ORIGINAL ARTICLE Prognostic gene expression signature for high-grade serous ovarian cancer J. Millstein1, T. Budden2,3, E. L. Goode4, M. S. Anglesio5,6,7, A. Talhouk5,7, M. P. Intermaggio2, H. S. Leong8, S. Chen9, W. Elatre10, B. Gilks5,6, T. Nazeran5, M. Volchek11, R. C. Bentley12, C. Wang13, D. S. Chiu5, S. Kommoss14, S. C. Y. Leung5, J. Senz5,6, A. Lum5, V. Chow5, H. Sudderuddin5, R. Mackenzie5, J. George15, AOCS Group8,16,17y, S. Fereday8,18, J. Hendley8,18, N. Traficante8,18, H. Steed19, J. M. Koziak20, M. Köbel21, I. A. McNeish22,23, T. Goranova24, D. Ennis22,23, G. Macintyre24, D. Silva De Silva24, T. Ramón y Cajal25, J. García-Donas26, S. Hernando Polo27, G. C. Rodriguez28, K. L. Cushing-Haugen29, H. R. Harris29,30, C. S. Greene31, R. A. Zelaya32, S. Behrens33, R. T. Fortner33, P. Sinn34, E. Herpel34,35, J. Lester36,37, J. Lubinski38, O. Oszurek38, A. Tołoczko38, C. Cybulski38, J. Menkiszak39, C. L. Pearce40,41, M. C. Pike41,42, C. Tseng43, J. Alsop44, V. Rhenius44, H. Song44, M. Jimenez-Linan45, A. M. Piskorz24, A. Gentry-Maharaj46, C. Karpinskyj46, M. Widschwendter47, N. Singh48, C. J. Kennedy17,49, R. Sharma50,51, P. R. Harnett17,52, B. Gao17,52, S. E. Johnatty16, R. Sayer49, J. Boros17,49, S. J. Winham13, G. L. Keeney53, S. H. Kaufmann54,55, M. C. Larson13, H. Luk56, B. Y. Hernandez56, P. J. Thompson57, L. R. Wilkens56, M. E. Carney58, B. Trabert59, J. Lissowska60, L. Brinton59, M. E. Sherman61, C. Bodelon59, S. Hinsley62, L. A. Lewsley62, R. Glasspool63, S. N. Banerjee64, E. A. Stronach22, P. Haluska54, I. Ray-Coquard65, S. Mahner66, B. Winterhoff67, D. Slamon68, D. A. Levine69,70, L. E. Kelemen71, J. Benitez72,73, J. Chang-Claude33,74, J. Gronwald38, A. H. Wu43, U. Menon46, M. T. Goodman57, J. M. Schildkraut75, N. Wentzensen59, R. Brown76, A. Berchuck77, G. Chenevix-Trench16, A. deFazio17,49, S. A. Gayther78, M. J. García73,79, M. J. Henderson80, M. A. Rossing29,30, A. Beeghly-Fadiel81, P. A. Fasching68,82, S. Orsulic36,37, B. Y. Karlan36,37, G. E. Konecny68, D. G. Huntsman5,6,7,83, D. D. Bowtell8,18, J. D. Brenton24, J. A. Doherty84, P. D. P. Pharoah44,85 & S. J. Ramus2,86* 1 Division of Biostatistics, Department of Preventive Medicine, Keck School of Medicine, University of Southern California, Los Angeles, USA; 2School of Women’s and Children’s Health, Faculty of Medicine, University of NSW Sydney, Sydney, Australia; 3CRUK Manchester Institute, The University of Manchester, Manchester, UK; 4 Department of Health Science Research, Division of Epidemiology, Mayo Clinic, Rochester, USA; 5British Columbia’s Ovarian Cancer Research (OVCARE) Program, BC Cancer, Vancouver General Hospital, and the University of British Columbia, Vancouver; 6Department of Pathology and Laboratory Medicine, University of British Columbia, Vancouver; 7Department of Obstetrics and Gynecology, University of British Columbia, Vancouver, Canada; 8Peter MacCallum Cancer Center, Melbourne, Australia; 9Center for Cancer Prevention and Translational Genomics, Samuel Oschin Comprehensive Cancer Institute, Cedars-Sinai Medical Center, Los Angeles; 10 Department of Pathology, Norris Comprehensive Cancer Center, Keck School of Medicine, University of Southern California, USA; 11Anatomical Pathology, Royal Women’s Hospital, Parkville, Australia; 12Department of Pathology, Duke University Hospital, Durham; 13Department of Health Science Research, Division of Biomedical Statistics and Informatics, Mayo Clinic, Rochester, USA; 14Department of Women’s Health, Tuebingen University Hospital, Tuebingen, Germany; 15The Jackson Laboratory for Genomic Medicine, Farmington, USA; 16Department of Genetics and Computational Biology, QIMR Berghofer Medical Research Institute, Brisbane, Queensland; 17Centre for Cancer Research, The Westmead Institute for Medical Research, The University of Sydney, Sydney; 18Sir Peter MacCallum Department of Oncology, The University of Melbourne, Parkville, Australia; 19Department of Obstetrics and Gynecology, Division of Gynecologic Oncology, Royal Alexandra Hospital, Edmonton; 20Alberta Health Services-Cancer Care, Calgary; 21Department of Pathology and Laboratory Medicine, University of Calgary, Foothills Medical Center, Calgary, Canada; 22Division of Cancer and Ovarian Cancer Action Research Centre, Department Surgery & Cancer, Imperial College London, London; 23Institute of Cancer Sciences, University of Glasgow, Glasgow; 24Cancer Research UK Cambridge Institute, University of Cambridge, Cambridge, UK; 25Medical Oncology Service, Hospital Sant Pau, Barcelona; 26HM Hospitales D Centro Integral Oncológico HM Clara Campal, Madrid; 27Medical Oncology Service, Hospital Universitario Funcacion Alcorcon, Alcorcón, Spain; 28Division of Gynecologic Oncology, NorthShore University HealthSystem, University of Chicago, Evanston; 29Program in Epidemiology, Division of Public Health Sciences, Fred Hutchinson Cancer Research Center, Seattle; 30Department of Epidemiology, University of Washington, Seattle; 31Department of Systems Pharmacology and Translational Therapeutics, Perelman School of Medicine, University of Pennsylvania, Philadelphia; 32Department of Genetics, Geisel School of Medicine at Dartmouth, Hanover, USA; 33Division of Cancer Epidemiology, German Cancer Research Center (DKFZ), Heidelberg; 34Department of Pathology, Institute of Pathology, University Hospital Heidelberg, Heidelberg; 35National Center for Tumor Diseases, University Hospital and German Cancer Research Center, Heidelberg, Germany; 36David Geffen School of Medicine, Department of Obstetrics and Gynecology, University of California at Los Angeles, Los Angeles; 37Women’s Cancer Program at the Samuel Oschin Comprehensive Cancer Institute, Cedars-Sinai Medical Center, Los Angeles, USA; 38Department of Genetics and Pathology, Pomeranian Medical University, Szczecin; 39Department of Gynecological Surgery and Gynecological Oncology of Adults and Adolescents, Pomeranian Medical University, Szczecin, Poland; 40Department of Epidemiology, University of Michigan School of Public Health, Ann Arbor; 41Department of Preventive Medicine, Keck School of Medicine, University of Southern California Norris Comprehensive Cancer Center, Los Angeles; 42Department of Epidemiology and Biostatistics, Memorial Sloan-Kettering Cancer Center, New York; 43Department of Preventive Medicine, Keck School of Medicine, University of Southern California, Los Angeles, USA; 44Centre for Cancer Genetic Epidemiology, Department of Oncology, University of Cambridge, Cambridge; 45Department of Pathology, Addenbrooke’s Hospital NHS Trust, Cambridge; 46MRC Clinical Trials Unit at UCL, Institute of Clinical Trials & Methodology, University College London, London; 47Department of Women’s Cancer, Institute for Women’s Health, University College London, London; 48Department of Pathology, Barts Health National Health Service Trust, London, UK; 49Department of Gynaecological Oncology, Westmead Hospital, Sydney; 50Pathology West ICPMR Westmead, Westmead Hospital, The University of Sydney, Sydney; 51University of Western Sydney at Westmead Hospital, Sydney; 52The Crown Princess Mary Cancer Centre Westmead, Sydney-West Cancer Network, Westmead Hospital, Sydney, Australia; 53Department of Laboratory Medicine and Pathology, Division of Anatomic Pathology, Mayo Clinic, Rochester; 54Department of Oncology, Mayo Clinic, *Correspondence to: Prof. Susan Ramus, School of Women’s and Children’s Health, Faculty of Medicine, Lowy Building, 2nd floor, UNSW Sydney, NSW 2052, Australia. Tel: +61-02-9385-1720 E-mail: s.ramus@unsw.edu.au (S.J. Ramus). y The AOCS Study members are listed in full in Appendix 1. 