A 54 Blood Glandular pattern. OV7 Granulosa Cell tumor Ungraded 1A 57 N/A Adult type, trabecular pattern. OV8 Mucinous Adenocarcinoma NS 1A 54 Blood Endocervical type. BRCA2 germline pathogenic variant. OV9 Mucinous Adenocarcinoma NS 1A 41 Blood Endocervical type. OV10 Mucinous Adenocarcinoma NS 1C 44 Blood Endocervical type. LMP neoplasm component. N/A, not available; NS, grading not standardized; LMP, low malignant potential. Samples used in exome sequencing. N/A, not available; NS, grading not standardized; LMP, low malignant potential. MutSigCV_1.4 was used to identify significantly mutated genes, but no genes were observed with significant q-value. Only three genes were observed with q-value <1: GOLGA6L1 (q = 0.62), PTEN (q = 0.73), and OR2T33 (q = 0.76). No single confident mutated position (<1% in 1000 Genomes and 238 TCC normal samples, passing manual review) was observed in more than three samples, and only PTEN stood out as a gene containing truncating mutations in three samples (Fig.  1A ). Mutations were mapped and compared to the COSMIC database ( http://cancer.sanger.ac.uk/cosmic ). Several COSMIC cancer genes were found to have various mutations in multiple samples, including PTEN , PIK3CA , BRCA2 , and ATM (Fig.  1A , Supplementary Table  4 ). Several known cancer mutations were seen in unique samples ( KRAS p.G12D, BRAF p.D594G, NF1 p.R1362*, and ZFAND1 p.R130*). Only one sample (OV8) had a TP53 mutation (p.G245D, seen 22x in TCGA breast, ovarian, and colorectal cancers). Figure 1 Mutation profile in non-serous ovarian cancer. ( A ) Mutation profiles showing mutations identified using tumor-normal pair analysis. Each column corresponds to exome sequencing data for one ovarian tumor. Rows correspond to presence or absence of mutation in the sample. ( B ) Mutation profiles showing mutations identified using tumor-only analysis. Each column corresponds to exome sequencing of one ovarian tumor. Rows correspond to presence or absence of mutation in the sample. ( C ) Mutation profiles showing mutations identified in target gene sequencing. Each column corresponds to one ovarian tumor. Rows correspond to presence or absence of mutation in the sample. Blue boxes highlight the high prevalence of specific mutations in certain histological subtypes. Sample histologies are shown in different colors representing endometrioid (blue), mucinous (yellow), clear cell (white), and granulosa cell (grey) histologies. Missense mutations and truncations are shown as red and black boxes, respectively. Star indicates samples with identified POLE proofreading domain mutations. ( D ) Frequency of the most common mutations in different tumor types and histologies. The radar graph display the frequency of various mutations across datasets, clockwise from top: ovarian serous cystadenocarcima (TCGA), serous endometrial uterine cancer (TCGA), non-serous ovarian tumors (exome sequencing, this study), non-serous ovarian tumors (target gene sequencing, this study), and non-hypermutable endometrial uterine cancers (TCGA). Height of the peak from the middle of the radar indicates the frequency of individuals with a mutation in the given gene. Mutation profile in non-serous ovarian cancer. ( A ) Mutation profiles showing mutations identified using tumor-normal pair analysis. Each column corresponds to exome sequencing data for one ovarian tumor. Rows correspond to presence or absence of mutation in the sample. ( B ) Mutation profiles showing mutations identified using tumor-only analysis. Each column corresponds to exome sequencing of one ovarian tumor. Rows correspond to presence or absence of mutation in the sample. ( C ) Mutation profiles showing mutations identified in target gene sequencing. Each column corresponds to one ovarian tumor. Rows correspond to presence or absence of mutation in the sample. Blue boxes highlight the high prevalence of specific mutations in certain histological subtypes. Sample histologies are shown in different colors representing endometrioid (blue), mucinous (yellow), clear