Neoadjuvant Chemotherapy Induces Genomic and Transcriptomic Changes in Ovarian Cancer

High-grade serous ovarian cancer (HGSOC) is the most common subtype of ovarian cancer with a high mortality rate. In recent years, the paradigm for the treatment of advanced stage HGSOC has shifted from a focus on primary debulking surgery followed by adjuvant chemotherapy to the more frequent use of neoadjuvant chemotherapy (NACT) followed by an interval debulking surgery. This approach has demonstrated non-inferior outcomes with reduced patient morbidity (1). Most large genomic studies of HGSOC, such as The Cancer Genome Atlas (TCGA), have focused on chemotherapy-naïve tumors and therefore cannot assess chemotherapy's impact on the tumor (2). NACT provides an opportunity to better understand the genomic and transcriptomic effects of chemotherapy by enabling direct comparison of paired untreated and treated tumor samples collected from the same patient.

Although HGSOC is one of the most chemotherapy-sensitive solid tumors, with 80% of women responding to their initial treatment, the majority have a recurrence of metastatic disease (3). There is currently no clinically useful way to predict a patient's response to NACT before treatment; however, genomic and transcriptomic studies have begun to reveal the changes that occur following NACT. In a study that compared 10 patients who received NACT with excellent response and 10 patients who received NACT with poor response, nonsense mutations in CSMD3 and PIK3CA were only seen in patients with poor response (4). Transcriptomic analysis of tumors post-NACT has found increased immune cell transcripts compared with pre-NACT samples, suggesting that chemotherapy response is associated with immune recruitment and remodeling (5). A study of patients with matched pre- and post-NACT tumor samples found that pre-NACT transcriptomes were enriched for cell growth pathways and post-NACT transcriptomes for drug transport and peroxisome pathways (6). Another study that employed targeted panel genomic and transcriptomic analyses found that NACT is associated with changes in cell-cycle progression and DNA damage response (7). A study focused on patients with minimal residual disease after NACT demonstrated upregulation of fatty acid oxidation and epithelial-to-mesenchymal transition signature genes in residual tumor cells (8).

We investigated a cohort of matched pre- and post-NACT tumors using RNA sequencing (RNA-seq) and targeted panel DNA sequencing to better understand the effect of NACT on the genome and transcriptome of HGSOC tumors. Our analyses found evidence of transcriptomic remodeling associated with NACT, as well as for genomic instability that gives rise to recurrent amplifications following NACT. This includes the identification of potentially clinically relevant roles for increased AP-1 signaling and SIK2 amplification in driving the initial responses of tumors to carboplatin and paclitaxel.

One hundred and fifty-three patients with Federation Internationale de Gynecolgie et d'Obstetrique (FIGO) 2013 advanced stage IIIC or IV HGSOC who received NACT between January 2007 and June 2019 were identified in the University of Chicago (Chicago, IL) ovarian cancer database. In accordance with the Common Rule, all patients in this database provided written informed consent under a University of Chicago Institutional Review Board–approved protocol for use of their archived tissue in future research studies. Fifty patients with high-quality tissue samples collected before and after NACT were selected for analysis. All patients received both carboplatin and paclitaxel for NACT. Patients underwent an interval debulking surgery consisting of hysterectomy, bilateral salpingo-oophorectomy, omentectomy, and other intra-abdominal procedures to remove the maximum amount of tumor. Caucasian women made up the majority of the cohort at 58%, followed by African American women at 28%, Hispanic women at 10%, and Asian women at 4%. Platinum response was defined as resistant if recurrence occurred within 6 months of primary therapy or sensitive if more than 6 months elapsed before disease progression. Pre- and post-chemotherapy samples were submitted to Tempus for DNA and RNA-seq.

