Application of Computational Toxicological Approaches in Supporting Human Health Risk Assessment, Project Summary


This summary describes some of the recent progress in computational toxicology that has been conducted by EPA’s National Center for Environmental Assessment (NCEA) and Human Health Risk Assessment research program. The primary objectives of this research were to: (1) apply computational or in silico approaches to support human health risk assessment; and (2) incorporate the latest technologies, such as chemoinformatics, transcriptomics and/or high-throughput in vitro assays, into human health assessment. This research provides alternative approaches to the risk assessment community in their efforts to conduct hazard identification and dose-response assessment, which are used in combination with site-specific exposure information to inform the characterization of risk to support regulatory and clean-up decisions.



This project has three parts. The first part focuses on developing a tiered strategy and applying computational toxicological approaches to support human health risk assessment by deriving a surrogate point-of-departure (e.g., NOAEL, LOAEL, etc.) using a test case study as proof of concept (Part I). The second part concentrates on utilizing genomics (transcriptomics) data in conjunction with Benchmark Dose (BMD) modeling by deriving transcriptomics-based PODs to be compared with known in-vivo-based points-of-departure (PODs) for several well-characterized IRIS chemicals (Part II). The third part demonstrates an in silico approach for evaluating a fraction-based risk assessment method for total petroleum hydrocarbon (TPH) mixtures (Part III).

Part I (tiered surrogate approach and proof of concept)

In general, hazard identification and dose-response assessment for chemicals of concern found in various environmental media are based on epidemiological and/or animal toxicity data. However, human health risk assessments are often requested for many compounds found at contaminated sites (e.g., Superfund sites) throughout the U.S. that have limited or no available toxicity information from either humans or animals. To address this issue, recent efforts have focused on expanding the use of SAR approaches to identify appropriate surrogates and/or predict toxicological phenotype(s) and associated adverse effect levels. A tiered approach based on three main types of surrogates (structural, metabolic, and toxicity-like) has been developed. In order to select the final surrogate chemical and its surrogate toxicity value(s), a weight-of-evidence approach based on the proposed rationale (e.g., human relevancy, metabolism, mode of action, etc.) is applied. A test case study, using actual toxicity data, was developed and evaluated as a proof-of-concept to support the tiered surrogate approach. Future work will include case studies demonstrating the utility of the surrogate approach under different scenarios for data-poor chemicals. In summary, the proposed surrogate approach provides a reasonable starting point for identifying potential toxic effects, target organs, and/or modes-of-action, and for selecting surrogate chemicals from which to derive toxicity values.

Part II (transcriptomics-based dose-response modeling)

The traditional approach for estimating toxicity values in quantitative chemical health assessment is often time and resource intensive. The extent and nature of the studies required using traditional approaches has limited the number of chemicals with published health assessments. This limitation is exacerbated because of new or existing chemicals without toxicity information. In this study, female mice were exposed for 13 weeks to multiple concentrations of five chemicals that tested positive in a 2-year cancer bioassay. Traditional histological and organ weight changes were evaluated, and gene expression microarray analysis was performed on the target tissues. Subsequently, the histological, organ weight changes, and the tumor incidences in the original cancer bioassay were analyzed using standard BMD methods to identify noncancer and cancer points of departure, respectively. Similarly, the dose-related changes in gene expression were analyzed using a BMD approach, and the responses were grouped based on cellular biological processes or molecular pathways. A comparison of the transcriptional BMD values with those for the traditional approach indicated a reasonably high degree of correlation for specific cellular biological processes. The correlation between the BMD values for the transcriptional and apical endpoints suggest that transcriptional BMD values may be used as potential PODs for health assessments. This approach may be applied to chemicals that are lacking toxicity data in the near future to support human health risk assessment.

Part III (in silico fraction-based mixtures risk assessment method)

Similar to single chemical assessments, computational approaches may be applied to health assessments of chemical mixtures as well. Previously, the Massachusetts Department of Environmental Protection (MADEP) and the Total Petroleum Hydrocarbon Criteria Working Group (TPHCWG) developed fraction-based approaches for assessing human health risks posed by total petroleum hydrocarbon (TPH) mixtures in the environment. Both organizations defined TPH fractions based on their expected environmental fate and by analytical chemical methods, which are more focused on expert judgment and physicochemical properties. MADEP and TPHCWG derived toxicity values for selected compounds within each fraction and used these as surrogates to assess hazard or risk of exposure to the whole fractions. Membership in a TPH fraction is generally defined by the number of carbon atoms in a compound and by a compound's equivalent carbon number index, which can predict its environmental fate. In this study, a systematic and objective re-evaluation of the assignment of TPHs to specific fractions using modeling and statistical methods, such as comparative molecular field analysis and hierarchical clustering, was proposed. The proposed approach is transparent and reproducible, and it reduces inherent reliance on judgment when toxicity information is limited. Re-evaluation of membership in these fractions is highly consistent (80% on average across various fractions) with the empirical approach of MADEP and TPHCWG. Furthermore, the results support the general methodology of mixture risk assessment to assess both cancer and noncancer toxicity values after the application of fractionation. The approach could be applied to other mixtures with defined characterization of fractions and/or components.


There are hundreds of chemicals in the environment – introduced to the air, water, or soil through consumer products and industry uses – to which the public may be exposed. For many of these chemicals, we do not have the data needed to assess the potential health risks that may arise from exposure to the chemicals. Traditional animal toxicity testing cannot keep up with the pace with the need to assess chemicals that have been or will be introduced to the environment. There is a pressing need to develop, evaluate and validate new alternative approaches to support human health risk assessment; these include in silico (via computer simulation) or computational approaches in addition to the current advances in molecular and systems biology technologies.

The work described here summarizes an effort to expand and evaluate the use of various computational approaches to support human health risk assessment. Several different approaches were developed and independently evaluated based on the types and purposes of assessments and availability of toxicity data. First, refinements and modifications were made to traditional structure-activity relationship (SAR; an approach to examine the relationship between a chemical or a molecular structure and its biological activity). These modifications provided sufficient scientific justifications for identifying hazard and characterizing dose-response relationships (see Part I in Summary below). In addition, because of recent advances in genomics, dose-response modeling based on transcriptomics (the expression level of mRNAs in a given cell population, often using high-throughput techniques based on DNA microarray technology) was compared with traditional endpoints (e.g., changes in organs or tissues, clinical chemistry, etc.) (see Part II in Summary below). Finally, because of the need to assess the potential toxicity of chemical mixtures as a whole, a refined strategy was developed, using molecular modeling and statistical analyses, to enhance established risk assessment methodologies for mixtures of chemicals (see Part III in Summary below). Overall, these three alternative or computational methods may be used for various types of assessments and may be used for future human health assessments of chemicals/mixtures with limited toxicity information.
Project Leader Management/Program Analyst
Nina Ching Y. Wang, NCEA-Cin
26 W. Martin Luther King Dr. (A-110)
Cincinnati, OH 45268
Ph: (513) 569-7752
Fax: (513) 487-2541
Bette Zwayer, NCEA-Cin
26 W. Martin Luther King Dr. (A-110)
Cincinnati, OH 45268
Ph: (513) 569-7575
Fax: (513) 569-7916