0923-7534/© 2020 The Authors. Published by Elsevier Ltd on behalf of European Society for Medical Oncology. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). 1240 https://doi.org/10.1016/j.annonc.2020.05.019 Volume 31 - Issue 9 - 2020 J. Millstein et al. Annals of Oncology Rochester; 55Department of Molecular Pharmacology and Experimental Therapeutics, Mayo Clinic, Rochester; 56Cancer Epidemiology Program, University of Hawaii Cancer Center, Honolulu; 57Samuel Oschin Comprehensive Cancer Institute, Cancer Prevention and Genetics Program, Cedars-Sinai Medical Center, Los Angeles; 58John A. Burns School of Medicine, Department of Obstetrics and Gynecology, University of Hawaii, Honolulu; 59Division of Cancer Epidemiology and Genetics, National Cancer Institute, Bethesda, USA; 60Department of Cancer Epidemiology and Prevention, M Sklodowska Curie National Research Institute of Oncology, Warsaw, Poland; 61 Department of Health Sciences Research, Mayo Clinic College of Medicine, Jacksonville, USA; 62Cancer Research UK Clinical Trials Unit, Institute of Cancer Sciences, University of Glasgow, Glasgow; 63Department of Medical Oncology, Beatson West of Scotland Cancer Centre and University of Glasgow, Glasgow; 64The Royal Marsden NHS Foundation Trust and Institute of Cancer Research, London, UK; 65Centre Leon Berard and University Claude Bernard Lyon 1, Lyon, France; 66Department of Obstetrics and Gynecology, University Hospital, LMU Munich, Munich, Germany; 67Department of Obstetrics, Gynecology and Women’s Health, University of Minnesota, Minneapolis; 68David Geffen School of Medicine, Department of Medicine Division of Hematology and Oncology, University of California at Los Angeles, Los Angeles; 69Gynecology Service, Department of Surgery, Memorial Sloan Kettering Cancer Center, New York; 70Gynecologic Oncology, Laura and Isaac Pearlmutter Cancer Center, NYU Langone Medical Center, New York; 71Hollings Cancer Center and Department of Public Health Sciences, Medical University of South Carolina, Charleston, USA; 72Centro de Investigación en Red de Enfermedades Raras (CIBERER), Madrid; 73Human Cancer Genetics Programme, Spanish National Cancer Research Centre (CNIO), Madrid, Spain; 74Cancer Epidemiology Group, University Cancer Center Hamburg (UCCH), University Medical Center Hamburg-Eppendorf, Hamburg, Germany; 75Department of Epidemiology, Rollins School of Public Health, Emory University, Atlanta, USA; 76Division of Cancer and Ovarian Cancer Action Research Centre, Department of Surgery and Cancer, Imperial College London, London, UK; 77Department of Gynecologic Oncology, Duke University Hospital, Durham, USA; 78Center for Bioinformatics and Functional Genomics and the Cedars Sinai Genomics Core, Cedars-Sinai Medical Center, Los Angeles, USA; 79Biomedical Network on Rare Diseases (CIBERER), Madrid, Spain; 80Children’s Cancer Institute, Lowy Cancer Research Centre, University of NSW Sydney, Sydney, Australia; 81Division of Epidemiology, Department of Medicine, Vanderbilt Epidemiology Center, Vanderbilt-Ingram Cancer Center, Vanderbilt University School of Medicine, Nashville, USA; 82 Department of Gynecology and Obstetrics, Comprehensive Cancer Center ER-EMN, University Hospital Erlangen, Friedrich-Alexander-University Erlangen- Nuremberg, Erlangen, Germany; 83Department of Molecular Oncology, BC Cancer Research Centre, Vancouver, Canada; 84Huntsman Cancer Institute, Department of Population Health Sciences, University of Utah, Salt Lake City, USA; 85Centre for Cancer Genetic Epidemiology, Department of Public Health and Primary Care, University of Cambridge, Cambridge, UK; 86Adult Cancer Program, Lowy Cancer Research Centre, University of NSW Sydney, Sydney, Australia Available online 28 May 2020 Background: Median overall survival (OS) for women with high-grade serous ovarian cancer (HGSOC) is w4 years, yet survival varies widely between patients. There are no well-established, gene expression signatures associated with prognosis. The aim of this study was to develop a robust prognostic signature for OS in patients with HGSOC. Patients and methods: Expression of 513 genes, selected from a meta-analysis of 1455 tumours and other candidates, was measured using NanoString technology from formalin-fixed paraffin-embedded tumour tissue collected from 3769 women with HGSOC from multiple studies. Elastic net regularization for survival analysis was applied to develop a prognostic model for 5-year OS, trained on 2702 tumours from 15 studies and evaluated on an independent set of 1067 tumours from six studies. Results: Expression levels of 276 genes were associated with OS (false discovery rate < 0.05) in covariate-adjusted single-gene analyses. The top five genes were TAP1, ZFHX4, CXCL9, FBN1 and PTGER3 (P < 0.001). The best performing prognostic signature included 101 genes enriched in pathways with treatment implications. Each gain of one standard deviation in the gene expression score conferred a greater than twofold increase in risk of death [hazard ratio (HR) 2.35, 95% confidence interval (CI) 2.02e2.71; P < 0.001]. Median survival [HR (95% CI)] by gene expression score quintile was 9.5 (8.3 to e), 5.4 (4.6e7.0), 3.8 (3.3e4.6), 3.2 (2.9e3.7) and 2.3 (2.1e2.6) years. Conclusion: The OTTA-SPOT (Ovarian Tumor Tissue Analysis consortium - Stratified Prognosis of Ovarian Tumours) gene expression signature may improve risk stratification in clinical trials by identifying patients who are least likely to achieve 5-year survival. The identified novel genes associated with the outcome may also yield opportunities for the development of targeted therapeutic approaches. Key words: formalin-fixed paraffin-embedded, gene expression, high-grade serous ovarian cancer, overall survival, prognosis INTRODUCTION The most informative prognostic factors for HGSOC are Epithelial ovarian cancer (EOC) causes w125 000 International Federation of Gynecology and Obstetrics (FIGO) deaths globally every year, and long-term survival rates stage, residual disease following debulking surgery,4 BRCA1 or have changed little in the past three decades.1 BRCA2 germline mutation5,6 and tumour-infiltrating Approximately 70% of women with EOC are diagnosed lymphocyte scores.7,8 Patients with HGSOC who carry a with advanced stage disease (stages III/IV), and fewer loss-of-function germline mutation in BRCA1 or BRCA2 have than 50% will survive more than 5 years.2 There are an increased sensitivity to platinum-based chemotherapy and five major EOC histotypes: high-grade serous, low-grade PARP inhibitor treatment9,10 and a medium-term survival serous, endometrioid, clear cell and mucinous.3 High- advantage.5 However, the frequent development of drug- grade serous ovarian cancer (HGSOC) comprises about resistant disease6 limits the effectiveness of current therapies. two-thirds of cases, is responsible for most deaths and Gene expression data have been used to define four is characterized by profound genomic and clinical tumour molecular subtypes of HGSOC (C1/mesenchymal, heterogeneity. C2/immune, C4/differentiated