cell (white), and granulosa cell (grey) histologies. Missense mutations and truncations are shown as red and black boxes, respectively. Star indicates samples with identified POLE proofreading domain mutations. ( D ) Frequency of the most common mutations in different tumor types and histologies. The radar graph display the frequency of various mutations across datasets, clockwise from top: ovarian serous cystadenocarcima (TCGA), serous endometrial uterine cancer (TCGA), non-serous ovarian tumors (exome sequencing, this study), non-serous ovarian tumors (target gene sequencing, this study), and non-hypermutable endometrial uterine cancers (TCGA). Height of the peak from the middle of the radar indicates the frequency of individuals with a mutation in the given gene. To ensure no mutations were missed due to local low coverage in the normal sample, we also identified genotypes using GATK on tumor samples alone. Mutations previously observed in COSMIC were retained. No further TP53 mutation was observed. Additional cancer mutations were detected in OV7 ( FOXL2 p.C134W) and OV9 ( KRAS p.G12V) (Fig.  1B ). We observed a somatic p.Q257H variant in BRAF but the functional consequence of this alteration is unclear. Although NRAS was well covered (97.4% bases covered ≥10× on average) in our dataset, no mutations were observed, suggesting it may not be a common driver of mucinous tumors. Interestingly, for sample OV10, the KRAS variant was only called in the tumor/normal analysis (Fig.  1A ) but not in tumor only analysis (Fig.  1B ). The variant base was present in 4/23 reads in the tumor and 0/11 reads in the normal. We confirmed this variant is present in heterozygosis by Sanger sequencing in OV10. GATK-based analysis of OV10 also detected a large insertion in PIK3CA , a missense variant in ARID1A (p.Ala532Thr) observed in only 2/25 tumor and 0/100 normal reads, and an NF1 large insertion that introduces a stopgain. Sanger sequencing did not confirm those changes suggesting that they may have been incorrectly called or that only a small percentage of cells in the tumor carry that variant. We also assessed germline variants in the exome sequencing data from the normal samples available from 9 cases (Supplementary Tables  9 and 10 ; except OV7) focusing on a set of genes implicated in ovarian cancer 34 . Only one clearly pathogenic variant, BRCA2 :c.6944_6947delTAAA (p.Ile2315Lysfs) (OV8) variant and few variants of uncertain clinical significance were identified (Supplementary Table  10 ). We confirmed the BRCA2 mutation using Sanger sequencing. A list of all non-synonymous variants called using a combination of Mutec and Strelka, including variants within the SeqCap EZ Exome v3 regions + 100 bp flanking sequence, is shown in Supplementary Table  11 . We excluded variants observed in 1000 Genomes Project (Phase 3) or in an internal dataset of 238 adjacent normal samples at ≥5% frequency. We expanded our investigation to include an additional 23 non-serous EOC tumors and five granulosa cell tumors with sequencing data covering 1,321 genes selected based on their suspected involvement in cancer, independently of the data obtained during exome sequencing. They were selected through a review of the literature, mutation databases and key pathways in tumorigenesis such as growth factor signaling, DNA damage response, p53 signaling, cell cycle control and apoptosis 35 (Table  2 , Supplementary Table  2 ). Approximately 1.4 GB of targeted gene sequencing data per sample was obtained by sequencing 1,321 genes covering 3.8 Mb (Supplementary Table  5 ). Mutations were cross referenced with COSMIC to identify driver mutations (Fig.  1C , Supplementary Table  6 ). Table 2 .Samples used in target gene sequencing. Case Histology Age at diagnosis TG1 Endometrioid carcinoma, squamous differentiation NA TG2 Endometrioid adenocarcinoma 51 TG3 Endometrioid adenocarcinoma 54 TG4 Endometrioid adenocarcinoma 71 TG5 Endometrioid adenocarcinoma 60 TG6 Endometrioid adenocarcinoma 56 TG7 Endometrioid