Genomic Classification of Serous Ovarian Cancer with Adjacent Borderline Differentiates RAS Pathway and TP53-Mutant Tumors and Identifies NRAS as an Oncogenic Driver

Recent molecular and pathologic studies have led to a significantly changed understanding of the subtypes of epithelial ovarian cancer (1). Serous epithelial ovarian carcinomas are the most common, and have been traditionally classified as high- or low-grade using morphologic criteria in either a two- or three-tier grading system (2, 3). Most high-grade serous carcinomas (HGSC) are thought to develop from secretory cells of the distal fallopian tube (4) and possibly ovarian inclusion cysts. TP53 mutations are almost ubiquitous in high-grade serous cancer and BRCA pathway disruption is common (5–8). In contrast, low-grade serous carcinomas (LGSC) typically have wild-type TP53 (9) and are thought to arise in a step-wise fashion from a subset of serous borderline tumors (SBT). Activating mutations of the MAPK pathway are present at a high rate in both SBTs and LGSCs (10–14).

Up to 60% of LGSCs are associated with SBT (2) and although more common in LGSC, a morphologic continuum between areas of SBT and invasive carcinoma has also been observed in HGSC cases. A review of 100 tumors reported areas of SBT in 2% of high-grade cases (2). Boyd and McCluggage described two cases (15). Dehari and colleagues (16) identified three cases of high-grade carcinoma associated with SBT (from 210 serous ovarian tumors) and in two cases mutation analysis showed the same KRAS mutations in both components, indicating a clonal relationship. SBT has been reported to recur as high-grade carcinoma, albeit in a small number of patients (17). Collectively, these studies indicate that HGSC with associated borderline histology are uncommon. It is not clear whether they represent progression from Ras-mutated borderline or low-grade serous tumors, or heterogeneity within the spectrum of HGSC.

LGSC tends to occur at a younger age and are usually poorly responsive to chemotherapy (18). Patients with advanced stage LGSC and residual disease after cytoreductive surgery have a high risk of recurrence and a high risk of cancer-related death, similar to women with HGSC (19, 20). The frequent presence of activating mutations in KRAS and BRAF suggests that treatment strategies that target the MAPK pathway may be of benefit to this subset of patients (21). The MEK inhibitor, selumetinib, has shown some activity in patients with LGSC (22) and dual blockade of MAPK and PI3K/AKT pathways may be an effective strategy (21, 23). However, the low incidence of LGSC represents a challenge to the conduct of definitive clinical trials, and there is an imperative to better define the subset of patients with ovarian cancer that may benefit from targeted treatment.

In this study, we sought to determine whether the presence of areas of adjacent borderline tumor was a reliable indicator of Ras-activated ovarian cancer, even in high-grade invasive cases. We present here the most comprehensive analysis to date of the clinical and molecular features of invasive ovarian cancer with coexisting regions of SBT. Our findings underscore the need to classify invasive ovarian tumors based on molecular features rather than morphology alone.

The study cohort was selected from review of 1,238 cases from the Australian Ovarian Cancer Study (AOCS) and a hospital-based series from Westmead Hospital, Sydney (Gynaecological Oncology Biobank at Westmead, GynBiobank). The AOCS recruited women with invasive or borderline ovarian, primary peritoneal, or fallopian tube cancer across Australia from 2002–2006 (http://www.aocstudy.org). Pathology review to confirm histologic subtype and to reassign as grade 1, 2 or 3 according to standardized criteria (3) was undertaken by a panel of specialist gynecological pathologists. At the time the cohort was selected for the current study, 572 cases had been reviewed, either by review of a single, representative slide, or review of a full set of diagnostic slides (136 cases). The review was undertaken using a synoptic report form and the presence of coexisting ovarian pathology, including borderline areas (categorized by a coarse papillary architecture with hierarchical branching, tufting, and mild nuclear atypia of the epithelial cells without invasion) was systematically recorded. Additional cases were selected from the GynBiobank (patients diagnosed between 1990 and 2010), based on diagnostic pathology reports and pathology review. Grade information available on these cases was also according to a grade 1–3 system in routine clinical use. Cases treated with neoadjuvant chemotherapy were excluded (n = 16).