and C5/proliferative).11,12 Volume 31 - Issue 9 - 2020 https://doi.org/10.1016/j.annonc.2020.05.019 1241 Annals of Oncology J. Millstein et al. Using transcriptome-wide data from fresh frozen tissues, Gene expression assay in tumour samples from study The Cancer Genome Atlas (TCGA) project used 215 tumours participants to identify an overall survival (OS) expression signature of FFPE tumour samples were processed with the NanoString 193 genes that has been validated on three other HGSOC nCounter technology at three different locations: Vancouver, gene expression datasets.12 Los Angeles and Melbourne. A control set of 48 FFPE tumour Despite these findings, gene expression biomarkers have samples was run at each location and the average intraclass not been implemented clinically owing to several important correlation coefficient was 0.987. Approximately 2% of the shortcomings. The majority of the individual markers samples were run in duplicate and the average Spearman’s comprising the 193 gene signature were not statistically correlation coefficient rs was 0.995. Single-patient classifi- significant across all studies, suggesting that the signature cation methods were used with reference samples to control may not be robust. The sample sizes in other discovery ef- for batch effects.18 The data in this publication have been forts have been too small for robust statistical inference.12 deposited in NCBI’s Gene Expression Omnibus19; GEO Series In addition, previous studies used fresh frozen samples, accession number GSE132342 (https://www.ncbi.nlm.nih. resulting in logistic and cost barriers to examining large gov/geo/query/acc.cgi?acc¼GSE132342). A total of 3329 clinically relevant datasets, and translation to the clinical samples passed quality control of which 3769 had survival setting. data and assessable gene expression for 513 genes. Data can The aim of this study was to identify a robust and clinic- be found in NCBI GEO: Accession numbers GSE132342 and ready prognostic HGSOC profile that can be applied to GPL26748. formalin-fixed paraffin-embedded (FFPE) tumour tissue. Overall survival analysis of individual genes Samples that contributed to the meta-analysis dataset (n ¼ PATIENTS AND METHODS 211) were removed from subsequent selected analyses to Twenty studies provided pretreatment FFPE tumour sam- enforce independence of study samples between the gene ples from 4071 women diagnosed with HGSOC selection and final survival analysis. Time-to-event analyses (supplementary Table S1, available at Annals of Oncology were carried out for OS with right censoring at 10 years and online). All HGSOC cases with available tissue were left truncation of prevalent cases. Associations between included. During this period, patients with HGSOC were log-transformed normalized gene expression and survival treated with chemotherapy (carboplatin and paclitaxel) af- time were tested using likelihood ratio tests with Cox pro- ter primary debulking surgery. Study protocols were portional hazards models adjusted for age, race and stage, approved by the respective Institutional Review Board/ and stratified by study. Patients with missing race or stage Ethics Approval Committee for each site (supplementary information were assigned to ‘unknown’ categories. Age Table S1, available at Annals of Oncology online). was modelled using a B-spline with a knot at the median A schematic of the overall study design is shown in age, which yielded a better fit than using knots at quartiles Figure 1. There were four main components: gene selection, or categorical variables. Stage was dichotomized into early gene expression assay, development of prognostic gene (FIGO stage I/II) and advanced (FIGO stage III/IV). Genes signature in a training set and validation of prognostic were scaled to have a standard deviation of one, so hazard signature in an independent validation set. ratios (HRs) correspond to a change of one standard devi- ation. A BenjaminieHochberg FDR of <0.05 was used to identify notable associations. Because the expression of Gene selection genes can be correlated, an analysis of correlated genes was Candidate prognostic genes were identified by carrying out performed using data from TCGA. Advanced stage ovarian an individual participant meta-analysis of six transcriptome- cancer usually has disease spread throughout the abdomen, wide microarray studies,11e16 which included tumour and therefore sensitivity analyses were performed to assess samples from 1455 participants. Association of gene effects of the anatomical location of tumour samples expression with OS was evaluated by Cox proportional included in the study by removing observations corre- hazards regression adjusted for molecular subtype sponding to samples known to be extraovarian (n ¼ 437). (supplementary Table S2, available at Annals of Oncology online). In total, 200 genes from the meta-analysis, most Prognostic signature development and validation achieving a permutation-based false discovery rate (FDR)17 Studies were initially randomized to training set (N ¼ 14) of <0.05, and an additional 313 candidate genes based on and validation set (N ¼ 6). The TRI study was randomized to the literature and unpublished data were selected the validation set, but, because 107 samples were part of (supplementary Tables S3 and S4, and supplementary the meta-analysis data used for gene selection, the study Figure S1, available at Annals of Oncology online; for was split, so those 107 samples were included in the model more details see supplementary Material, available at training dataset. Thus 2702 samples from 15 studies were Annals of Oncology online). Five genes, RPL19, ACTB, PGK1, used for model training and 1067 samples from 6 studies SDHA and POLR1B, were included as house-keeping genes were used for validation (supplementary Table S1, available for normalization. at Annals of Oncology online). In the training set, four 1242 https://doi.org/10.1016/j.annonc.2020.05.019 Volume 31 - Issue 9 - 2020 Volume 31 J. Millstein et al. - Issue 9 - 2020 https://doi.org/10.1016/j.annonc.2020.05.019 Figure 1. Schematic of study design. *The TRI study was split across the training and validation sets due to 107 samples overlapping with the meta-analysis. GWAS, genome-wide association studies; HGSOC, high-grade serous ovarian cancer. Annals of Oncology 1243 Annals of Oncology J. Millstein et al. modelling approaches (stepwise regression, elastic net Development of a novel prognostic gene signature regularized regression, boosting and random survival for- The four predictive modelling approaches that were evalu- ests) were applied to construct competing gene ated in the training data using 10-fold cross-validation expressionebased biomarkers. Each was evaluated in the yielded median AUCs that ranged from 0.69 to 0.73 for 2- training data using 10-fold cross-validation for its prognostic year OS and 0.69 to 0.74 for 5-year survival value for OS at 2 and 5 years of follow-up using an area (supplementary Figure S2, available at Annals of Oncology under the curve (AUC) measure derived from receiver online) with better prediction of 5-year OS than of 2-year operator characteristic analysis (see supplementary OS. The elastic net approach yielded the highest median Material, available at Annals of Oncology online, for addi- AUC for