adenocarcinoma 61 TG8 Endometrioid adenocarcinoma 58 TG9 Endometrioid adenocarcinoma 54 TG10 Endometrioid adenocarcinoma 50 TG11 Endometrioid adenocarcinoma 61 TG12 Endometrioid adenocarcinoma 52 TG13 Endometrioid adenocarcinoma 51 TG14 Endometrioid adenocarcinoma 79 TG15 Granulosa Cell NA TG16 Granulosa Cell NA TG17 Granulosa Cell 61 TG18 Granulosa Cell 73 TG19 Granulosa Cell 66 TG20 Mucinous cystadenocarcinoma 78 TG21 Mucinous cystadenocarcinoma 30 TG22 Mucinous, carcinoma 55 TG23 Mucinous, carcinoma 51 TG24 Clear cell, carcinoma 62 TG25 Clear cell, carcinoma 64 TG26 Clear cell, carcinoma 53 TG27 Clear cell adenocarcinoma 63 TG28 Clear cell, carcinoma 63 .Samples used in target gene sequencing. We found 11 unique TP53 mutations, seen in 10/28 (36%) tumors. TP53 mutations were prevalent in endometrioid (6/14) and mucinous (3/4) subtypes but less frequent in clear cell (1/5) and absent in granulosa cell (0/5). We also observed the presence of FOXL2 p.C134W mutation in 4/5 (80%) of the additional granulosa cell samples. Although other FOXL2 mutations were observed in different samples, the p.C134W driver mutation was only observed in the granulosa cell subtype. KRAS mutations were observed in 3/4 (75%) of additional mucinous samples and 3/14 (21%) of additional endometrioid samples. ARID1A and CTNNB1 were mutated more frequently in endometrioid tumors (6/14, 43% and 7/14, 50% respectively), further suggesting these are important contributors to that tumor type 31 . ARID1A truncating mutations were also recurrently observed in 2/5 (40%) clear cell tumors. We compared mutation patterns observed in our set of 9 whole exome ovarian tumors and our set of 23 targeted gene ovarian tumors to TCGA studies on serous ovarian 8 and endometrial cancers 36 at genes known to be mutated in these diseases (Fig.  1D ). Serous ovarian cancer is marked by very high mutation rate of TP53 (96%) and of few other cancer genes. Serous endometrial also has a high TP53 mutation rate of 74%, as well as PIK3CA (36%), FBXW7 (26%), and PPP2R1A (23%). TP53 mutation rates were much lower in our non-serous OC samples: 10% in ES, and 36% in our TGS cohort. PIK3CA mutation rates were, on average, higher in our non-serous OC samples compared to serous cancers. We observed higher mutation rates in each of the following genes in non-serous ovarian samples compared to serous ovarian and serous endometrial: PTEN , PIK3R1 , CTNNB1 , ARID1A , and KRAS . Many of these genes are frequently mutated in non-ultramutated endometrioid uterine cancer (Fig.  2 ), particularly PTEN , PIK3CA , PIK3R1 , CTNNB1 , ARID1A , and KRAS 36 . Figure 2 Mutation signatures of non-serous ovarian tumors. ( A ) Incidence of most probable mutation signatures in each sample by raw mutation count (left) or mutation fraction (right). The inferred mutation signature is listed below sample names. Parentheses indicate lower confidence in the assigned signature due to lower mutation counts. ( B ) Mutation distribution similar to Signature 10 ( POLE proofreading domain dysfunction) from 38 . ( C ) Mutation distribution similar to Signature 18 from 38 . Mutation signatures of non-serous ovarian tumors. ( A ) Incidence of most probable mutation signatures in each sample by raw mutation count (left) or mutation fraction (right). The inferred mutation signature is listed below sample names. Parentheses indicate lower confidence in the assigned signature due to lower mutation counts. ( B ) Mutation distribution similar to Signature 10 ( POLE proofreading domain dysfunction) from 38 . ( C ) Mutation distribution similar to Signature 18 from 38 . One endometrioid sample (OV4) was observed to have very high mutation count (Fig.  1 ) (Supplementary Table  3 ). Further investigation revealed that this sample contained a POLE p.V411L mutation in the proofreading domain that has been previously observed in endometrioid uterine carcinoma 36 and colorectal cancer 37 . Two additional POLE mutations were observed in endometrioid ovarian samples in our expanded TGS cohort: p.P286R (TG3) and another p.V411L (TG10). Examining the pattern of nucleotide changes, i.e . the mutational signatures, in a tumor can suggest mechanisms that may have contributed to tumorigenesis. All three endometrioid ovarian tumors containing POLE proofreading domain mutations consisted almost entirely of the previously reported POLE mutation signature (TCT > TAT, TCG > TTG, TTT > TGT; Signature 10) ( http://cancer.sanger.ac.uk/cosmic/signatures ) 38 (Fig.  2A,B ). This signature was visible in the tumor-only analysis despite the lower somatic mutation specificity we have previously observed (Teer et al . unpublished observation). In ES samples, we observed signature 1b “Age” (likely age-related cytosine-deamination) in all samples and signature 18, which has unknown etiology, as the predominant signature in two samples of different histologies (OV10: mucinous and OV6: endometrioid) (Fig.  3A,C ). Figure 3 Mutational landscape across non-serous ovarian tumors. Left: ( A ) Circos plot display detected somatic alterations in Exome Sequencing (ES) tumors. Right: From outmost to inner most ring, chromosome ideogram, number of samples with mutations (light grey, increasing towards outside), number of samples with amplification (light red, increasing towards outside), numbers of samples with deletion (light blue, increasing towards inside). Dark blue slices highlight recurrent features: counterclockwise from top-left: PIK3CA mutations, amplification in MYC region, lack of mutations or deletions in TP53 region. Mutational landscape across non-serous ovarian tumors. Left: ( A ) Circos plot display detected somatic alterations in Exome Sequencing (ES) tumors. Right: From outmost to inner most ring, chromosome ideogram, number of samples with mutations (light grey, increasing towards outside), number of samples with amplification (light red, increasing towards outside), numbers of samples with deletion (light blue, increasing towards inside). Dark blue slices highlight recurrent features: counterclockwise from top-left: PIK3CA mutations, amplification in MYC region, lack of mutations or deletions in TP53 region. Tumor specific copy number alterations (CNA) were examined in the ES dataset. A median of 25 events was observed in each sample (range 5–192) covering a median of 76,303,334 bases (range 1,818,739–661,026,741) (Supplementary Table  7 ). Although many events were observed in OV7, we suspected these were unreliable due to the very low coverage in the sample from normal plasma, and did not consider this sample further. As many of the alterations were large (median = 236,356 bases, range 1,291–47,327,598 bases), genes that have been previously described as cancer drivers 39 were identified in the CNA regions. Between 0 and 13 driver genes fell in copy number altered regions in each sample (Supplementary Table  8 ). Driver genes in amplified segments included FGFR3 , MYC , KRAS , EXT1 , and PTEN . Genes in deleted segments included STK11 , NF2 , CHEK2 , PMS2 , CDKN2A , and FANCA (Fig.  3 ). No sample contained a copy number alteration overlapping TP53 , BRCA1 or BRCA2 (Fig.  3 , Supplementary Table  8 ). Recent TCGA genomic classifications of high-grade serous ovarian adenocarcinoma and endometrial carcinomas revealed high levels of somatic CNAs in high grade serous ovarian and serous endometrial tumors, including recurrent amplifications in MYC , and CCNE1 . We observed fewer CNAs in non-serous ovarian tumors, including only 2 samples with a MYC amplification and no samples with a CCNE1 amplification. Interestingly, sample OV3, an endometrioid tumor from the youngest individual in our cohort (diagnosed at 28 years of age) had the most CNAs. This tumor had the third lowest number of mutations with no identified common cancer mutations (Supplementary Table  4 ; Fig.  1 ). Altered driver genes in OV3 were mainly amplified, and included NOTCH2 , ALK , PIK3R1 , and MET . Two tumor suppressor genes were also amplified: PTEN and CDKN2A . Only one deletion was observed, which contained PMS2 .