The final study population consisted of three groups (i) 102 patients with serous epithelial ovarian cancer (EOC) with coexisting serous borderline tumor (designated SBT–EOC to denote their mixed features); (ii) 104 patients with serous carcinoma without coexisting SBT, i.e., a subset of AOCS cases which had centralized pathology review of all diagnostic slides in which no coexisting SBT was identified; (iii) 53 patients with SBT only, with no indication of invasive disease, randomly selected from SBT cases with frozen tissue specimens available enrolled in AOCS. See Supplementary Information and Supplementary Fig. S1 for additional details on cohort selection. The study was approved by the Western Sydney Local Health District Human Research Ethics Committee [Reference: HREC2001/9/4.15(1308)].

Cryopreserved tumor samples were sectioned, stained, and reviewed by a gynecological pathologist (R. Sharma) to confirm the presence of SBT and/or invasive cancer. Samples with <70% tumor were enriched by needle- or laser-microdissection. Where coexisting borderline and invasive disease was present on the same section, each area was isolated by microdissection. RNA and DNA were extracted using AbsolutelyRNA Microprep Kit (Stratagene, Agilent Technologies) and DNeasy Blood and Tissue Kit (Qiagen), respectively. RNA quality was assessed using Agilent BioAnalyzer (Agilent Technologies). For some cases, DNA was extracted from deparaffinized sections of formalin-fixed paraffin embedded (FFPE) specimens using a modified protocol of the QIAmp DNA FFPE Tissue Kit (Qiagen). See Supplementary Methods for further details on sample preparation from cryopreserved and paraffin-embedded sections.

Inclusion of paired cases in specific molecular analyses described below was dictated largely by the presence of borderline and invasive regions in the frozen or FFPE material available for research. Supplementary Table S1 summarizes the samples included in each analysis.

DNA was obtained from both the borderline tumor and invasive regions of some SBT–EOC (“paired” samples) in which both morphologic regions were accessible in single fresh frozen or FFPE sections. In other cases, either borderline or invasive regions (“unpaired” samples) were obtained from a given sample (details in Supplementary Table S1). DNA from paired samples was screened for mutations in oncogenes and tumor suppressor genes using the Sequenom/OncoMap3 platform, including OM3 Core (34 genes) and extended set (117 genes; Supplementary Table S2). All candidate mutations were validated using multibase extension homogenous mass-extend chemistry (24, 25). DNA from paired and unpaired SBT–EOC cases were screened for mutations in TP53 (26), KRAS (27), BRAF (28), and ERBB2 (H. Do and A. Dobrovic; unpublished data) using high resolution melt (HRM) analysis and validated by Sanger sequencing (Supplementary Table S3). Mutations in NRAS were detected by Sanger sequencing. Details of mutation analyses are available in Supplementary Information.

DNA from paired and unpaired SBT–EOC cases were analyzed for copy number aberrations (CNA) using Illumina 610-Quad SNP or OmniExpress SNP arrays (Supplementary Table S1). The assays were performed according to the manufacturer's instructions (Illumina) at the Australian Genome Research Facility (AGRF, University of Queensland, St Lucia, Queensland, Australia). Details of SNP array assays are provided in Supplementary Information and raw data are available from GEO (GSE57280). Genome alteration print (GAP) method was used to identify the absolute segmental copy numbers in tumor profiles by estimating the amount of normal contamination, as well as the subclonal structure of the tumor cell population based on the SNP array data (29; Supplementary Fig. S2). Further details and examples are shown in Supplementary Information.

The gene expression profile of paired SBT–EOC cases (Supplementary Table S1) was determined using Affymetrix Hu-Gene ST 1.0 arrays (Affymetrix) according to manufacturer's instructions. Raw data are available from GEO (GSE57280). Data were normalized using the RMA method available in the R package (30). Gene set enrichment analysis was also performed using all available genes sets from MSigDB (Broad Institute; ref. 31).

