Gene batteries and synexpression groups applied in a multivariate statistical approach to dose–response analysis of toxicogenomic data

Abstract

Univariate statistical analyses have limited strength when employed in low-dose toxicogenomic studies, due to diminished magnitudes and frequencies of gene expression responses, compounded by high data dimensionality. Analysis using co-regulated gene sets and a multivariate statistical test based upon ranks of expression were explored as means to improve statistical confidence and biological insight at low-doses. Sixteen gene regulatory groups were selected in order to investigate murine hepatic gene expression changes following low-dose oral exposure to the beta-adrenergic agonist, isoproterenol (IPR). Gene sets in this focussed analysis included well-defined gene batteries and synexpression groups with co-regulated responses to toxin exposures and linkage of chronic responses to adverse outcomes. Significant changes of target gene expression within Nfkb, Stat3 and 5′ terminal oligopryrimidine (5′TOP) batteries, as well as the acute phase and angiogenesis synexpression groups, were detected at IPR doses 100-fold lower than doses producing significant individual gene expression values. IPR-induced changes in these target gene groups were confirmed using a similar analysis of rat toxicogenomic data from published IPR-induced cardiotoxicity studies. Cumulative expression differences within gene sets were useful as aggregated metrics for benchmark dose calculations. The results supported the conclusion that toxicologically-relevant, co-regulated genes provide an effective means to reduce microarray dimensionality, thereby providing substantial statistical and interpretive power for quantitative analysis of low-dose, toxin-induced gene expression changes.