Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

A Robust Gene Expression Prognostic Signature for Overall Survival in High-Grade Serous Ovarian Cancer

A Robust Gene Expression Prognostic Signature for Overall Survival in High-Grade Serous Ovarian... Hindawi Journal of Oncology Volume 2019, Article ID 3614207, 12 pages https://doi.org/10.1155/2019/3614207 Research Article A Robust Gene Expression Prognostic Signature for Overall Survival in High-Grade Serous Ovarian Cancer 1,2 3 4 5 6 Yue Zhao, Shao-Min Yang, Yu-Lan Jin, Guang-Wu Xiong, Pin Wang, 7 7 2 7 Antoine M. Snijders, Jian-Hua Mao, Xiao-Wei Zhang , and Bo Hang Department of Gynecology, e First Affiliated Hospital, Nanjing Medical University, Nanjing 210000, China Department of Obstetrics and Gynecology, Peking University ird Hospital, Beijing 100191, China Department of Pathology, School of Basic Medical Sciences, ird Hospital, Peking University Health Science Center, Beijing 100191, China Department of Pathology, Beijing Obstetrics and Gynecology Hospital, Capital Medical University, Beijing Maternal and Child Health Care Hospital, Beijing 100026, China Women & Children Health Center, e ird Affiliated Hospital of Chongqing Medical University, Chongqing 401120, China Department of Gastroenterology, Nanjing Drum Tower Hospital, Nanjing University Medical School, Nanjing, Jiangsu 210008, China Biological Systems and Engineering Division, Lawrence Berkeley National Laboratory, Berkeley, CA 94720, USA Correspondence should be addressed to Xiao-Wei Zhang; jiangsuzhaoy@126.com and Bo Hang; bo_hang@lbl.gov Received 26 March 2019; Accepted 17 July 2019; Published 7 November 2019 Academic Editor: Pierfrancesco Franco Copyright © 2019 Yue Zhao et al. 2is is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. 2e objective of this research was to develop a robust gene expression-based prognostic signature and scoring system for predicting overall survival (OS) of patients with high-grade serous ovarian cancer (HGSOC). Transcriptomic data of HGSOC patients were obtained from six independent studies in the NCBI GEO database. Genes significantly deregulated and as- sociated with OS in HGSOCs were selected using GEO2R and Kaplan–Meier analysis with log-rank testing, respectively. Enrichment analysis for biological processes and pathways was performed using Gene Ontology analysis. A resampling/cross- validation method with Cox regression analysis was used to identify a novel gene expression-based signature associated with OS, and a prognostic scoring system was developed and further validated in nine independent HGSOC datasets. We first identified 488 significantly deregulated genes in HGSOC patients, of which 232 were found to be significantly associated with their OS. 2ese genes were significantly enriched for cell cycle division, epithelial cell differentiation, p53 signaling pathway, vasculature development, and other processes. A novel 11-gene prognostic signature was identified and a prognostic scoring system was developed, which robustly predicted OS in HGSOC patients in 100 sampling test sets. 2e scoring system was further validated successfully in nine additional HGSOC public datasets. In conclusion, our integrative bioinformatics study combining transcriptomic and clinical data established an 11-gene prognostic signature for robust and reproducible pre- diction of OS in HGSOC patients. 2is signature could be of clinical value for guiding therapeutic selection and individualized treatment. patients are diagnosed with OC at an advanced stage. 1. Introduction Globally, more than 239,000 women are diagnosed with OC Ovarian cancer (OC) represents the most lethal gynaeco- and 152,000 succumb to this disease each year [2]. logical malignancy and the fifth leading cause of death in OC has been shown to have considerable complexity and women, with a 5-year survival rate around 10% [1]. Due to heterogeneity in biology, drug response, and survival time lack of early screening and diagnostic measures, most [3], representing a major obstacle for its precision medicine 2 Journal of Oncology practice. OCs of epithelial origin constitute approximately (GSE32063, GSE19829 GPL570, GSE30161, GSE3149, OV- 90% of all the cases, whereas ovarian sex cord stromal tumor, AU-ICGC, GSE14764, GSE9891, GSE 17260, and ovarian germ cell tumor, and secondary tumor of ovarian GSE32062) were used for independent validation of the gene metastasis (e.g., Krukenberg tumor) are less frequent [4]. signature and prognostic scoring system. High-grade serous ovarian carcinoma (HGSOC) is the most predominant in epithelial OCs, accounting for 70–80% of 2.2. Statistical Analysis. By employing a 1.5-fold change OC deaths [5]. 2e majority of HGSOCs can be grouped into cutoff and adjusted p-value<0.05, the differentially expressed four subtypes based on gene overexpression levels specific genes between normal versus HGSOC tissues were identified for each subtype: mesenchymal, immunoreactive, differen- with GEO2R. Differentially expressed genes associated with tiated, and proliferative [6]. OS in patients with HGSOC were selected using KM survival HGSOC has been characterized by both genetic alter- analysis (Kaplan–Meier plotter (http://kmplot.com)) with a ations, including inherited BRCA gene mutations, TP53 hazard ratio (HR) with 95% confidence intervals and log-rank mutations, DNA damage, chromosomal instability [6, 7], p value cutoff for each gene at 0.05 [19]. and changes in RNA and miRNA expression and methyl- ation status [8]. Microarray and next-generation sequencing technologies have become vital tools for identifying these 2.3. Gene Ontology Pathway Analysis and Network Construction. Metascape (http://www.metascape.org) was changes genomewide, providing novel opportunities for the identification of biomarkers for diagnosis, prognosis, ther- used to assess overrepresentation of Gene Ontology cate- gories in biological networks [20]. Cytoscape 3.4.0 (http:// apeutic targets, and treatment response. For instance, many multigene biomarkers based on transcription patterns have www.cytoscape.org) was applied to generate and visualize the gene coexpression networks, to better understand the been associated with prognosis across tumor types [9–14]. A number of groups have sought to use genomewide gene biological processes enriched, as well as their relationships in the form of a network instead of the tabular text format expression data to identify multigene signatures aimed at [21, 22]. Note that KEGG pathway (http://www.genome.jp/ predicting clinical outcomes, therapy responses, and sub- kegg), GO Biological Processes (http://geneontology.org), types in OC [13–18]. Many existing signatures were gen- Reactome Gene Sets (http://www.reactome.org), and erated using partial genome annotations, limited number of patients, or used targeted gene selecting. 2us, it is very CORUM (http://mips.helmholtz-muenchen.de/corum) were ontology sources of gene network, pathway, and process much warranted to identify and develop clinically valuable gene signatures for OC prognosis, especially when based on enrichment analysis. comprehensive and unbiased whole-genome data. In this study, we employed a multistep bioinformatic 2.4. Gene Expression Signature-Based Prognostic Risk Score. strategy that uses omics information and clinical data to Clinical data of HGSOC patients were obtained from the build a gene expression prognostic scoring system in TCGA dataset (http://cancergenome.nih.gov), with which HGSOC. We previously developed this approach to identify a biomarker panel associated with OS was reachable. 100 and successfully validate a 53-gene signature associated with random selections of 307 patients from TCGA were OS of gastric cancer [11] and a 27-gene signature for lung conducted and used as training sets. 2e remaining pa- adenocarcinoma [12]. Here, we used fifteen publicly avail- tients for each selection were used as test sets to validate able datasets of HGSOCs; six were used to identify an 11- the reliability of the identified biomarker panel for gene signature associated with patient prognosis using Cox prognosis. regression analysis and cross-validation. We then used nine Forward conditional Cox regressions using SPSS were independent HGSOC datasets to validate the prognostic carried out on each of the 100 training sets in order to isolate scoring system and signature’s performance. Moreover, in the biomarker panel. Selected genes were recorded and those comparison with an existing 5-gene expression signature for that appeared in at least 20% of 100 training sets were in- ovarian serous cystadenocarcinoma (CAC) [15], we showed cluded in our biomarker panel. Subsequently, Cox re- that our signature was superior in determining overall gression was repeated on all 100 training sets using our 11- survival for this type of epithelial ovarian carcinoma. gene signature as covariates and using the forced entry (enter) method to obtain the coefficient values for each 2. Materials and Methods biomarker. 100 coefficients for every gene in the biomarker panel were then obtained, and the average of them was used 2.1. Patient Datasets. To broadly mine all the available in- to estimate the true coefficient of each gene. A formula was formation on HGSOCs, we have screened and used 15 in- created to act as the prognostic scoring system, and all the dependent datasets in the current study. Six public datasets patients can get their scores accordingly: from the NCBI Gene Expression Omnibus (GSE18520, GSE26712, GSE40595, GSE38666, GSE27651, and GSE2328) 􏽘(gene i coefficient)∗ (gene i expression level). (1) provided the HGSOC gene transcript data to identify genes i�1 differentially expressed between tumor and normal ovarian tissues. TCGA HGSOC data were used to identify the gene 2e patients in the training sets were ranked by their signature and develop the prognostic scoring system for prognostic scores and divided into three equal-sized cohorts. predicting OS of patients. Nine additional datasets 2e corresponding prognostic scores at cut points were Journal of Oncology 3 recorded and averaged as the true cut point scores, with <0.05; Figure 2(b) and Table S3). 2e hazard ratio (HR) for which the patients in the test sets were also split into three 82 genes was <1 (higher gene expression associated with good prognosis), which are referred as protective genes, groups: “good”, “intermediate”, and “poor” groups. Dif- ferences in OS among the three groups in all the test sets whereas 150 genes had a HR >1 (higher gene expression were determined by constructing Kaplan–Meier plots and associated with poor prognosis), which are considered risk performing log-rank tests. genes. 3.3. Gene Ontology (GO) Analysis of Prognostic Genes in 2.5. Validation in Independent Datasets and Comparison with HGSOC. To understand the potential biological functions of an Existing Signature. 