Grantee Research Project Results

2024 Progress Report: A tiered hybrid experimental-computational strategy for rapid risk assessment of complex environmental mixtures using novel analytical and toxicological methods

EPA Grant Number: R840450
Title: A tiered hybrid experimental-computational strategy for rapid risk assessment of complex environmental mixtures using novel analytical and toxicological methods
Investigators: Rusyn, Ivan , Wright, Fred A. , Zhou, Yihui , Chiu, Weihsueh A
Institution: Texas A & M University , North Carolina State University
EPA Project Officer: Aja, Hayley
Project Period: September 1, 2022 through May 2, 2025
Project Period Covered by this Report: September 1, 2023 through August 31,2024
Project Amount: $750,000
RFA: Development of Innovative Approaches to Assess the Toxicity of Chemical Mixtures Request for Applications (RFA) (2022) RFA Text |  Recipients Lists
Research Category: Health Effects , Endocrine Disruptors , New Approach Methods (NAMs) , Human Health , Safer Chemicals , Mixtures , Chemical Safety for Sustainability , Predictive Toxicology , CSS

Objective:

Evaluation of composition and hazards of chemical mixtures, or complex products classified as UVCBs (unknown variable composition or biological substances), presents a multitude of challenges. These include the presence of unknown constituents, a limited basis for grouping additive and independent components, and the lack of toxicity data on most constituents, and whole mixtures or UVCBs. Recent disasters resulting in redistribution of complex chemical mixtures in the environment have provided our group an opportunity to develop new approaches to rapid composition and hazard characterization to inform rapid decision-making. Our long-term goal is to ensure timely risk-based assessment of mixtures and UVCBs that ensures human health protection from the toxicity of known/unknown components. We will accomplish this through integration of novel toxicological (i.e., human cell-based assays), analytical (i.e., ion mobility spectrometry-mass spectrometry), and modeling (i.e., interaction, mediation and dose reconstruction) methods. We plan to demonstrate the application of these methods in the context of rapid risk assessment/management of sites contaminated with uncharacterized chemical mixtures.

The main outcomes of this project will be a suite of analytical, in vitro, and computational methods and tools that can be applied in a tiered strategy for rapid quantitative characterization of the composition and hazards of complex environmental mixtures and UVCBs. Overall, the project is pursuing the following specific aims.

Aim 1: To determine grouping of chemical mixture components for assessment of hazard(s) through integration of multi-phenotype/multi-tissue bioactivity data from a compendium of human induced pluripotent stem cell (iPSC)-derived cells.

Aim 2: To develop approaches for prioritization of components in whole mixtures or UVCBs that are likely to contribute most to joint toxicity through mediation and interaction methods that integrate multi-dimensional analytical and in vitro bioactivity data.

Aim 3: To evaluate prediction of joint toxicity of whole or defined mixtures and UVCBs through novel probabilistic additivity models of grouped (Aim 1) and prioritized (Aim 2) components.

Aim 4: To demonstrate the integration of the proposed exposure, toxicological and modeling methods into a tiered hybrid experimental-computational strategy for rapid risk assessment of complex environmental mixtures and UVCBs.

Progress Summary:

Specific Aim 1: Grouping Chemical Mixtures Using iPSC-Derived Cell Bioactivity

We are advancing methods to group chemical mixture components for hazard assessment by integrating multi-phenotype and multi-tissue bioactivity data from human induced pluripotent stem cell (iPSC)-derived cells. Focusing on diverse libraries of per- and poly-fluoroalkyl substances (PFAS) and polycyclic aromatic hydrocarbons (PAHs) in UVCBs (unresolved complex mixtures from petroleum refining), we collaborate with US EPA-NCCTE to address the high concern surrounding PFAS contaminants. Despite limited existing data, our experiments aim to generate essential safety information for regulatory decisions on exposure limits. Recent publications highlight our progress:

