Development of a toxicogenomics signature for genotoxicity using a dose‐optimization and informatics strategy in human cells

Abstract

The development of in vitro molecular biomarkers to accurately predict toxicological effects has become a priority to advance testing strategies for human health risk assessment. The application of in vitro transcriptomic biomarkers promises increased throughput as well as a reduction in animal use. However, the existing protocols for predictive transcriptional signatures do not establish appropriate guidelines for dose selection or account for the fact that toxic agents may have pleiotropic effects. Therefore, comparison of transcriptome profiles across agents and studies has been difficult. Here we present a dataset of transcriptional profiles for TK6 cells exposed to a battery of well-characterized genotoxic and non-genotoxic chemicals. The experimental conditions applied a new dose optimization protocol that was based on evaluating expression changes in several well-characterized stress-response genes using quantitative real-time PCR in preliminary dose-finding studies. The subsequent microarray-based transcriptomic analyses at the optimized dose revealed responses to the test chemicals that were typically complex, often exhibiting substantial overlap in the transcriptional responses between a variety of the agents making analysis challenging. Using the nearest shrunken centroids method we identified a panel of 65 genes that could accurately classify toxicants as genotoxic or non-genotoxic. To validate the 65-gene panel as a genomic biomarker of genotoxicity, the gene expression profiles of an additional three well-characterized model agents were analyzed and a case study demonstrating the practical application of this genomic biomarker-based approach in risk assessment was performed to demonstrate its utility in genotoxicity risk assessment.

Using a gene expression biomarker to identify DNA damage‐inducing agents in microarray profiles

“…We utilized gene expression data derived in a recent study in which TK6 cells were exposed to 45 chemicals for 4 hr followed by analysis using nCounter (Li et al, ) (The list of chemicals is found in Supplemental File 1). As part of this study, the original training set of 27 treatments (26 chemicals and γ‐rays) used to derive the TGx‐DDI biomarker (Li et al, ) were examined for gene expression effects in TK6 cells using nCounter. The 27 chemicals included 13 true positives and 14 true negatives.…”

“…Our method for identification of chemicals that cause DNA damage is outlined in Figure A and required the following: (a) a list of TGx‐DDI biomarker genes with associated fold‐change values; (b) gene expression profiles of statistically filtered genes (also called biosets); and (c) a method to compare the biomarker to each bioset. These components are described below. The TGx‐DDI biomarker is a list of differentially expressed genes (DEGs) that were consistently increased or decreased after exposure to 13 DDI agents but not non‐DDI agents in TK6 cells (Li et al, ). The biomarker includes fold‐change values associated with each gene, derived from the average differences in expression across the 13 DDI agents. Statistically filtered gene lists were analyzed in a commercially available gene expression database called BaseSpace Correlation Engine (BSCE) ( https://www.illumina.com/products/by‐type/informatics‐products/basespace‐correlation‐engine.html; formally NextBio). The TGx‐DDI gene biomarker was uploaded to the BSCE database and compared to the biosets used in this analysis to assess correlation using the Running Fisher algorithm (Kupershmidt et al, ).…”

“…The TGx‐DDI biomarker (originally called the TGx‐28.65 biomarker) was developed using DNA microarray expression data and a machine learning algorithm called the nearest shrunken centroid model that identified 64 genes (65 probes) representing pathways associated with early responses to DNA damage. The chemical training set included 28 agents: 13 DDI including both direct and non‐direct acting, and 15 non‐DDI agents (Li et al, ). The DDI compounds covered a broad range of mechanisms including DNA alkylating agents, DNA strand breaking agents, topoisomerase inhibitors, nucleotide antimetabolites, and heavy metals.…”

“…The DDI compounds covered a broad range of mechanisms including DNA alkylating agents, DNA strand breaking agents, topoisomerase inhibitors, nucleotide antimetabolites, and heavy metals. Initial studies indicated that the TGx‐DDI biomarker differentiates DDI and non‐DDI compounds with 100% accuracy (Li et al, ). In follow up studies, the efficiency of the biomarker for accurately classifying compounds that require metabolic activation was demonstrated using different metabolic activating systems and TK6 cells (Buick et al, ; Yauk et al, ).…”

“…In follow up studies, the efficiency of the biomarker for accurately classifying compounds that require metabolic activation was demonstrated using different metabolic activating systems and TK6 cells (Buick et al, ; Yauk et al, ). In addition, using data from a microarray platform not used in the original studies (Affymetrix) that measured gene expression in the metabolically competent human liver cell line HepaRG, the biomarker correctly classified 15 out of 15 chemicals (Buick et al, ; Li et al, ; Yauk et al, ). More recently, it was demonstrated that the biomarker can accurately classify 45 test agents across a broad set of chemical classes using the nCounter high‐throughput cell‐based testing platform and probability analysis, principal components analysis, and two‐dimensional clustering (Li et al, ).…”