Mechanisms of intrinsic resistance of serous ovarian cancers to standard treatment with carboplatin and paclitaxel are poorly understood. Seventeen primary serous ovarian cancers classified as responders or nonresponders to standard treatment were screened with DigiWest protein array analysis for 279 analytes. Histone methyl transferase EZH2, an interaction partner of ataxia telangiectasia mutated (ATM), was found as one of the most significantly represented proteins in responsive tumors. Survival analysis of 616 patients confirmed a better outcome in patients with high EZH2 expression, but a worse outcome in patients with low EZH2 and high-ATM–expressing tumors compared with patients with low EZH2 and low-ATM–expressing tumors. A proximity ligation assay further confirmed an association between ATM and EZH2 in tumors of patients with an increased disease-free survival. Knockdown of EZH2 resulted in treatment-resistant cells, but suppression of both EZH2 and ATM, or ATM alone, had no effect. DigiWest protein analysis of EZH2-knockdown cells revealed a decrease in proteins involved in mitotic processes and checkpoint regulation, suggesting that deregulated ATM may induce treatment resistance.
Ovarian cancer is a malignancy with high mortality rates, with to date, no successful molecular characterization strategies. Our study uncovers in a comprehensive approach the involvement of checkpoint regulation via ATM and EZH2, potentially providing a new therapeutic perspective for further investigations.
Ovarian cancer is the eighth most commonly diagnosed cancer among women in the world and the seventh leading cause of cancer-related death among women (1), with serous carcinomas being the most common type of ovarian cancer (2). The detection of disease often takes place at a late stage, with therapy consisting of the surgical removal of the tumor and an adjuvant chemotherapy including carboplatin and paclitaxel (3, 4). Carboplatin intercalates into DNA and inhibits DNA replication in highly proliferating cancer cells (5). Paclitaxel binds to tubulin structures of the microtubule, thereby preventing the disassembly during mitosis, a process needed for proper cell division (6). Approximately 10% of the patients with ovarian cancer show an early disease progression (7). A progress or relapse within 6 months of therapy is considered as “resistant,” with 1 year of disease-free survival (DFS) being clinically classified as “sensitive” to platinum treatment (8).
The molecular basis for this intrinsic resistance of ovarian cancer remains, to date, unknown, with various attempts to identify candidate proteins in the primary tumor tissue that could potentially predict therapeutic outcome yielding, arguably, very few results (9). Comprehensive studies on DNA-sequencing data obtained from hundreds of tumors identified somatic TP53 mutations in 96% of ovarian cancer (10). In addition, BRCA mutations were assigned to tumors being sensitive to a platinum-based therapy (11). Most of these studies, however, focused primarily on platinum resistance, rendering them obsolete as the current standard treatment is a combination of platinum-based therapeutics and paclitaxel. Conventional chemotherapeutic agents functionally require proliferative cancer cells, this is why cells that can temporarily arrest their growth are able to evade the treatment. This has led to the emergence of cell-cycle checkpoint inhibitors (PARP, ATR, CHK1, and Wee1) in the treatment of cancers (12). One protein which is involved in DNA repair and checkpoint induction is the serine/threonine kinase ATM (ataxia telangiectasia mutated; ref. 13). The histone-lysine N-methyltransferase EZH2 (Enhancer Of Zeste 2 Polycomb Repressive Complex 2 Subunit) was recently identified as one of ATM's numerous targets (14).
This study utilizes a novel method for highly sensitive protein analysis, enabling detection of differentially activated signal transduction pathways in therapy responsive and nonresponsive tumors treated with carboplatin and paclitaxel. Elucidating intrinsic molecular mechanisms could potentially lead to novel targets for the treatment of resistant serous ovarian carcinomas.
Materials and Methods
For protein array analysis and DNA sequencing, primary serous tumors from 17 patients with ovarian cancer were used. All patients were under the age of 75 and treated adjuvantly with carboplatin and paclitaxel. Eight patients were classified as non-responders, i.e. they suffered from relapse or metastasis, or they had died in between 12 months from date of surgery. A group of nine patients was classified as responders due to disease-free survival of more than 12 months after surgery. Formalin-fixed, paraffin-embedded (FFPE) tissues (n = 2) of 34 patients were used as independent patient cohort for IHC staining and proximity ligation assay. Twenty-two patients were classified as responders, 12 as nonresponders according to the same criteria.
DigiWest protein analysis
Fresh frozen tumor tissue (1–3 mm3) was used for protein array analysis via DigiWest technology (15). Raw data were logarithmized and analyzed using MultiExperimentViewer Software 4.9.0. Tumor samples assigned as nonresponsive were compared with responsive samples using the Wilcoxon–Mann–Whitney test. Significant proteins or protein modifications (P ≤ 0.05) and patient samples were arranged by hierarchical clustering according to Pearson correlation using an average linkage method. Pathway analysis was performed using browser-based PANTHER overrepresentation test, based on the Gene Ontology database. Biological processes were assigned using Fisher exact test and a FDR below 0.05.
Next-generation sequencing—based mutation detection
Coding regions of the genes ATM, BRCA1, BRCA2, BRIP1, CHEK2, MRE11, PALB2, RAD51C, RAD51D, and TP53 from 17 ovarian cancers were analyzed using capture- or amplicon-based next-generation sequencing (NGS) on Illumina MiSeq with a coverage of more than 50×.
Reads were mapped using Homo sapiens reference genome GRCh37/hg19. Variants were named according to the Human Genome Variation Society. Variants of more than 18% allele frequency were taken into account. Mutations were stated as deleterious if annotated as pathogenic by ClinVar. Exome Aggregation Consortium (ExAC) population frequencies were listed by ClinVar as well.
In silico analysis
Publicly available datasets of mRNA expression were used for survival analysis with log rank test based on data provided by Kaplan–Meier plotter (www.kmplot.com; ref. 16). The available datasets GSE14764, GSE26193, GSE30161, GSE32062, GSE63885, GSE9891, and The Cancer Genome Atlas (TCGA) were utilized, comprising 616 patients. All tumors were of serous histology, and all patients received a chemotherapy containing platinum and taxol. Patients were censored if no event occurred during the observation period. A follow-up threshold of 60 months was set. Automated dichotomization for EZH2 and ATM alone was performed using best cutoff selection. Survival analysis with EZH2 high and low and ATM high and low samples was performed on median dichotomization. Log rank comparisons were conducted pairwise for any constellation.
IHC on FFPE tissues was performed using primary antibodies against EZH2 (1:50, RRID: AB_10694683), ATM (1:100, RRID: AB_725574), Rb pS807/811 (1:100, RRID: AB_11178658), and ERK1/2 pT202/Y204 (1:400, RRID: AB_2315112). Rating was done using the immunoreactive score (IRS; percentage of positive cells × intensity of staining) to compare patient groups (17). Log rank survival analysis was done after dichotomization to lower quartile.
Proximity ligation assay
Proximity ligation assay (PLA) was performed using Duolink In Situ Detection Reagents FarRed (DUO92013, Sigma-Aldrich) for fixed tissues and Duolink In Situ Detection Reagents Orange (DUO92102, Sigma-Aldrich) was used for cell lines. Procedure was done according to the manufacturer's advice using antibodies against EZH2 (1:150, RRID: AB_10694683) and ATM pS1981 (1:500, RRID: AB_725573). In tissues, counting of spots was done on a predefined area and normalized to tumor content of the samples. Group comparison was done using Wilcoxon–Mann–Whitney test. Log rank survival analysis was performed with DFS. Lower quartile was used for dichotomization.
The A2780 cells were treated with sublethal concentrations of agents. After performing the PLA, spots per nucleus were counted. The comparison with nontreated cells was performed using one-way ANOVA with Dunnett multiple comparison test.