• Tsai et al. (2024) evaluated 26 diverse PFAS using iPSC-derived hepatocytes and cardiomyocytes, observing minimal effects in hepatocytes but significant negative chronotropy in cardiomyocytes for 8 PFAS. Transcriptomic analyses revealed stress and extracellular matrix pathways in hepatocytes and contractility pathways in cardiomyocytes. Most PFAS exhibited bioactivity-to-exposure ratios (BER) greater than one, indicating potential risks beyond exposure estimates.
• Ford et al. (2024a) expanded the study to 56 PFAS across eight subclasses using six human cell types, finding cell-specific activities and no clear class-specific groupings except for some structural trends. Approximately 20% of PFAS had margins of exposure (MOE) below one, prioritizing them for further study.
• Ford et al. (2024b) assessed variability in cardiomyocyte responses from 16 donors, identifying 46 PFAS with concentration-response effects and significant inter-individual variability. Most PFAS had toxicodynamic variability within a factor of ten, with MOEs above 100, highlighting certain PFAS as potential cardiotoxic risks.

Overall, these studies demonstrate that a comprehensive panel of human cell-based models can effectively derive bioactivity estimates for PFAS, facilitating comparisons with human biomonitoring data and underscoring the complexity of structure-bioactivity relationships.

Specific Aim 2: Prioritizing Mixture Components Through Mediation and Interaction Methods

This aim focuses on developing prioritization strategies for components in whole mixtures or UVCBs that contribute most to joint toxicity. We integrate multi-dimensional analytical data with in vitro bioactivity data using mediation and interaction analyses. Key activities and advancements include:

• Analytical Characterization with IMS-MS: Collaborating with Cordova et al. (2023), we utilized ion mobility spectrometry-mass spectrometry (IMS-MS) to analyze 195 crude oil samples, assigning molecular formulas and identifying polycyclic aromatic hydrocarbon biomarkers. IMS-MS proved effective for rapid, high-throughput fingerprinting and exposure characterization.
• Multivariate Prediction Models: We refined a penalized regression tool incorporating a randomization component to generate robust p-values through permutation and multiple testing corrections (Benjamini-Hochberg procedure). This tool, validated with leave-one-out cross-validation, has been applied in Ford et al. (2024a,b) and will be extended to additional case studies.
• Mediation Analysis: Utilizing data from Tsai et al. (2024), we are dissecting the relative influences of analytical data (e.g., polycyclic aromatic rings) and gene expression on bioactivity. Although highly multivariate mediation methods faced challenges with multiple comparisons, standard approaches like the Sobel test showed promise. Initial results support the utility of mediation analysis for understanding component interactions, with comprehensive results expected in Year 3.

These efforts enhance our ability to prioritize mixture components based on their toxicological contributions, leveraging advanced statistical and analytical techniques.

Specific Aim 3: Predicting Joint Toxicity with Additivity and Mass-Balance Models

Aim 3 seeks to evaluate the joint toxicity of mixtures and UVCBs using probabilistic additivity models for grouped and prioritized components. Additionally, we aim to improve the characterization of in vitro bioavailability through mass-balance models. Progress includes:

• Data Collection for Defined Mixtures: We are conducting experiments using over 1,000 compounds across several iPSC-derived organotypic cell lines to create sufficiently similar defined mixtures (Aim 3.1).
• Mass-Balance Model Evaluation: We assessed four published in vitro mass-balance models against experimental data for various organic chemicals, including pesticides, PFAS, PBDEs, and pharmaceuticals. Preliminary results show good performance in predicting free fractions and media concentrations (correlation coefficients 0.7–0.9) but lower accuracy for cell concentrations (r < 0.5). Chemical properties primarily influenced media concentrations, while cell type and other parameters affected cellular concentrations. A manuscript detailing these findings is in preparation (Lin et al., in prep).

These activities are crucial for accurately predicting mixture bioactivity and understanding the bioavailability of individual chemicals within complex mixtures.