both 2- and 5-year OS and was selected for final tional details). The best performing method, elastic net development of the signature. Using the model on the full regularized regression, was applied to the full training set to training dataset resulted in a prognostic signature of 101 determine the final gene signature and scoring method, genes in addition to age and stage (supplementary which was then evaluated using the independent testing Table S10, available at Annals of Oncology online). Of set. All models were constrained to include age and stage, these, 66 genes were associated with OS (FDR < 0.05) in where age was modelled as categorical based on quartiles the single gene models. There was no obvious subset of of the training dataset with groups aged <53, 53e59, 60e signature genes that performed as well or nearly as well as 66, and 67. Stage was modelled as described earlier for the full 101 gene signature (supplementary Figure S3, the OS individual gene analysis. available at Annals of Oncology online). Performance of the signature including age and stage was AUC 0.69 [95% confidence interval (CI) 0.65e0.73] and AUC RESULTS 0.75 (95% CI 0.72e0.78) for 2- and 5-year OS, respectively (Figures 2 and 3; supplementary Figure S4, available at Association of expression of individual genes with OS in Annals of Oncology online). This was substantially better HGSOC than age and stage alone with AUC 0.61 (95% CI 0.57e0.65) In a gene-by-gene analysis of the full dataset adjusted for and AUC 0.62 (95% CI 0.59e0.67) for 2- and 5-year OS, age, race and stage, and stratified by study, 276 of the 513 respectively, particularly for the 5-year OS outcome with selected genes were associated with OS (FDR < 0.05). Of non-overlapping 95% CI. One standard deviation change in these, 138 were selected from the meta-analysis of six the gene expression score was associated with an HR of published microarray studies (supplementary Table S2, 2.35 [(95% CI 2.02e2.71); P ¼ 5.1  1031], and median available at Annals of Oncology online)11e16 and 144 from survival [HR (CI)] varied substantially across quintiles of the candidate gene approaches (supplementary Tables S5 and gene expression score [9.5 (8.3 to e), 5.4 (4.6e7.0), 3.8 S6, available at Annals of Oncology online). HRs for one (3.3e4.6), 3.2 (2.9e3.7) and 2.3 (2.1e2.6) years, respec- standard deviation change in gene expression ranged from tively, from smallest to largest quintile (Q1-Q5); Table 2]. 0.84 to 1.19, with multiple genes exhibiting associations at For a subset of cases, there was clinical and experimental very stringent significance levels (e.g. 19 genes with P < data for known prognostic factors. All samples had molec- 1 108; supplementary Tables S5 and S6, available at ular subtype classification (Talhouk et al.20); residual disease Annals of Oncology online). The five most significant genes was known for 1771 cases, primary chemotherapy treat- were TAP1, ZFHX4, CXCL9, FBN1 and PTGER3 (Table 1). We ment for 687, germline BRCA mutation status for 904 and did not find extensive evidence of high co-expression be- nuclear CD8 tumour-infiltrating lymphocyte counts8 for tween these five genes and genes measured in TCGA project 1111 (supplementary Table S11, available at Annals of (supplementary Table S7, available at Annals of Oncology Oncology online). When examined by quintile of gene online). In sensitivity analyses we found that excluding expression score there were differences, as expected, for samples from omentum and other extra-ovarian sites did each of the known prognostic factors, including age and not substantially affect the results (supplementary Tables S8 stage that were included in the model (Table 3). However, and S9, available at Annals of Oncology online). in sensitivity analyses, applying the signature to specific Table 1. Hazard ratios and 95% CIs for top five prognostic genes in covariate-adjusted single-gene analyses Gene HR (95% CI) P Selection Correlated genea rs 18 TAP1 0.84 (0.80e0.87) 8.3  10 Meta PSMB9 0.89 ZFHX4 1.19 (1.14e1.25) 1.4  1015 Meta LOC100192378 0.74 CXCL9 0.85 (0.82e0.88) 1.8  1015 Meta and candidate CXCR6 0.89 FBN1 1.18 (1.13e1.24) 4.2  1014 Candidate SPARCb 0.91 PTGER3 1.18 (1.13e1.24) 1.2  1013 Meta COL8A1 0.67 CI, confidence interval; HR, hazard ratio. a Most correlated gene according to Spearman’s rank correlation coefficient, rs, computed in The Cancer Genome Atlas (TCGA) Ovarian Serous Cystadenocarcinoma RNA-seq dataset. b SPARC was included in this project and was less significant. 1244 https://doi.org/10.1016/j.annonc.2020.05.019 Volume 31 - Issue 9 - 2020 J. Millstein et al. Annals of Oncology proteinecoupled receptor) ligand binding. Four other 1.0 pathways were related to the cell cycle and mitosis, with the remaining enriched for fibroblast growth factor receptor 0.8 (FGFR) and epidermal growth factor receptor (ERBB) sig- nalling, and one pathway related to homologous combina- tion repair. 0.6 TP DISCUSSION 0.4 In a large-scale study of patients with HGSOC, we identified AUC Model Timepoint a 101-gene expression signature able to predict clinically 0.61 −− 2 year relevant differences in OS. Using methods that are both 0.2 0.62 −− 5 year economical and applicable to standard clinical sampling 0.69 Signature 2 year techniques, we showed that the signature performs sub- 0.75 Signature 5 year stantially better than age and stage alone for prognosis of 0.0 0.0 0.2 0.4 0.6 0.8 1.0 both 2- and 5-year OS. The number of patients and samples FP included in this study is an order of magnitude greater than previous comparable studies of gene expression and OS in Figure 2. Receiver operator characteristic (ROC) curves for prognostic per- formance of the gene expression signature in independent high-grade serous patients with HGSOC.12,21,22 Thus, we have been able to ovarian cancer patients (testing data). more precisely quantify the prognostic value of gene There was no overlap between studies or patient data used to develop models expression. (training data) and construct ROC curves and calculate area under the curve (AUC) values shown here (testing data). All models included age and stage as We report definitive associations between OS and described in Methods section. TP denotes the true positive rate (sensitivity) and expression of 276 genes. Of the five most significant genes FP denotes the false positive rate (1  specificity). (TAP1, ZFHX4, CXCL9, FBN1 and PTGER3), four have been previously reported to be associated with survival in patient groups, a robustness of stratification was demon- HGSOC. The top prognostic gene, TAP1, is involved in the strated, suggesting that the prognostic power of the antigen-presenting pathway. Expression was reduced in signature is not explained by the individual factors, residual metastatic HGSOC, positively associated with OS,23 as disease, treatment, BRCA status or CD8 score (Figure 3 and observed here, and linked to tumour regression in response supplementary Figures S5eS7, available at Annals of to treatment.24 Further, hypomethylation of TAP1 was Oncology online). The signature score showed modest dif- associated with improved time to disease recurrence.25 ferences by molecular subtype (supplementary Figure S8, CXCL9 is a chemokine that mediates the recruitment of T available at Annals of Oncology online), and adjusting for cells to solid tumours.26 High expression of intratumoural molecular subtype in the Cox analysis resulted in only minor CXCL9 was associated with higher OS27 and higher lym- changes to the HR estimates for signature quintiles phocytic infiltration, which is also a robust prognostic factor (Table 2). The signature was shown to be prognostic within in HGSOC8,11,28 and a feature of the immunoreactive a homogenous group of 316 stage IIIC cases with no re- HGSOC molecular subtype.11 CXCL9 has also been proposed sidual disease, within early stage cases (FIGO IA and IB) and as a therapeutic target due to evidence that it inhibits within patients whose samples were collected from the angiogenesis and promotes antitumour adaptive immu- omentum (supplementary Figures S9 and S10, available at nity.29e31 Strikingly, the signature was able to further refine Annals of Oncology online). Analysis of the signature score prognostic groups within patients with high tumour- for paired ovarian and omental tissue from 42 of the cases infiltrating lymphocyte counts suggesting that CXCL9 and showed a highly significant Pearson’s correlation coefficient, TAP1 expression may be strong indicators of immune r ¼ 0.79 (P ¼ 5.4  1010; supplementary Figure S11, competency in HGSOC. available at Annals of Oncology online). FBN1 is an extracellular matrix protein previously found A gene set enrichment analysis was performed for the to be a biomarker associated with early recurrence in pa- 101 genes in the signature, as well as for genes correlated tients with ovarian cancer who are initially sensitive to with signature genes achieving r2 > 0.75 (supplementary chemotherapy32 and strongly correlated with desmoplasia Table S12, available at Annals of Oncology online). For the in HGSOC. The prostaglandin E2 receptor PTGER3 is correlated gene analysis, the three most significant path- expressed in ovarian tumour cells and is associated with ways involved the immune system, including the adaptive relapse-free survival.33 By contrast, ZFHX4 does not have immune system and cytokine signalling. A further 10 im- previous associations with HGSOC. mune pathways were significantly enriched and included Associations between the expression of specific genes in interferon signalling, innate immune system and TCR sig- tumour tissues and OS in patients with HGSOC may suggest nalling and antigen presentation pathways. Restricting to new drug targets and lead to insights into biological varia- the signature genes only, there was also enrichment in the tion in treatment response. For example, cases in the Q5 immune system, but the top two pathways were PI-3K quintile with the poorest outcome had increased expression (phosphoinositide 3-kinase) cascade and GPCR (G of IGF2, FGFR1 and MYC, a possible argument for the use of Volume 31 - Issue 9 - 2020 https://doi.org/10.1016/j.annonc.2020.05.019 1245 Annals of Oncology J. Millstein et al. Figure 3. KaplaneMeier curves of overall survival for patients (A) in the training and (B) testing sets. Patients were assigned to quintiles (Q1eQ5) of the signature score including age and stage. Shaded areas indicate 95% confidence regions, only included for plots representing larger sample sizes. Because of limited sample size, the following plots represent all such patients in the entire dataset, training or testing: (C) no macroscopic residual disease after debulking surgery, (D) primary chemotherapy treatment 4 cycles of intravenous (IV) carboplatin area under the curve (AUC) 5 or 6 and paclitaxel 135 or 175 mg/m2 every 3 weeks (actual dose known or presumed), (E) BRCA1 or BRCA2 germline mutation and (F) CD8 > 19. IGFR1, FGFR, bromodomain (MYC) or a combination of subgroups. The signature will be incorporated into future PARP and CDK4/6 inhibitors (MYC).34 More immediately, prospective clinical trials to determine if it can predict the signature may help clinicians identify patients most in response to specific treatments. need of intervention, such as patients that could potentially Measurement of the signature required standard FFPE benefit from neoadjuvant chemotherapy (NACT). Alterna- tissue used in routine histopathology. In addition, data tively, in clinical trials it could be used to stratify randomi- preprocessing and normalization were conducted on an zation by patients’ risk, thereby reducing heterogeneity individual level, thus translatable to a general patient within subgroups and increasing heterogeneity between population. That is, 5-year OS prognosis of future patients Table 2. Hazard ratios and 95% CIs for quintiles of the gene expression signature score in validation data Quintile N Deaths Median survivala HR (95% CI) Adjusted for age and stage Adjusted for molecular subtype age and stage HR (95% CI) HR (95% CI) Q1 214 81 9.47 (8.32 to e) 0.44 (0.33e0.58) 0.34 (0.22e0.55) 0.37 (0.23e0.59) Q2 213 117 5.38 (4.63e6.97) 0.73 (0.57e0.93) 0.71 (0.55e0.91) 0.74 (0.58e0.96) Q3 213 145 3.80 (3.34e4.60) Q4 213 158 3.23 (2.85e3.68) 1.56 (1.25e1.96) 1.56 (1.24e1.97) 1.56 (1.24e1.96) Q5 214 179 2.27 (2.09e2.62) 2.23 (1.78e2.78) 2.11 (1.67e2.67) 2.07 (1.63e2.61) CI, confidence interval; HR, hazard ratio. a Median survival (95% CI) in years for patients in the validation set. 1246 https://doi.org/10.1016/j.annonc.2020.05.019 Volume 31 - Issue 9 - 2020 J. Millstein et al. Annals of Oncology Table 3. Clinical data for the 3769 patients that passed quality control and the percentage of patients in each quintile of the gene expression score Total Q1 Q2 Q3 Q4 Q5 P value N 3769 754 754 753 754 754 Median survival (years) 4.1 9.5 5.4 3.8 3.2 2.3 % 5-year survival 41 75 57 39 25 10 Age median 63 58 57 61 64 70 Age range 25e89 39e78 25e86 36e82 27e89 39e86 Age quartile Q1 894 30.8 31.3 20.0 13.4 4.5 <1  1050 Age quartile Q2 838 21.5 20.0 22.9 21.2 14.3 Age quartile Q3 961 16.0 20.2 21.4 23.6 18.7 Age quartile Q4 1076 13.5 10.4 16.4 21.3 38.5 FIGO stage I/II 607 97.4 2.6 0.0 0.0 0.0 <1  1050 FIGO stage III/IV 3067 3.8 23.0 24.1 24.4 24.6 Primary chemoa 1 136 16.2 22.1 23.5 19.1 19.1 0.163 Primary chemoa 2 190 16.3 20.0 21.6 22.1 20.0 Primary chemoa 3 361 11.1 16.9 22.4 20.5 29.1 Residual disease: No 614 32.4 22.1 17.8 15.5 12.2 <1  1050 Residual disease: Yes 1157 6.0 19.2 24.1 24.5 26.2 Germline BRCA1 mutation 130 23.8 31.5 26.2 11.5 6.9 2.22  107 Germline BRCA2 mutation 71 28.2 26.8 18.3 18.3 8.5 Germline no BRCA1/2 mutation 663 19.6 16.7 18.7 20.7 24.3 CD8 TIL score 0 192 19.8 14.6 12.5 21.4 31.8 2.46  1014 CD8 TIL score 1e2 186 18.3 14.0 18.8 21.5 27.4 CD8 TIL score 3e19 515 19.8 24.1 20.8 17.9 17.5 CD8 TIL score >19 218 34.4 31.2 16.5 11.5 6.4 Molecular subtype C1.MES 1105 5.4 10.4 20.7 27.4 36.0 <1  1050 Molecular subtype C2.IMM 907 23.2 28.8 21.2 16.2 10.7 Molecular subtype C4.DIF 1144 32.6 25.5 17.9 12.8 11.2 Molecular subtype C5.PRO 613 18.1 14.0 20.7 25.8 21.4 FIGO stage IA and IB 111 96.4 3.6 0.0 0.0 0.0 <1  1050 FIGO stage IIIC 1979 3.1 23.7 24.6 24.1 24.6 <1  1050 FIGO stage IIIC residual disease: No 316 6.3 31.0 24.4 20.9 17.4 6.24  1045 FIGO stage IIIC residual disease: Yes 846 2.6 21.5 25.3 24.6 26.0 Q1 is the quintile with the best survival and Q5 the worst survival. Samples with missing data are reported in supplementary Table S11, available at Annals of Oncology online. P values for BRCA1/2 mutation status were calculated for BRCA1 or BRCA2 mutation versus no mutation. FIGO, International Federation of Gynecology and Obstetrics; TIL, tumour-infiltrating lymphocyte. a Treatment: 1 ¼ known to have received first-line chemotherapy treatment of 4 cycles of IV carboplatin AUC 5 or 6 and paclitaxel 135 or 175 mg/m2 every 3 weeks. 