2e 11-gene biomarker panel was the 232 genes significantly associated with OS in HGSOC further validated in nine independent datasets (Table S1). patients, we conducted Gene Ontology (GO) analysis using New coefficients for the 11 genes were obtained from Cox Metascape and found significant enrichment of many cel- regression. Prognostic scores for all patients were calculated, lular process and pathway-related genes associated with and patients were ranked based on their scores and divided cancer development including cell division, epithelial cell into three equal-sized cohorts. Kaplan–Meier analysis and a differentiation, p53 signaling pathway, and vasculature de- log-rank test were conducted to determine differences in velopment (Figure 3(a) and Table S4). 2e interconnectivity survival, as previously described [11, 12]. We compared the performance of our 11-gene signature of related GO terms was visualized using Cytoscape where individual GO terms are displayed as nodes connected based with a recently published 5-gene signature for prognosis of on similarity (Figure 3(b)). ovarian serous CAC [15]. A multivariate Cox regression analysis was conducted with the 5 genes on the same 100 training sets as described above for our inner validation. 3.4. Establishment of an 11-Gene Prognostic Scoring System in Coefficients for each of the 5 genes used in [15] and scores of HGSOCs. Figure 4(a) shows the strategy we employed to all 307 patients were calculated as above. 2en patients were isolate a prognostic biomarker signature and to develop a divided into tertiles (good, intermediate, and poor) based on scoring system based on the 232 genes that were found to be their prognostic scores, and the cut point scores were significantly associated with OS in HGSOC patients. We first recorded and averaged. Kaplan–Meier analysis was per- used a random resampling method to split 307 patients from formed, and a log-rank test was used to demonstrate dif- the TCGA dataset into 100 training (200 patients) and 100 ferences in OS among different groups for all test sets. testing (107 patients) sets. 2e training sets were then used to isolate a prognostic signature, and the testing sets were used 3. Results for validation. First, we performed a multivariate Cox re- gression analysis in all 100 training sets to identify statis- 3.1. Identification of Deregulated Genes in HGSOCs. To tically significant independent genes within the 232 genes for identify genes that are consistently deregulated in HGSOC, predicting OS. Genes that recurred in at least 20% of 100 we performed a meta-analysis and compared gene transcript training were assembled into an 11-gene signature: levels in six publically available datasets containing tran- RAD51AP1, CADPS2, DSE, ITGB8, PDE10A, GALNT10, scriptomic data for both HGSOC and normal ovarian tissues SNX1, MTHFD2, C9orf16, PYCR1, and ARL4 (Table S5). For (n � 397 from GSE18520, GSE26712, GSE40595, GSE38666, each of the 11 genes in the signature, gene function and GSE27651, and GSE2328) using GEO2R. For each dataset, known roles in ovarian and other cancers are summarized in we compared HGSOC gene expression to gene expression in Table S6. normal ovarian tissues (Figure 1). A prognostic score for each cancer patient was used to 2e criteria for significant differential expression for assess each patient’s risk of death and was defined as the each gene were set to a 1.5-fold change and adjusted p-value linear combination of logarithmically transformed gene <0.05. A total of 562 probe IDs (260 downregulated and 302 expression levels weighted by average Cox regression co- upregulated) were consistently up- or downregulated across efficients (Table S7) obtained from 100 training datasets all six datasets, representing 488 unique genes (222 down- [11, 12]. 2e prognostic scores were assigned for all patients regulated and 266 upregulated) (Figure 1 and Table S2). in both training and test sets. In each training set, the pa- tients were divided into tertiles based on their prognostic score. 2e cutpoint scores were recorded and averaged for 3.2. Analysis of Deregulated Genes and Overall Survival of HGSOCs. 2e prognostic value for each of the 488 each of 100 training sets. Based on the average scores, each deregulated genes individually in HGSOC patients was test set was split into three groups, i.e., good, intermediate, evaluated in a large public clinical database which integrates and poor. We then performed Kaplan–Meier and log-rank gene expression and patient survival using Kaplan–Meier test analysis to determine significant differences in OS survival analysis (Figure 2(a)). 2e effects of high or low among different groups for all test sets (Figure 4(b)). 2e expression levels of these genes on OS were assessed using hazard ratios (HR) for the “intermediate” and “poor” groups in comparison with the “good” groups were calculated for Kaplan–Meier survival analysis and compared statistically using the log-rank test, with representative genes shown in each test set. In 99% of the test sets, patients in the “poor” groups had a significantly shorter OS than those in the Figure 2(b). 2e results showed that 232 out of the 488 genes were significantly associated with OS (adjusted p-value “good” groups (HR confidence interval above “1”) 4 Journal of Oncology GSE26712 (11262) GSE27651 (11959) GSE40595 GSE6008 (37330) 1182 245 (6733) 130 84 GSE38666 1654 326 (17903) 409 1398 80 108 222 475 280 112 39 49 77 145 392 590 1207 392 189 216 110 164 67 84 6551 968 125 101 GSE18520 (24049) Figure 1: Human datasets of ovarian cancer and normal sample tissues. Samples were obtained from six independent gene transcript datasets containing HGSOC and normal ovarian cases. To identify genes (common probe IDs) consistently deregulated in HGSOC, a fold- change cutoff of 1.5 and adjusted p-value <0.05 were used for each dataset. (Figure 4(c), top panel), while in more than 60% of the test performing a multivariate Cox regression analysis using the sets, patients in the “intermediate” groups showed a sig- same strategy described in Figure 4 where for 100 training nificantly shorter OS than those in the “good” groups sets, coefficients for each of the 5 genes and scores of all the (Figure 4(c), bottom panel). 2ese results validated the 307 patients were calculated. discriminative ability of this 11-gene signature and prog- Figure 6 shows the HR and 95% confidence interval for the “intermediate” and “poor” groups in comparison with nostic scoring system to stratify patients with good or worse prognosis. the “good” groups in the 100 test sets. For the 5-gene panel, in 90% of the testing sets, patients in the “poor” groups had a significantly shorter OS than those in the “good” groups. For 3.5. Independent Validation of the 11-Gene Scoring System. the “intermediate” groups vs. “good” groups, only in 12% of To further validate our 11-gene signature, we tested it in nine the testing sets, patients showed a significantly shorter OS. In independent OC datasets (Table S1). Prognostic scores for all comparison, for our 11-gene signature, these two numbers patients were calculated and patients were ranked based on are 99% and 61%, respectively. In addition, the median HR their score. Significant differences were identified using of the 11-gene signature was on average 1.46-fold higher in Kaplan–Meier analysis across all nine datasets between the “intermediate” vs. “good” groups and 1.73-fold higher in patient cohorts of “good” and “poor” prognosis. Patients the “poor” vs. “good” groups compared to the 5-gene sig- with a high prognostic score had a significantly shorter OS nature (Figure 6). 2ese results indicate that the 11-gene than those patients who scored low (p< 0.05) (Figure 5). 2e signature has discriminative ability for determining OS in HR values range from 1.94 to 9.76 (Table S1) We conclude ovarian CAC patients, which is also significantly superior to that the 11-gene prognostic scoring system reproducibly the 5-gene panel. predicts overall survival of HGSOC patients. 4. Discussion 3.6. Comparison with an Existing Prognostic Signature. We compared the performance of our 11-gene signature Identification and development of reliable predictive bio- markers and new therapeutic targets are critical for sig- with a recently published 5-gene expression signature pre- dicting clinical outcome of ovarian serous CAC [15] by nificantly improving cancer patient outcomes. 2e Journal of Oncology 5 GFPT2 (205100_at) HSD17B6 (37512_at) 1.0 1.0 590 probe IDs HR = 1.58 (1.31−1.89) HR = 1.47 (1.25−1.72) Log-rank P = 7.3e −07 Log-rank P = 1.8e −06 0.8 0.8 0.6 0.6 Filter genes with the same direction fold change 0.4 0.4 0.2 0.2 0.0 0.0 488 genes 50 100 150 200 250 0 50 100 150 200 250 (562 probe IDs) Time (months) Time (months) Number at risk Number at risk Low 320 93 25 4 1 0 Low 463 124 27 5 1 0 High 887 174 24 4 1 0 High 744 143 22 3 1 0 Expression Expression Low Low Filter genes associated High High with OS ALDH1A2 (207016_s_at) PART1 (205833_s_at) 1.0 1.0 HR = 1.44 (1.23−1.68) HR = 0.67 (0.56−0.8) Log-rank P = 4.9e −06 Log-rank P = 1.2e −05 0.8 0.8 232 genes 0.6 0.6 0.4 0.4 0.2 0.2 Functional annotation 0.0 0.0 50 100 150 200 250 0 50 100 150 200 250 Time (months) Time (months) Number at risk Number at risk Low 507 135 27 4 2 0 Low 895 180 34 7 1 0 High 700 132 22 4 0 0 High 312 87 15 1 1 0 Expression Expression Low Low High High (a) (b) Figure 2: Identification of genes associated with prognostic function in HGSOC. (a) 2e 562 consistently deregulated probe IDs identified represent 488 genes in the cancer patients. 2rough Kaplan–Meier survival analysis, 232 genes were found to be significantly associated with overall survival of HGSOC patients. Functional annotation was carried out for the 232 genes. (b) Examples of Kaplan–Meier survival curves for four individual genes significantly associated with overall survival in HGSOC patients, which was divided into two groups to maximize the difference in survival using log-rank testing between groups. We used HR and log-rank p-value for the curve comparison between the groups. objective of this work was to use a multistep bioinformatics CAC. Taken together, the 11-gene signature could be of analytic strategy we developed previously [11, 12] to an- translational value for clinical use. We are currently alyze six publicly available omics and clinical datasets to working on the development of a multiplex high- generate a robust prognostic signature for patients with throughput assay to facilitate the clinical use of the sig- HGSOC. We first identified 232 genes associated with OS nature. To date, there are still no clinically useful prognostic biomarkers/scores in OC. However, two multigene ex- that served as candidate markers to provide a prediction of the prognosis of HGSOC patients. Eventually, we selected pression-based scores, the Oncotype DX 21-gene breast an 11-gene prognostic signature and scoring system cancer assay developed by Genomic Health [9, 23] and the showing strong discriminative power to separate patients MammaPrint 70-gene breast cancer recurrence assay by with good or poor survival. Moreover, the results were Agendia [24], have been utilized to guide treatment de- independently validated in each of the nine independent cisions, such as for adjuvant chemotherapy in breast cancer HGSOC datasets. We also demonstrated that our 11-gene [23]. 