BioMed Research International

BioMed Research International / 2016 / Article
Special Issue

Integrated Analysis of Multiscale Large-Scale Biological Data for Investigating Human Disease 2016

View this Special Issue

Research Article | Open Access

Volume 2016 |Article ID 6945304 | https://doi.org/10.1155/2016/6945304

Li-Wei Liu, Qiuhao Zhang, Wenna Guo, Kun Qian, Qiang Wang, "A Five-Gene Expression Signature Predicts Clinical Outcome of Ovarian Serous Cystadenocarcinoma", BioMed Research International, vol. 2016, Article ID 6945304, 6 pages, 2016. https://doi.org/10.1155/2016/6945304

A Five-Gene Expression Signature Predicts Clinical Outcome of Ovarian Serous Cystadenocarcinoma

Academic Editor: Jialiang Yang
Received16 Apr 2016
Accepted25 May 2016
Published05 Jul 2016

Abstract

Ovarian serous cystadenocarcinoma is a common malignant tumor of female genital organs. Treatment is generally less effective as patients are usually diagnosed in the late stage. Therefore, a well-designed prognostic marker provides valuable data for optimizing therapy. In this study, we analyzed 303 samples of ovarian serous cystadenocarcinoma and the corresponding RNA-seq data. We observed the correlation between gene expression and patients’ survival and eventually established a risk assessment model of five factors using Cox proportional hazards regression analysis. We found that the survival time in high-risk patients was significantly shorter than in low-risk patients in both training and testing sets after Kaplan-Meier analysis. The AUROC value was 0.67 when predicting the survival time in testing set, which indicates a relatively high specificity and sensitivity. The results suggest diagnostic and therapeutic applications of our five-gene model for ovarian serous cystadenocarcinoma.

1. Introduction

Ovarian serous cystadenocarcinoma is a common female genital cancer that causes more deaths than any other cancer of the female reproductive system. According to Global Cancer Statistics, approximately 230,000 women are diagnosed with ovarian cancer every year, and an estimated 150,000 women die of this disease annually [1]. Ovarian serous cystadenocarcinoma, a type of epithelial ovarian cancer, accounts for about 90% of all ovarian cancers [2]. Studies suggest that the risk factors for the disease include nulliparity, early menarche, late menopause, and family history [3]. Since the disease is often asymptomatic, the majority of patients are diagnosed at an advanced stage, with tumor invasion. Studies showed that the 5-year survival of stage I patients is greater than 90%, while that of patients in stages III to IV is less than 20% [4, 5]. The recent increase in the incidence of ovarian cancer has attracted the interest and attention of researchers worldwide.

With the development of sequencing technology, the research focus has been on the study of signature analysis for prognostic monitoring of ovarian cancer [612]. Microarray studies require precise design of probes despite the currently available and well-studied biomarkers for ovarian cancers. Other studies using miRNAs as biomarkers also suggest the limited value for clinical application, and miRNA therapy is still not clinically feasible. Compared with the foregoing methods, gene expression markers not only possess higher practical value, but also yield higher accuracy.

Here, we analyzed 303 clinical samples of ovarian serous cystadenocarcinoma and the corresponding RNA-seq data. We determined the relationship between gene expression data and survival time, in an effort to develop effective and accurate biomarkers for outcome prediction and personalized treatment.

2. Materials and Methods

2.1. Patient Samples and Gene Expression Data

We collected data from a total of 587 samples of serous cystadenocarcinoma (April 2016) from TCGA (http://cancergenome.nih.gov/) and finally used 303 samples (Table S1, at Supplementary Material available online at http://dx.doi.org/10.1155/2016/6945304) in this study after excluding 284 samples with unknown survival time or insufficient gene expression data. The 303 samples were assigned into 13 batches and randomly allocated to training and testing sets. The prognostic marker model was established with a training set containing 8 batches (batches 9, 11–15, and 17-18) with 168 samples and validated using a testing set, comprising 5 batches (batches 19–22, 24, and 409) with 135 samples.