Formalin-fixed paraffin-embedded tissues were sectioned (4 μm), stained with monoclonal antibodies for progesterone receptor A (clone 16, Leica Microsystems) and progesterone receptor B (clone SAN27, Leica Microsystems) using methods previously described (32). A gynecological pathologist (R. Sharma) identified a minimum of five high-power fields in the SBT and invasive compartments of stained tumors for quantitation using the Aperio Positive Pixel algorithm (Aperio Technologies). Data were presented as percentage positive pixels.

Progression-free survival was defined as the time between the date of histologic diagnosis and the first confirmed sign of disease recurrence or progression based on definitions developed by the Gynaecological Cancer Intergroup, as previously described (33). In the majority of cases, the date of progression was assigned using CA125 criteria. In cases where CA125 was not a marker of disease, or progression preceded an increase in CA125, relapse was based on imaging (appearance of new lesion), or, in a minority of cases, global deterioration in health status attributable to the disease, or death. Overall survival was calculated from the date of histologic diagnosis to the date of death and censored at last contact date if the patient was alive.

Associations between clinical variables were determined using the χ² test for significance or ANOVA. Differences between progression-free or overall survival were statistically assessed using Cox regression and Kaplan–Meier curves with log-rank test. Statistical analyses were performed using Stata 10.0 (StataCorp LP).

We reviewed the diagnostic pathology reports of 967 serous invasive ovarian cancer cases from the AOCS and identified 33 with coexisting borderline histology, including nine grade 3 cases. We suspected that coexisting borderline was less likely to be reported when high-grade disease was present, as the latter is the most clinically relevant feature. Systematic pathology review had been undertaken for 572 AOCS patients. Details of coexisting pathology, including borderline tumors, was specifically requested on the pathology review form and we identified 31 of 572 (5.4%) cases with borderline regions. Of the 572 reviews, most were undertaken on a single, representative slide; however, 136 were on a full set of diagnostic slides, and of these, 16 were found to have borderline regions, 7 of which were only identified through pathology review. This suggests that the frequency of coexisting borderline may be as high as 12% (16/136) in serous ovarian cancer cases and is frequently unreported in HGSC. An additional 38 invasive cancer cases with adjacent borderline were identified from the GynBiobank based on diagnostic reports, results from multidisciplinary team meetings, and/or pathology review (Supplementary Fig. S1). Our systematic review of this very large cohort of serous ovarian carcinoma cases showed that serous borderline histology could be found in association with grade 1, 2, or 3 invasive disease (Table 1), in contrast to the prevailing view that it is primarily restricted to LGSC. Figure 1 shows the typical appearance of coexisting borderline histology with invasive cancer across grade 1–3 cancers.

To determine whether serous tumors with or without borderline histology had distinct clinical behaviors, we considered clinicopathologic features, including progression-free and overall survival, in SBT–EOC versus a comparator cohort of 104 patients with serous EOC without coexisting SBT, as determined by centralized pathology review of all diagnostic blocks. Patients with SBT–EOC were younger than patients with EOC without adjacent SBT (P = 4 × 10⁻⁴; Table 1), were more likely to be early stage (P < 0.001; Table 1) and had lower pretreatment CA125 levels (P < 0.026; Table 1; and Supplementary Fig. S3). The difference in pretreatment CA125 between SBT–EOC and EOC without adjacent SBT was independent of grade (P < 0.012). In addition, patients with SBT–EOC had a longer median progression-free (P < 0.0001) and overall survival (P < 0.0001) compared with EOC with no adjacent SBT (Table 1; Supplementary Fig. S4). However, the presence of borderline regions (i.e., SBT–EOC vs. EOC) was not a significant predictor of survival in multivariate Cox regression including age, stage, grade, and residual disease.

