Citation:

Lewis, R., T. Hill III, AND C. Corton. A set of six Gene expression biomarkers and their thresholds identify rat liver tumorigens in short-term assays. TOXICOLOGY. Elsevier Science Ltd, New York, NY, 443:152547, (2020). https://doi.org/10.1016/j.tox.2020.152547

Impact/Purpose:

The “gold-standard” in identifying carcinogens is currently the 2-year rodent bioassay, but due to the resources needed to assess a chemical ($2-4M USD; 800 rodents; histopathological analysis of more than 40 tissues), only ~1,500 commercial chemicals have been examined (Bucher et al., 2004; Gold et al., 2005; Waters et al., 2010). New resource-efficient methods are needed that can identify a chemical’s carcinogenic potential in shorter term assays, determine the dose of human-relevant risk and the boundaries of exposure. We have recently described a successful predictive approach to chemical-induced liver cancer in rats by optimizing our analysis within six major adverse outcome pathways (AOP) for liver cancer (Rooney, 2018). Rodent liver cancer often occurs through multiple AOPs that include genotoxicity and cytotoxicity, as well as activation of one or more receptors such as: aryl hydrocarbon receptor (AhR), constitutive activated/androstane receptor (CAR), estrogen receptor (ER), or peroxisome proliferator-activated receptor α (PPARα). Our AOP-driven approach utilizes discrete biomarkers comprised of genes responsive to chemicals known to act through the MIEs for six specific AOPs. This biomarker gene expression pattern is compared to a statistically-filtered gene list derived from the livers of treated rats vs. control rats, and the statistical correlation with individual biomarkers is calculated using the Running Fisher test. For our gene expression biomarkers, the identification of chemicals that work through individual MIEs had balanced accuracies of 92-98% (Corton et al., 2019; Rooney, 2018). The six biomarker correlation values were combined using a ToxPi analysis to generate a single value for each chemical-dose-time comparison. The highest values were dominated with those with tumorigenic outcomes. Using the ToxPi approach, the balance accuracy approached 96% for identification of rat liver tumorigens from microarray data including identification of 36 out of 37 tumorigenic chemicals in our microarray compendium (Corton, et al., 2019). Therefore, we believe this predictive approach for tumorigen identification could be applied to new datasets from transcriptomics studies. In a separate computational study of our biomarker approach, we found that discrete KE thresholds based on gene expression biomarkers, individual gene induction, liver to body weight, or clinical chemistry values re predictive for rat liver cancer (Hill et al., 2019; JC, Submitted). We analyzed archival data from two large studies of chemicals in rats to identify chemical-agnostic indicators from short-term studies (< 28d) linked to an increased incidence of liver tumors. We found that use of the thresholds derived from either gene expression biomarkers or a set of 12 individual genes (Hill, et al., 2019) or from various combinations of liver weight to body weight and clinical chemistry markers (JC, Submitted) were predictive of liver tumors at up to 97% balanced accuracy. While this approach appears to be promising to identify liver tumorigens in short-term exposure studies, additional examples are needed to independently validate the approach. To address the application issue to novel data streams and effect thresholds, this study applies ToxPi and threshold predictive methodology to identification of liver tumorigens in a previously unexamined dataset. The dataset source was a separate arm of the DrugMatrix rat study, and describes genomic liver response to a library of 42 chemicals using the Affymetrix full-genome array platform. We also evaluated the predictive accuracy of the biomarker approach if only a more limited subset of genes were analyzed. To do this, the gene expression output from the BioSpyder rat 1500+ platform interrogating ~2600 genes was examined across a smaller set of chemicals.

Description:

Traditional methods for cancer risk assessment are retrospective, resource-intensive, and not feasible for the vast majority of environmental chemicals. In earlier studies, we used a set of six biomarkers to accurately identify liver tumorigens in transcript profiles derived from chemically-treated rats using either a Toxicological Priority Index (ToxPi) approach or using derived biomarker thresholds for cancer. The biomarkers consisting of 7-113 genes are used to predict the most common liver cancer molecular initiating events: genotoxicity, cytotoxicity and activation of the xenobiotic receptors AhR, CAR, ER, and PPARα. In the present study, we apply and evaluate the performance of these methods for cancer prediction in an independent rat liver study of 44 chemicals (6 h-7d exposures) examined by Affymetrix arrays. In the first approach, ToxPi ranking of biomarker scores consistently gave the highest scores to tumorigenic chemical-dose pairs; balanced accuracies for identification of liver tumorigenic chemicals were up to 89 %. The second approach used tumorigenic thresholds derived in the present study or from our earlier study that were set at the maximum value for chemical-dose exposures without detectable liver tumor outcomes. Using these thresholds, balanced accuracies were up to 90 %. Both approaches identified all tumorigenic chemicals. Almost all of the tumorigenic chemicals activated more than one MIE. We also compared biomarker responses between two types of profiling platforms (Affymetrix full-genome array, TempO-Seq 1500+ array containing ∼2600 genes) and found that the lack of the full set of biomarker genes on the 1500+ array resulted in decreased ability to identify chemicals that activate the MIEs. Overall, these results demonstrate that predictive approaches based on the 6 biomarkers could be used in short-term assays to identify chemicals and their doses that induce liver tumors, the most common endpoint in rodent bioassays.

URLs/Downloads:

Record Details: