Sample matching strategies in gene expression studies of brain tissues

J Li, et al.
Society for Neuroscience annual meeting. 2005.

Abstract

Gene expression patterns in brain tissues are strongly influenced by the severity of cellular stress at the time of death. This effect, if not carefully monitored, may lead to spurious findings of differential expression. A common strategy to deal with this is to match cases and controls by tissue pH and the clinically recorded agonal condition. We found, however, that these conventional indicators were sometimes at odds with the observed expression patterns, and that pre-selection by these criteria may still result in significant case-control differences that could be simply explained as uncontrolled remnants of the agonal effect. The problem is analogous to the one in genetic association studies, where race and ancestry are often flawed proxies for the complex environmental and genetic factors. In this report, we describe a post hoc matching strategy that involves constructing an agonal stress rating of individual samples based on microarray data, and applying this rating in sample selection. We argue that brain pH and clinical records are imperfect surrogates of the accompanying medical condition, and that the data-derived ratings provide a more proximate assessment of the state of agonal stress in individual brain regions, and represent a strong control for this important confounder. The expression changes in low pH samples are similar across diverse regions of the brain. This canonical pattern can therefore be used for detecting the presence of residual confounding in most case-control studies involving brain tissues. The authors are members of the Pritzker Neuropsychiatric Disorders Research Consortium. An IP agreement exists between the universities and the Pritzker Neuropsychiatric Disorders Research Fund L.L.C..