Formalin-fixed paraffin-embedded (FFPE) tissue blocks were cut into 10 μm sections from the patient's pre-NACT biopsy and interval debulking surgery. Samples were assessed by two gynecologic pathologists (R.R. Lastra, D. Chapel) to confirm high-grade histology, stage, and chemotherapy response score (CRS). Samples were also evaluated by a pathologist to ensure at least 20% tumor cellularity and some specimens were macrodissected to increase tumor cellularity. Germline DNA was collected at the time of surgery from peripheral blood mononuclear cells or FFPE sections free of tumor.

Targeted panel sequencing was performed on FFPE samples using the Tempus xT panel, a hybrid capture next-generation sequencing panel that encompasses clinically relevant and known cancer driver genes (Supplementary Table S1). Total tumor nucleic acid was extracted by digesting sections with proteinase K. Each tumor and germline sample had 100 ng of DNA mechanically sheared by a Covaris ultrasonicator to an average size of 200 bp. Libraries were then prepared using KAPA Hyper Prep Kit, hybridized to the xT probe set, and were then amplified with KAPA HiFi HotStart ReadyMix. Remaining nucleic acid was purified to RNA with DNase-I digestion. One-hundred ng of RNA was input to the Kapa RNA Hyperprep kit hybridized to IDT xGen Exome Research Panel version 1 probes. Amplification of cDNA was performed with KAPA HiFi HotStart ReadyMix. Sequencing to an average target depth of 500× was performed on Illumina HiSeq 4000 (version 2) or NovaSeq6000 (version 3). For quality control, each sample was required to have 95% of target bases sequenced to at least 300×.

Germline data were filtered using the Broad institute hard filter recommendations. Briefly, sequence quality was evaluated by FastQC (v0.11.8, RRID:SCR_014583) and sequences trimmed with Trimmomatic (v0.36, RRID:SCR_011848). Trimmed sample sequence files were mapped to GRCh38 with BWA MEM (v0.7.17, RRID:SCR_010910) and merged. Duplicates were marked with sambamba (v0.6.8) and base quality score recalibration performed using the GATK 4.1.6 tools BaseRecalibrator and ApplyBQSR (RRID:SCR_001876). Germline variants of known clinical interest in ovarian cancer were selected and cBioportal's (SCR_014555) MutationMapper used to generate lollipop plots. Somatic data were annotated with ANNOVAR and reads passing filters were selected for analysis. This included mutations with a variant allele frequency (VAF) of >20% and population variant allele frequency of <5% in women. To estimate mutational burden, exonic single-nucleotide variants with a VAF > 5% that were not in the COSMIC database (RRID:SCR_002260) were included. Copy-number variations (CNV) were detected from panel sequencing data using SynthEx (v1.0.5) and significantly enriched CNVs detected using GISTIC2.0 (v2.0.23, RRID:SCR_000151).

RNA-seq data were aligned and quantified using Kallisto (v0.46.1, RRID:SCR_016582) against human genome build hg38. Raw counts were normalized using reads per kilobase million (RPKM). Given high interpatient heterogeneity, GFOLD was used to determine which genes were differentially expressed in before and after chemotherapy samples. GFOLD (0.01) was used, meaning in 99% of cases the fold change of a gene is above the calculated GFOLD value. A cut off of >±2 was used to consider expression significantly altered by chemotherapy. STRING database (RRID:SCR_005223) was used for extracting and visualizing gene ontology (GO) terms. R version 4.0.2 (RRID:SCR_000432) with consensusOV package was used to determine molecular subtypes; Complex Heatmap (RRID:SCR_017270) and ggplot2 (RRID:SCR_014601) packages were used for visualizations. Genes in the AP-1 pathway were extracted from the Pathway Interaction Database (PID_AP1_PATHWAY; M167). Kaplan–Meier survival curves were generated with the Kaplan–Meier Plotter using HGSOC datasets GSE14767, GSE15622, GSE18520, GSE19829, GSE23554, GSE26193, GSE26712, GSE26751, GSE30161, GSE3149, GSE51373, GSE63885, GSE65986, GSE9891, and TCGA.

