Integrating the Ribonucleic Acid Sequencing Data From Various Studies for Exploring the Multiple Sclerosis-Related Long Noncoding Ribonucleic Acids and Their Functions

Abstract

Multiple sclerosis (MS) is a chronic fatal central nervous system (CNS) disease involving in complex immunity dysfunction. Recently, long noncoding RNAs (lncRNAs) were discovered as the important regulatory factors for the pathogenesis of MS. However, these findings often cannot be repeated and confirmed by the subsequent studies. We considered that the small-scale samples or the heterogeneity among various tissues may result in the divergence of the results. Currently, RNA-seq has become a powerful approach to quantify the abundances of lncRNA transcripts. Therefore, we comprehensively collected the MS-related RNA-seq data from a variety of previous studies, and integrated these data using an expression-based meta-analysis to identify the differentially expressed lncRNA between MS patients and controls in whole samples and subgroups. Then, we performed the Jensen-Shannon (JS) divergence and cluster analysis to explore the heterogeneity and expression specificity among various tissues. Finally, we investigated the potential function of identified lncRNAs for MS using weighted gene co-expression network analysis (WGCNA) and gene set enrichment analysis (GSEA), and 5,420 MS-related lncRNAs specifically expressed in the brain tissue were identified. The subgroup analysis found a small heterogeneity of the lncRNA expression profiles between brain and blood tissues. The results of WGCNA and GSEA showed that a potential important function of lncRNAs in MS may be involved in the regulation of ribonucleoproteins and tumor necrosis factor cytokines receptors. In summary, this study provided a strategy to explore disease-related lncRNAs on genome-wide scale, and our findings will be benefit to improve the understanding of MS pathogenesis.

Keywords: function analysis; long non-coding ribonucleic acids; meta-analysis; multiple sclerosis; ribonucleic acid sequencing.

Figure 1

The flow chart of selecting the RNA sequencing (RNA-seq) datasets and studies which are used to identify the multiple sclerosis-related long noncoding RNAs.

Figure 2

The results of heterogeneity test and meta-analysis for all samples and subgroups. (A) The expression level of the significantly differentially expressed long noncoding RNAs (lncRNAs) in each study after meta-analysis. The random effect model was used for 157 lncRNAs with a significant heterogeneity, while the fixed effect model was used for 5,263 non-heterogeneous lncRNAs. The details can be clearly viewed by enlarging the electronic version. (B) The forest plot for the meta-analysis of the lncRNA NONHSAG108980.1 which is the most significant result associated with an increased risk of MS (SMD = 0.59, 95% CI = 0.40−0.78, P = 1.89×10⁻⁹). (C) The bar plot showing the results of heterogeneity test in each group. For all samples, the proportion of lncRNAs with a significant heterogeneity is not high (about 2.90%), and this percentage is further decreased to about 1.99 and 1.20% in blood and brain, respectively. (D) The Venn diagram exhibiting the overlap among the significantly differentially expressed lncRNAs that are identified using brain tissues, blood tissues, and all samples.

Figure 3

The tissue specificity of the multiple sclerosis-related long noncoding RNAs (lncRNAs) based on expression data from NONCODE database. (A) Tissue specific expression measured by Jensen-Shannon divergence. The distributions of the maximal tissue specificity scores showed the high tissue specificity of the differentially expressed lncRNAs identified using whole (blue), brain (green), and blood sample (red), respectively. The (B) to (D) showed the hierarchical clustering heatmap for expression of these lncRNAs in primary human tissues and cell lines. These differentially expressed lncRNAs identified using whole (B), brain (C), and blood sample (D) are all highly specifically expressed in brain tissue. The Manhattan distance was used to perform all of the three cluster analyses.

Figure 4

The co-expression network analysis of the differentially expressed long noncoding RNAs (lncRNAs) and protein-coding genes. (A) The clustering dendrogram of these co-expressed lncRNAs and protein-coding genes. There are 15 clustered modules in the hierarchical clustering dendrogram which is constructed by a dynamic cut-tree algorithm. These clustered modules are marked as 15 different colors, respectively, i.e., yellow, turquoise, tan, salmon, red, purple, pink, midnight blue, magenta, green yellow, green, cyan, brown, blue, and black. (B) The heatmap for the association of each module with the disease states, platforms, and tissue types. Each cell represents a module, and contains the correlation r and corresponding P value (in brackets). Panels (C) to (E) show the results of correlation between the module membership and the gene significance in MEyellow, MEred, and MEcyan, respectively. The results of other modules were described in Supplementary Figure S3 .

Figure 5

The Gene Ontology and Kyoto Encyclopedia of Genes and Genomes pathway enrichment in the three most significant modules for multiple sclerosis. (A) The enrichment for MEbrown. The protein-coding genes co-expressed with the MS-related lncRNAs in this module are mainly involved in leukocyte and interleukin-related immune response. (B) The enrichment for MEpink. The co-expressed protein-coding genes in this module are mainly associated with the intercellular junction and signaling transmission. (C) The enrichment for MEyellow. The co-expressed protein-coding genes in this module are mainly related to the ribonucleoprotein.