Cell culturing and experiments were performed at 37°C and 5% CO2 without antibiotics. A2780 (RRID: CVCL_0134), A2780cis (RRID: CVCL_1942), and IGROV1 cells (RRID: CVCL_1304) were cultured in RPMI1640 with 10% FCS. Caov3 cells (RRID: CVCL_0201) were cultured in DMEM with 10% FCS, and Ovcar3 cells (RRID: CVCL_0465) in RPMI1640 medium with 0.01 mg/mL insulin and 20% FCS. PEO1 (RRID: CVCL_2686) and PEO4 cells (RRID: CVCL_2690) were cultured in RPMI1640 with 2 mmol/L sodium pyruvate and 10% FCS. SKOV3 cells (RRID: CVCL_0532) were grown in modified McCoy 5a Medium with 10% FCS. All cell lines were regularly tested for Mycoplasma infection. Recent external authentication by short tandem repeat analysis confirmed the origin of the cells.
Compounds and cell viability assay
Carboplatin (carboplatin hospira, stock 10 mg/mL, Pfizer Pharma) and paclitaxel (taxol, NeoTaxan containing the active component paclitaxel; stock 6mg/mL, Hexal AG) were used to treat cell line cells to determine the resistance status.
For the viability assay, cells were seeded at a density of 7,500 cells in 96-well plates. A range of treatment concentrations were generated by successive dilution series, in accordance to the treatment ratio of carboplatin to paclitaxel (2.63:1) found in patients with ovarian cancer during treatment (18). Cells were exposed to compounds for 3 days and endpoint measurements were performed using ATP Bioluminescence Assay Kit HS II (Roche), according to the manufacturer's protocol. The IC50 values were calculated using nonlinear regression analysis in GraphPad Prism. The detection limit for IC50 calculation was 0.5 nmol/L.
Western blot analysis
Equal protein loads of whole-cell lysates were run on SDS-PAGE gels, transferred to polyvinylidene difluoride membranes, and probed with primary antibodies against EZH2 (1:1,000, RRID: AB_10694683), ATM (1:3,000, RRID: AB_725574), and GAPDH (1:1,000, RRID: AB_783595). Multiplex fluorescent labeling was performed using secondary donkey antibodies against goat (AF546, RRID: AB_2534103) or rabbit (AF647, RRID: AB_2536183) antibodies. Target proteins were analyzed densitometrically and normalized to the GAPDH signal. The mean IC50 value for every cell line was then used for Spearman correlation analysis.
Knockdown of EZH2 and ATM
Knockdown of EZH2 and ATM was performed using siRNA Qiagen FlexiTube GeneSolution GS2146 for EZH2 and GS472 for ATM (Qiagen). Transfection was carried out with HiPerFect Transfection Reagent (Qiagen) according to the manufacturer's guidelines. Samples for checking the effectiveness of the knockdown were collected first on day 3 before renewing small interfering RNA (siRNA) content and the start of the viability assays, and then also on day 6 after endpoint measurement. Nontargeting siRNAs were used as negative control (Qiagen). Values of each concentration were compared with negative control siRNA using two-way ANOVA with Bonferroni posttest.
Image processing was performed using ImageJ 1.50e software. Heatmaps and protein array comparisons were performed using MultiExperimentViewer software 4.9.0. Other statistical analysis and data representation were performed using GraphPad Prism version 5.03 for Windows.
A detailed methods description can be found in the Supplementary Materials and Methods.
This study was carried out in accordance with the Good Clinical Practice guidelines and was approved by the Ethics Committee of the Medical Faculty of the Eberhard Karls University Tübingen (04/2007; 266/1998; and 397/2006) and Charité Berlin (EK207/2003). All patients enrolled in this study provided written informed consent.