We focused particularly on the clinicopathologic characteristics and survival of grade 3 SBT–EOC to see whether these lacked the characteristics of typical platinum-responsive, grade 3 serous carcinoma (HGSC). As we had seen with the overall cohort, grade 3 SBT–EOC were more often early stage (P < 0.01, Table 2), although the majority had late stage disease. There was no difference in age, pretreatment CA125 levels, residual disease, or median progression-free survival of grade 3 tumors with or without associated SBT (Table 2). While there was a trend for grade 3 SBT–EOC cases to have better overall survival, this did not reach statistical significance (Table 2; Supplementary Fig. S5). Therefore, the phenotypic difference between tumors with or without associated borderline histology was not matched by marked differences in clinical outcome.

To determine whether the coexistence of borderline and invasive regions represented independent collision tumors, or whether they had a common origin, we analyzed gene expression, DNA copy number, and mutations in targeted genes. A subset of SBT–EOC cases with available biospecimens (n = 58) was used for molecular analysis. The clinicopathologic characteristics of this subset did not differ from the whole cohort (Supplementary Table S4).

Genomic copy number alterations were determined in paired borderline and invasive samples from 13 patients. Copy number variation (CNV) in paired samples were virtually identical, clearly indicating a common origin (Fig. 2). Copy number profiles from SBT–EOC cases (n = 36; paired cases as above and the invasive component of an additional 23 SBT–EOC) were heterogeneous, ranging from remarkably limited to highly aberrant. Grade 1 cases had relatively low-level CNV and few breakpoints, consistent with previous findings (refs. 34–36; Figs. 2–4). Grade 3 cases were highly aberrant, with a high number of breakpoints, typical of HGSC (Figs. 3 and 4). Grade 2 cases segregated almost equally into those with either high or low level CNV (Figs. 3 and 4).

To better define drivers of low- and high-grade tumors, we performed mutation profiling of 14 paired cases by HRM analysis of hotspot mutations in KRAS, BRAF, ERBB2, and TP53 and also screened 124 commonly mutated cancer genes using mass spectrometry analysis (OncoMap3 Extended; Supplementary Table S2) in seven microdissected pairs. Across paired cases, mutually exclusive mutations were detected in 13 of 14 cases (KRAS, n = 3; NRAS, n = 3; TP53, n = 7 mutations; Table 3) including activating NRAS mutations (Q61K and Q61R), not commonly reported in serous ovarian cancer. Consistent with a clonal relationship within pairs, mutations found in 11 of 13 cases were identical across each component.

KRAS, ERBB2, and BRAF mutations have been reported previously in LGSC and SBT (10–14) and the identification of NRAS added another member of the Ras signaling pathway that is commonly mutated in LGSC. To gain a better estimate of frequency, we performed mutation analysis of the invasive region of a further 44 SBT–EOC cases, 53 purely SBT, and 53 EOC cases without adjacent borderline (Table 4; and Supplementary Information). We identified a further 2 NRAS mutations in SBT–EOC, so that combined with the 3 initial cases, the frequency of activating NRAS mutations was 5 of 58 (9%) invasive tumors with adjacent borderline malignancy. Whereas BRAF and KRAS mutations were found in both SBT and SBT–EOC tumors (Supplementary Tables S5A and S5B), no NRAS mutations were found in any of the 53 purely SBT tumors tested. NRAS mutations were found in both grade 1 and grade 2 tumors (Table 5 and Supplementary Tables S5A and S5B). Patients with an NRAS mutation were older, had a higher proportion of late stage disease, and more residual disease than patients with KRAS or BRAF mutations; however, these trends did not reach statistical significance (Table 5). Pretreatment serum CA125 levels were significantly lower in patients with NRAS-mutated tumors (Table 5; P < 0.001). Together, these data suggest that mutations in Ras pathway genes may not be equivalent in relation to their role in ovarian cancer pathogenesis.

