Evaluating adaptive stress consensus gene signatures using transcriptomics 

Chemicals are believed to produce adverse effects through two main routes, either by  disrupting a defined target and signal cascade, termed specific, or by interfering with basal  cellular processes associated with cell stress response, termed non-specific. Around half  (529/1063) of the chemicals tested in the EPAs ToxCast project elicit non-specific cytotoxicity  underscoring the need for new approach methodologies (NAMs) capable of assessing the  activities of these chemicals. One approach for screening non-specific activities of chemicals  is to measure the perturbation of adaptive stress response pathways, which are responsible  for maintaining homeostasis across diverse cell types and can be measured using high throughput transcriptomics. Accurately quantifying chemical-induced adaptive stress  response activation from transcriptomic data is challenging due to the cross-talk between  pathways, which results in overlap between target genes assigned to a stress response  system. Here we present a new approach based on gene set enrichment analysis of signature  sets, that is more selective than current techniques. First, we describe our approach for  finding consensus stress response gene signatures for DNA damage, the unfolded protein  response, heat shock, hypoxia, metal stress response, and oxidative stress response pathways. Consensus signatures were constructed from 48 gene signatures sourced from the Molecular Signatures Database that were associated with stress response pathways. Second, we built a database comprised of 18 reference perturbagens and 32 transcriptomic  profiles for the six classic stress responses systems sourced from the Gene Expression  Omnibus (GEO). A further 32 transcriptomic profiles were also included in the database to  serve as negative examples of stress responses. Third, we used gene set enrichment analysis  (GSEA) to score the activity of a signature in a transcriptomic profile. The diagnostic ability  of each signature as a classifier of a designated stress response system was assessed  against all other contributing signature sets as well as all other consensus signature sets by comparing area under the curve of the receiver operating characteristic curve. Consensus  signature sets were found to better or equally diagnose contributing signature sets in all stress response categories, with an average 10% increase in area under the curve relative  to all other contributing signature sets. Furthermore, consensus signature sets were more  discriminatory when compared to unassigned stress response profiles, resulting in an  approximate two-fold improvement relative to competing consensus signature sets. These  findings suggest the potential utility of adaptive stress response signatures and  transcriptomics data for efficiently evaluating the non-specific activities of chemical  perturbagens. 

The views expressed in this presentation are those of the author[s] and do not necessarily  reflect the views or policies of the U.S. Environmental Protection Agency.