2 ¼ known to have received first-line chemotherapy treatment of 4 cycles of IV carboplatin and paclitaxel three times weekly but at doses presumed to be carboplatin AUC 5 or 6 and paclitaxel 135 or 175 mg/m2. 3 ¼ all remaining cases with chemo regimens that do not fit criteria 1 or 2 and include unknown or no chemotherapy. can be evaluated against the patient population reported patients had NACT during the calendar period in which here by (i) following the same steps described here for these samples were collected. Because the signature ap- generating the normalized gene expression data, (ii) pears to be prognostic in omentum samples, future studies computing an individual signature score and (iii) assigning may assess the value in NACT patients, using pretreatment an HR based on the score or comparing it with the reported omental biopsies or post-treatment tumour samples. Future quintiles (supplementary Material, available at Annals of work will also address if the signature can predict platinum- Oncology online). NanoString gene expression is highly refractory patients. reproducible as seen by our quality control metrics We have developed a robust prognostic signature for (supplementary Material, available at Annals of Oncology HGSOC that can be used to stratify patients and identify online) and the FDA approval of the Prosigna test for breast those in need of alternative treatments. Gene set enrich- cancer. ment analysis applied to the signature indicates an impor- The question of heterogeneity by ancestry or ethnicity tant role for the immune system in OS and supports further was beyond the scope of this study but should be pursued investigation of immune therapy in ovarian cancer. More in future research. Another important question is whether generally, the identification here of high-confidence prog- molecular subtype can improve biomarker performance. A nostic genes may lead to new hypotheses for targeted substantial proportion of signature genes were identified by treatments. the subtype-adjusted meta-analysis, suggesting that the strong performance of the signature is not solely attribut- ACKNOWLEDGEMENTS able to differences among molecular subtypes. In addition, We thank all the study participants who contributed to this all of the individual genes used in the molecular subtype study and all the researchers, clinicians and technical and classification were included in development of the administrative staff who have made this work possible. This signature. project received technical and data management support Although the cases received chemotherapy, the FFPE from OVCARE’s core units, including the Cheryl Brown samples used in this study were chemo-naïve, as few Ovarian Cancer Outcomes Unit and the Genetic Pathology Volume 31 - Issue 9 - 2020 https://doi.org/10.1016/j.annonc.2020.05.019 1247 Annals of Oncology J. Millstein et al. Evaluation Centre, and statistical analysis support from the (NCATS) [grant number UL1TR000124]. HRH is supported by Biostatistics Core of the Norris Comprehensive Cancer the National Institutes of Health/National Cancer Institute Center. The AOV study recognizes the valuable contributions [grant number K22 CA193860]. OVCARE (including the VAN from Mie Konno, Shuhong Liu, Michelle Darago, Faye study) receives core funding through the BC Cancer Foun- Chambers and the staff at the Tom Baker Cancer Centre dation (no grant number applicable) and The VGHþUBC Translational Laboratories. We thank Olivier Tredan and Hospital Foundation (authors AT, BG, DGH and MSA, no Pierre Heudel as investigators on the TRIO14 study and grant number applicable). The AOV study is supported by Sandrine Berge-Montamat as assistant for clinical research. the Canadian Institutes of Health Research [grant number The Australian Ovarian Cancer Study gratefully acknowl- MOP-86727]. The Gynaecological Oncology Biobank at edges additional support from Ovarian Cancer Australia and Westmead, a member of the Australasian Biospecimen the Peter MacCallum Foundation. The AOCS also acknowl- Network-Oncology group, was funded by the National edges the cooperation of the participating institutions in Health and Medical Research Council Enabling [grant Australia and acknowledges the contribution of the study numbers ID 310670, ID 628903] and the Cancer Institute nurses, research assistants and all clinical and scientific NSW [grant numbers ID 12/RIG/1-17, 15/RIG/1-16]. The collaborators to the study. The complete AOCS Study Group Australian Ovarian Cancer Study Group was supported by can be found at www.aocstudy.org. We thank all of the the U.S. Army Medical Research and Materiel Command women who participated in these research programs. [grant number DAMD17-01-1-0729], The Cancer Council Victoria, Queensland Cancer Fund, The Cancer Council New South Wales, The Cancer Council South Australia, The Can- FUNDING cer Council Tasmania and The Cancer Foundation of This work was funded by the National Institutes of Western Australia (Multi-State Applications 191, 211 and Health/National Cancer Institute (NCI) Grants to SJR [grant 182) and the National Health and Medical Research Council number R01CA172404] and J.A. Doherty and MAR [grant of Australia [grant numbers NHMRC, ID199600, ID400413, number R01CA168758], the Canadian Institutes for Health ID400281]. BriTROC-1 was funded by Ovarian Cancer Action Research (Proof-of-Principle I program, no grant number (to IAM and JDB) [grant number 006] and supported by applicable) and the United States Department of Defense Cancer Research UK [grant numbers A15973, A15601, Ovarian Cancer Research Program [grant number A18072, A17197, A19274, A19694] and the National Insti- OC110433]. J. Millstein and SJR received support from tute for Health Research Cambridge and Imperial Biomed- National Institutes of Health/National Cancer Institute ical Research Centres (no grant number applicable). SEARCH [grant number P30CA014089] and J. Millstein received was supported by Cancer Research UK [grant number support from NIH/National Cancer Institute award number A16561]. The University of Cambridge receives salary sup- P01CA196569. MSA receives funding from the Janet port for PDPP from the NHS Clinical Academic Reserve (no D. Cottrelle Foundation Scholar’s program managed by the grant number applicable). Samples from the Mayo Clinic BC Cancer Foundation (no grant number applicable). JG was were collected and provided with support of the National partially supported by the National Institutes of Health/ Institutes of Health/National Cancer Institute (NCI) P50 National Cancer Institute [grant number P30CA034196]. CW CA136393 (ELG, GLK, SHK, MES). S. Orsulic was funded by was a Career Enhancement Awardee of the Mayo Clinic the Department of Defense Award W81XWH-17-1-0144. SPORE in Ovarian Cancer [grant number P50 CA136393]. MRC Clinical Trials Unit at UCL receives funding from The MJH received funding from Cancer Australia (1067110), Eve Appeal (The Oak Foundation) with investigators sup- DGH receives support from the Dr Chew Wei Memorial