2ese two tests represent the first prognostic gene signature has higher predictive power compared to an expression assays that have successfully passed multiple existing prognostic panel developed for ovarian serous independent clinical trials. Probability Probability Probability Probability 6 Journal of Oncology GO:0051301: cell division GO:0001568: blood vessel development GO:0071229: cellular response to acid chemical R-HSA-3560782: Diseases associated with glycosaminoglycan metabolism GO:0043009: chordate embryonic development GO:0033273: response to vitamin GO:0072001: renal system development GO:0071407: cellular response to organic cyclic compound GO:0030855: epithelial cell differentiation GO:0000079: regulation of cyclin-dependent protein serine/threonine kinase activity hsa04115: p53 signaling pathway R-HSA-109582: Hemostasis GO:1902850: microtubule cytoskeleton organization involved in mitosis GO:1905114: cell surface receptor signaling pathway involved in cell-cell signaling GO:0035690: cellular response to drug GO:0010039: response to iron ion GO:0048863: stem cell differentiation GO:0008285: negative regulation of cell proliferation GO:0045664: regulation of neuron differentiation GO:0007423: sensory organ development (a) GO:0051301: cell division GO:0071229: cellular response to acid chemical GO:0043009: chordate embryonic development GO:0072001: renal system development GO:0030855: epithelial cell differentiation hsa04115: p53 signaling pathway GO:1902850: microtubule cytoskeleton organization involved in mitosis GO:0035690: cellular response to drug GO:0048863: stem cell differentiation GO:0045664: regulation of neuron differentiation GO:0001568: blood vessel development R-HSA-3560782: Diseases associated with glycosaminoglycan metabolism GO:0033273: response to vitamin GO:0071407: cellular response to organic cyclic compound GO:0000079: regulation of cyclin-dependent protein serine/threonine kinase activity R-HSA-109582: Hemostasis GO:1905114: cell surface receptor signaling pathway involved in cell-cell signaling GO:0010039: response to iron ion GO:0008285: negative regulation of cell proliferation GO:0007423: sensory organ development 0 2 4 6 8 10 12 14 16 –log 10 (p) (b) Figure 3: Gene Ontology analysis of 232 genes associated with OS. (a) Network layout of the clusters generated with the complete list of the 232 OS-associated genes in HGSOC. Each node represents one enriched term, where its size is proportional to the number of genes associated with each term, and its color representing its cluster identity (i.e., nodes of the same color belong to the same cluster). All similar terms with a kappa similarity score>0.3 are connected by edges (the thicker the edge, the higher the similarity). One term from each cluster was selected to describe the general function of each cluster. Created by Metascape (http://metascape.org). (b) Top 20 most significant GO categories associated with the 232 genes. Microarray and next-generation sequencing technolo- novel panel of prognostic biomarkers is the first step in gies broadened the accessibility of large cancer genomewide developing a practically valuable assay/score in a clinical expression profiles. Taking advantage of these unbiased setting. 2e next steps include multicenter clinical trials and genomewide approaches, we established multigene signa- prospective trials that allow further validation of the efficacy tures for predictive and prognostic purposes, including the and accuracy of the signature, in order to make a successful 11-gene signature described in this study. To discover a clinical translation. It should be mentioned that microarray Journal of Oncology 7 TCGA 307 patients Random sampling Training set Test set (200 patients) (107 patients) Multivariate cox regression Identify genes selected into Cox regression model Identify 11-gene signature for OS Prognostic score for OS Validation (a) 1.0 1.0 Log-rank (Mantel-Cox) Log-rank (Mantel-Cox) Chi-square = 44.03 Chi-square = 37.66 p < 0.0001 p < 0.0001 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0.1 0.1 0 20 40 60 80 100 120 140 0 25 50 75 100 125 Time (months) Time (months) Bad Bad Good Good Intermediate Intermediate (b) Figure 4: Continued. Survival 100 times Survival 100 times 8 Journal of Oncology Poor vs. Good 0 20 40 60 80 100 120 100 independent samplings HR 95% CI Intermediate vs. Good 0 20 40 60 80 100 120 100 independent samplings HR 95% CI (c) Figure 4: Strategy to generate an 11-gene prognostic signature and its performance evaluation. (a) We employed multivariate Cox re- gression analysis on 100 training sets through random sampling for the 232 genes and identified 11 genes selected into our Cox regression model. Such a signature was used to generate a prognostic scoring system, which was further validated using 100 randomly assembled test sets. (b) Representative Kaplan–Meier overall survival curves in two test sets. 2ese curves were separated into tertiles according to the prognostic score calculated using the 11-gene signature. (c) HR values and their 95% confidence interval across the 100 test sets, calculated using a Cox model based on the prognostic score comparing poor vs. good (top) and intermediate vs. good (bottom). data-based analyses have generated many single and mul- Ovarian cancer, like many other cancers, occurs through the accumulation of genetic alterations, which can result in tiple gene biomarkers/signatures associated with prognosis of specific types of cancers including OC. For OC, several deregulation of gene expression. So far, there is still limited prognostic signatures have been developed based on dif- information on the genes that are associated with prognosis ferent platforms, as described before. While these signatures of OC. Table S6 summarizes the known functions for each can predict OC survival, some of them were developed based gene in the 11-gene panel in tumor development and on limited patient numbers or conducted within a single prognostic relevance. Of them, six have already been im- medical center. In addition, signatures developed in earlier plicated in the development and progression of HGSOCs in years were either based on incomplete genome annotations previously published studies [25–31]. Six genes (RAD51AP1, or based solely on existing knowledge. Nevertheless, we DSE, ITGB8, GALNT10, SNX1, and MTHFD2) were re- ported to provide useful prognostic information about the expect that with ongoing and future prospective studies, some of these preclinical biomarker signatures, including the survival in various types of cancer [25, 27, 28, 32–36], in- 11-gene signature described here, will be fully evaluated for cluding three genes (RAD51AP1, ITGB8, and GALNT10) their value in the clinical settings. which were reported to be prognostic for OC [25, 28, 32]. Hazard ratio Hazard ratio Journal of Oncology 9 GSE32063 GSE3149 HR: 4.76 (1.81–12.53) HR: 3.2 (1.77–5.81) 1.0 1.0 p = 4.4E –04 p = 5.389E –05 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0.0 0.0 0 20406080 100 0 50 100 150 20 19 18 15 12 9 9 8 6 3 2 62 40 26 20 12 6 2 20 18 15 9 8 4 1 1 1 1 1 62 41 24 9 4 2 0 Time (months) Time (months) (a) (b) GSE30161 GSE19829 HR: 3.3 (1.59–6.84) HR: 2.9 (1.08–7.8) 1.0 1.0 p = 7.2E –04 p = 0.024 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0.0 0.0 0 20 40 60 80 100 0 50 100 150 200 250 300 350 29 25 24 20 14 14 13 12 8 7 5 5 3 1 1 14 14 11 97655333 29 25 18 13 12 10 5 3 2 1 0 0 0 0 0 14 12 996320000 Time (months) Time (months) (c) (d) Figure 5: Validation of the 11-gene signature using four independent ovarian cancer cohorts. We analyzed the Kaplan–Meier plots generated for the four cohorts used by applying the 11-gene signature. 2e patient cohort was split by the median based on the prognostic index, and the log-rank p-values of the curve comparison between the risk groups and HR are shown. 2e HR values and 95% confidence intervals were calculated using Cox survival analysis. RAD51AP1, which encodes an RAD51 accessory protein, its expression is an independent biomarker for predicting participates in the homologous recombination DNA damage unfavorable survival of patients with HGSOCs [27]. In- response pathway. 2e finding in this study is in agreement tegrative network analysis of TCGA data has shown that with DNA repair defects in HGSOCs. Upregulation of GALNT10 was highly predictive for the OS of ovarian cancer RAD51AP1 predicted poorer OS in patients with ovarian patients [28]. 2ere is evidence that SNX1 may play a role in cancer [25]. DSE (SART2) gene has been shown to be fre- tumorigenesis and its downregulation is significantly corre- quently upregulated in human brain tumors and other types lated with poor OS of colon cancer patients [29]. MTHFD2 is of cancer [37]. Moreover, elevated DSE expression in glioma a gene associated with cancer development, and its high expression is associated with poor prognosis of many types of is associated with a worse tumor grade and poor OS [37]. Elevated levels were also detected in cervical, ovarian, and cancer, for example [36–38]. Five of these genes, CADPS2, MTHFD2, PDE10A, PYCR1, and ARL4, have never been endometrial cancers [34]. ITGB8 encodes a β-subunit of integrin, and integrins play a regulatory role on cancer cells reported to have a role in OC (Table S6). Interestingly, the through survival- and metastasis-related signaling pathways genes in the multigene panels reported in the literature, in- [34]. Upregulation of ITGB8 has been shown in several types cluding our 11-gene signature, are rarely overlapping, which of cancers, including HGSOCs. In addition, it was found that may reflect the disparity in tumor samples, microarray Probability Probability Probability Probability 10 Journal of Oncology p < 0.0001 p < 0.0001 (a) (b) Figure 6: Comparison of HR and 95% confidence interval between the 11-gene and 5-gene signatures. For both 11-gene and 5-gene signatures, the HR of all the 100 test sets was calculated using a Cox model based on the prognostic score between groups (poor vs. good: top; intermediate vs. good: bottom). 2e differences between the two signatures were significant for both the poor vs. good groups and the intermediate vs. good groups (p< 0.0001). designs, database selection, and analytical approaches. 2e Acknowledgments genes in this signature may be novel potential therapeutic 2is work was supported by grants from the National Key targets for HGSOCs. Research and Development Program for Reproductive 2e genes included in our signature might also be po- Health and Major Birth Defects Control and Prevention tential biomarkers or targets for the treatment of OC. (2017YFC1002004) and China Scholarship Council Personalized treatment is often highlighted in today’s (201606010313). clinical practice, where the molecular features such as ge- netic background of an individual patient’s tumor determine the prime treatment modalities. For example, Prexasertib Supplementary Materials (LY2606368), a cell cycle checkpoint kinase 1 and 2 in- Table S1: summary of independent validation of the 11-gene hibitor, showed clinical activity and was tolerable in HGSOC signature in 9 datasets. Table S2: list of genes that are patients with BRCA wild-type disease [39]. consistently deregulated in HGSOC across six datasets using In conclusion, as the most lethal gynaecological malig- criteria: adjusted p< 0.05 and fold change>1.5. Table S3: the nancy, OC is undoubtedly a challenge for patients, medical impact of deregulated genes on overall survival (OS). Genes practitioners, and researchers. In this study, with an unbiased significantly associated with OS are highlighted in yellow. multistep bioinformatics analytic strategy, we identified an Table S4: top 20 altered gene clusters identified in the 232 11-gene prognostic biomarker panel which robustly and deregulated genes that are significantly associated with OS. accurately predicts overall survival in patients with HGSOCs. Gene set enrichment analysis was conducted using Meta- Gene Ontology analysis revealed several important enriched scape (http://metascape.org). Count represents the number cellular processes and pathways in HGSOCs. Together, our of genes with membership in the given ontology term. “%” results pave the way for developing a clinical assay for guiding represents the percentage of total 232 genes associated with therapeutic selection and individualized treatment. OS that are found in the given ontology term. Log10 (p) is the p-value in log base 10. Table S5: frequency by which each Abbreviations gene appeared in the Cox regression model among 100 resampling training sets. 2e signature genes are highlighted OC: Ovarian cancer in yellow. Table S6: the function and role of 11 genes in the HGSOC: High-grade serous ovarian cancer prognostic signature in normal and HGSOC cells. Table S7: CAC: Cystadenocarcinoma the average Cox regression coefficient for each gene used to HR: Hazard ratio calculate the prognostic score. (Supplementary Materials) K-M plot: Kaplan–Meier plot OS: Overall survival TCGA: 2e Cancer Genome Atlas. References [1] R. L. Siegel, K. D. Miller, and A. Jemal, “Cancer statistics, Data Availability 2017,” CA: A Cancer Journal for Clinicians, vol. 67, no. 1, pp. 7–30, 2017. 2e data used to support the findings of this study are [2] B. M. Reid, J. B. Permuth, and T. A. Sellers, “Epidemiology of available from the corresponding author upon request. ovarian cancer: a review,” Cancer Biology & Medicine, vol. 14, no. 1, pp. 9–32, 2017. [3] P. T. Kroeger Jr., and R. Drapkin, “Pathogenesis and het- Conflicts of Interest erogeneity of ovarian cancer,” Current Opinion in Obstetrics 2e authors declare that they have no conflicts of interest. and Gynecology, vol. 29, no. 1, pp. 26–34, 2017. HR Poor vs. good 11-gene 5-gene HR Intermediate vs. good 11-gene 5-gene Journal of Oncology 11 [4] J. Prat, “New insights into ovarian cancer pathology,” Annals [21] G. Bindea, B. Mlecnik, H. Hackl et al., “ClueGO: a cytoscape of Oncology, vol. 23, no. 10, pp. x111–x117, 2012. plug-in to decipher functionally grouped gene ontology and [5] R. Vang, I.-M. Shih, and R. J. Kurman, “Ovarian low-grade pathway annotation networks,” Bioinformatics, vol. 25, no. 8, and high-grade serous carcinoma,” Advances in Anatomic pp. 1091–1093, 2009. Pathology, vol. 16, no. 5, pp. 267–282, 2009. [22] I. H. Goenawan, K. Bryan, and D. J. Lynn, “DyNet: visuali- [6] Cancer Genome Atlas Research Network, “Integrated geno- zation and analysis of dynamic molecular interaction net- mic analyses of ovarian carcinoma,” Nature, vol. 474, works,” Bioinformatics, vol. 32, no. 17, pp. 2713–2715, 2016. no. 7353, pp. 609–615, 2011. [23] J. A. Sparano, R. J. Gray, D. F. Makower et al., “Adjuvant [7] U. A. Matulonis, A. K. Sood, L. Fallowfield et al., “Ovarian chemotherapy guided by a 21-gene expression assay in breast cancer,” Nature Reviews Disease Primers, vol. 2, no.1, p.16061, cancer,” New England Journal of Medicine, vol. 379, no. 2, pp. 111–121, 2018. [8] G. E. Konecny, C. Wang, H. Hamidi et al., “Prognostic and [24] F. Cardoso, L. J. van’t Veer, J. Bogaerts et al., “70-gene sig- therapeutic relevance of molecular subtypes in high-grade nature as an aid to treatment decisions in early-stage breast serous ovarian cancer,” JNCI: Journal of the National Cancer cancer,” New England Journal of Medicine, vol. 375, no. 8, Institute, vol. 106, no. 10, article dju249, 2014. pp. 717–729, 2016. [9] S. Paik, S. Shak, G. Tang et al., “A multigene assay to predict [25] D. Chudasama, V. Bo, M. Hall et al., “Identification of cancer recurrence of tamoxifen-treated, node-negative breast can- biomarkers of prognostic value using specific gene regulatory cer,” New England Journal of Medicine, vol. 351, no. 27, networks (GRN): a novel role of RAD51AP1 for ovarian and pp. 2817–2826, 2004. lung cancers,” Carcinogenesis, vol. 39, no. 3, pp. 407–417, [10] J. R. Kratz, J. He, S. K. Van Den Eeden et al., “A practical molecular assay to predict survival in resected non-squamous, [26] S. Tanaka, N. Tsuda, K. Kawano et al., “Expression of tumor- non-small-cell lung cancer: development and international rejection antigens in gynecologic cancers,” Japanese Journal of validation studies,” e Lancet, vol. 379, no. 9818, pp. 823– Cancer Research, vol. 91, no. 11, pp. 1177–1184, 2000. 832, 2012. [27] J. He, Y. Liu, L. Zhang, and H. Zhang, “Integrin subunit beta 8 [11] P. Wang, Y. Wang, B. Hang et al., “A novel gene expression- (ITGB8) upregulation is an independent predictor of un- based prognostic scoring system to predict survival in gastric favorable survival of high-grade serous ovarian carcinoma cancer,” Oncotarget, vol. 7, no. 34, pp. 55343–55351, 2016. patients,” Medical Science Monitor, vol. 24, pp. 8933–8940, [12] E. G. Chen, P. Wang, H. Lou et al., “A robust gene expression- based prognostic risk score predicts overall survival of lung [28] Q. Zhang, J. E. Burdette, and J. P. Wang, “Integrative network adenocarcinoma patients,” Oncotarget, vol. 9, no. 6, analysis of TCGA data for ovarian cancer,” BMC Systems pp. 6862–6871, 2018. Biology, vol. 8, no. 1, p. 1338, 2014. [13] D. Spentzos, D. A. Levine, M. F. Ramoni et al., “Gene ex- [29] L. N. Nguyen, M. S. Holdren, A. P. Nguyen et al., “Sorting pression signature with independent prognostic significance nexin 1 down-regulation promotes colon tumorigenesis,” in epithelial ovarian cancer,” Journal of Clinical Oncology, Clinical Cancer Research, vol. 12, no. 23, pp. 6952–6959, 2006. vol. 22, no. 23, pp. 4648–4658, 2004. [30] W. Ju, B. C. Yoo, I. J. Kim et al., “Identification of genes with [14] T. Bonome, D. A. Levine, J. Shih et al., “A gene signature differential expression in chemoresistant epithelial ovarian predicting for survival in suboptimally debulked patients with cancer using high-density oligonucleotide microarrays,” ovarian cancer,” Cancer Research, vol. 68, no. 13, pp. 5478– Oncology Research Featuring Preclinical and Clinical Cancer 5486, 2008. erapeutics, vol. 18, no. 2-3, pp. 47–56, 2009. [15] L. W. Liu, Q. Zhang, W. Guo, K. Qian, and Q. Wang, “A five- [31] J. Wang, C. Chen, H. F. Li, X. L Jiang, and L. Zhang, “In- gene expression signature predicts clinical outcome of ovarian vestigating key genes associated with ovarian cancer by in- serous cystadenocarcinoma,” BioMed Research International, tegrating affinity propagation clustering and mutual vol. 2016, Article ID 6945304, 6 pages, 2016. information network analysis,” European Review for Medical [16] R. W. Tothill, A. V. Tinker, J. George et al., “Novel molecular and Pharmacological Sciences, vol. 20, no. 12, pp. 2532–2540, subtypes of serous and endometrioid ovarian cancer linked to clinical outcome,” Clinical Cancer Research, vol. 14, no. 16, [32] L. Liu, Y. Xiong, W. Xi et al., “Prognostic role of N-Ace- pp. 5198–5208, 2008. tylgalactosaminyltransferase 10 in metastatic renal cell car- [17] B. Győrffy, A. Lanczky, and Z. Szallasi, “Implementing an cinoma,” Oncotarget, vol. 8, no. 9, pp. 14995–15003, 2017. online tool for genome-wide validation of survival-associated [33] X.-Y. Zhan, Y. Zhang, E. Zhai, Q.-Y. Zhu, and Y. He, “Sorting biomarkers in ovarian-cancer using microarray data from nexin-1 is a candidate tumor suppressor and potential 1287 patients,” Endocrine-Related Cancer, vol. 19, no. 2, prognostic marker in gastric cancer,” PeerJ, vol. 6, p. e4829, pp. 197–208, 2012. [18] G. P. Sfakianos, E. S. Iversen, R. Whitaker et al., “Validation of [34] J. Koseki, M. Konno, A. Asai et al., “Enzymes of the one- ovarian cancer gene expression signatures for survival and carbon folate metabolism as anticancer targets predicted by subtype in formalin fixed paraffin embedded tissues,” Gy- survival rate analysis,” Scientific Reports, vol. 8, no. 1, p. 303, necologic Oncology, vol. 129, no. 1, pp. 159–164, 2013. [19] B. Gyorffy, P. Surowiak, J. Budczies, and A. Lanczky, ´ “Online [35] H. Lin, B. Huang, H. Wang et al., “MTHFD2 overexpression survival analysis software to assess the prognostic value of predicts poor prognosis in renal cell carcinoma and is as- biomarkers using transcriptomic data in non-small-cell lung sociated with cell proliferation and vimentin-modulated cancer,” PLoS One, vol. 8, no. 12, Article ID e8224, 2013. [20] S. Tripathi, M. O. Pohl, Y. Zhou et al., “Meta- and orthogonal migration and invasion,” Cellular Physiology and Bio- integration of influenza “OMICs” data defines a role for UBR4 chemistry, vol. 51, no. 2, pp. 991–1000, 2018. in virus budding,” Cell Host & Microbe, vol. 18, no. 6, [36] K. Noguchi, M. Konno, J. Koseki et al., “2e mitochondrial pp. 723–735, 2015. one-carbon metabolic pathway is associated with patient 12 Journal of Oncology survival in pancreatic cancer,” Oncology Letters, vol. 16, no. 2, pp. 1827–1834, 2018. [37] W. C. Liao, C. K. Liao, Y. H. Tsai et al., “DSE promotes aggressive glioma cell phenotypes by enhancing HB-EGF/ ErbB signaling,” PLoS One, vol. 13, no. 6, Article ID e0198364, [38] R. Ata and C. N. Antonescu, “Integrins and cell metabolism: an intimate relationship impacting cancer,” International Journal of Molecular Sciences, vol. 18, no. 1, p. 189, 2017. [39] J.-M. Lee, J. Nair, A. Zimmer et al., “Prexasertib, a cell cycle checkpoint kinase 1 and 2 inhibitor, in BRCA wild-type re- current high-grade serous ovarian cancer: a first-in-class proof-of-concept phase 2 study,” e Lancet Oncology, vol. 19, no. 2, pp. 207–215, 2018. MEDIATORS of INFLAMMATION The Scientific Gastroenterology Journal of World Journal Research and Practice Diabetes Research Disease Markers Hindawi Hindawi Publishing Corporation Hindawi Hindawi Hindawi Hindawi www.hindawi.com Volume 2018 http://www www.hindawi.com .hindawi.com V Volume 2018 olume 2013 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 International Journal of Journal of Immunology Research Endocrinology Hindawi Hindawi www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 Submit your manuscripts at www.hindawi.com BioMed PPAR Research Research International Hindawi Hindawi www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 Journal of Obesity Evidence-Based Journal of Journal of Stem Cells Complementary and Ophthalmology International Alternative Medicine Oncology Hindawi Hindawi Hindawi Hindawi Hindawi www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2013 Parkinson’s Disease Computational and Behavioural Mathematical Methods AIDS Oxidative Medicine and in Medicine Neurology Research and Treatment Cellular Longevity Hindawi Hindawi Hindawi Hindawi Hindawi www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Journal of Oncology Hindawi Publishing Corporation

A Robust Gene Expression Prognostic Signature for Overall Survival in High-Grade Serous Ovarian Cancer

Download PDF
13 pages

Loading next page...