OVCAR-8 cells were obtained from Marcus Peter (Northwestern University, Chicago, IL, RRID: CVCL_IM65), TYK-nu cells from Gottfried Konecny (Ronald Reagan UCLA Medical Center, RRID:CVCL_1776), OVCAR-4 cells from Charles River Fredrick National Laboratory for Cancer Research (RRID:CVCL_1627), and PEO-1 cells from Scott Kaufmann (Mayo Clinic, RRID:CVCL_2686). OVCAR-8 cells were grown in DMEM (10-013-CV; Corning) with 10% FBS (35-010-CV, FBS; Thermo Fisher Scientific), 1% MEM nonessential amino acids (25-025-CI; Corning) and 1% MEM Vitamins (25-020-CI; Corning). TYK-nu cells were grown in MEM (10-022-CV; Corning) with 10% FBS, 1% MEM nonessential amino acids (25-025-CI; Corning), and 1% MEM Vitamins (25-020-CI; Corning). OVCAR-4 cells were grown in RPMI with 10% FBS (35-010-CV, FBS; Thermo Fisher Scientific). PEO-1 cells were grown in DMEM (10-013-CV; Corning) with 10% FBS (35-010-CV, FBS; Thermo Fisher Scientific), 1% MEM nonessential amino acids (25-025-CI; Corning), and 1% MEM Vitamins (25-020-CI; Corning). Cells were grown in 5% CO₂ incubator at 37°C. All cells were Mycoplasma negative and their identity verified regularly by short tandem repeat profiling (IDEXX; OVCAR-8 and TYK-nu Feb 2020, OVCAR-4 June 2019, PEO-1 July 2019). Cells were passaged 2–10 times after unfreezing prior to performing experiments.

c-Jun antibody (1:1,000, ab31419, Abcam, RRID:AB_731605) or c-Jun antibody (1:1,000, 2315, Cell Signaling Technology, RRID:AB_490780), c-Fos antibody (1:500, sc-166940, Santa Cruz Biotechnology, RRID:AB_10609634), anti-thrombospondin 1 (THBS1) antibody (1:100, sc-59887, Santa Cruz Biotechnology, RRID:AB_793045), GAPDH antibody (1:1,000, 2118, Cell Signaling Technology, RRID:AB_561053), SIK2 antibody (1:1,000, 6919S, Cell Signaling Technology, RRID:AB_10830063), phospho-Akt (Ser473) antibody (1:1,000, 9271, Cell Signaling Technology, RRID:AB_329825), and AKT antibody (1:1,000, 9272, Cell Signaling Technology, RRID:AB_329827).

OVCAR-8 cells (i) treated with 5 μmol/L of SIK2 inhibitor (HG-9-91-01, MedChem Express, HY-15776) for the indicated time, (ii) with stable knockout of SIK2, or (iii) OVCAR-8 cells transfected with SIK2 siRNA or NT siRNA (30 nmol/L; 24 hours, Dharmacon), were treated with carboplatin (100 μmol/L; 72 hours) or paclitaxel (100 nmol/L; 24 hours). Cells were then trypsinized and washed with staining buffer (BD, 554657). A total of 500,000 cells were resuspended in binding buffer (Apoptosis Detection Kit, BioLegend, 640932) and stained with Annexin V (APC) and PI (PE) for 15 minutes at room temperature. Flow cytometry analysis was then carried out using CytoFLEX Flow Cytometer (Beckmann Coulter). Data analysis was carried out using FlowJo 10.7.1 (RRID:SCR_008520).

OVCAR-8 cells were seeded at 10,000 cells per well in a black-wall 96-well plate and allowed to adhere for 24 hours before addition of treatment [paclitaxel (100 nmol/L), carboplatin (100 μmol/L), SIK2 inhibitor (5 μmol/L)]. After 24 or 48 hours, nuclei were visualized by addition of Hoechst 33258 (1:5,000) and wells fluorescently imaged with a Nikon Eclipse Ti2 to extract the Hoechst-positive area fraction.

