Current estimates of the size of worldwide and domestic chemical inventories approach hundreds of thousands of chemicals, with increasing trends in future chemical production and release. Relatively few of the chemicals in commerce, or those found in the environment, various waste streams, and the human body, have traditional toxicity data and fewer have human health assessments. Given historical, current, and future trends in chemical production and the lack of toxicity testing data available to inform human health assessments, the U.S. Environmental Protection Agency (EPA) is frequently faced with making decisions with limited or no data when evaluating potential human health risks.

Transcriptomics is the large-scale measurement of gene expression changes, and its application to toxicology enables broad characterization of the biological processes and pathways that may be impacted following exposure to a chemical. The technology and analysis methods for characterizing transcriptomic responses have matured and moved beyond the research laboratory into regulatory application. A literature review was conducted to evaluate the potential for using transcriptomic points of departure (PODs) from short-term in vivo studies in rodents to predict apical PODs from traditional in vivo toxicity studies. The literature survey included over 140 chemicals with diverse properties tested in 33 independent studies with varying experimental designs. The results of the literature survey demonstrated that transcriptomic benchmark dose (BMD) and benchmark dose lower confidence bound (BMDL) values, when integrated at a gene set level, were consistently concordant with BMD and BMDL values for apical responses in traditional subchronic and chronic rodent toxicity studies. The transcriptomic and apical dose concordance was robust across different exposure durations, exposure routes, species, sex, target tissues, physicochemical properties, toxicokinetic half-lives, and technology platforms. For the 40 chemicals with reported chronic rodent bioassay results, the Pearson’s correlation coefficient was 0.820 with a log₁₀ root-mean-square difference (RMSD) of 0.593 (log₁₀ mg/kg-day) and a median absolute ratio of 2.4 ± 1.0 (Median Absolute Deviation; MAD). The RMSD value is similar to the range of inter-study standard deviation estimates for the lowest observable adverse effect levels (LOAELs) for systemic toxicity in repeated dose studies, approximated as residual root-mean-square error (RMSE) in log10-mg/kg-day units [0.45-0.56; (Pham et al. 2020)]. The results suggest that the error associated with the concordance between the transcriptomic BMD values versus non-cancer and cancer apical BMD values is approximately equivalent to the inter-study variability in the repeated dose toxicity study itself.

Building on the results of the literature survey, the EPA Office of Research and Development (ORD) evaluated a process to derive transcriptomic PODs based on recommendations in the peer-reviewed National Toxicology Program (NTP) Approach to Genomic Dose Response Modeling report (NTP 2018). The methodology utilizes a 5-day, repeated dose in vivo study in male and female rats with an extended dose response series. The transcriptomic dose response modeling follows a stepwise process that utilizes BMD modeling approaches that are commonly employed in chemical risk assessment. In the evaluation process, a comprehensive series of analyses was performed to identify and support the choices and parameters used in each step of the transcriptomic dose response modeling process to promote detection of true signal, maximize reproducibility, and minimize false signal. A combination of parameters was identified that resulted in a Pearson correlation coefficient and log₁₀ RMSD for the transcriptomic and chronic apical BMD values of 0.910 and 0.567, respectively. The median absolute ratio of the transcriptomic BMD and chronic apical BMD values was 3.2 ± 1.9 (MAD). The inter-study transcriptomic BMD standard deviation for a subset of chemicals that were independently replicated was 0.242 (log₁₀ mg/kg-day). The overall estimated family-wise error rate for identifying a gene set level BMD was 0.006.

To provide context for the transcriptomic and apical BMD concordance, a statistical analysis was conducted to derive a lower bound of the expected mean squared difference (MSD) given inter-study variances in both the transcriptomic and apical responses. The results showed that the MSD of the transcriptomic and apical BMD concordance for the top performing combination of parameters (ranging from 0.285 - 0.387, depending on chemical replicates used) falls within the expected range of 0.267 - 0.617 (log₁₀ mg/kg-day)² when considering inter-study variances. The results of the analysis suggest that the error associated with the concordance between the transcriptomic BMD values versus non-cancer and cancer apical BMD values is approximately equivalent to the combined inter-study variability associated with the 5-day transcriptomic study and the two-year rodent bioassay.

The overall conclusions from the literature survey, evaluation of the transcriptomic dose response analysis methods, and the statistical comparison of the concordance with inter-study variances support the use of transcriptomic PODs from 5-day, repeated dose in vivo rodent studies in quantitative human health assessments. The application is supported based on multiple studies demonstrating transcriptomics as a reliable method to measure changes in gene expression; extensive peer review of the study design and dose response analysis methods in the individual publications and NTP report; availability of peer-reviewed software for reproducible application across datasets; broad application of the use of these dose response analysis methods in the government, academic, and private sectors; and historical precedence of the underlying dose response modeling methods in risk assessment. The application is further supported based on the performance of the method in approximating an apical POD from two-year toxicity studies, an inter-study variability that is consistent with those estimated for repeated chronic toxicity studies, and low family-wise error rate.