 
/lp/hindawi-publishing-corporation/a-robust-gene-expression-prognostic-signature-for-overall-survival-in-XLsILTg05A

References (43)

Publisher
Hindawi Publishing Corporation
Copyright
Copyright © 2019 Yue Zhao et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
ISSN
1687-8450
eISSN
1687-8469
DOI
10.1155/2019/3614207
Publisher site
See Article on Publisher Site

Abstract

Hindawi Journal of Oncology Volume 2019, Article ID 3614207, 12 pages https://doi.org/10.1155/2019/3614207 Research Article A Robust Gene Expression Prognostic Signature for Overall Survival in High-Grade Serous Ovarian Cancer 1,2 3 4 5 6 Yue Zhao, Shao-Min Yang, Yu-Lan Jin, Guang-Wu Xiong, Pin Wang, 7 7 2 7 Antoine M. Snijders, Jian-Hua Mao, Xiao-Wei Zhang , and Bo Hang Department of Gynecology, e First Affiliated Hospital, Nanjing Medical University, Nanjing 210000, China Department of Obstetrics and Gynecology, Peking University ird Hospital, Beijing 100191, China Department of Pathology, School of Basic Medical Sciences, ird Hospital, Peking University Health Science Center, Beijing 100191, China Department of Pathology, Beijing Obstetrics and Gynecology Hospital, Capital Medical University, Beijing Maternal and Child Health Care Hospital, Beijing 100026, China Women & Children Health Center, e ird Affiliated Hospital of Chongqing Medical University, Chongqing 401120, China Department of Gastroenterology, Nanjing Drum Tower Hospital, Nanjing University Medical School, Nanjing, Jiangsu 210008, China Biological Systems and Engineering Division, Lawrence Berkeley National Laboratory, Berkeley, CA 94720, USA Correspondence should be addressed to Xiao-Wei Zhang; jiangsuzhaoy@126.com and Bo Hang; bo_hang@lbl.gov Received 26 March 2019; Accepted 17 July 2019; Published 7 November 2019 Academic Editor: Pierfrancesco Franco Copyright © 2019 Yue Zhao et al. 2is is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. 2e objective of this research was to develop a robust gene expression-based prognostic signature and scoring system for predicting overall survival (OS) of patients with high-grade serous ovarian cancer (HGSOC). Transcriptomic data of HGSOC patients were obtained from six independent studies in the NCBI GEO database. Genes significantly deregulated and as- sociated with OS in HGSOCs were selected using GEO2R and Kaplan–Meier analysis with log-rank testing, respectively. Enrichment analysis for biological processes and pathways was performed using Gene Ontology analysis. A resampling/cross- validation method with Cox regression analysis was used to identify a novel gene expression-based signature associated with OS, and a prognostic scoring system was developed and further validated in nine independent HGSOC datasets. We first identified 488 significantly deregulated genes in HGSOC patients, of which 232 were found to be significantly associated with their OS. 2ese genes were significantly enriched for cell cycle division, epithelial cell differentiation, p53 signaling pathway, vasculature development, and other processes. A novel 11-gene prognostic signature was identified and a prognostic scoring system was developed, which robustly predicted OS in HGSOC patients in 100 sampling test sets. 2e scoring system was further validated successfully in nine additional HGSOC public datasets. In conclusion, our integrative bioinformatics study combining transcriptomic and clinical data established an 11-gene prognostic signature for robust and reproducible pre- diction of OS in HGSOC patients. 2is signature could be of clinical value for guiding therapeutic selection and individualized treatment. patients are diagnosed with OC at an advanced stage. 1. Introduction Globally, more than 239,000 women are diagnosed with OC Ovarian cancer (OC) represents the most lethal gynaeco- and 152,000 succumb to this disease each year [2]. logical malignancy and the fifth leading cause of death in OC has been shown to have considerable complexity and women, with a 5-year survival rate around 10% [1]. Due to heterogeneity in biology, drug response, and survival time lack of early screening and diagnostic measures, most [3], representing a major obstacle for its precision medicine 2 Journal of Oncology practice. OCs of epithelial origin constitute approximately (GSE32063, GSE19829 GPL570, GSE30161, GSE3149, OV- 90% of all the cases, whereas ovarian sex cord stromal tumor, AU-ICGC, GSE14764, GSE9891, GSE 17260, and ovarian germ cell tumor, and secondary tumor of ovarian GSE32062) were used for independent validation of the gene metastasis (e.g., Krukenberg tumor) are less frequent [4]. signature and prognostic scoring system. High-grade serous ovarian carcinoma (HGSOC) is the most predominant in epithelial OCs, accounting for 70–80% of 2.2. Statistical Analysis. By employing a 1.5-fold change OC deaths [5]. 2e majority of HGSOCs can be grouped into cutoff and adjusted p-value<0.05, the differentially expressed four subtypes based on gene overexpression levels specific genes between normal versus HGSOC tissues were identified for each subtype: mesenchymal, immunoreactive, differen- with GEO2R. Differentially expressed genes associated with tiated, and proliferative [6]. OS in patients with HGSOC were selected using KM survival HGSOC has been characterized by both genetic alter- analysis (Kaplan–Meier plotter (http://kmplot.com)) with a ations, including inherited BRCA gene mutations, TP53 hazard ratio (HR) with 95% confidence intervals and log-rank mutations, DNA damage, chromosomal instability [6, 7], p value cutoff for each gene at 0.05 [19]. and changes in RNA and miRNA expression and methyl- ation status [8]. Microarray and next-generation sequencing technologies have become vital tools for identifying these 2.3. Gene Ontology Pathway Analysis and Network Construction. Metascape (http://www.metascape.org) was changes genomewide, providing novel opportunities for the identification of biomarkers for diagnosis, prognosis, ther- used to assess overrepresentation of Gene Ontology cate- gories in biological networks [20]. Cytoscape 3.4.0 (http:// apeutic targets, and treatment response. For instance, many multigene biomarkers based on transcription patterns have www.cytoscape.org) was applied to generate and visualize the gene coexpression networks, to better understand the been associated with prognosis across tumor types [9–14]. A number of groups have sought to use genomewide gene biological processes enriched, as well as their relationships in the form of a network instead of the tabular text format expression data to identify multigene signatures aimed at [21, 22]. Note that KEGG pathway (http://www.genome.jp/ predicting clinical outcomes, therapy responses, and sub- kegg), GO Biological Processes (http://geneontology.org), types in OC [13–18]. Many existing signatures were gen- Reactome Gene Sets (http://www.reactome.org), and erated using partial genome annotations, limited number of patients, or used targeted gene selecting. 2us, it is very CORUM (http://mips.helmholtz-muenchen.de/corum) were ontology sources of gene network, pathway, and process much warranted to identify and develop clinically valuable gene signatures for OC prognosis, especially when based on enrichment analysis. comprehensive and unbiased whole-genome data. In this study, we employed a multistep bioinformatic 2.4. Gene Expression Signature-Based Prognostic Risk Score. strategy that uses omics information and clinical data to Clinical data of HGSOC patients were obtained from the build a gene expression prognostic scoring system in TCGA dataset (http://cancergenome.nih.gov), with which HGSOC. We previously developed this approach to identify a biomarker panel associated with OS was reachable. 100 and successfully validate a 53-gene signature associated with random selections of 307 patients from TCGA were OS of gastric cancer [11] and a 27-gene signature for lung conducted and used as training sets. 2e remaining pa- adenocarcinoma [12]. Here, we used fifteen publicly avail- tients for each selection were used as test sets to validate able datasets of HGSOCs; six were used to identify an 11- the reliability of the identified biomarker panel for gene signature associated with patient prognosis using Cox prognosis. regression analysis and cross-validation. We then used nine Forward conditional Cox regressions using SPSS were independent HGSOC datasets to validate the prognostic carried out on each of the 100 training sets in order to isolate scoring system and signature’s performance. Moreover, in the biomarker panel. Selected genes were recorded and those comparison with an existing 5-gene expression signature for that appeared in at least 20% of 100 training sets were in- ovarian serous cystadenocarcinoma (CAC) [15], we showed cluded in our biomarker panel. Subsequently, Cox re- that our signature was superior in determining overall gression was repeated on all 100 training sets using our 11- survival for this type of epithelial ovarian carcinoma. gene signature as covariates and using the forced entry (enter) method to obtain the coefficient values for each 2. Materials and Methods biomarker. 100 coefficients for every gene in the biomarker panel were then obtained, and the average of them was used 2.1. Patient Datasets. To broadly mine all the available in- to estimate the true coefficient of each gene. A formula was formation on HGSOCs, we have screened and used 15 in- created to act as the prognostic scoring system, and all the dependent datasets in the current study. Six public datasets patients can get their scores accordingly: from the NCBI Gene Expression Omnibus (GSE18520, GSE26712, GSE40595, GSE38666, GSE27651, and GSE2328) 􏽘(gene i coefficient)∗ (gene i expression level). (1) provided the HGSOC gene transcript data to identify genes i�1 differentially expressed between tumor and normal ovarian tissues. TCGA HGSOC data were used to identify the gene 2e patients in the training sets were ranked by their signature and develop the prognostic scoring system for prognostic scores and divided into three equal-sized cohorts. predicting OS of patients. Nine additional datasets 2e corresponding prognostic scores at cut points were Journal of Oncology 3 recorded and averaged as the true cut point scores, with <0.05; Figure 2(b) and Table S3). 2e hazard ratio (HR) for which the patients in the test sets were also split into three 82 genes was <1 (higher gene expression associated with good prognosis), which are referred as protective genes, groups: “good”, “intermediate”, and “poor” groups. Dif- ferences in OS among the three groups in all the test sets whereas 150 genes had a HR >1 (higher gene expression were determined by constructing Kaplan–Meier plots and associated with poor prognosis), which are considered risk performing log-rank tests. genes. 3.3. Gene Ontology (GO) Analysis of Prognostic Genes in 2.5. Validation in Independent Datasets and Comparison with HGSOC. To understand the potential biological functions of an Existing Signature. 