2.2. Statistical Analysis

Initially, we screened the samples by excluding those with unclear survival time or status. We retained only those genes expressed in more than half of the samples for further analysis. The expression level was then determined by logarithmic transformation and univariate Cox regression analysis. The significance of genes with value less than 0.001 was evaluated using random forests. We selected 100 genes of the largest importance to perform multivariate Cox’s analysis. Considering the practicality of clinical testing, we established 75,287,520 models with variables ranging from one to five genes using Cox proportional hazards regression analysis [35]. Further, all the 75,287,520 models were subjected to Receiver Operating Characteristic (ROC) analysis and the model with the largest area was selected.

Kaplan-Meier analysis was then conducted in both training and testing groups to validate the efficiency of the model. In order to test the independence and reproducibility of our model, we divided the samples into different datasets according to their ages and disease stages. We then performed Kaplan-Meier analyses and ROC analyses in each condition with IBM SPSS Statistics 22 (http://www.ibm.com/analytics/us/en/technology/spss/).

3. Results

3.1. Sample Characteristics

According to the screening criteria described, we randomly allocated the 303 samples with explicit survival time, survival state, and expression data into training and testing sets for modeling and validation, respectively. The median age of diagnosis in the selected patients was 58 years, the median survival time was 949 days, and the median survival of late-stage patients was 1069 days. A single patient was found in clinical stage I and 21, 241, and 38 patients were in stages II, III, and IV, respectively. The clinical stages of two patients were unknown (Table 1).


CharacteristicPatients
Training setTesting setTotal

Age at diagnosis (years)
 Median575958
 Range34–8730–8730–87
Vital status
 Living5862120
 Dead11073183
Follow-up (days)
 Median1018883949
 Median (dead)11559191069
Clinical stage
 Stage I011
 Stage II81321
 Stage III14299241
 Stage IV182038
 Unknown022

3.2. Obtain Genes Associated with Survival Time

Subsequently, we constructed 75,287,520 models comprising factors from 1 to 5 based on the 100 genes with the highest significance in the random forest method. The survival risk score of each patient was calculated according to the corresponding risk formula in each model, and the ROC curves were drawn. We extracted a batch of 5 genes (GPR128, AGXT, CYTH3, C10orf76, and TSPAN9) (Table 2) with the largest AUROC using the following formula: risk score = (0.0796 × expression point of GPR128) + (0.3451 × expression point of AGXT) + (0.3402 × expression point of CYTH3) + (0.6198 × expression point of C10orf76) + (0.2534 × expression point of TSPAN9). All of these genes were reported previously (Table 3). The CYTH3 gene was expressed in the liver alone, playing a key role in regulating protein sorting and membrane trafficking [21]. Its use as a prognostic molecular marker in liver disease is also discussed. TSPAN9 is probably directly related to the proliferation of cancer cells. Other genes not directly correlated with the development of cancer may affect metabolism via signal transduction and indirectly affect the development of cancer.


Gene name valueHazard ratioCoefficientVariable importanceRelative importance

GPR128|848730.000921.08280.07960.00090.2478
AGXT|1890.000381.41210.3450.00050.1442
CYTH3|92650.000481.40520.34020.00050.1432
C10orf76|795910.000371.85850.61980.00090.2446
TSPAN9|108670.00081.28840.25340.00020.0518


ChromosomalStart siteEnd siteFunction

GPR128chr3100328433100414323Playing important role in the transduction of intercellular signals across the plasma membrane; related to weight gain and intestinal contraction frequency in mouse [1316].

AGXTchr2240868479240880502Expressing proteins involved in glyoxylate detoxification in the peroxisomes; its mutation causes primary hyperoxaluria type I, a severe inborn error of metabolism [1720].

CYTH3chr761617766272644Mediating the regulation of protein sorting and membrane trafficking; related to HCC (hepatocellular carcinoma) tissues and could serve as prognostic factor [2124].

C10orf76chr10101845599102056193Currently unknown; a recent study suggested the loss of C10orf76 resulted in the upregulation of several genes [2529].

TSPAN9chr1230773553286564Mediating signal transduction events that play a role in the regulation of cell development, activation, growth, and motility; associated with adhesion receptors of the integrin family and regulates integrin-dependent cell migration [3034].