The appearance of near identical copy number and mutation profiles in paired samples demonstrated a shared origin but failed to identify genes that may drive transformation from borderline to invasive disease. To investigate this, we performed gene expression analysis of borderline and invasive regions from seven paired SBT–EOC samples. Supervised analysis (Supplementary Materials and Methods) revealed 127 differentially expressed genes (P < 0.01; Supplementary Table S6), albeit with a relaxed false discovery rate (5%) to allow for the small sample size. Genes with the largest fold change (≥1.5) included Progesterone Receptor (PGR), which was more highly expressed in the borderline compartment, and SPTSSB, DDIT4, COL12A1, and FAM153C, which were more highly expressed in the invasive regions. PGR was selected for validation. It is expressed in most SBT (37) and in approximately 30% of serous EOC (38) and had the largest fold change in comparing paired samples. Consistent with expression array results, immunohistochemistry of progesterone receptor A and B (PRA and PRB) in 10 FFPE samples of SBT–EOC revealed strong differential expression in SBT versus invasive regions for both progesterone receptor isoforms (Fig. 5 and see Supplementary Table S7). These findings suggest that areas of SBT are distinguished from coexisting EOC by patterns of gene expression that may also give insight to the biologic basis of this differential morphology.

There is an increasing awareness of the molecular differences between various ovarian cancer histologic subtypes and the need for subtype-specific therapies. Here, we sought to clarify the nature of serous carcinoma with adjacent borderline tumor, and to determine whether this pathologic feature could be a reliable clinical indicator of Ras-driven cancers. We analyzed 102 cases of SBT–EOC identified from >1,200 patients with serous invasive ovarian cancer. Full pathology review of 136 cases suggested that SBT–EOC may comprise approximately 12% of serous invasive tumors, which is higher than previous estimates (2, 16, 17). The distribution of histologic grades among the 102 SBT–EOC was approximately equal (Table 1). The relatively high proportion of grade 3 cases was surprising, as serous borderline tumors are thought to be more often associated with low-grade invasive cancers, although adjacent SBT in high-grade serous invasive cancers has been reported in a few cases (2, 15–17, 39, 40).

The type I/II model of ovarian cancer proposes that a characteristic feature of type I tumors is that they arise on a borderline background from a distinct noninvasive (borderline or proliferating) precursor (41), whereas type II/HGSC tumors arise de novo. Although this model provided a strong thesis to support recognition of ovarian cancers as distinct molecular entities, the clinical implications of this classification have become less clear following identification of precursors of HGSC, so called serous tubal intraepithelial carcinoma; ref. 4). Furthermore, our data demonstrate that high-grade tumors can also display regions with a type I/“borderline” appearance, but that these tumors have the molecular characteristics of type II/high-grade carcinomas. It appears that a more precise classification of serous cancers involves low and high molecular grades, classified on the basis of aggregate features of copy number change, TP53, and Ras pathway mutations (Fig. 4). The ability to accurately classify tumors is of most clinical importance for grade 2 invasive serous cancers. We find that grade 2 tumors contain two distinct molecular subtypes, consistent with either Ras-driven or typical, TP53-mutated HGSC, which are not readily distinguishable based on histology. This, and similar molecular data may be needed to resolve the classification of grade 2 tumors into “low” or “high” grade (39).

Although KRAS/BRAF and TP53/BRCA mutations are very common in low and high molecular grade tumors, respectively, they are not seen in all cases, possibly as other pathway-related events, such as the NRAS mutations, remain to be defined. Of the changes that characterize low and high molecular grades, pathogenic TP53 mutations appear to be the most indicative of molecular subtype. In our series, TP53 mutations were almost always accompanied by the extensive DNA copy number change typical of HGSC (42) and were absent from all Ras-mutant tumors and tumors with limited copy number change. Nevertheless, we recommend classification of serous tumors based on a combination of mutation in Ras pathway and TP53 genes, the extent of DNA copy number change, and histologic appearance.

