Differential co-expression analysis reveals a novel prognostic gene module in ovarian cancer

Abstract

Ovarian cancer is one of the most significant disease among gynecological disorders that women suffered from over the centuries. However, disease-specific and effective biomarkers were still not available, since studies have focused on individual genes associated with ovarian cancer, ignoring the interactions and associations among the gene products. Here, ovarian cancer differential co-expression networks were reconstructed via meta-analysis of gene expression data and co-expressed gene modules were identified in epithelial cells from ovarian tumor and healthy ovarian surface epithelial samples to propose ovarian cancer associated genes and their interactions. We propose a novel, highly interconnected, differentially co-expressed, and co-regulated gene module in ovarian cancer consisting of 84 prognostic genes. Furthermore, the specificity of the module to ovarian cancer was shown through analyses of datasets in nine other cancers. These observations underscore the importance of transcriptome based systems biomarkers research in deciphering the elusive pathophysiology of ovarian cancer, and here, we present reciprocal interplay between candidate ovarian cancer genes and their transcriptional regulatory dynamics. The corresponding gene module might provide new insights on ovarian cancer prognosis and treatment strategies that continue to place a significant burden on global health.

Figure 1

Differentially expressed genes (DEGs) in ovarian cancer datasets. (a) The table represents the numbers (and percentages in parenthesis) of differentially expressed genes between laser micro-dissected epithelial cells from ovarian tumor (CEPI) and ovarian surface epithelia (OSE) samples. The direction of regulation (i.e., up or down-regulation) of the genes was also specified. (b) Venn diagram of DEGs across the six ovarian cancer associated datasets. (c) Statistically significant KEGG pathways obtained through the pathway enrichment analysis of the 698 mutual DEGs of all datasets (adjusted p-value < 0.05).

Figure 2

Characteristics of co-expression networks and belonging modules in laser micro-dissected epithelial cells from ovarian tumor (CEPI) and ovarian surface epithelia (OSE) samples. (a) Topological properties of the co-expression network in CEPI samples (CNC). (b) Topological properties of the co-expression network in OSE samples (CNO). (c) Top four co-expressed modules of CNC representing the diseased state. (d) Top four co-expressed modules of CNO representing the healthy state. DEGs were represented as nodes, and the statistically significant co-expression associations between DEGs were represented as edges.

Figure 3

Differential co-expression module (CMC) of ovarian cancer consisting of 84 genes. (a) The topological features of the CMC representing the dense co-expression pattern in diseased samples. (b) The topological features of the CMC in healthy samples.

Figure 4

The biological processes, pathways and chromosomal locations associated with the prognostic genes. (a) The distribution of the prognostic genes into biological processes or molecular pathways. Process or pathway annotations of genes were obtained from GeneCards database. (b) The distribution of the prognostic genes into the “signaling by GPCR” subcategory. (c) The distribution of the prognostic genes into metabolism related pathways. (d) Distribution of the prognostic genes into chromosomal locations.

Figure 5

Prognostic performance of the module. (a) Clustering of samples using principle components (PC1, PC2, PC3) of the expression matrix of module genes. (b) Kaplan-Meier analysis of ovarian cancer dataset using the clusters identified through differentially co-expressed module. The p-values are computed using log-rank test.

Figure 6

Transcriptional regulators of the prognostic genes. (a) Bar graph representing the distribution of transcription factors (TFs) which regulate the prognostic genes. (b) Venn diagram of the prognostic genes regulated by the most significant (top 3) TFs. (c) Bar graph representing the distribution of microRNAs (miRNAs) regulating the prognostic genes. (d) Venn diagram of the prognostic genes regulated by the most significant (top 3) miRNAs.

Figure 7

Differential expression of the prognostic genes in different tumor tissues. Bar graph represents the coverage of the module in each tumor sample, i.e. the ratio of the number of differentially expressed prognostic genes to the number of all genes in the prognostic module.