3.3. Test the Predictive Ability of the Constructed Model Using Testing Set

After constructing the five-variable model with training set, we performed a Kaplan-Meier survival analysis of both training and testing sets to determine its prognostic value. In the training set, by calculating each patient’s risk score using the model, we divided the patients into two groups, designated as high-risk () and low-risk groups (), based on their risk scores. The average survival time of patients in the low-risk group was 1,443 days, longer than in the high-risk group, which was 892 days. Kaplan-Meier analysis indicated a significant difference () between the high-risk and low-risk groups in survival time [Figure 1(a)]. The prognosis of high-risk group appeared worse than that of the low-risk group, indicating that our model successfully distinguished the risk pattern. The higher risk tended to result in shorter survival time. Similar results of Kaplan-Meier analysis were found in the test group [Figure 1(b)], suggesting that our model was universally applicable in determining the risk level and predicting the survival of patients.

In order to further confirm the prognostic value of our model in predicting the survival time, we performed ROC analysis of the test group, setting 3 years as the cut-off, and calculated the risk score as the variable. The AUROC value of 0.670 (Figure 2) indicated a relatively high specificity and sensitivity.

3.4. The Independence and Reproducibility of the Five-Gene Model

The survival of patients is associated with their age, clinical stage, and other factors. To determine the independence of our model, we conducted a multivariate Cox regression analysis using age and disease stages. We found that the five-gene model was independent of age and disease stage (Table 4).


VariablesUnivariable modelMultivariable model
HR (95% CI) valueHR (95% CI) value

Training group
 Five-gene model2.672 (1.801–3.965)<0.0012.536 (1.832–3.509)<0.001
 Age1.683 (1.153–5.457)0.0071.013 (0.994–1.031)0.173
Testing group
 Five-gene model2.248 (1.397–3.620)0.0012.224 (1.379–3.586)0.001
 Age1.224 (0.772–1.941)0.3891.153 (0.726–1.830)0.546

Training group
 Five-gene model2.672 (1.801–3.965)<0.0012.725 (1.821–4.078)<0.001
 Stage1.080 (0.670–1.741)0.7520.883 (0.541–1.442)0.62
Testing group
 Five-gene model2.248 (1.397–3.620)0.0012.385 (1.387–3.562)<0.001
 Stage1.032 (0.580–1.461)0.4530.685 (0.432–1.238)0.428

Further Kaplan-Meier analysis and ROC analysis were then conducted (Table 5). We merged the training and testing sets into an overall dataset, which was divided into two separate groups by age 57. The Kaplan-Meier analysis revealed that, in both groups, patients in low-risk group survived longer than in the high-risk group (). Similar results were obtained with the groups of patients at different disease stages (stages I and II were merged because of limited specimen) except stage IV (Figure S1), which may be attributed to the relatively small sample size. However, the AUROC of this group was rather high. These analyses established that our model was independent of other risk factors and successfully distinguished low risk from high risk in each dataset.


Prognostic factorGroupKaplan-Meier valueAUROCs

Age ≤57 (146)<0.0010.653
>57 (157)0.0010.683

StageI, II (22)0.0180.625
III (241)<0.0010.664
IV (38)<0.10.778

4. Discussion

Ovarian serous cystadenocarcinoma is a common female genital cancer. Due to the absence of early-stage clinical symptoms and effective diagnosis, most patients were diagnosed with advanced disease. Further, due to the lack of effective treatment, the management of epithelial ovarian cancer is passive. Developing reliable prognostic molecular markers provides meaningful guidance for a reasonable and effective management program.

In this study, we analyzed 303 clinical samples of ovarian serous cystadenocarcinoma and the corresponding RNA-seq data, observed the correlation between gene expression and survival time, and eventually established a risk assessment model based on five factors. Two of these genes (TSPAN9 [3034], CYTH3 [2124]) were directly correlated with cancer, with CYTH3 identified as a biomarker in liver cancer.

By calculating each patient’s risk score, we found that each set showed significant differences in survival time between low-risk and high-risk groups, indicating that the model accurately predicted the mortality risk. The AUROC value in testing group is 0.670, representing a relatively high specificity and sensitivity.

