Correlation network analysis of cortical mRNA transcripts in an AD case-control study.

Abstract

The genetic basis for Alzheimer Disease is heterogenous and not fully understood. Correlation pattern analysis in cortical tissue can help identify relationships between known and unknown AD associated genes and neuropathologic processes. Correlation network analysis is effective in identifying such modules. We used mRNA data from cortical tissue to form a discovery dataset of 123 AD cases and 130 controls. Modules were identified by edge thresholding a correlation network and tested for significance with Barnard's exact test on a validation set of 52 cases and 56 controls. Our method uses vector-valued correlations to capture expression heterogeneity allowing for the calculation of contingency table statistics and a notion of module inverses. This facilitates the identification of strong genetic patterns with differential effects. One risk module of particular interest was selected for further analysis. We evaluated synchronicity qualitatively with visualizations and quantitatively via correlation matrix. The module inverse was similarly analyzed. A significant ( Bonferroni Corrected padj = 1.067x10-4, OR = 9.35 ) module of 14 mRNA probes was identified containing high expressions of genes known to be AD associated, and several genes that do not appear in the AD literature at all. These genes span several important processes such as Ca2+ regulation, Aβ formation, apoptosis, and DNA repair. Both visualizations and the correlation matrix show high synchronicity with mean PCC = 0.96589. The module and its inverse were also evaluated on the full data set yielding OR = 6.55 for the module and 1/OR = 6.22 for the protective inverse (validated with monte-carlo association trials). This suggests the module can differentiate cases and controls with equal power. Brain region was the only significant covariate for cases (p = 0.0241), which is due to a lower rate of module expression in the frontal lobe. We found a highly synchronized gene expression pattern in cortical tissue that is highly (and equally) powerful in differentiating AD patients and controls. The module includes several genes known to the AD literature, and several that are not. High correlations with the known AD genes indicate the newly identified genes are worth further study.