Our mutation analyses revealed that Ras pathway mutations may not be equivalent in pathogenicity in serous ovarian tumors. We found pathogenic NRAS mutation in 5 of 58 (8%) SBT–EOC cases. The Catalogue of Somatic Mutations in Cancer lists 3 of 279 (1%) ovarian specimens with an NRAS mutation, including one cell line derived from an undifferentiated ovarian cancer and two serous ovarian cancer specimens from TCGA including one with coexisting borderline pathology mentioned in the diagnostic pathology report (ref. 8; accessed 15 June, 2014). NRAS mutations were found in 6% of malignant ascites/washings from patients with ovarian cancer (43); in 1 case (2.5%) of HGSC (44); and one case of LGSC from which a cell line has also been derived (45). In the paired samples tested, both the invasive and adjacent borderline compartments contained the same NRAS mutation. However, we did not identify any NRAS mutations in a cohort of 53 SBT without invasive disease. In contrast, mutations in BRAF and KRAS were found more often in SBT than in invasive disease. These data suggest that NRAS may be an important oncogene for the progression of SBT to frankly invasive disease. Indeed, NRAS is a recognized oncogene in other cancer types, including leukemia and melanoma (46, 47), and mutant NRAS was identified in an in vitro retroviral expression library screen for ovarian oncogenes (48). The role of NRAS mutations in ovarian cancer warrants further investigation given emerging, individualized therapy targeting either NRAS specifically or its downstream effectors (49).

There are several potential models of the evolution of tumors with adjacent borderline features that are consistent with the molecular data presented here. The first is that borderline tumors arise in association with a spectrum of alterations and specific molecular “drivers” predetermine pathogenesis. Specifically, a borderline tumor that is associated with a BRAF^V600E mutation may be more likely to remain a borderline tumor and in only rare cases progress to carcinoma, whereas a borderline tumor that is associated with a TP53 or NRAS mutation, is on an obligate path to carcinoma; those with NRAS mutations resulting in grade 1 or grade 2 features, while those with TP53 mutations result in high-grade tumors with genomic instability and widespread copy number alterations. An alternative hypothesis for the development of high-grade serous invasive cancer with adjacent SBT is that the SBT component is not a precursor lesion, but is part of the invasive tumor that morphologically resembles SBT, representing a better differentiated component (15). If so, we would expect to see no genome-level difference (point mutation, copy number change) between the SBT and invasive components of SBT–EOC tumors, but may see differences in gene expression consistent with differentiation. We favor the latter explanation, as copy number and point mutation changes were shared, whereas expression of a small number of genes including progesterone receptor differed, between the SBT and EOC components of the same tumor. In particular, we found no evidence that HGSC with adjacent SBT arise through progression from low-grade tumors. If so, we would have expected that high-grade SBT–EOC would have the same spectrum of mutations present in low-grade SBT-EOC, plus additional genomic changes that render them high-grade. Our data are not consistent with this hypothesis, as we found the presence of KRAS/BRAF/NRAS mutations and TP53 mutations to be mutually exclusive. Similarly, the presence of numerous regions of genomic change in adjacent borderline regions of grade 3 cases argues against progression of grade 1 or “typical” borderline disease to high-grade invasive cancer.

In conclusion, identification of an apparent borderline component in serous ovarian carcinoma is not a reliable indicator of Ras pathway–activated tumors. Molecular features including the degree of genomic copy number aberrations and mutation profile, dividing serous carcinomas into two “molecular grades”, may provide a more robust classification than histologic features alone. Dichotomization of serous ovarian carcinomas based on underlying molecular aberrations may have important implications for the selection of patients with ovarian cancer that may benefit from novel Ras/MAPK pathway–targeted agents, in contrast with patients with TP53 mutated tumors, who are more likely to respond to standard platinum-based chemotherapy. The identification of NRAS mutations in a proportion of SBT– EOC tumors suggest that NRAS may be an oncogenic driver in ovarian cancer that warrant further study.

No potential conflicts of interest were disclosed.

Conception and design: C. Emmanuel, G.V. Wain, R. Balleine, M.S. Anglesio, L.E. MacConaill, C.L. Clarke, D.D.L. Bowtell, A. deFazio

Development of methodology: C. Emmanuel, L.E. MacConaill, A. Dobrovic, C.L. Clarke

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C. Emmanuel, Y.-E. Chiew, P. Russell, C. Kennedy, S. Fereday, J. Hung, L. Galletta, R. Hogg, G.V. Wain, A. Brand, M.S. Anglesio, L.E. MacConaill, E. Palescandolo, S.M. Hunter, I. Campbell, A. Dobrovic, S.Q. Wong, H. Do, C.L. Clarke, P.R. Harnett, D.D.L. Bowtell, A. deFazio