In conclusion, our gene expression biomarkers can be used for accurate patient risk assessment, demonstrating practical value in predicting clinical outcomes. Our results are based on the samples derived from 303 individuals. Expanding sample size, especially including early-stage cancer patients, will further improve the prognostic value of the model.

Competing Interests

The authors declare that there is no conflict of interests regarding the publication of this paper.

Authors’ Contributions

Li-Wei Liu and Qiuhao Zhang contributed equally to this work.

Acknowledgments

This work was supported by the National Natural Science Foundation of China (31471200). The authors are grateful to the High Performance Computing Center (HPCC) of Nanjing University for doing the numerical calculations in this paper on its IBM Blade cluster system.

Supplementary Materials

Figure S1: Kaplan-Meier curves with two-sided log rank test show correlation between five-gene model and survival time in certain groups. In each set, by calculating each patient’s risk score out of the model, we divided the patients into two groups, named as high risk group and low risk group, based on their risk scores. Kaplan-Meier analysis was then performed and significant difference (p<0.001) was found between high risk and low risk group in the level of survival time except in stage IV. (a) patients under the age of 57, (b) patients over the age of 57, (c) patients from stages I and II, (d) patients from stage III, (e) patients from stage IV. Table S1: Clinical sample information of 303 ovarian serous cystuadenocarcinoma patients.