2e 11-gene biomarker panel was the 232 genes significantly associated with OS in HGSOC further validated in nine independent datasets (Table S1). patients, we conducted Gene Ontology (GO) analysis using New coefficients for the 11 genes were obtained from Cox Metascape and found significant enrichment of many cel- regression. Prognostic scores for all patients were calculated, lular process and pathway-related genes associated with and patients were ranked based on their scores and divided cancer development including cell division, epithelial cell into three equal-sized cohorts. Kaplan–Meier analysis and a differentiation, p53 signaling pathway, and vasculature de- log-rank test were conducted to determine differences in velopment (Figure 3(a) and Table S4). 2e interconnectivity survival, as previously described [11, 12]. We compared the performance of our 11-gene signature of related GO terms was visualized using Cytoscape where individual GO terms are displayed as nodes connected based with a recently published 5-gene signature for prognosis of on similarity (Figure 3(b)). ovarian serous CAC [15]. A multivariate Cox regression analysis was conducted with the 5 genes on the same 100 training sets as described above for our inner validation. 3.4. Establishment of an 11-Gene Prognostic Scoring System in Coefficients for each of the 5 genes used in [15] and scores of HGSOCs. Figure 4(a) shows the strategy we employed to all 307 patients were calculated as above. 2en patients were isolate a prognostic biomarker signature and to develop a divided into tertiles (good, intermediate, and poor) based on scoring system based on the 232 genes that were found to be their prognostic scores, and the cut point scores were significantly associated with OS in HGSOC patients. We first recorded and averaged. Kaplan–Meier analysis was per- used a random resampling method to split 307 patients from formed, and a log-rank test was used to demonstrate dif- the TCGA dataset into 100 training (200 patients) and 100 ferences in OS among different groups for all test sets. testing (107 patients) sets. 2e training sets were then used to isolate a prognostic signature, and the testing sets were used 3. Results for validation. First, we performed a multivariate Cox re- gression analysis in all 100 training sets to identify statis- 3.1. Identification of Deregulated Genes in HGSOCs. To tically significant independent genes within the 232 genes for identify genes that are consistently deregulated in HGSOC, predicting OS. Genes that recurred in at least 20% of 100 we performed a meta-analysis and compared gene transcript training were assembled into an 11-gene signature: levels in six publically available datasets containing tran- RAD51AP1, CADPS2, DSE, ITGB8, PDE10A, GALNT10, scriptomic data for both HGSOC and normal ovarian tissues SNX1, MTHFD2, C9orf16, PYCR1, and ARL4 (Table S5). For (n � 397 from GSE18520, GSE26712, GSE40595, GSE38666, each of the 11 genes in the signature, gene function and GSE27651, and GSE2328) using GEO2R. For each dataset, known roles in ovarian and other cancers are summarized in we compared HGSOC gene expression to gene expression in Table S6. normal ovarian tissues (Figure 1). A prognostic score for each cancer patient was used to 2e criteria for significant differential expression for assess each patient’s risk of death and was defined as the each gene were set to a 1.5-fold change and adjusted p-value linear combination of logarithmically transformed gene <0.05. A total of 562 probe IDs (260 downregulated and 302 expression levels weighted by average Cox regression co- upregulated) were consistently up- or downregulated across efficients (Table S7) obtained from 100 training datasets all six datasets, representing 488 unique genes (222 down- [11, 12]. 2e prognostic scores were assigned for all patients regulated and 266 upregulated) (Figure 1 and Table S2). in both training and test sets. In each training set, the pa- tients were divided into tertiles based on their prognostic score. 2e cutpoint scores were recorded and averaged for 3.2. Analysis of Deregulated Genes and Overall Survival of HGSOCs. 2e prognostic value for each of the 488 each of 100 training sets. Based on the average scores, each deregulated genes individually in HGSOC patients was test set was split into three groups, i.e., good, intermediate, evaluated in a large public clinical database which integrates and poor. We then performed Kaplan–Meier and log-rank gene expression and patient survival using Kaplan–Meier test analysis to determine significant differences in OS survival analysis (Figure 2(a)). 2e effects of high or low among different groups for all test sets (Figure 4(b)). 2e expression levels of these genes on OS were assessed using hazard ratios (HR) for the “intermediate” and “poor” groups in comparison with the “good” groups were calculated for Kaplan–Meier survival analysis and compared statistically using the log-rank test, with representative genes shown in each test set. In 99% of the test sets, patients in the “poor” groups had a significantly shorter OS than those in the Figure 2(b). 2e results showed that 232 out of the 488 genes were significantly associated with OS (adjusted p-value “good” groups (HR confidence interval above “1”) 4 Journal of Oncology GSE26712 (11262) GSE27651 (11959) GSE40595 GSE6008 (37330) 1182 245 (6733) 130 84 GSE38666 1654 326 (17903) 409 1398 80 108 222 475 280 112 39 49 77 145 392 590 1207 392 189 216 110 164 67 84 6551 968 125 101 GSE18520 (24049) Figure 1: Human datasets of ovarian cancer and normal sample tissues. Samples were obtained from six independent gene transcript datasets containing HGSOC and normal ovarian cases. To identify genes (common probe IDs) consistently deregulated in HGSOC, a fold- change cutoff of 1.5 and adjusted p-value <0.05 were used for each dataset. (Figure 4(c), top panel), while in more than 60% of the test performing a multivariate Cox regression analysis using the sets, patients in the “intermediate” groups showed a sig- same strategy described in Figure 4 where for 100 training nificantly shorter OS than those in the “good” groups sets, coefficients for each of the 5 genes and scores of all the (Figure 4(c), bottom panel). 2ese results validated the 307 patients were calculated. discriminative ability of this 11-gene signature and prog- Figure 6 shows the HR and 95% confidence interval for the “intermediate” and “poor” groups in comparison with nostic scoring system to stratify patients with good or worse prognosis. the “good” groups in the 100 test sets. For the 5-gene panel, in 90% of the testing sets, patients in the “poor” groups had a significantly shorter OS than those in the “good” groups. For 3.5. Independent Validation of the 11-Gene Scoring System. the “intermediate” groups vs. “good” groups, only in 12% of To further validate our 11-gene signature, we tested it in nine the testing sets, patients showed a significantly shorter OS. In independent OC datasets (Table S1). Prognostic scores for all comparison, for our 11-gene signature, these two numbers patients were calculated and patients were ranked based on are 99% and 61%, respectively. In addition, the median HR their score. Significant differences were identified using of the 11-gene signature was on average 1.46-fold higher in Kaplan–Meier analysis across all nine datasets between the “intermediate” vs. “good” groups and 1.73-fold higher in patient cohorts of “good” and “poor” prognosis. Patients the “poor” vs. “good” groups compared to the 5-gene sig- with a high prognostic score had a significantly shorter OS nature (Figure 6). 2ese results indicate that the 11-gene than those patients who scored low (p< 0.05) (Figure 5). 2e signature has discriminative ability for determining OS in HR values range from 1.94 to 9.76 (Table S1) We conclude ovarian CAC patients, which is also significantly superior to that the 11-gene prognostic scoring system reproducibly the 5-gene panel. predicts overall survival of HGSOC patients. 4. Discussion 3.6. Comparison with an Existing Prognostic Signature. We compared the performance of our 11-gene signature Identification and development of reliable predictive bio- markers and new therapeutic targets are critical for sig- with a recently published 5-gene expression signature pre- dicting clinical outcome of ovarian serous CAC [15] by nificantly improving cancer patient outcomes. 2e Journal of Oncology 5 GFPT2 (205100_at) HSD17B6 (37512_at) 1.0 1.0 590 probe IDs HR = 1.58 (1.31−1.89) HR = 1.47 (1.25−1.72) Log-rank P = 7.3e −07 Log-rank P = 1.8e −06 0.8 0.8 0.6 0.6 Filter genes with the same direction fold change 0.4 0.4 0.2 0.2 0.0 0.0 488 genes 50 100 150 200 250 0 50 100 150 200 250 (562 probe IDs) Time (months) Time (months) Number at risk Number at risk Low 320 93 25 4 1 0 Low 463 124 27 5 1 0 High 887 174 24 4 1 0 High 744 143 22 3 1 0 Expression Expression Low Low Filter genes associated High High with OS ALDH1A2 (207016_s_at) PART1 (205833_s_at) 1.0 1.0 HR = 1.44 (1.23−1.68) HR = 0.67 (0.56−0.8) Log-rank P = 4.9e −06 Log-rank P = 1.2e −05 0.8 0.8 232 genes 0.6 0.6 0.4 0.4 0.2 0.2 Functional annotation 0.0 0.0 50 100 150 200 250 0 50 100 150 200 250 Time (months) Time (months) Number at risk Number at risk Low 507 135 27 4 2 0 Low 895 180 34 7 1 0 High 700 132 22 4 0 0 High 312 87 15 1 1 0 Expression Expression Low Low High High (a) (b) Figure 2: Identification of genes associated with prognostic function in HGSOC. (a) 2e 562 consistently deregulated probe IDs identified represent 488 genes in the cancer patients. 2rough Kaplan–Meier survival analysis, 232 genes were found to be significantly associated with overall survival of HGSOC patients. Functional annotation was carried out for the 232 genes. (b) Examples of Kaplan–Meier survival curves for four individual genes significantly associated with overall survival in HGSOC patients, which was divided into two groups to maximize the difference in survival using log-rank testing between groups. We used HR and log-rank p-value for the curve comparison between the groups. objective of this work was to use a multistep bioinformatics CAC. Taken together, the 11-gene signature could be of analytic strategy we developed previously [11, 12] to an- translational value for clinical use. We are currently alyze six publicly available omics and clinical datasets to working on the development of a multiplex high- generate a robust prognostic signature for patients with throughput assay to facilitate the clinical use of the sig- HGSOC. We first identified 232 genes associated with OS nature. To date, there are still no clinically useful prognostic biomarkers/scores in OC. However, two multigene ex- that served as candidate markers to provide a prediction of the prognosis of HGSOC patients. Eventually, we selected pression-based scores, the Oncotype DX 21-gene breast an 11-gene prognostic signature and scoring system cancer assay developed by Genomic Health [9, 23] and the showing strong discriminative power to separate patients MammaPrint 70-gene breast cancer recurrence assay by with good or poor survival. Moreover, the results were Agendia [24], have been utilized to guide treatment de- independently validated in each of the nine independent cisions, such as for adjuvant chemotherapy in breast cancer HGSOC datasets. We also demonstrated that our 11-gene [23]. 