ported by the National Institute for Health Research Uni- Professorship in Gynecologic Oncology, (no grant number versity College London Hospitals Biomedical Research applicable) the Canada Research Chairs program (Research Centre and MRC core funding (MR_UU_12023). Chair in Molecular and Genomic Pathology, no grant num- ber applicable), and the Janet D. Cottrelle Foundation (no grant number applicable). MW receives funding from the DISCLOSURE European Union’s Horizon 2020 European Research Council BYK served on Invitae Corporation’s Advisory Board from Programme [grant number H2020 BRCA-ERC] under Grant 2017 to 2018. IAM has acted on the Advisory Boards for Agreement No. 742432, as well as from the charity The Eve AstraZeneca, Clovis Oncology, Tesaro, Carrick Therapeutics Appeal (https://eveappeal.org.uk/, no grant number appli- and Takeda. His institution receives funding from AstraZe- cable) and support from the National Institute for Health neca. RG is on the Advisory Boards for AstraZeneca, Tesaro, Research (NIHR, no grant number applicable) and the Clovis and Immunogen and does consultancy work for University College London Hospitals (UCLH) Biomedical SOTIO. She has received support to attend conferences Research Centre (no grant number applicable). GEK is from AstraZeneca, Roche and Tesaro. Her institution has supported by the Miriam and Sheldon Adelson Medical received research funding from Boehringer Ingelheim and Research Foundation (no grant number applicable). BYK is Lilly/Ignyta and she is the national co-ordinating investi- funded by the American Cancer Society Early Detection gator for the UK for trials sponsored by AstraZeneca and Professorship [grant number SIOP-06-258-01-COUN] and Tesaro and site principal investigator for trials sponsored by the National Center for Advancing Translational Sciences AstraZeneca, Tesaro, Immunogen, Pfizer, Lilly and Clovis. 1248 https://doi.org/10.1016/j.annonc.2020.05.019 Volume 31 - Issue 9 - 2020 J. Millstein et al. Annals of Oncology PAF has received grants from Novartis, BioNtech and 15. Konecny GE, Haluska P, Janicke F, et al. A phase II, multicenter, ran- Cepheid as well as personal fees from Novartis, Roche, domized, double-blind, placebo-controlled trial of ganitumab or pla- cebo in combination with carboplatin/paclitaxel as front-line therapy Pfizer, Celgene, Daiichi-Sankyo, TEVA, Astra Zeneca, Merck for optimally debulked primary ovarian cancer: the TRIO14 trial. J Clin Sharp & Dohme, Myelo Therapeutics, MacroGenics, Eisai Oncol. 2014;32:5529. and Puma during the conduct of the study. JDB has acted on 16. Konecny GE, Wang C, Hamidi H, et al. Prognostic and therapeutic Advisory Boards for AstraZeneca and has received support relevance of molecular subtypes in high-grade serous ovarian cancer. from GSK to attend conferences. His institution receives J Natl Cancer Inst. 2014;106:dju249. 17. Millstein J, Volfson D. Computationally efficient permutation-based funding from AstraZeneca and Aprea. UM has shares in confidence interval estimation for tail-area FDR. Front Genet. 2013;4: Abcodia Ltd. Sandra Orsulic and Beth Y. Karlan have patents 179. on predictive gene signatures in ovarian cancer 18. Talhouk A, Kommoss S, Mackenzie R, et al. Single-patient molecular (US010253368 and EU2908913). All remaining authors have testing with NanoString nCounter data using a reference-based strat- declared no conflicts of interest. egy for batch effect correction. PLoS One. 2016;11:e0153844. 19. Edgar R, Domrachev M, Lash AE. Gene Expression Omnibus: NCBI gene expression and hybridization array data repository. Nucleic Acids Res. 2002;30:207e210. 20. Talhouk A, George J, Wang C, et al. Development and validation of the REFERENCES gene-expression Predictor of high-grade-serous Ovarian carcinoma 1. Vaughan S, Coward JI, Bast Jr RC, et al. Rethinking ovarian cancer: molecular subTYPE (PrOTYPE). Clin Cancer Res. 2020. https://doi.org/ recommendations for improving outcomes. Nat Rev Cancer. 2011;11: 10.1158/1078-0432.CCR-20-0103. 719e725. 21. Jin C, Xue Y, Li Y, et al. A 2-protein signature predicting clinical outcome 2. Torre LA, Trabert B, DeSantis CE, et al. Ovarian cancer statistics, 2018. in high-grade serous ovarian cancer. Int J Gynecol Cancer. 2018;28:51e CA Cancer J Clin. 2018;68:284e296. 58. 3. Bowtell DD, Bohm S, Ahmed AA, et al. Rethinking ovarian cancer II: 22. Mankoo PK, Shen R, Schultz N, et al. Time to recurrence and survival in reducing mortality from high-grade serous ovarian cancer. Nat Rev serous ovarian tumors predicted from integrated genomic profiles. Cancer. 2015;15:668e679. PLoS One. 2011;6:e24709. 4. du Bois A, Reuss A, Pujade-Lauraine E, et al. Role of surgical outcome 23. Nymoen DA, Hetland Falkenthal TE, Holth A, et al. Expression and as prognostic factor in advanced epithelial ovarian cancer: a combined clinical role of chemoresponse-associated genes in ovarian serous exploratory analysis of 3 prospectively randomized phase 3 multi- carcinoma. Gynecol Oncol. 2015;139:30e39. center trials: by the Arbeitsgemeinschaft Gynaekologische Onkologie 24. Jimenez-Sanchez A, Memon D, Pourpe S, et al. Heterogeneous tumor- Studiengruppe Ovarialkarzinom (AGO-OVAR) and the Groupe d’Inves- immune microenvironments among differentially growing metastases tigateurs Nationaux Pour les Etudes des Cancers de l’Ovaire (GINECO). in an ovarian cancer patient. Cell. 2017;170:927e938.e20. Cancer. 2009;115:1234e1244. 25. Wang C, Cicek MS, Charbonneau B, et al. Tumor hypomethylation at 5. Bolton KL, Chenevix-Trench G, Goh C, et al. Association between 6p21.3 associates with longer time to recurrence of high-grade serous BRCA1 and BRCA2 mutations and survival in women with invasive epithelial ovarian cancer. Cancer Res. 2014;74:3084e3091. epithelial ovarian cancer. JAMA. 2012;307:382e390. 26. Gorbachev AV, Kobayashi H, Kudo D, et al. CXC chemokine ligand 9/ 6. Candido-dos-Reis FJ, Song H, Goode EL, et al. Germline mutation in monokine induced by IFN-gamma production by tumor cells is critical BRCA1 or BRCA2 and ten-year survival for women diagnosed with for T cell-mediated suppression of cutaneous tumors. J Immunol. epithelial ovarian cancer. Clin Cancer Res. 2015;21:652e657. 2007;178:2278e2286. 7. Goode EL, Block MS, Kalli KR, et al. Dose-response association of CD8þ 27. Bronger H, Singer J, Windmuller C, et al. CXCL9 and CXCL10 predict tumor-infiltrating lymphocytes and survival time in high-grade serous survival and are regulated by cyclooxygenase inhibition in advanced ovarian cancer. JAMA Oncol. 2017;3:e173290. serous ovarian cancer. Br J Cancer. 2016;115:553e563. 8. Zhang L, Conejo-Garcia JR, Katsaros D, et al. Intratumoral T cells, 28. Ovarian Tumor Tissue Analysis (OTTA) Consortium, Goode EL, recurrence, and survival in epithelial ovarian cancer. N Engl J Med. Block MS, et al. Dose-response association of CD8þ tumor-infiltrating 2003;348:203e213. lymphocytes and survival time in high-grade serous ovarian cancer. 9. Pujade-Lauraine E, Ledermann JA, Selle F, et al. Olaparib tablets as JAMA Oncol. 2017;3:e173290. maintenance therapy in patients with platinum-sensitive, relapsed 29. Tokunaga R, Zhang W, Naseem M, et al. CXCL9, CXCL10, CXCL11/CXCR3 ovarian cancer and a BRCA1/2 mutation (SOLO2/ENGOT-Ov21): a axis for immune activation e a target for novel cancer therapy. Cancer double-blind, randomised, placebo-controlled, phase 3 trial. Lancet Treat Rev. 2018;63:40e47. Oncol. 