FFPE tissue sections (5 μm) were deparaffinized with xylene and rehydrated with a graded series of ethanol. For THBS1, the slides were stained using Leica Bond RX automated stainer. Epitope retrieval solution I (Leica Biosystems, AR9640) was used for 20-minute treatment. Anti-THBS1 (1:100) was applied on tissue sections for one-hour incubation and the antigen-antibody binding was detected with Bond polymer refine detection (Leica Biosystems, DS9800). For trichrome stain, slides were placed in Bouin solution at 55°C–66°C for 1 hour and then washed in running water until yellow color disappeared. Next, slides were stained with Weigert Iron Hematoxylin for 3 minutes and washed in running water for 3 minutes. Then, slides were stained with Biebrich Scarlet-Acid Fuschin solution for 3 minutes and rinsed with deionized water. Finally, slides were differentiated in phosphotungstic-phosphomolybdic acid for 3 minutes and immersed slides into aniline blue solution for 5 minutes.

Cells were lysed using SDS lysis buffer, and protein estimation was carried out using a BCA Protein Assay Kit (Thermo Fisher Scientific, 23224). Equal protein amounts were resolved using 4%–20% SDS gel and transferred onto nitrocellulose membrane. Signals were detected using 1:3,000 dilutions of either horseradish peroxide–linked anti-mouse (Cell Signaling Technology, 7076, RRID:AB_330924) or anti-rabbit (Cell Signaling Technology, 7074, RRID:AB_2099233) secondary antibodies and subsequently developed with ECL substrate (Bio-Rad, 1705061) or Pico (Thermo Fisher Scientific, 34579). Images were acquired using G: Box (Syngene).

A total of 200,000 OVCAR-8 cells in 6-well plates were cotransfected with AP-1 reporter (1.5 μg pAP-1-Luc plasmid containing seven AP-1 enhancer elements; Agilent) and pRL-CMV-Renilla plasmids (0.1 ng; Promega) using 6 μL of Lipofectamine 2000. Cells were treated with paclitaxel or carboplatin at indicated treatment for 24 hours 2 days after transfection. Cells were lysed, and luminescence measured using a Berthold Technologies Lumat LB 9507 per manufacturer's instructions (Promega Dual Luciferase Reporter Assay).

For inhibitor assays, 10,000 cells were seeded into 96-well plates on day 1 and treated with chemotherapy and inhibitor on day 2. MTT assay (Thiazolyl blue tetrazolium bromide, M2128, Sigma) was performed 72 hours after treatment. TYK-nu and OVCAR-8 cells were treated with constant ratios of drug combinations (carboplatin and SR11302; paclitaxel and SR11302) in serial dilutions and combination indices calculated with Compusyn software.

The data generated in this study are available within the article and its Supplementary Data files. Expression profile data analyzed in this study were obtained from Gene Expression Omnibus at GSE14767, GSE15622, GSE18520, GSE19829, GSE23554, GSE26193, GSE26712, GSE26751, GSE30161, GSE3149, GSE51373, GSE63885, GSE65986, GSE9891.

We identified a cohort of 50 consecutive patients from our ovarian cancer database with advanced stage IIIC and IV HGSOC treated with neoadjuvant carboplatin and paclitaxel. These women were not eligible for primary debulking surgery because of their significant tumor burden on imaging and clinical exam. Tumor tissue was biopsied before NACT and at the interval debulking surgery. After removal of cases that were exceptional responders (9) with minimal tumor tissue at the second surgery and tumors that failed library preparation or sequencing (Materials and Methods), the final patient cohort consisted of 32 paired DNA and 20 paired RNA samples from pre- and post-NACT, with 18 patients contributing both paired DNA and RNA, 14 patients only contributing DNA, and 2 patients only contributing RNA (Fig. 1A; Supplementary Fig. S1A; Table 1; Supplementary Table S2 and S3).