Here, we performed exome sequencing of ten incident cases of non-serous stage 1 ovarian neoplasms to determine somatic mutations and copy number alterations. We also performed target gene sequencing for 1,321 genes in 28 non-serous tumors to explore the mutation landscape of non-serous ovarian tumors. A limitation of our ES cohort study is a relatively small sample size and reported frequencies should be taken with caution. Despite this limitation, the availability of normal matched DNA allowed an improved identification of relevant variants. We also used an additional independent cohort. The data integration and comparison with publically available data led to five main conclusions. First, non-serous ovarian tumors seem to be distinct from serous subtypes in their prevalence of TP53 pathogenic variants. While previous studies have reported mutations in 96% of high grade serous epithelial tumors 8 , only one ES sample (OV8; mucinous) had a TP53 mutation and no samples had deletions in the region. We examined the fraction of bases with depth of coverage ≥ 10 across the TP53 gene, and found an average of 73.7% per sample, suggesting our coverage was sufficient to have observed mutations. In the TGS cohort, mutations were prevalent in mucinous (4/7, 57%), endometrioid (6/20, 30%) and clear cell (1/5, 20%) subtypes but still markedly lower than high grade serous. This is line with previous studies of clear cell carcinomas reporting 10–15% TP53 mutation frequency 9 , 40 . Interestingly, comparative studies of TP53 mutations in uterine cancer also reports widely different mutation frequencies between the endometrioid (11.4–17%) and serous carcinomas (>90%) 36 , 41 . Second, a small but significant fraction of endometrioid tumors display mutations in the proofreading domain of POLE , and mutations are not found in other non-serous subtypes. One ES sample (OV4; endometrioid) harbored a POLE mutation in the proofreading domain, and showed a mutational signature matching a previously reported signature of POLE deficiency 38 . In the TGS set we observed recurrent POLE mutations in 14% (2/14) of endometrioid ovarian cancers. A screen for POLE mutations in a study of 251 Chinese samples of different subtypes of ovarian carcinomas, identified mutations in 3 out of 37 (8.1%) patients with ovarian endometrioid carcinoma, but not observed in the other subtypes 42 . Interestingly, POLE hotspot mutations were found in 7% (17/248) of uterine carcinomas, all of endometrioid histology 36 . Third, we confirmed the high prevalence (5/6; 83%) of pathognomonic mutations in FOXL2 in adult granulosa cell tumors. This is consistent with existing reports showing a prevalence ranging from 70.4% to 100% 43 – 50 . Fourth, different non-serous subtypes have distinct mutation profiles. KRAS mutations are observed in significant fraction of mucinous and, to a lesser extent, in endometrioid but much less frequently (0.6%; 2/316) in high grade serous ovarian tumors 8 . A total of 71% (5/7) of mucinous samples carried mutations in KRAS , confirming its high frequency in this subtype 51 – 53 . We also observed 20% (4/20) of endometrioid tumors with mutations in KRAS . Mutations in PIK3CA and ARID1A are the most frequent alteration in clear cell carcinomas found in combination in 2/5 (40%) of all CCC tumors in the present study. Sequencing of 97 OCCC tumors identified a high frequency of mutations in PIK3CA (33%) and in TP53 (15%) 9 . Finally, the finding that deletion in genes implicated in the DNA damage response are relatively common suggest that therapeutic approaches targeting these pathways, such as the use of PARP inhibitors may be provide effective options in treatment. In summary, our results show that non-serous ovarian tumors are mutationally distinct from serous ovarian tumors. Interestingly, a similar but not identical mutational profile can be found in endometrioid ovarian and endometrioid uterine tumors. These similarities support the link between endometrioid ovarian cancers and endometriosis 31 . Our findings also offer molecular evidence to support continued distinction of non-serous ovarian cancers by histology, and may offer the opportunity for future work to investigate subtype-specific interventions targeting unique molecular alterations.