Mitotic proteins are increased in responsive tumors
Seventeen primary tumors of patients with ovarian cancer were obtained for this study. Eight of these patients with a DFS of less than 12 months subsequent to surgery were classified as “nonresponders,” whereas 9 patients, exceeding the 12 months of DFS, were classified as “responders” (Supplementary Table S1). The screening of the tumors was performed with DigiWest technology, an antibody-based protein array analysis, to identify differentially activated signal transduction pathways in tumors between the two groups of patients. A total of 279 proteins and protein modifications (Supplementary Table S2) were quantified in each tumor and statistically compared between responders and nonresponders. A Wilcoxon–Mann–Whitney test revealed 30 differentially phosphorylated and expressed proteins (Fig. 1A; Supplementary Fig. S1). In responders, a significant increase of phosphorylated retinoblastoma protein (Rb pS807/811), EZH2 and the cell-cycle protein Cyclin B1 (Supplementary Table S3) were observed, whereas in nonresponders phosphorylated ribosomal protein S6 (pS235/236) was significantly increased. Hierarchical clustering of the significantly differential expressed and modified proteins revealed an activated MAPK/ERK pathway leading to an activation of ribosomal protein S6 in nonresponders, whereas proteins prominent in cell-cycle progression and mitosis, that is, pRb, Cyclin B1, PLK1, and TOP2 were found predominantly elevated or phosphorylated in responders. A pathway analysis linked protein clusters of the heatmap to cell-cycle processes.
Detection of DNA variants in ovarian cancers
The same 17 primary tumors of patients with ovarian cancer were screened to identify alterations in the coding regions of genes expressing DNA damage repair proteins, presumably involved in repair of DNA damages caused by platinum in cells.
Thus, the mutational status of the genes ATM, BRCA1, BRCA2, BRIP1, CHEK2, MRE11, PALB2, RAD51C, RAD51D, and TP53 was determined to associate potential changes in signaling pathways to variations in these genes. Alterations leading to missense, nonsense, frameshift, or splice site mutations were taken into account. A total of 94% of the samples harbored a known deleterious TP53 mutation (Supplementary Table S4). Variants of ATM were accumulated in nonresponder's tumors. There is, however, no evidence for a functional relevance regarding therapy responsiveness except for one variant classified as a polymorphism (ATM p.Asp1853Asn), which is associated with a decreased progression-free survival (PFS) in patients treated with cisplatin and paclitaxel (19). Variants in the other genes (BRCA1, BRCA2, BRIP1, PALB2, RAD51C, and RAD51D) could not be assigned to one of the two groups. No variants were found in CHEK2 or MRE11.
As ATM is associated with cell-cycle progression, and because it is known to interact with one of the candidate proteins EZH2 (14), the ATM/EZH2 protein ratio of every sample was determined. This revealed high ratios in nonresponders (Supplementary Fig. S2). Further experiments were conducted to elucidate the cooperation of both proteins in serous ovarian cancers.
Worse outcome in patients with ovarian cancer with low EZH2 and high ATM mRNA expression in tumors
In silico validation was obtained using mRNA data of the TCGA and the Genomic Spatial Event datasets from patients who underwent platinum- and paclitaxel-based chemotherapy. Data analyses confirmed a significant correlation between a better PFS in patients and high EZH2 mRNA levels (Fig. 1B). The median survival of patients with EZH2-high–expressing tumors was 3 months longer than in patients with low-EZH2–expressing tumors [19.00 vs. 16.00 months, respectively; HR, 0.77; 95% confidence interval (CI), 0.62–0.95]. A high ATM mRNA expression could not be linked to a better survival (Fig. 1C). Interestingly, by separating patients into EZH2-high and -low expressing groups, PFS significantly worsened in patients with tumors low in EZH2, but high in ATM expression (Fig. 1D). The median survival differed by nearly 6 months (EZH2low/ATMlow:19.98 months vs. EZH2low/ATMhigh:14.03 months; HR, 0.66; 95% CI, 0.50–0.88). A comparison of ATM subgroups within EZH2-high–expressing tumors revealed no significant differences (Fig. 1E).