Three patients (10%) carried a BRCA1 and 1 patient (4%) a BRCA2 germline mutation (Supplementary Fig. S1B); both BRCA1 patients were chemotherapy sensitive and the BRCA2 patient was chemotherapy resistant. Patients were nearly evenly split between platinum sensitive (>6 months to recurrence after the end of chemotherapy) and resistant (<6 months to recurrence) disease. Whole transcriptome t-SNE analysis established that clustering of samples was driven by pre- or post-chemotherapy status, but not chemotherapy sensitivity (Fig. 1B). Increased epithelial marker expression was not correlated with a lower pathologic tumor content (Supplementary Fig. S1C). We determined each patient's TCGA molecular subtype (10) pre- and post-NACT and found that 50% of patients underwent a subtype change (Fig. 1C) consistent with previous reports (6). There was no correlation between chemotherapy resistance and pre-NACT TCGA subtype (P = 0.9), post-NACT subtype (P = 0.4), or change in subtype (P = 0.6). As HGSOC molecular subtypes were initially identified from chemotherapy-naïve patient specimens and are imperfect prognostic indicators (2), these data suggest that tumors are highly plastic and surviving tumor cells adapt to chemotherapy by changing gene expression (11).

Analyzing all patients, 77 genes were upregulated after chemotherapy and 15 genes were downregulated (Fig. 2A). To identify differentially expressed genes (DEG) associated with chemotherapy response, we performed a paired analysis of each patient's pre- and post-NACT sample using GFOLD (Supplementary Tables S4–S8). When samples were clustered on the basis of read counts, clustering occured by before or after chemotherapy status and not chemotherapy response. However, when clustering was based on DEGs, the samples did cluster by response as well as chemotherapy status (Fig. 2A; Supplementary Fig. S2A). We validated previous observations that FOS, NR4A1, NR4A3, NTRK2, SFRP2, and SFRP4 are upregulated following NACT (7). When categorizing patients by platinum sensitivity, we identified 141 genes exclusively upregulated in chemotherapy-resistant patients compared with only seven genes in chemotherapy-sensitive patients (Supplementary Table S9), consistent with findings in triple-negative breast cancer that shares many genomic features with HGSOC (12). Genes downregulated following NACT included several genes associated with DNA replication and the cell cycle (e.g., MKI67, MCM10, TOP2A, and CENP). Upregulated genes included extracellular matrix genes (e.g., COL14A1, THBS1; Fig. 2B), metabolic genes (e.g., ALDH1A1, FABP4, AOX1), cytokines (e.g., CCL14 and IL6), and transcriptional regulators. In particular, genes involved in the AP-1 transcription factor network, including FOS, FOSL1, and FOSB, showed consistently high expression following NACT (Fig. 2C; Supplementary Fig. S2B and S2C). Supporting a functional role of AP-1 family members in chemotherapy resistance, we found that the core AP-1 genes were more highly expressed in chemoresistant patients (Supplementary Fig. S2D; Supplementary Table S10). Consistent with our findings, Jiménez-Sánchez and colleagues (5) identified similar increases in the expression of AP-1 pathway genes in platinum resistant patients following chemotherapy using a microarray (Supplementary Fig. S2E).

On the basis of the consistently elevated expression of AP-1 family members in the chemoresistant post-treatment samples, we further investigated their functional role in mediating chemoresistance. The activator protein 1 (AP-1) transcription factor is a dimeric leucine-zipper protein composed of elements from the JUN, FOS, ATF, and MAF protein families, which play important roles in cell proliferation and invasion (13). While AP-1 expression is predominantly tumor promoting, some family members can act as tumor suppressors depending on heterodimer composition, cell type, and tumor microenvironment (14). Treatment of HGSOC cell lines with carboplatin and paclitaxel led to an increase in total and phosphorylated c-Jun (Fig. 2D; Supplementary Fig. S3A and S3B). Platinum-resistant cell lines generated by long-term, in vitro treatment of OVCAR-4, TYK-nu, and PEO1 cells with carboplatin had significantly elevated levels of c-Jun phosphorylation (Fig. 2E). Paclitaxel, but not carboplatin, increased AP-1 transcriptional activity, as assessed with a luciferase reporter system (Fig. 2F; Supplementary Fig. S3C). Treatment of ovarian cancer cells with an AP-1 inhibitor [SR11302 (15) ] combined with paclitaxel or carboplatin led to a synergistic increase in cell death (Fig. 2G; Supplementary Fig. S3D), which was present at lower kill fractions for paclitaxel than carboplatin. Moreover, elevated expression of both c-Jun and c-Fos are consistently associated with worse overall and progression-free survival in several large-scale transcriptomic analyses of patients with epithelial ovarian cancer (Supplementary Fig. S3E–S3H). Combined, these data suggest that both carboplatin and taxane-based chemotherapies select cells with elevated AP-1 transcriptional activity.