Epithelial Ovarian Cancer (EOC) is a heterogeneous disease with five major histologic types 1 , 2 : high-grade serous carcinoma, which accounts for the approximately 70% of all EOC cases, and clear cell (10%), endometrioid (10%), mucinous (3%), and low-grade serous carcinomas (<5%). These subtypes differ in terms of histopathology, morphology, and genomic alterations and are considered distinct diseases, which require careful diagnosis and specific therapies that are critical for successful treatment 3 . Ovarian granulosa cell tumors do not have epithelial origin but are the most common sex cord-stromal tumor and represent 5% of all ovarian malignancies 4 . Based on molecular profiles, disease development and prognosis, the different histological subtypes can be hierarchically grouped into type I and type II 5 . Type I tumors are slow growing and encompass low-grade serous, low-grade endometrioid, mucinous and clear cell carcinomas characterized by mutations in KRAS , BRAF , and PIK3CA and by the absence of TP53 mutations 6 . Type II tumors are highly aggressive and encompass high-grade serous tumors as a typical representative, as well as high-grade endometrioid and undifferentiated carcinomas with disruption of TP53 7 . Type II tumors have expression profiles that cluster separately from type I tumors and are characterized by genomic instability, extremely aggressive clinical progression and poor prognosis 8 – 11 . A comprehensive view of the germline variation associated with cancer risk 12 – 22 and somatic alterations in high-grade serous ovarian carcinoma has recently emerged (Supplementary Table  1 ) 8 , 23 – 25 . These tumors were characterized by somatic TP53 mutations in 96% of all samples and low prevalence mutations in nine additional genes such as NF1 , BRCA1 , BRCA2 , RB1 and CDK12 . Recurrent large copy number alterations were observed (8 gains and 22 losses) and included genes such as MYC and KRAS (gain) and PTEN , RB1 , and NF1 (loss). Many of these larger gains and losses were observed in the majority of tumors, highlighting the genomic instability of high-grade serous ovarian carcinoma. In addition, analyses of complete genomes for sixteen low stage high grade serous EOC revealed frequent TP53 mutations, tetraploidy and homologous recombination repair defects 26 . However, other non-serous forms of ovarian cancer contribute to significant morbidity and mortality yet few systematic studies have been conducted to characterize their germline and somatic mutational profile. Recently, genome-wide association studies identified three susceptibility loci and analysis of somatic alterations revealed a more detailed view of mucinous ovarian cancer 27 , 28 . Most published studies, however, report germline or somatic targeted sequencing of a limited number of genes 9 , 29 – 31 , or exome studies with a limited number of samples 32 (Supplementary Table  1 ). A comparison between endometrioid uterine and endometrioid ovarian carcinomas using select exon capture of mutation profiles revealed that despite a possible common origin in endometriosis these tumors display differences in their mutation profile 33 . In order to further explore the genomic landscape of non-serous ovarian cancer, we conducted exome sequencing of ten non-serous (endometrioid, mucinous, and granulosa) tumors and performed targeted gene sequencing of exons in 1,321 genes for 28 additional tumor samples including the clear-cell subtype. We report mutational and copy number profiles indicating that non-serous ovarian cancer is genomically distinct from serous EOC and displays similarities to non-serous endometrial uterine cancer.

For the Exome Sequencing (ES) cohort all subjects were women between the ages of 28 and 63 (median 51) from Moffitt Cancer Center. Tumor and matched normal samples were retrieved from Moffitt’s Total Cancer Care tissue bank. Tumor samples for sequencing were selected to have >80% tumor cellularity. Six endometrioid, three mucinous, and one granulosa cell ovarian tumors were sequenced (Table  1 ). The study was approved by the Institutional Review board of the University of South Florida and all methods were performed in accordance with the relevant guidelines and regulations. For the Target Gene Sequencing (TGS) cohort all samples were from the Total Cancer Care tissue bank. Patients were women between the ages of 30 and 79 (median 60) diagnosed in 1993–2012. Tumors underwent macrodissection to achieve >80% tumor