Colocalization of ATM and EZH2 in tumors of responders
IHC staining of FFPE tissue microarrays from 34 patients with ovarian cancer (Fig. 2A; Supplementary Fig. S3) displayed no differences in DFS between EZH2-high and EZH2-low–scored samples using log rank survival analysis (Fig. 2C). A significant correlation was, however, detected between scores of EZH2 and ATM or EZH2 and pRb (Supplementary Fig. S4). The low signal intensities of the staining in tissues could not validate phospho-ERK1/2 levels, probably due to the higher sensitivity of DigiWest protein quantification method.
Following up on the cooperation of EZH2 and ATM, a proximity ligation assay (PLA) was performed to visualize the vicinity of the two proteins in tissues. Consecutive sections of the same 34 tumors used in IHC were analyzed using antibodies against EZH2 and the active, phosphorylated form S1981 of the ATM protein, with only incidences of the two proteins in close proximity yielding a signal (Fig. 2B; Supplementary Fig. S5). A log rank survival analysis further revealed a significant level of correlation between tissues with low PLA signals and a decreased DFS (Fig. 2D). The numbers of EZH2-pATM PLA signals correlated to the immunoreactive scores of EZH2 (Fig. 2E), but not to that of the ATM protein in these tissues (Supplementary Fig. S6).
Resistance levels correlate inversely to EZH2 and ATM protein levels in ovarian cancer cells
EZH2 and ATM total protein amounts were quantified in untreated cells of eight ovarian cancer cell lines using Western blot analysis (Fig. 3A). The cell lines were additionally analyzed with regards to their responsiveness to treatment with either carboplatin (CP) or paclitaxel (TX), or a combination of the two (CPTX). Viability was measured by ATP assay and the IC50 values were calculated for each cell line. Comparison of these resistance levels indicate similar responsiveness in cell lines being treated with paclitaxel alone or the CPTX (Fig. 3B), but resistance to carboplatin was not connected to CPTX (Supplementary Fig. S7).
The normalized protein levels were correlated to the determined IC50 values using Spearman correlation. The results indicate that low levels of the EZH2 protein are significantly associated with higher resistance to CPTX and paclitaxel, whereas no trend was observed with carboplatin (Fig. 3C; Supplementary Fig. S8). Low ATM levels also significantly correlated with increased resistance to CPTX and paclitaxel (Fig. 3D; Supplementary Fig. S9), yet again, no significant trend was observed with carboplatin-resistant cell lines.
Decreased EZH2 protein expression increases resistance in ATMhigh ovarian cancer cells
The functional relevance of EZH2 and ATM proteins in resistance behavior was verified by their downregulation followed by treatment of the modified cells with either paclitaxel, carboplatin or CPTX. The A2780 cell line, which exhibited high levels of EZH2 and ATM similar to that found for the responsive ovarian cancer group, was classified as a representative of the responder class with which the siRNA knockdown experiments could be conducted. ATM and EZH2 were depleted either separately, using siATM and siEZH2, respectively, or simultaneously (siEZH2 + siATM, Fig. 4A). DigiWest analysis of the 30 differentially expressed proteins in responsive compared with nonresponsive tumors indicated a decrease of cell cycle and mitotic proteins in nonresponders. EZH2 depletion mainly caused a reduction of these proteins in A2780 cells as well (Fig. 4B). Responsiveness to drug treatment was subsequently analyzed relative to A2780 cells treated with control siRNA (siNC). Notably, EZH2-depleted cells showed a significantly increased tolerance to carboplatin, paclitaxel, and to the CPTX (Fig. 4C, D, and E, respectively). In contrast, knocking down ATM significantly increased the sensitivity of the cells to carboplatin treatment (Fig. 4C), whereas no discernable change of sensitivity was observed for paclitaxel or CPTX treatments. Interestingly, no change was observed in simultaneous EZH2/ATM knockdown cells exposed to carboplatin, paclitaxel, or CPTX treatments compared with control cells.
