Genomic convergence and network analysis approach to identify candidate genes in Alzheimer's disease

Abstract

Background: Alzheimer's disease (AD) is one of the leading genetically complex and heterogeneous disorder that is influenced by both genetic and environmental factors. The underlying risk factors remain largely unclear for this heterogeneous disorder. In recent years, high throughput methodologies, such as genome-wide linkage analysis (GWL), genome-wide association (GWA) studies, and genome-wide expression profiling (GWE), have led to the identification of several candidate genes associated with AD. However, due to lack of consistency within their findings, an integrative approach is warranted. Here, we have designed a rank based gene prioritization approach involving convergent analysis of multi-dimensional data and protein-protein interaction (PPI) network modelling.

Results: Our approach employs integration of three different AD datasets- GWL,GWA and GWE to identify overlapping candidate genes ranked using a novel cumulative rank score (SR) based method followed by prioritization using clusters derived from PPI network. SR for each gene is calculated by addition of rank assigned to individual gene based on either p value or score in three datasets. This analysis yielded 108 plausible AD genes. Network modelling by creating PPI using proteins encoded by these genes and their direct interactors resulted in a layered network of 640 proteins. Clustering of these proteins further helped us in identifying 6 significant clusters with 7 proteins (EGFR, ACTB, CDC2, IRAK1, APOE, ABCA1 and AMPH) forming the central hub nodes. Functional annotation of 108 genes revealed their role in several biological activities such as neurogenesis, regulation of MAP kinase activity, response to calcium ion, endocytosis paralleling the AD specific attributes. Finally, 3 potential biochemical biomarkers were found from the overlap of 108 AD proteins with proteins from CSF and plasma proteome. EGFR and ACTB were found to be the two most significant AD risk genes.

Conclusions: With the assumption that common genetic signals obtained from different methodological platforms might serve as robust AD risk markers than candidates identified using single dimension approach, here we demonstrated an integrated genomic convergence approach for disease candidate gene prioritization from heterogeneous data sources linked to AD.

Figure 1

Flow chart describing the entire work flow. Integrated data analysis was performed on three genome wide datasets to identify overlapping 108 AD putative candidate genes which were ranked by using cumulative rank based scoring method. These genes were further used to create a PPI and identify overlapping proteins among 108 and proteins from CSF and plasma proteome. PPI was then used to create a layered network based on the subcellular localization information of 640 genes, to identify clusters using MCL algorithm and to retrieve functional annotation using DAVID web tool.

Figure 2

Venn diagram of putative overlapping AD candidate genes among different genome wide datasets. The venn diagram represents the genes in the three individual datasets and overlapping 108 putative AD target genes identified by integrated analysis of the three datasets.

Figure 3

Layered Protein-Protein Interaction network (PPI) of 108 proteins. A layered network based on the subcellular localization of 640 proteins in the PPI was created. The nodes representing functionally important genes were highlighted in the layered network using colour codes - green (genes forming hub nodes in clusters (7), occurring in top 15 of ranked genes (108) and also present in putative biomarker dataset (38)); cyan (genes forming hub nodes in clusters, occurring in 108 AD genes and also present in putative biomarker dataset); yellow (genes occurring both in cluster hub and in 108 ranked genes); pink (genes forming hub nodes in clusters and occurring in top 15 of ranked genes); blue (remaining 59 from 108 list); red (38 biomarkers from AD, CSF and plasma overlap); grey (remaining proteins from 640 candidates).

Figure 4

Important clusters obtained from clustering of 640 proteins using MCL algorithm in clusterMaker. (a-f) Biologically significant gene clusters were identified from PPI using MCL algorithm. The nodes representing functionally important genes were coloured in the pattern described for the layered network.

Figure 5

Clustering of GO terms: significantly over represented top 11 functionally annotated clusters from biological process, cellular component and molecular function of 108 proteins.

Figure 6

Putative AD specific biomarkers. The venn diagram depicts overlap among putative 108 AD proteins, proteins from CSF and plasma proteome.

Figure 7

Prioritized putative PD candidate genes and overlap with prioritized AD candidate genes. (a) The venn diagram represents the genes in the three individual datasets and overlapping 59 putative PD target genes identified by integrated analysis of the three datasets. (b) The venn diagram represents the overlapping genes among AD and PD putative target genes.