cellularity in subsequent molecular studies. Samples were grouped by subtype: fourteen endometrioid, four mucinous, five clear cell, and five granulosa cell tumors were target sequenced (Table  2 ). All patients in both cohorts have provided informed consent under the Total Cancer Care protocol. We conducted a pathology review of all 10 cases used in exome sequencing. Details of the findings are presented in Table  1 . Pathology reviews performed on hematoxylin and eosin (H&E) stained sections from a formalin-fixed paraffin-embedded (FFPE) block made during tissue acquisition at Moffitt’s Tissue Bank. Samples for sequencing were derived from the same dissected tumor samples used for FFPE and pathology review. For samples OV2, OV6, OV8, and OV9, original H&E sections were not available and new cuts were made from the original corresponding FFPE blocks and stained. In these cases it is possible that the section may not represent the local tumor section used for sequencing. Representative images are shown in Supplementary Figure  3 . DNAs from snap frozen ovarian tumor and from normal uterus were isolated using Qiagen DNeasy Blood & Tissue Kit (Qiagen). DNA from blood was isolated using Qiagen Gentra Puregene Blood Kit or Qiagen PAXgene blood DNA kit. Paired-end libraries were constructed and whole exome capture was performed using Roche NimbleGen SeqCap EZ v3 (Roche). Captured libraries were sequenced using 100 base-pair reads on an Illumina HiScanSQ platform (Illumina, Inc., San Diego, CA) following the manufacturer’s protocols. Reads were aligned to the human genome reference hs37d5 using the Burrows-Wheeler Aligner (BWA) 54 . The Genome Analysis ToolKit (GATK) 55 was used for insertion/deletion realignment, and quality score recalibration. Somatic mutations were detected with either Strelka 56 or MuTect 57 . Strelka 1.0.13 was run with default settings except the following: ssnvNoise = 0.00000005, sindelNoise = 0.0000001. MuTect 1.1.4 was run with default settings except: max_alt_alleles_in_normal_count = 3, max_alt_allele_in_normal_fraction = 0.05. ANNOVAR 58 was used to annotate variants. Final read depth averaged 39.2×, and a median of ~87% of bases across samples had ≥10 reads. Coverage in each sample across genes of interest can be seen in Supplementary Figure  2 . Copy number alterations were detected using EXCAVATOR 59 . Circos was used for visualization of somatic alterations 60 . Blood was unavailable for case OV7, and although plasma was subjected to exome capture, sequencing coverage was low (0.2% target bases having ≥10 reads) and duplicate rates were high (87%) preventing sensitive tumor/normal mutation detection. Case OV7 was analyzed with GATK, and mutations were enriched by filtering out known variants. We also examined normal DNA to identify germline mutations implicated in ovarian cancer in breast/ovarian cancer families or in Lynch syndrome 30 , 61 – 63 : ATM, BARD1, BRCA1 , BRCA2 , BRIP1, CDH1, CHEK2, MLH1, MRE11A, MSH2, MSH6, MUTYH, NBN, PALB2, PMS1 , PMS2, PTEN, RAD50 , RAD51C , RAD51D, STK11 , and TP53 . Target gene capture was performed using SureSelect custom designs (Agilent Technologies, Inc., Santa Clara, CA) targeting 1,321 genes (Supplementary Table  2 ). Sequencing was performed using an Illumina GAIIx sequencing platform (Illumina, Inc., San Diego, CA) at the Beijing Genomics Institute (BGI, Shenzhen, China). Sequences were aligned to the human genome reference hs37d5 using the Burrows-Wheeler Aligner (BWA) 54 . The Genome Analysis ToolKit (GATK) 55 was used for insertion/deletion realignment, quality score recalibration, and variant identification. ANNOVAR 58 was used to annotate mutations. Matched normal samples were not available for comparison so somatic mutations were enriched via population filtering, including 1000 Genomes and 238 unmatched normal samples. Final read depth averaged ~140x across the targeted regions and a median of ~94% of bases across samples had ≥10 reads (Supplementary Figure  2 ). Mutation frequencies were downloaded from cBioportal 64 , 65 using the “Ovarian Serous Cystadenocarcinoma 8 and “Uterine Corpus Endometrial Carcinoma 36 . Mutational signatures were derived from the ES samples using previously described methods 38 . Curated and updated ICGC (International Cancer Genomics Consortium) signatures can be found at http://cancer.sanger.ac.uk/cosmic/signatures .

Supplementary Data Supplementary Tables 2–11 Supplementary Data Supplementary Tables 2–11