From this, it can be surmised that EZH2 reduction permits resistance only in the presence of ATM by lowering mitotic proteins through checkpoint activation, but not if EZH2 and ATM are both reduced. Of note, the DigiWest analysis of additional checkpoint proteins revealed a dependence on EZH2 of cell-cycle proteins responsible for G2–M-phase transition (Supplementary Fig. S10).
Induction of EZH2 and phospho-ATM colocalization in ovarian cancer cells
Having identified an cooperation between EZH2 and phospho-ATM in responsive ovarian cancers, A2780 cells with high expressions of EZH2 and ATM, were exposed to chemotherapy and the EZH2-ATM colocalization was measured during treatment with sublethal concentrations of carboplatin, paclitaxel, or CPTX by PLA (Fig. 4F; Supplementary Fig. S11). The PLA assays revealed a significant increase in EZH2-pATM signals in cells treated with paclitaxel alone and with CPTX around the nucleus, whereas carboplatin alone did not induce any PLA signals between EZH2 and phospho-ATM (Fig. 4G).
To identify characteristics of intrinsic resistance in serous ovarian cancers, tumors were initially categorized, depending on the patients' DFS period, as either responsive or nonresponsive to CPTX combined treatments. We chose as a cutoff for the stratification of patients into nonresponsive, a DFS period of less than 12 months from the date of surgery, which corresponds nearly to the clinical definition of resistant (relapse occurring 6 months from the end of chemotherapy; ref. 20).
Tumors of both groups were compared using the protein profiles of 279 analytes by DigiWest protein array technology. The array revealed an increase of proteins with functions in mitosis and proliferation in therapy responsive tumors, whereas increased activated MAPK/ERK signaling proteins in nonresponsive tumors indicate an activation of the G2–M checkpoint upon DNA damage (21). One of the most significantly increased proteins in the responders' tumors was the histone methyl transferase EZH2. EZH2 was found elevated among various tumor entities (22, 23) and is associated with cell proliferation. The protein acts primarily in the polycomb repressor complex 2 (PRC2) as a methyl transferase catalyzing trimethylation on lysine27 of histone H3. This results in epigenetic silencing of genes, which are important in developmental processes (24). However, in analyzing H3K27me3 in tumor samples by the antibody-based DigiWest technology, no link was found to therapy response. The complexity of EZH2's functions and interactions in the tumor biology remains unclear, with EZH2 serving as a favorable prognostic marker in colorectal cancer (25), yet as an unfavorable marker for renal, liver cancer, and melanoma (26–28). Moreover, a loss of EZH2 was linked to multidrug resistance in acute myeloid leukemia (29).
Our study revealed significantly upregulated EZH2 levels correlating with an increased DFS period utilizing a highly sensitive method for protein quantification. The greater sensitivity of DigiWest detection system over that of traditional methods was highlighted by the failure of IHC on tissues to detect an increase in EZH2 expression, whereas mRNA in silico data using a large patient cohort of 616 patients supported the DigiWest's EZH2 results.
Both responsive and nonresponsive tumors were additionally submitted to NGS panel sequencing of various DNA-damage repair and tumor suppressor genes. Sequencing confirmed previously published findings (10) of an overall high frequency of TP53 mutations (96%). No discernable pattern could, however, be observed for BRCA1 or BRCA2 mutations. Of note, NGS revealed an accumulation of ATM variants in nonresponsive tumors. ATM serine/threonine kinase is known to be activated by double-strand DNA breaks, initiating cell-cycle checkpoints and DNA repair mechanisms (30, 31). The detected mutations were located outside of ATM's functional domains, suggesting a potential regulatory role in ATM activation or binding. One of the variants, ATM p.Asp1853Asn, has been classified as a polymorphism with a population frequency of approximately 11% (ExAC). Supporting our data, this variant has been linked to a worse outcome in a study of 225 patients treated with cisplatin and paclitaxel (19).