Panel sequencing of cancer-associated genes identified between one and 14 nonsynonymous mutations per patient; 12 genes were mutated in 2 or more patients (Fig. 3A; Supplementary Tables S11 and S12). No gene demonstrated a recurrent gain in mutation following NACT, suggesting that NACT does not recurrently alter specific genes. There were no significant differences in average mutational burden pre- and post-NACT and mutational burden was not associated with chemotherapy resistance or CRS (Supplementary Fig. S4A). Patients showed both an increase and decrease in mutational burden following NACT, which is in contrast to breast cancer, which demonstrates either no change or decrease in mutations (16). Consistent with other studies, we found that the most common mutational class was driven by the aging-related C>T transition signature caused by spontaneous deamination of 5-methylcytosine (Fig. 3B; ref. 9). As expected, TP53 mutations were the primary recurrent mutation found, identified in all but 1 patient. Consistent with TCGA and other analyses, we identified mutations in NF1 (9%), BRCA1 (3%), and BRCA2 (3%) (2) in several patients; other mutations present in multiple patients included APOB, MTOR, CREBBP, NRG1, PAX3, and SALL2. Reduced NF1 expression is associated with decreased 5-year survival and chemotherapy resistance in multiple cancer types, while APOB (9%) is recurrently mutated in hepatocellular carcinoma, and MTOR (9%) mutations are common in renal cell cancers. CREBBP mutations in acute lymphoblastic leukemia, NRG1 fusions in non–small cell lung cancer, PAX3 overexpression in gliomas, and epigenetic silencing of SALL2 in breast cancer are all genomic events that have been associated with chemotherapy resistance (17).

Copy-number profiles of patient samples generally recapitulated those identified in TCGA and other analyses (2), with frequent amplification of chromosome 3q and 8q and deletion of chromosomes 16 and 17 (Fig. 4A; Supplementary Table S13). In general, both pre- and post-NACT specimens were characterized by widespread genomic instability. To identify recurrently amplified and deleted chromosomal regions, we performed CNV analysis with GISTIC2.0 (18) software tool. We identified several regions that were recurrently amplified or deleted following NACT, emphasizing that NACT leads to the induction or selection of conserved copy-number alterations in HGSOC. The most prominent were amplification of 2q33.1 and 11q23.1 and deletion of 19p13.3 and 2q35 (Fig. 4B). Relatively few to no copy-number gains or losses were associated with response to NACT or associated with recurrence 1 year after NACT, independent of whether pre- or post-NACT samples were interrogated (Supplementary Fig. S4B). The most significant amplification was noted in 11q23.1, which encompasses salt-inducible kinase 2 (SIK2), a serine/threonine kinase that is associated with paclitaxel and platinum resistance in ovarian cancer (19, 20) (Supplementary Table S14). Genomic amplification of SIK2 has not been described as a mechanism driving SIK2 upregulation, but our analysis shows that the expression of SIK2 is strongly associated with copy-number status (P < 0.0001; Fig. 4C; Supplementary Fig. S4C). The treatment of OVCAR-8 HGSOC cells with carboplatin or paclitaxel and a SIK2 inhibitor led to a significant synergistic increase in cell death (Fig. 4D). Addition of SIK2 inhibitor HG-9–91–01 (19) or blocking SIK2 with siRNA against SIK2 inhibited AKT activity (Supplementary Fig. S4D). Genomic amplification of SIK2 in response to NACT may be a novel mechanism by which HGSOC evolves resistance to cytotoxic therapies.