  1. Supplementary Material

References

  1. R. L. Siegel, K. D. Miller, and A. Jemal, “Cancer statistics, 2016,” CA—Cancer Journal for Clinicians, vol. 66, no. 1, pp. 7–30, 2016. View at: Publisher Site | Google Scholar
  2. Cancer Genome Atlas Research Network, “Integrated genomic analyses of ovarian carcinoma,” Nature, vol. 474, no. 7353, pp. 609–615, 2011. View at: Publisher Site | Google Scholar
  3. Y. Lee, A. Miron, R. Drapkin et al., “A candidate precursor to serous carcinoma that originates in the distal fallopian tube,” The Journal of Pathology, vol. 211, no. 1, pp. 26–35, 2007. View at: Publisher Site | Google Scholar
  4. T. Meyer and G. J. S. Rustin, “Role of tumour markers in monitoring epithelial ovarian cancer,” British Journal of Cancer, vol. 82, no. 9, pp. 1535–1538, 2000. View at: Publisher Site | Google Scholar
  5. D. M. Gershenson, C. C. Sun, K. H. Lu et al., “Clinical behavior of stage II-IV low-grade serous carcinoma of the ovary,” Obstetrics & Gynecology, vol. 108, no. 2, pp. 361–368, 2006. View at: Publisher Site | Google Scholar
  6. T. R. Adib, S. Henderson, C. Perrett et al., “Predicting biomarkers for ovarian cancer using gene-expression microarrays,” British Journal of Cancer, vol. 90, no. 3, pp. 686–692, 2004. View at: Publisher Site | Google Scholar
  7. X. Yu, X. Zhang, T. Bi et al., “MiRNA expression signature for potentially predicting the prognosis of ovarian serous carcinoma,” Tumor Biology, vol. 34, no. 6, pp. 3501–3508, 2013. View at: Publisher Site | Google Scholar
  8. N. Jin, H. Wu, Z. Miao et al., “Network-based survival-associated module biomarker and its crosstalk with cell death genes in ovarian cancer,” Scientific Reports, vol. 5, Article ID 11566, 2015. View at: Publisher Site | Google Scholar
  9. M. Schwede, D. Spentzos, S. Bentink et al., “Stem cell-like gene expression in ovarian cancer predicts type II subtype and prognosis,” PLoS ONE, vol. 8, no. 3, Article ID e57799, 2013. View at: Publisher Site | Google Scholar
  10. K. P. Prahm, G. W. Novotny, C. Høgdall, and E. Høgdall, “Current status on microRNAs as biomarkers for ovarian cancer,” APMIS, vol. 124, no. 5, pp. 337–355, 2016. View at: Publisher Site | Google Scholar
  11. P. Mapelli, E. Incerti, F. Fallanca, L. Gianolli, and M. Picchio, “Imaging biomarkers in ovarian cancer: the role of 18F-FDG PET/CT,” The Quarterly Journal of Nuclear Medicine and Molecular Imaging, vol. 60, no. 2, pp. 93–102, 2016. View at: Google Scholar
  12. I. Sedláková, J. Laco, J. Tošner, and J. Špaček, “Prognostic significance of Pgp, MRP1, and MRP3 in ovarian cancer patients,” Ceska Gynekologie, vol. 80, no. 6, pp. 405–413, 2015. View at: Google Scholar
  13. A. Chase, T. Ernst, A. Fiebig et al., “TFG, a target of chromosome translocations in lymphoma and soft tissue tumors, fuses to GPR128 in healthy individuals,” Haematologica, vol. 95, no. 1, pp. 20–26, 2010. View at: Publisher Site | Google Scholar
  14. R. Fredriksson, D. E. I. Gloriam, P. J. Höglund, M. C. Lagerström, and H. B. Schiöth, “There exist at least 30 human G-protein-coupled receptors with long Ser/Thr-rich N-termini,” Biochemical and Biophysical Research Communications, vol. 301, no. 3, pp. 725–734, 2003. View at: Publisher Site | Google Scholar
  15. T. K. Bjarnadóttir, R. Fredriksson, P. J. Höglund, D. E. Gloriam, M. C. Lagerström, and H. B. Schiöth, “The human and mouse repertoire of the adhesion family of G-protein-coupled receptors,” Genomics, vol. 84, no. 1, pp. 23–33, 2004. View at: Publisher Site | Google Scholar
  16. Y. Suzuki, R. Yamashita, M. Shirota et al., “Sequence comparison of human and mouse genes reveals a homologous block structure in the promoter regions,” Genome Research, vol. 14, no. 9, pp. 1711–1718, 2004. View at: Publisher Site | Google Scholar
  17. R. Montioli, E. Oppici, M. Dindo et al., “Misfolding caused by the pathogenic mutation G47R on the minor allele of alanine: glyoxylate aminotransferase and chaperoning activity of pyridoxine,” Biochimica et Biophysica Acta (BBA)—Proteins and Proteomics, vol. 1854, no. 10, pp. 1280–1289, 2015. View at: Publisher Site | Google Scholar
  18. E. Oppici, R. Montioli, and B. Cellini, “Liver peroxisomal alanine:glyoxylate aminotransferase and the effects of mutations associated with primary hyperoxaluria type I: an overview,” Biochimica et Biophysica Acta, vol. 1854, no. 9, pp. 1212–1219, 2015. View at: Publisher Site | Google Scholar
  19. N. Miyata, J. Steffen, M. E. Johnson, S. Fargue, C. J. Danpure, and C. M. Koehler, “Pharmacologic rescue of an enzyme-trafficking defect in primary hyperoxaluria 1,” Proceedings of the National Academy of Sciences of the United States of America, vol. 111, no. 40, pp. 14406–14411, 2014. View at: Publisher Site | Google Scholar