2ese two tests represent the first prognostic gene signature has higher predictive power compared to an expression assays that have successfully passed multiple existing prognostic panel developed for ovarian serous independent clinical trials. Probability Probability Probability Probability 6 Journal of Oncology GO:0051301: cell division GO:0001568: blood vessel development GO:0071229: cellular response to acid chemical R-HSA-3560782: Diseases associated with glycosaminoglycan metabolism GO:0043009: chordate embryonic development GO:0033273: response to vitamin GO:0072001: renal system development GO:0071407: cellular response to organic cyclic compound GO:0030855: epithelial cell differentiation GO:0000079: regulation of cyclin-dependent protein serine/threonine kinase activity hsa04115: p53 signaling pathway R-HSA-109582: Hemostasis GO:1902850: microtubule cytoskeleton organization involved in mitosis GO:1905114: cell surface receptor signaling pathway involved in cell-cell signaling GO:0035690: cellular response to drug GO:0010039: response to iron ion GO:0048863: stem cell differentiation GO:0008285: negative regulation of cell proliferation GO:0045664: regulation of neuron differentiation GO:0007423: sensory organ development (a) GO:0051301: cell division GO:0071229: cellular response to acid chemical GO:0043009: chordate embryonic development GO:0072001: renal system development GO:0030855: epithelial cell differentiation hsa04115: p53 signaling pathway GO:1902850: microtubule cytoskeleton organization involved in mitosis GO:0035690: cellular response to drug GO:0048863: stem cell differentiation GO:0045664: regulation of neuron differentiation GO:0001568: blood vessel development R-HSA-3560782: Diseases associated with glycosaminoglycan metabolism GO:0033273: response to vitamin GO:0071407: cellular response to organic cyclic compound GO:0000079: regulation of cyclin-dependent protein serine/threonine kinase activity R-HSA-109582: Hemostasis GO:1905114: cell surface receptor signaling pathway involved in cell-cell signaling GO:0010039: response to iron ion GO:0008285: negative regulation of cell proliferation GO:0007423: sensory organ development 0 2 4 6 8 10 12 14 16 –log 10 (p) (b) Figure 3: Gene Ontology analysis of 232 genes associated with OS. (a) Network layout of the clusters generated with the complete list of the 232 OS-associated genes in HGSOC. Each node represents one enriched term, where its size is proportional to the number of genes associated with each term, and its color representing its cluster identity (i.e., nodes of the same color belong to the same cluster). All similar terms with a kappa similarity score>0.3 are connected by edges (the thicker the edge, the higher the similarity). One term from each cluster was selected to describe the general function of each cluster. Created by Metascape (http://metascape.org). (b) Top 20 most significant GO categories associated with the 232 genes. Microarray and next-generation sequencing technolo- novel panel of prognostic biomarkers is the first step in gies broadened the accessibility of large cancer genomewide developing a practically valuable assay/score in a clinical expression profiles. Taking advantage of these unbiased setting. 2e next steps include multicenter clinical trials and genomewide approaches, we established multigene signa- prospective trials that allow further validation of the efficacy tures for predictive and prognostic purposes, including the and accuracy of the signature, in order to make a successful 11-gene signature described in this study. To discover a clinical translation. It should be mentioned that microarray Journal of Oncology 7 TCGA 307 patients Random sampling Training set Test set (200 patients) (107 patients) Multivariate cox regression Identify genes selected into Cox regression model Identify 11-gene signature for OS Prognostic score for OS Validation (a) 1.0 1.0 Log-rank (Mantel-Cox) Log-rank (Mantel-Cox) Chi-square = 44.03 Chi-square = 37.66 p < 0.0001 p < 0.0001 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0.1 0.1 0 20 40 60 80 100 120 140 0 25 50 75 100 125 Time (months) Time (months) Bad Bad Good Good Intermediate Intermediate (b) Figure 4: Continued. Survival 100 times Survival 100 times 8 Journal of Oncology Poor vs. Good 0 20 40 60 80 100 120 100 independent samplings HR 95% CI Intermediate vs. Good 0 20 40 60 80 100 120 100 independent samplings HR 95% CI (c) Figure 4: Strategy to generate an 11-gene prognostic signature and its performance evaluation. (a) We employed multivariate Cox re- gression analysis on 100 training sets through random sampling for the 232 genes and identified 11 genes selected into our Cox regression model. Such a signature was used to generate a prognostic scoring system, which was further validated using 100 randomly assembled test sets. (b) Representative Kaplan–Meier overall survival curves in two test sets. 2ese curves were separated into tertiles according to the prognostic score calculated using the 11-gene signature. (c) HR values and their 95% confidence interval across the 100 test sets, calculated using a Cox model based on the prognostic score comparing poor vs. good (top) and intermediate vs. good (bottom). data-based analyses have generated many single and mul- Ovarian cancer, like many other cancers, occurs through the accumulation of genetic alterations, which can result in tiple gene biomarkers/signatures associated with prognosis of specific types of cancers including OC. For OC, several deregulation of gene expression. So far, there is still limited prognostic signatures have been developed based on dif- information on the genes that are associated with prognosis ferent platforms, as described before. While these signatures of OC. Table S6 summarizes the known functions for each can predict OC survival, some of them were developed based gene in the 11-gene panel in tumor development and on limited patient numbers or conducted within a single prognostic relevance. Of them, six have already been im- medical center. In addition, signatures developed in earlier plicated in the development and progression of HGSOCs in years were either based on incomplete genome annotations previously published studies [25–31]. Six genes (RAD51AP1, or based solely on existing knowledge. Nevertheless, we DSE, ITGB8, GALNT10, SNX1, and MTHFD2) were re- ported to provide useful prognostic information about the expect that with ongoing and future prospective studies, some of these preclinical biomarker signatures, including the survival in various types of cancer [25, 27, 28, 32–36], in- 11-gene signature described here, will be fully evaluated for cluding three genes (RAD51AP1, ITGB8, and GALNT10) their value in the clinical settings. which were reported to be prognostic for OC [25, 28, 32]. Hazard ratio Hazard ratio Journal of Oncology 9 GSE32063 GSE3149 HR: 4.76 (1.81–12.53) HR: 3.2 (1.77–5.81) 1.0 1.0 p = 4.4E –04 p = 5.389E –05 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0.0 0.0 0 20406080 100 0 50 100 150 20 19 18 15 12 9 9 8 6 3 2 62 40 26 20 12 6 2 20 18 15 9 8 4 1 1 1 1 1 62 41 24 9 4 2 0 Time (months) Time (months) (a) (b) GSE30161 GSE19829 HR: 3.3 (1.59–6.84) HR: 2.9 (1.08–7.8) 1.0 1.0 p = 7.2E –04 p = 0.024 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0.0 0.0 0 20 40 60 80 100 0 50 100 150 200 250 300 350 29 25 24 20 14 14 13 12 8 7 5 5 3 1 1 14 14 11 97655333 29 25 18 13 12 10 5 3 2 1 0 0 0 0 0 14 12 996320000 Time (months) Time (months) (c) (d) Figure 5: Validation of the 11-gene signature using four independent ovarian cancer cohorts. We analyzed the Kaplan–Meier plots generated for the four cohorts used by applying the 11-gene signature. 2e patient cohort was split by the median based on the prognostic index, and the log-rank p-values of the curve comparison between the risk groups and HR are shown. 2e HR values and 95% confidence intervals were calculated using Cox survival analysis. RAD51AP1, which encodes an RAD51 accessory protein, its expression is an independent biomarker for predicting participates in the homologous recombination DNA damage unfavorable survival of patients with HGSOCs [27]. In- response pathway. 2e finding in this study is in agreement tegrative network analysis of TCGA data has shown that with DNA repair defects in HGSOCs. Upregulation of GALNT10 was highly predictive for the OS of ovarian cancer RAD51AP1 predicted poorer OS in patients with ovarian patients [28]. 2ere is evidence that SNX1 may play a role in cancer [25]. DSE (SART2) gene has been shown to be fre- tumorigenesis and its downregulation is significantly corre- quently upregulated in human brain tumors and other types lated with poor OS of colon cancer patients [29]. MTHFD2 is of cancer [37]. Moreover, elevated DSE expression in glioma a gene associated with cancer development, and its high expression is associated with poor prognosis of many types of is associated with a worse tumor grade and poor OS [37]. Elevated levels were also detected in cervical, ovarian, and cancer, for example [36–38]. Five of these genes, CADPS2, MTHFD2, PDE10A, PYCR1, and ARL4, have never been endometrial cancers [34]. ITGB8 encodes a β-subunit of integrin, and integrins play a regulatory role on cancer cells reported to have a role in OC (Table S6). Interestingly, the through survival- and metastasis-related signaling pathways genes in the multigene panels reported in the literature, in- [34]. Upregulation of ITGB8 has been shown in several types cluding our 11-gene signature, are rarely overlapping, which of cancers, including HGSOCs. In addition, it was found that may reflect the disparity in tumor samples, microarray Probability Probability Probability Probability 10 Journal of Oncology p < 0.0001 p < 0.0001 (a) (b) Figure 6: Comparison of HR and 95% confidence interval between the 11-gene and 5-gene signatures. For both 11-gene and 5-gene signatures, the HR of all the 100 test sets was calculated using a Cox model based on the prognostic score between groups (poor vs. good: top; intermediate vs. good: bottom). 2e differences between the two signatures were significant for both the poor vs. good groups and the intermediate vs. good groups (p< 0.0001). designs, database selection, and analytical approaches. 