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C. Emmanuel, Y.-E. Chiew, J. George, D. Etemadmoghadam, P. Russell, S. Fereday, G.V. Wain, L.E. MacConaill, E. Palescandolo, S.M. Hunter, S.Q. Wong, H. Do, C.L. Clarke, D.D.L. Bowtell, A. deFazio

Writing, review, and/or revision of the manuscript: C. Emmanuel, Y.-E. Chiew, J. George, D. Etemadmoghadam, R. Sharma, P. Russell, C. Kennedy, S. Fereday, J. Hung, R. Hogg, G.V. Wain, A. Brand, R. Balleine, M.S. Anglesio, L.E. MacConaill, E. Palescandolo, I. Campbell, A. Dobrovic, S.Q. Wong, H. Do, C.L. Clarke, P.R. Harnett, D.D.L. Bowtell, A. deFazio

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C. Emmanuel, Y.-E. Chiew, C. Kennedy, J. Hung

Study supervision: I. Campbell, D.D.L. Bowtell, A. deFazio

The authors acknowledge the contribution of the Australian Ovarian Cancer Study (AOCS) nurses and research assistants and thank all of the women who participated in the study. The authors gratefully acknowledge a donation from Stephanie Boldeman in support of the AOCS and acknowledge the AOCS Management Group: D.L. Bowtell (Peter MacCallum Cancer Centre), G. Chenevix-Trench, A. Green, P. Webb (QIMR), A. deFazio (Westmead lnstitute for Cancer Research), D. Gertig (University of Melbourne). The authors also gratefully acknowledge the cooperation of the following institutions: New South Wales: John Hunter Hospital, North Shore Private Hospital, Royal Hospital for Women, Royal North Shore Hospital, Royal Prince Alfred Hospital, Westmead Hospital; Queensland: Mater Misericordiae Hospital, Royal Brisbane and Women's Hospital, Townsville Hospital, Wesley Hospital; South Australia: Flinders Medical Centre, Queen Elizabeth II, Royal Adelaide Hospital; Tasmania: Royal Hobart Hospital; Victoria: Freemasons Hospital, Mercy Hospital for Women, Monash Medical Centre, Royal Women's Hospital; Western Australia: King Edward Memorial Hospital, St John of God Subiaco Hospital, Sir Charles Gairdner Hospital, Western Australia Research Tissue Network (WARTN). We gratefully acknowledge Silke Kantimm, who performed immunohistochemistry for progesterone receptor. The authors also gratefully acknowledge Dr. Douglas Levine for providing advice regarding data from the TCGA, and they acknowledge the use of data generated by the TCGA Research Network: http://cancergenome.nih.gov/.

The Australian Ovarian Cancer Study was supported by the U.S. Army Medical Research and Materiel Command under DAMD17-01-1-0729, The Cancer Council Victoria, Queensland Cancer Fund, The Cancer Council New South Wales, The Cancer Council South Australia, The Cancer Foundation of Western Australia, The Cancer Council Tasmania, and the National Health and Medical Research Council of Australia (NHMRC; ID400413, ID400281). The Gynaecological Oncology Biobank at Westmead is funded by Cancer Institute NSW and is a member bank of the Australasian Biospecimens Network-Oncology, funded by NHMRC (ID310670, ID628903). A. de Fazio is funded by the University of Sydney Cancer Research Fund, and A. de Fazio and P.R. Harnett are funded by the Cancer Institute NSW through the Sydney-West Translational Cancer Research Centre.

This work was also supported by a grant from the Cancer Council of New South Wales (CCNSW RG10-05). This work was also supported by the Victorian Breast Cancer Research Consortium (VBCRC), the NHMRC (ID 628630), Cancer Australia (1004673), and the Emer Casey Foundation. H. Do was funded by a fellowship from the Cancer Council of Victoria, and S.Q. Wong was supported by the Melbourne Melanoma Project, funded by the Victorian Cancer Agency Translational Research Program. L. MacConaill and E. Palescandolo were funded by the Dana-Farber Cancer Institute.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.