  20. R. Montioli, A. Roncador, E. Oppici et al., “S81L and G170R mutations causing Primary Hyperoxaluria type I in homozygosis and heterozygosis: an example of positive interallelic complementation,” Human Molecular Genetics, vol. 23, no. 22, pp. 5998–6007, 2014. View at: Publisher Site | Google Scholar
  21. Y. Fu, J. Li, M.-X. Feng et al., “Cytohesin-3 is upregulated in hepatocellular carcinoma and contributes to tumor growth and vascular invasion,” International Journal of Clinical and Experimental Pathology, vol. 7, no. 5, pp. 2123–2132, 2014. View at: Google Scholar
  22. A. W. Malaby, B. van den Berg, and D. G. Lambright, “Structural basis for membrane recruitment and allosteric activation of cytohesin family Arf GTPase exchange factors,” Proceedings of the National Academy of Sciences of the United States of America, vol. 110, no. 35, pp. 14213–14218, 2013. View at: Publisher Site | Google Scholar
  23. C. Pilling, K. E. Landgraf, and J. J. Falke, “The GRP1 PH domain, like the AKT1 PH domain, possesses a sentry glutamate residue essential for specific targeting to plasma membrane PI(3,4,5)P3,” Biochemistry, vol. 50, no. 45, pp. 9845–9856, 2011. View at: Publisher Site | Google Scholar
  24. M.-B. Poirier, G. Hamann, M.-E. Domingue, M. Roy, T. Bardati, and M.-F. Langlois, “General receptor for phosphoinositides 1, a novel repressor of thyroid hormone receptor action that prevents deoxyribonucleic acid binding,” Molecular Endocrinology, vol. 19, no. 8, pp. 1991–2005, 2005. View at: Publisher Site | Google Scholar
  25. M. K. Wojczynski, M. Li, L. F. Bielak et al., “Genetics of coronary artery calcification among African Americans, a meta-analysis,” BMC Medical Genetics, vol. 14, no. 1, article 75, 2013. View at: Publisher Site | Google Scholar
  26. C. A. Rietvelda, T. Eskoc, and G. Davies, “Common genetic variants associated with cognitive performance identified using the proxy-phenotype method,” Proceedings of the National Academy of Sciences, vol. 111, no. 38, pp. 13790–13794, 2014. View at: Publisher Site | Google Scholar
  27. A. Grupe, Y. Li, C. Rowland et al., “A scan of chromosome 10 identifies a novel locus showing strong association with late-onset Alzheimer disease,” The American Journal of Human Genetics, vol. 78, no. 1, pp. 78–88, 2006. View at: Publisher Site | Google Scholar
  28. E. D. Neto, R. G. Correa, S. Verjovski-Almeida et al., “Shotgun sequencing of the human transcriptome with ORF expressed sequence tags,” Proceedings of the National Academy of Sciences of the United States of America, vol. 97, no. 7, pp. 3491–3496, 2000. View at: Publisher Site | Google Scholar
  29. A. Castello, B. Fischer, K. Eichelbaum et al., “Insights into RNA biology from an atlas of mammalian mRNA-binding proteins,” Cell, vol. 149, no. 6, pp. 1393–1406, 2012. View at: Publisher Site | Google Scholar
  30. T. Yamaguchi, H. Nakaoka, K. Yamamoto et al., “Genome-wide association study of degenerative bony changes of the temporomandibular joint,” Oral Diseases, vol. 20, no. 4, pp. 409–415, 2014. View at: Publisher Site | Google Scholar
  31. J. Kotha, C. Zhang, C. M. Longhurst et al., “Functional relevance of tetraspanin CD9 in vascular smooth muscle cell injury phenotypes: a novel target for the prevention of neointimal hyperplasia,” Atherosclerosis, vol. 203, no. 2, pp. 377–386, 2009. View at: Publisher Site | Google Scholar
  32. M. B. Protty, N. A. Watkins, D. Colombo et al., “Identification of Tspan9 as a novel platelet tetraspanin and the collagen receptor GPVI as a component of tetraspanin microdomains,” Biochemical Journal, vol. 417, no. 1, pp. 391–401, 2009. View at: Publisher Site | Google Scholar
  33. V. Serru, P. Dessen, C. Boucheix, and E. Rubinstein, “Sequence and expression of seven new tetraspans,” Biochimica et Biophysica Acta (BBA)—Protein Structure and Molecular Enzymology, vol. 1478, no. 1, pp. 159–163, 2000. View at: Publisher Site | Google Scholar
  34. F. Berditchevski, “Complexes of tetraspanins with integrins: more than meets the eye,” Journal of Cell Science, vol. 114, no. 23, pp. 4143–4151, 2001. View at: Google Scholar
  35. A. A. Margolin, E. Bilal, E. Huang et al., “Systematic analysis of challenge-driven improvements in molecular prognostic models for breast cancer,” Science Translational Medicine, vol. 5, no. 181, Article ID 181re1, 2013. View at: Publisher Site | Google Scholar

Copyright © 2016 Li-Wei Liu 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.


More related articles

 PDF Download Citation Citation
 Download other formatsMore
 Order printed copiesOrder
Views999
Downloads353
Citations

Related articles