2e Acknowledgments genes in this signature may be novel potential therapeutic 2is work was supported by grants from the National Key targets for HGSOCs. Research and Development Program for Reproductive 2e genes included in our signature might also be po- Health and Major Birth Defects Control and Prevention tential biomarkers or targets for the treatment of OC. (2017YFC1002004) and China Scholarship Council Personalized treatment is often highlighted in today’s (201606010313). clinical practice, where the molecular features such as ge- netic background of an individual patient’s tumor determine the prime treatment modalities. For example, Prexasertib Supplementary Materials (LY2606368), a cell cycle checkpoint kinase 1 and 2 in- Table S1: summary of independent validation of the 11-gene hibitor, showed clinical activity and was tolerable in HGSOC signature in 9 datasets. Table S2: list of genes that are patients with BRCA wild-type disease [39]. consistently deregulated in HGSOC across six datasets using In conclusion, as the most lethal gynaecological malig- criteria: adjusted p< 0.05 and fold change>1.5. Table S3: the nancy, OC is undoubtedly a challenge for patients, medical impact of deregulated genes on overall survival (OS). Genes practitioners, and researchers. In this study, with an unbiased significantly associated with OS are highlighted in yellow. multistep bioinformatics analytic strategy, we identified an Table S4: top 20 altered gene clusters identified in the 232 11-gene prognostic biomarker panel which robustly and deregulated genes that are significantly associated with OS. accurately predicts overall survival in patients with HGSOCs. Gene set enrichment analysis was conducted using Meta- Gene Ontology analysis revealed several important enriched scape (http://metascape.org). Count represents the number cellular processes and pathways in HGSOCs. Together, our of genes with membership in the given ontology term. “%” results pave the way for developing a clinical assay for guiding represents the percentage of total 232 genes associated with therapeutic selection and individualized treatment. OS that are found in the given ontology term. Log10 (p) is the p-value in log base 10. Table S5: frequency by which each Abbreviations gene appeared in the Cox regression model among 100 resampling training sets. 2e signature genes are highlighted OC: Ovarian cancer in yellow. Table S6: the function and role of 11 genes in the HGSOC: High-grade serous ovarian cancer prognostic signature in normal and HGSOC cells. Table S7: CAC: Cystadenocarcinoma the average Cox regression coefficient for each gene used to HR: Hazard ratio calculate the prognostic score. (Supplementary Materials) K-M plot: Kaplan–Meier plot OS: Overall survival TCGA: 2e Cancer Genome Atlas. References [1] R. L. Siegel, K. D. Miller, and A. Jemal, “Cancer statistics, Data Availability 2017,” CA: A Cancer Journal for Clinicians, vol. 67, no. 1, pp. 7–30, 2017. 2e data used to support the findings of this study are [2] B. M. Reid, J. B. Permuth, and T. A. Sellers, “Epidemiology of available from the corresponding author upon request. ovarian cancer: a review,” Cancer Biology & Medicine, vol. 14, no. 1, pp. 9–32, 2017. [3] P. T. Kroeger Jr., and R. Drapkin, “Pathogenesis and het- Conflicts of Interest erogeneity of ovarian cancer,” Current Opinion in Obstetrics 2e authors declare that they have no conflicts of interest. and Gynecology, vol. 29, no. 1, pp. 26–34, 2017. HR Poor vs. good 11-gene 5-gene HR Intermediate vs. good 11-gene 5-gene Journal of Oncology 11 [4] J. Prat, “New insights into ovarian cancer pathology,” Annals [21] G. Bindea, B. Mlecnik, H. Hackl et al., “ClueGO: a cytoscape of Oncology, vol. 23, no. 10, pp. x111–x117, 2012. plug-in to decipher functionally grouped gene ontology and [5] R. Vang, I.-M. Shih, and R. J. Kurman, “Ovarian low-grade pathway annotation networks,” Bioinformatics, vol. 25, no. 8, and high-grade serous carcinoma,” Advances in Anatomic pp. 1091–1093, 2009. Pathology, vol. 16, no. 5, pp. 267–282, 2009. [22] I. H. Goenawan, K. Bryan, and D. J. Lynn, “DyNet: visuali- [6] Cancer Genome Atlas Research Network, “Integrated geno- zation and analysis of dynamic molecular interaction net- mic analyses of ovarian carcinoma,” Nature, vol. 474, works,” Bioinformatics, vol. 32, no. 17, pp. 2713–2715, 2016. no. 7353, pp. 609–615, 2011. [23] J. A. Sparano, R. J. Gray, D. F. Makower et al., “Adjuvant [7] U. A. Matulonis, A. K. Sood, L. Fallowfield et al., “Ovarian chemotherapy guided by a 21-gene expression assay in breast cancer,” Nature Reviews Disease Primers, vol. 2, no.1, p.16061, cancer,” New England Journal of Medicine, vol. 379, no. 2, pp. 111–121, 2018. [8] G. E. Konecny, C. Wang, H. Hamidi et al., “Prognostic and [24] F. Cardoso, L. J. van’t Veer, J. Bogaerts et al., “70-gene sig- therapeutic relevance of molecular subtypes in high-grade nature as an aid to treatment decisions in early-stage breast serous ovarian cancer,” JNCI: Journal of the National Cancer cancer,” New England Journal of Medicine, vol. 375, no. 8, Institute, vol. 106, no. 10, article dju249, 2014. pp. 717–729, 2016. [9] S. Paik, S. Shak, G. Tang et al., “A multigene assay to predict [25] D. Chudasama, V. Bo, M. Hall et al., “Identification of cancer recurrence of tamoxifen-treated, node-negative breast can- biomarkers of prognostic value using specific gene regulatory cer,” New England Journal of Medicine, vol. 351, no. 27, networks (GRN): a novel role of RAD51AP1 for ovarian and pp. 2817–2826, 2004. lung cancers,” Carcinogenesis, vol. 39, no. 3, pp. 407–417, [10] J. R. Kratz, J. He, S. K. Van Den Eeden et al., “A practical molecular assay to predict survival in resected non-squamous, [26] S. Tanaka, N. Tsuda, K. Kawano et al., “Expression of tumor- non-small-cell lung cancer: development and international rejection antigens in gynecologic cancers,” Japanese Journal of validation studies,” e Lancet, vol. 379, no. 9818, pp. 823– Cancer Research, vol. 91, no. 11, pp. 1177–1184, 2000. 832, 2012. [27] J. He, Y. Liu, L. Zhang, and H. Zhang, “Integrin subunit beta 8 [11] P. Wang, Y. Wang, B. Hang et al., “A novel gene expression- (ITGB8) upregulation is an independent predictor of un- based prognostic scoring system to predict survival in gastric favorable survival of high-grade serous ovarian carcinoma cancer,” Oncotarget, vol. 7, no. 34, pp. 55343–55351, 2016. patients,” Medical Science Monitor, vol. 24, pp. 8933–8940, [12] E. G. Chen, P. Wang, H. Lou et al., “A robust gene expression- based prognostic risk score predicts overall survival of lung [28] Q. Zhang, J. E. Burdette, and J. P. Wang, “Integrative network adenocarcinoma patients,” Oncotarget, vol. 9, no. 6, analysis of TCGA data for ovarian cancer,” BMC Systems pp. 6862–6871, 2018. Biology, vol. 8, no. 1, p. 1338, 2014. [13] D. Spentzos, D. A. Levine, M. F. Ramoni et al., “Gene ex- [29] L. N. Nguyen, M. S. Holdren, A. P. Nguyen et al., “Sorting pression signature with independent prognostic significance nexin 1 down-regulation promotes colon tumorigenesis,” in epithelial ovarian cancer,” Journal of Clinical Oncology, Clinical Cancer Research, vol. 12, no. 23, pp. 6952–6959, 2006. vol. 22, no. 23, pp. 4648–4658, 2004. [30] W. Ju, B. C. Yoo, I. J. Kim et al., “Identification of genes with [14] T. Bonome, D. A. Levine, J. Shih et al., “A gene signature differential expression in chemoresistant epithelial ovarian predicting for survival in suboptimally debulked patients with cancer using high-density oligonucleotide microarrays,” ovarian cancer,” Cancer Research, vol. 68, no. 13, pp. 5478– Oncology Research Featuring Preclinical and Clinical Cancer 5486, 2008. erapeutics, vol. 18, no. 2-3, pp. 47–56, 2009. [15] L. W. Liu, Q. Zhang, W. Guo, K. Qian, and Q. Wang, “A five- [31] J. Wang, C. Chen, H. F. Li, X. L Jiang, and L. Zhang, “In- gene expression signature predicts clinical outcome of ovarian vestigating key genes associated with ovarian cancer by in- serous cystadenocarcinoma,” BioMed Research International, tegrating affinity propagation clustering and mutual vol. 2016, Article ID 6945304, 6 pages, 2016. information network analysis,” European Review for Medical [16] R. W. Tothill, A. V. Tinker, J. George et al., “Novel molecular and Pharmacological Sciences, vol. 20, no. 12, pp. 2532–2540, subtypes of serous and endometrioid ovarian cancer linked to clinical outcome,” Clinical Cancer Research, vol. 14, no. 16, [32] L. Liu, Y. Xiong, W. Xi et al., “Prognostic role of N-Ace- pp. 5198–5208, 2008. tylgalactosaminyltransferase 10 in metastatic renal cell car- [17] B. Győrffy, A. Lanczky, and Z. Szallasi, “Implementing an cinoma,” Oncotarget, vol. 8, no. 9, pp. 14995–15003, 2017. online tool for genome-wide validation of survival-associated [33] X.-Y. Zhan, Y. Zhang, E. Zhai, Q.-Y. Zhu, and Y. He, “Sorting biomarkers in ovarian-cancer using microarray data from nexin-1 is a candidate tumor suppressor and potential 1287 patients,” Endocrine-Related Cancer, vol. 19, no. 2, prognostic marker in gastric cancer,” PeerJ, vol. 6, p. e4829, pp. 197–208, 2012. [18] G. P. Sfakianos, E. S. Iversen, R. Whitaker et al., “Validation of [34] J. Koseki, M. Konno, A. Asai et al., “Enzymes of the one- ovarian cancer gene expression signatures for survival and carbon folate metabolism as anticancer targets predicted by subtype in formalin fixed paraffin embedded tissues,” Gy- survival rate analysis,” Scientific Reports, vol. 8, no. 1, p. 303, necologic Oncology, vol. 129, no. 1, pp. 159–164, 2013. [19] B. Gyorffy, P. Surowiak, J. Budczies, and A. Lanczky, ´ “Online [35] H. Lin, B. Huang, H. Wang et al., “MTHFD2 overexpression survival analysis software to assess the prognostic value of predicts poor prognosis in renal cell carcinoma and is as- biomarkers using transcriptomic data in non-small-cell lung sociated with cell proliferation and vimentin-modulated cancer,” PLoS One, vol. 8, no. 12, Article ID e8224, 2013. [20] S. Tripathi, M. O. Pohl, Y. Zhou et al., “Meta- and orthogonal migration and invasion,” Cellular Physiology and Bio- integration of influenza “OMICs” data defines a role for UBR4 chemistry, vol. 51, no. 2, pp. 991–1000, 2018. in virus budding,” Cell Host & Microbe, vol. 18, no. 6, [36] K. Noguchi, M. Konno, J. Koseki et al., “2e mitochondrial pp. 723–735, 2015. one-carbon metabolic pathway is associated with patient 12 Journal of Oncology survival in pancreatic cancer,” Oncology Letters, vol. 16, no. 2, pp. 1827–1834, 2018. [37] W. C. Liao, C. K. Liao, Y. H. Tsai et al., “DSE promotes aggressive glioma cell phenotypes by enhancing HB-EGF/ ErbB signaling,” PLoS One, vol. 13, no. 6, Article ID e0198364, [38] R. Ata and C. N. Antonescu, “Integrins and cell metabolism: an intimate relationship impacting cancer,” International Journal of Molecular Sciences, vol. 18, no. 1, p. 189, 2017. [39] J.-M. Lee, J. Nair, A. Zimmer et al., “Prexasertib, a cell cycle checkpoint kinase 1 and 2 inhibitor, in BRCA wild-type re- current high-grade serous ovarian cancer: a first-in-class proof-of-concept phase 2 study,” e Lancet Oncology, vol. 19, no. 2, pp. 207–215, 2018. MEDIATORS of INFLAMMATION The Scientific Gastroenterology Journal of World Journal Research and Practice Diabetes Research Disease Markers Hindawi Hindawi Publishing Corporation Hindawi Hindawi Hindawi Hindawi www.hindawi.com Volume 2018 http://www www.hindawi.com .hindawi.com V Volume 2018 olume 2013 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 International Journal of Journal of Immunology Research Endocrinology Hindawi Hindawi www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 Submit your manuscripts at www.hindawi.com BioMed PPAR Research Research International Hindawi Hindawi www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 Journal of Obesity Evidence-Based Journal of Journal of Stem Cells Complementary and Ophthalmology International Alternative Medicine Oncology Hindawi Hindawi Hindawi Hindawi Hindawi www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2013 Parkinson’s Disease Computational and Behavioural Mathematical Methods AIDS Oxidative Medicine and in Medicine Neurology Research and Treatment Cellular Longevity Hindawi Hindawi Hindawi Hindawi Hindawi www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018

Journal

Journal of Oncology – Hindawi Publishing Corporation

Published: Nov 7, 2019

There are no references for this article.