Drug repositioning for dengue haemorrhagic fever by integrating multiple omics analyses

Abstract

To detect drug candidates for dengue haemorrhagic fever (DHF), we employed a computational drug repositioning method to perform an integrated multiple omics analysis based on transcriptomic, proteomic, and interactomic data. We identified 3,892 significant genes, 389 proteins, and 221 human proteins by transcriptomic analysis, proteomic analysis, and human-dengue virus protein-protein interactions, respectively. The drug candidates were selected using gene expression profiles for inverse drug-disease relationships compared with DHF patients and healthy controls as well as interactomic relationships between the signature proteins and chemical compounds. Integrating the results of the multiple omics analysis, we identified eight candidates for drug repositioning to treat DHF that targeted five proteins (ACTG1, CALR, ERC1, HSPA5, SYNE2) involved in human-dengue virus protein-protein interactions, and the signature proteins in the proteomic analysis mapped to significant pathways. Interestingly, five of these drug candidates, valparoic acid, sirolimus, resveratrol, vorinostat, and Y-27632, have been reported previously as effective treatments for flavivirus-induced diseases. The computational approach using multiple omics data for drug repositioning described in this study can be used effectively to identify novel drug candidates.

Figure 1

Schematic overview of the integrative analysis of omics data performed in this study to detect potential drug candidates. Firstly, signature genes, signature proteins, and interactions between human proteins and dengue virus proteins were selected from GEO datasets, and by reviewing the proteome literature and human–dengue virus PPI literature, respectively. Secondly, disease-specific pathways for the transcriptome and proteome data were detected by GSEA using the signature genes and proteins identified in the first step. Then, a human–dengue virus PPI network was constructed by integrating the interaction data from the PPI literature. Thirdly, drug candidates were detected in the transcriptomic, proteomic, and interactomic layers by CMap and STITCH searches. Fourthly, a set of drug candidates for DHF was identified by integrating the significant molecules, pathways, and drug candidates obtained in each layer. Finally, we identified common significant molecules and pathways for each omics analysis and layer, and identified the drug candidates targeting the human proteins that interact with dengue virus proteins in a targeted mechanism-based method. These methods are explained in detail in the main text.

Figure 2

Venn diagram of the signature genes detected in the transcriptomic analysis. The number of signature genes from three GEO gene expression data (GSE18090, GSE25226, GSE38246) are shown in the green, light blue, light green circle, respectively. The numbers of signature genes that overlap between these data are also shown.

Figure 3

Human protein and dengue virus protein interaction network. Blue nodes represent the dengue viral proteins and are labelled with the corresponding gene names. Pink nodes represent the human proteins and are labelled with the corresponding UniProt ID. The black edges show the interactions between human proteins and dengue viral proteins as determined by our interactomic analysis of the experimental data.

Figure 4

Venn diagram of signature genes, signature proteins, and human proteins that interact with dengue virus proteins. The number of signature genes obtained from the gene expression data with significant differences between DHF patients and normal controls is shown in the blue circle. The number of signature proteins obtained from the protein expression data is shown in the orange circle. The number of human proteins that interact with dengue virus proteins in human–virus PPIs is shown in the green circle. The numbers of gene products and proteins that overlap between the groups are also shown.

Figure 5

Venn diagram of the significant pathways in the transcriptomic and proteomic analyses. The numbers of significant pathways identified by GSEAs of the transcriptome and proteome are shown in the blue and orange circles, respectively. The number of pathways that overlap between the two analyses is also shown.

Figure 6

Histograms showing the normalized distribution probability of pathways identified by transcriptomics alone, proteomics alone, and the combination of transcriptomics and proteomics. The bars show the proportion of each group among the total number of pathway types identified by Reactome (blue), PWO (orange), or KEGG (green). The top panel shows the distribution of the 115 common pathways identified by both transcriptomics and proteomics GSEAs (T&P). The middle panel shows the distribution of the 304 pathways identified only by the transcriptomics GSEA (T). The bottom panel shows the distribution of the 140 pathways identified only by the proteomics GSEA (P). The sum of all the groups in each section is 1.0 for all three panels.

Figure 7

Venn diagram for the drug candidates identified by the transcriptomic, proteomic, and interactomic analyses. The numbers of drug candidates identified by the transcriptomic, proteomic, and interactomic analyses are shown in the blue, orange, and green circles, respectively. The numbers of drug candidates that overlap between each analysis type are also shown.

Figure 8

Schematic diagram of the filtering method used to narrow down the drug candidates. 13 drugs were identified as the intersection of the drug candidates by transcriptomic, proteomic and interactomic analyses (Fig. 7). 11 proteins were identified as the intersection between signature proteins of proteomic analysis and human proteins in human-viral PPI (Fig. 4). 115 pathways were identified as the intersection between significant pathways determined by GSEA of transcriptomic analysis and that of proteomic analysis (Fig. 5). Firstly, nine proteins were selected as the intersection between proteins interacted with 13 drugs and 11 signature proteins. These nine proteins were interacted with eight drugs out of 13 drugs. Secondly, seven proteins were selected as the intersection between proteins participating in 115 pathways and 11 signature proteins. These seven proteins were mapped in 43 pathways out of 115 pathways. Finally, five proteins were selected as the intersection between nine proteins interacted with 13 drugs and seven proteins participating in 43 pathways. These five proteins were interacted with eight drugs, and were participating in 33 pathways out of 43 pathways. Then, eight drugs were identified as candidates for use in future drug repositioning.

Figure 9

Chord diagram of the likely relationships among drug candidates, proteins, and pathways identified by the multiple omics analyses. Based on the drug repositioning method, eight drugs, five proteins, and 33 pathways were selected as potential candidates for use in the development of treatments for DHF. The drug candidates, proteins, and pathways are shown in green, orange, and cyan, respectively. Connections show the interactions among the drug candidates, proteins, and pathways. The Reactome, PWO, and KEGG pathways are shown in blue, orange, and green, respectively.