Li and colleagues (14) reported the phosphorylation of EZH2 on S734 or S652 by ATM, blocking its complex assembly. Furthermore, it was shown that, among others, the PRC2 complex is needed for the proper induction of DNA double-strand repair mediated by ATM (32). Using a proximity ligation assay, we were not only able to confirm a colocalization of EZH2 and ATM, but additionally show an increased level of PLA signals in tumors responsive to therapy, as depicted by the increased DFS of patients.
A further in vitro investigation of the EZH2–ATM relation was conducted. Ovarian cancer cell lines were screened for their response to combinational or single-agent treatment with carboplatin and paclitaxel. High IC50 values of CPTX corresponded to paclitaxel, but not to carboplatin mono treatment suggesting an impaired cell division as a prerequisite to develop resistance mechanisms against CPTX. EZH2 knockdown alone, but not in combination with ATM silencing, desensitized A2780 cells to CPTX. This is potentially indicative of an inhibitory effect of EZH2 on ATM. Checkpoint signaling upon DNA damage relies on ATM signaling through CHK2 and ATR signaling through CHK1 (33). The lack of mitotic proteins in the nonresponsive tumors, as it is indicated in the DigiWest analysis, implies an altered regulation in cell-cycle progression. This was additionally confirmed by the DigiWest analysis of EZH2-knockdown cells.
We hypothesize that, upon treatment with carboplatin and paclitaxel, EZH2 inhibits ATM-dependent G2–M block, leading to increased proliferation and targetable cells (Supplementary Fig. S12). Loss of EZH2 or impediment of its binding by structural changes in EZH2 or ATM could lead to a dysregulation and overactivation of ATM, which in turn instigates checkpoint arrest and potential resistance to chemotherapy. Survival analysis also indicated a relation of high ATM and low EZH2 expression with worse PFS in a larger patient cohort of 616 patients. The use of synthetic lethality mechanisms in this context would be a promising approach to improve the efficacy of a treatment. Future studies should consider the increased cell-cycle checkpoint arrest in nonresponsive ovarian cancer cells and evaluate treatment options to overcome this barrier to improve the therapeutic outcome of patients with ovarian cancer.
Disclosure of Potential Conflicts of Interest
Y. Beiter is an employee at NMI TT GmbH. E.I. Braicu has provided expert testimony for Roche Pharma, AstraZeneca, Tesaro, Clovis, and Incyte. T. Fehm is a consultant for Novartis, Roche, Pfizer, AstraZeneca, and Daichii Sankyo. No potential conflicts of interest were disclosed by the other authors.
Conception and design: J. Naskou, T. Fehm, M.F. Templin, H. Neubauer
Development of methodology: J. Naskou, Y. Beiter, R. van Rensburg
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J. Naskou, Y. Beiter, E. Honisch, M. Rudelius, J. Gottstein, L. Walter, E.I. Braicu, J. Sehouli, S. Darb-Esfahani, A. Staebler, S. Brucker, T. Fehm
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Naskou, R. van Rensburg, M. Rudelius, M. Schlensog, T. Fehm, M.F. Templin
Writing, review, and/or revision of the manuscript: J. Naskou, Y. Beiter, R. van Rensburg, E. Honisch, M. Schlensog, L. Walter, E.I. Braicu, J. Sehouli, S. Darb-Esfahani, A. Staebler, D. Wallwiener, I. Beyer, D. Niederacher, T. Fehm, M.F. Templin, H. Neubauer
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): E.I. Braicu, J. Sehouli, S. Darb-Esfahani, A.D. Hartkopf, S. Brucker, D. Wallwiener
Study supervision: H. Neubauer
Tumor samples were obtained from the Institutional Tumor Bank of the University of Tübingen and the Tumor Bank Ovarian Cancer Network (TOC) of the Charité Berlin. J. Naskou, Y. Beiter, and M.F. Templin were supported by the German Federal Ministry of Education and Research (BMBF; grant FKZ 03V798 VIP – Validation of Innovative Potential). We would like to thank Ewa Breitinger (Natural and Medical Sciences Institute at the University of Tübingen, Reutlingen, Germany) for her technical assistance in the DigiWest analysis.
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.