Exploring functions of long noncoding RNAs across multiple cancers through co-expression network

Abstract

In contrast to protein-coding genes, long-noncoding RNAs (lncRNAs) are much less well understood, despite increasing evidence indicating a wide range of their biological functions, and possible roles in various cancers. Based on public RNA-seq datasets of four solid cancer types, we here utilize Weighted Correlation Network Analysis (WGCNA) to propose a strategy for exploring the functions of lncRNAs altered in more than two cancer types, which we call onco-lncRNAs. Results indicate that cancer-expressed lncRNAs show high tissue specificity and are weakly expressed, more so than protein-coding genes. Most of the 236 onco-lncRNAs we identified have not been reported to have associations with cancers before. Our analysis exploits co-expression network to reveal that onco-lncRNAs likely play key roles in the multistep development of human cancers, covering a wide range of functions in genome stability maintenance, signaling, cell adhesion and motility, morphogenesis, cell cycle, immune and inflammatory response. These observations contribute to a more comprehensive understanding of cancer-associated lncRNAs, while demonstrating a novel and efficient strategy for subsequent functional studies of lncRNAs.

Figure 1

Distributions and expression levels of expressed genes in the four cancer types. The venn plot of 14,470 expressed protein-coding genes (a) and 2,902 expressed lncRNA genes (b). (c) The proportion of expressed protein-coding genes (blue) and lncRNA genes (red) appearing in different number of cancers. The x axis depicts the number of cancer types. The y-axis depicts the proportion of expressed genes, which is the ratio between the counts of expressed genes appearing in different number of cancers and the total counts of all expressed genes. (d) The expression levels (log2(FPKM + 0.1)) of expressed protein-coding genes (blue) and lncRNAs (red) in each cancer type.

Figure 2

Hierarchical clustering based on expression profiles of significantly differentially expressed lncRNA genes (DELs) from each cancer type. (a) The heatmap of 357 DELs in bladder cancer. (b) The heatmap of 321 DELs in prostate cancer. (c) The heatmap of 267 DELs lung adenocarcinoma. (d) The heatmap of 375 DELs in breast cancer. The intensity of the color scheme is scaled to expression values (log2(FPKM + 0.1)) which are Z-score standardized per gene. The color bar above the heatmap represents the sample groups, and red indicates tumor sample, and blue represents normal sample.

Figure 3

Heatmap of the eleven well-characterized onco-lncRNAs across all the four cancers. Red denotes up-regulation, and blue indicates down-regulation. Blank represents the gene which is not significantly dysexpressed in cancer.

Figure 4

Barplot of representative GO biological process (BP) terms of 12 modules. Three representative GO BP terms were chosen from top 10 significant terms for each module. The y-axis depicts names of BP terms, and the x-axis depicts −log10 (P-value). Bar color denotes the module color. The red dotted line denotes p-value of 0.05.

Figure 5

Network and expression levels of 51 top hub genes in brown module. (a) Cytoscape network visualization of 51 genes, in which only edges with weight (w) above a threshold of 0.2 are displayed. The red nodes denote lncRNA genes. The green and blue nodes both denote protein-coding genes, but the green color stands for genes with function in maintaining genomic stability. (b) The expression levels of 51 genes in each cancer type. The y axis represents the average FPKM value of samples in each group. And the blue and red colors denote protein-coding genes and onco-lncRNAs, respectively.