2017;18:1274e1284. 30. Xiao P, Guo Y, Zhang H, et al. Myeloid-restricted ablation of Shp2 re- 10. Moore K, Colombo N, Scambia G, et al. Maintenance olaparib in pa- strains melanoma growth by amplifying the reciprocal promotion of tients with newly diagnosed advanced ovarian cancer. N Engl J Med. CXCL9 and IFN-gamma production in tumor microenvironment. 2018;379:2495e2505. Oncogene. 2018;37:5088e5100. 11. Tothill RW, Tinker AV, George J, et al. Novel molecular subtypes of 31. Zhang R, Tian L, Chen LJ, et al. Combination of MIG (CXCL9) chemokine serous and endometrioid ovarian cancer linked to clinical outcome. gene therapy with low-dose cisplatin improves therapeutic efficacy Clin Cancer Res. 2008;14:5198e5208. against murine carcinoma. Gene Ther. 2006;13:1263e1271. 12. Cancer Genome Atlas Research Network. Integrated genomic analyses 32. Zhang W, Ota T, Shridhar V, et al. Network-based survival analysis re- of ovarian carcinoma. Nature. 2011;474:609e615. veals subnetwork signatures for predicting outcomes of ovarian cancer 13. Bonome T, Levine DA, Shih J, et al. A gene signature predicting for treatment. PLoS Comput Biol. 2013;9:e1002975. survival in suboptimally debulked patients with ovarian cancer. Cancer 33. Reinartz S, Finkernagel F, Adhikary T, et al. A transcriptome-based Res. 2008;68:5478e5486. global map of signaling pathways in the ovarian cancer microenvi- 14. Karlan BY, Dering J, Walsh C, et al. POSTN/TGFBI-associated stromal ronment associated with clinical outcome. Genome Biol. 2016;17:108. signature predicts poor prognosis in serous epithelial ovarian cancer. 34. Konecny GE. Combining PARP and CDK4/6 inhibitors in MYC driven Gynecol Oncol. 2014;132:334e342. ovarian cancer. EBioMedicine. 2019;43:9e10. Volume 31 - Issue 9 - 2020 https://doi.org/10.1016/j.annonc.2020.05.019 1249 Annals of Oncology J. Millstein et al. 6 APPENDIX 1. AOCS STUDY GROUP Ovarian Cancer Action Research Centre, Department of Surgery and Cancer, Imperial College London, London, UK; Management Group 7 Centre for Cancer Research, The Westmead Institute for D Bowtell1,3,4,5,6, G Chenevix-Trench2, A Green2, P Webb2, Medical Research, Sydney, Australia; 8The University of A DeFazio7,8,9, D Gertig10 Sydney, Sydney, Australia; 9Department of Gynaecological Oncology, Westmead Hospital, Sydney, Australia; 10Mel- Project and Data Managers bourne School of Population and Global Health, University N Traficante1, S Fereday1, S Moore2, J Hung7, K Harrap2, T of Melbourne, Parkville, Australia; 11John Hunter Hospital, Sadkowsky2, N Pandeya2 Lookout Road, New Lambton, Australia; 12Royal Hospital for Women, Barker Street, Randwick, Australia; 13Royal North Research Nurses and Assistants Shore Hospital, St Leonards, Australia; 14Royal Prince Alfred Hospital, Camperdown, Australia; 15Royal Adelaide Hospital, M Malt2, A Mellon11, R Robertson11, T Vanden Bergh12, M North Terrace, Adelaide, Australia; 16Royal Hobart Hospital, Jones12, P Mackenzie12, J Maidens13, K Nattress14, YE Hobart, Australia; 17Monash Medical Centre, Clayton, Chiew7, A Stenlake9, H Sullivan9, B Alexander2, P Ashover2, S Australia; 18Western Australian Research Tissue Network Brown2, T Corrish2, L Green2, L Jackman2, K Ferguson2, K (WARTN), St John of God Pathology, Osborne Park, Martin2, A Martyn2, B Ranieri2, J White15, V Jayde16, P Australia; 19Women and Infant’s Research Foundation, King Mamers17, L Bowes1, L Galletta1, D Giles1, J Hendley1, K Edward Memorial Hospital, Subiaco, Australia; 20St John of Alsop1, T Schmidt18, H Shirley18, C Ball19, C Young19, S God Hospital, Subiaco, Australia; 21Canberra Hospital, Viduka18, Hoa Tran18, Sanela Bilic18, Lydia Glavinas18, Julia Yamba Drive, Garran, Australia; 22Bankstown Cancer Centre, Brooks20 Bankstown Hospital, Bankstown, Australia; 23Northern Haematology & Oncology Group, Integrated Cancer Centre, Clinical and Scientific Collaborators Wahroonga, Australia; 24Illawarra Shoalhaven Local Health R Stuart-Harris21, F Kirsten22, J Rutovitz23, P Clingan24, A District, Wollongong Hospital, Wollongong, Australia; Glasgow24, A Proietto11, S Braye11, G Otton11, J Shannon25, T 25 Nepean Hospital, Kingswood, Australia; 26Newcastle Bonaventura26, J Stewart26, S Begbie27 M Friedlander28 D Mater Misericordiae Hospital, Waratah, Australia; 27Port Bell13, S Baron-Hay13, A Ferrier13,a, G Gard13, D Nevell13, N Macquarie Base Hospital, Port Macquarie, Australia; Pavlakis13, S Valmadre13, B Young13, C Camaris12, R 28 Prince of Wales Clinical School, University of New South Crouch12, L Edwards12, N Hacker12, D Marsden12, G Rob- Wales, Australia; 29St George Hospital, Kogarah, Australia; ertson12, P Beale14, J Beith14, J Carter14, C Dalrymple14, R 30 St Vincent’s Hospital, Darlinghurst, Australia; 31Wagga Houghton14, P Russell14, M Links29, J Grygiel30, J Hill31, A Wagga Base Hospital, Wagga Wagga, Australia; 32Crown Brand8,32, K Byth32, R Jaworski33, P Harnett8,32, R Princess Mary Cancer Centre, Westmead Hospital, West- Sharma8,33, G Wain32, B Ward34, D Papadimos34, A Cran- mead, Sydney, Australia; 33Department of Pathology, don35, M Cummings35, K Horwood35, A Obermair35, L Per- Westmead Clinical School, Westmead Hospital, The Uni- rin35, D Wyld35, J Nicklin35,36, M Davy15, MK Oehler15, C versity of Sydney, Australia; 34Mater Misericordiae Hospital, Hall15, T Dodd15, T Healy37, K Pittman37, D Henderson37, J Raymond Terrace, South Brisbane, Australia; 35The Royal Miller39, J Pierdes39, P Blomfield16, D Challis16, R McIn- Brisbane and Women’s Hospital, Butterfield Street, Herston, tosh16, A Parker16, B Brown40, R Rome40, D Allen41, P Australia; 36Wesley Hospital, Auchenflower, Australia; Grant41, S Hyde41, R Laurie41, M Robbie41, D Healy17, T 37 Burnside Hospital, Toorak Gardens, Australia; 38Flinders Jobling17, T Manolitsas17, J McNealage17, P Rogers17, B Medical Centre, Bedford Park, Australia; 39Queen Elizabeth Susil17, E Sumithran17, I Simpson17, K Phillips1, D Rischin1, S Hospital, Woodville South, Australia; 40Freemasons Hospital, Fox1, D Johnson1, S Lade1, M Loughrey1, N O’Callaghan1, W East Melbourne, Australia; 41Mercy Hospital for Women, Murray1, P Waring3, V Billson42, J Pyman42, D Neesham42, M Heidelberg, Australia; 42The Royal Women’s Hospital, Park- Quinn42, C Underhill43, R Bell44, LF Ng45, R Blum46, V ville, Australia; 43Border Medical Oncology, Wodonga, Ganju47, I Hammond19, Y Leung19, A McCartney19, a, M Australia; 44Andrew Love Cancer Centre, Geelong, Australia; Buck48, I Haviv49, D Purdie2, D Whiteman2, N Zeps18 45 Ballarat Base Hospital, Ballarat, Australia; 46Bendigo 1 Peter MacCallum Cancer Centre, Melbourne, Australia; Health Care Group, Bendigo, Australia; 47Peninsula Health, 2 QIMR Berghofer Medical Research Institute, Brisbane, Frankston, Australia; 48Mount Hospital, Perth, Australia; Australia; 3Department of Pathology, University of Mel- 49 Faculty of Medicine, Bar-Ilan University, Safed, Israel bourne, Parkville, Australia; 4Sir Peter MacCallum Cancer The seven people in bold are named authors on the Centre Department of Oncology, University of Melbourne, manuscript. Parkville, Australia; 5Department of Biochemistry and Mo- a Deceased. lecular Biology, University of Melbourne, Parkville, Australia; 1250 https://doi.org/10.1016/j.annonc.2020.05.019 Volume 31 - Issue 9 - 2020