Using a combination of transcriptomic and genomic analyses, we found that NACT is associated with significant transcriptomic remodeling, including upregulation of AP-1 transcriptional networks. Although tumors continue to evolve mutations during NACT, no recurrently mutated genes associated with NACT were identified in our patient cohort. In contrast, we found evidence that NACT is associated with recurrent copy-number alterations, including amplification of 11q23.1, which is associated with increased expression of SIK2. The selection criteria for NACT is inconsistent, but patients triaged to NACT typically have a higher disease burden, increased ascites, and are more medical complex than primary debulking surgery candidates (1). Although our study provides insight into the molecular mechanisms underlying chemotherapy response in HGSOC, fully understanding the adaptive and evolutionary processes underlying the response of tumor cells to NACT will require an analysis of clonal dynamics and epigenetic processes. In particular, single-cell techniques will be useful to understand NACT-associated transcriptional remodeling by unraveling the selection pressure of NACT on pre-existing tumor cell populations from the induction of specific gene expression programs in tumor cells (or stromal) subpopulations.

Our data suggest that AP-1 activity is induced by chemotherapy and may contribute to the survival of cancer cells in response to cytotoxic agents. AP-1 is a druggable target for which multiple classes of small-molecule inhibitors have been developed. Several AP-1 inhibitors have undergone clinical testing for inflammatory conditions, including rheumatoid arthritis. Therefore, it is possible to consider therapeutically targeting AP-1 in the context of NACT for the purpose of overcoming chemoresistance. In addition, work from our group and others have demonstrated that SIK2 is a therapeutically tractable mediator of platinum and paclitaxel sensitivity in ovarian cancer. Indeed, a small-molecule inhibitor of SIK2 (GRN-300) is currently under investigation for the treatment of recurrent ovarian, primary peritoneal, and fallopian tube cancers (NCT04711161).

M. Javellana reports nonfinancial support from Tempus during the conduct of the study. J. Heide reports grants from DFG (German Research Foundation) during the conduct of the study. D.B. Chapel reports grants from Ovarian Cancer Research Alliance outside the submitted work. S. Yamada reports other support from Bears Care and Harris Family during the conduct of the study and personal fees from ABOG outside the submitted work. A.A. Ahmed reports personal fees from Singula Bio Ltd outside the submitted work. E. Lengyel reports other support from Tempus during the conduct of the study and grants from AbbVie and Arsenal Bioscience outside the submitted work. No disclosures were reported by the other authors.

M. Javellana: Conceptualization, data curation, formal analysis, investigation, visualization, writing–original draft. M.A. Eckert: Conceptualization, formal analysis, supervision, funding acquisition, investigation, visualization, writing–original draft. J. Heide: Investigation, writing–original draft. K. Zawieracz: Investigation, writing–review and editing. M. Weigert: Investigation, writing–review and editing. S. Ashley: Investigation. E. Stock: Investigation, writing–review and editing. D. Chapel: Investigation, writing–review and editing. L. Huang: Software, formal analysis. S.D. Yamada: Writing–review and editing. A.A. Ahmed: Investigation, writing–review and editing. R.R. Lastra: Investigation, writing–review and editing. M. Chen: Software, formal analysis, supervision, writing–review and editing. E. Lengyel: Conceptualization, resources, supervision, funding acquisition, writing–review and editing.

The authors thank Gail Isenberg for help with editing the article. They appreciate Martin Miller Ph.D. Cancer Systems Biology Lab, CRUK, Cambridge University for sharing RNA-seq results. This project was supported by a grant from the NCI (R01CA169604 to E. Lengyel) and funding from Tempus for panel sequencing (E. Lengyel and M. Eckert).

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.