Specific Aim 4: Integrating Exposure, Toxicological, and Modeling Methods for Risk Assessment

Aim 4 aims to integrate exposure data, toxicological insights, and modeling approaches into a hybrid strategy for the rapid risk assessment of complex environmental mixtures and UVCBs. Key outputs include:

• Comparison of Analytical Approaches: Upon completion of Aims 1-2, we will compare targeted and untargeted analytical methods for identifying mixture components (Aim 4.1) and evaluate interaction and mediation models for component prioritization (Aim 4.2).
• Bioactivity-Based Grouping: We have focused on refining probabilistic methodologies for rapid risk assessment, leveraging findings from Aim 4.3. For example:
  • Lin et al. (2024): Developed a workflow combining high-throughput in vitro testing with in silico analysis to predict proarrhythmic potential, validated through cross-validation and external datasets. This approach is adaptable to mixture testing, enhancing sensitivity and specificity in risk predictions.
  • Probabilistic and Bayesian Case Studies:
    • Lu et al. (2023): Integrated probabilistic dose-response modeling with biomonitoring data for the mycotoxin deoxynivalenol.
    • Lu et al. (2024a): Extended this approach to 19 Superfund priority chemicals.
    • Lu et al. (2024b): Improved in vitro-to-in vivo concordance using Bayesian model averaging, allometric scaling, and human-relevant in vitro assays, thereby increasing confidence in our in vitro-based risk assessment methods.
• Key Characteristics Approach: We analyzed the utility of the “key characteristics” framework for carcinogen evaluation within IARC Monographs (Rusyn and Wright, 2024). By examining 73 agents across 19 monographs, we identified that mechanistic data from in vivo animal, in vitro animal, and human studies were crucial for linking agents to specific carcinogenic mechanisms. Large-scale systematic in vitro testing, like ToxCast, was effective in excluding key characteristics. This analysis supports the integration of diverse data streams, particularly human in vitro data, to enhance cancer hazard classifications.

These integrated strategies facilitate a comprehensive and rapid assessment of complex mixtures, leveraging both experimental and computational advancements to inform risk management decisions effectively.

In conclusion, across all four Specific Aims, significant progress has been made in developing and refining methods for grouping chemical mixtures, prioritizing hazardous components, predicting joint toxicity, and integrating diverse data streams for comprehensive risk assessment. Collaborative efforts with regulatory agencies and the publication of multiple studies underscore the robustness and applicability of our approaches. Future work will continue to enhance these methodologies, ensuring they provide reliable, actionable insights for managing environmental health risks associated with complex chemical exposures.

Future Activities:

In Specific Aim 1, we will continue experiments with the in vitro models and both individual chemicals and mixtures. Manny of the experiments with PFAS mixtures have been completed and we anticipate publishing the results next year. These data also will be used for studies in Aims 2 and 3.

In Specific Aim 2, we will finalize our machine learning prediction models for responses and predictors with standard methods for mediation analysis, with analytical predictors such as polycyclic aromatic content used as the base causal predictors, gene expression used as potential mediating factors, and iPSC cell assay bioactivity as responses. These approaches will be used to rank the relative utility of combinations of cell types for which gene expression data are available, and for which bioactivity phenotypes available. We note that the literature is often unclear on the relative utility of cell type data that may not match tissues of interest for toxicants with understood modes of action. This work will thus extend the comparison of multiple cell types we reported (Tsai et al., 2023), which had focused primarily on the utility of iPSC cell types in prediction of pre-defined UVCB manufacturing-based categories.

In Specific Aim 3, we will complete our evaluation of mass-balance models to be better characterize in vitro bioavailability in a single chemical setting and begin exploring application to chemical mixtures. Additionally, we will complete evaluation of additivity models from the PFAS mixture experiments conducted in Aim 1. Once both in vitro testing and IMS-MS characterization of the 1000+ chemical library is complete, we will use those results to design sufficiently similar mixtures as outlined in Aim 3.1.

In Specific Aim 4, once the in vitro bioactivity and IMS-MS characterizations of the 1000+ chemical library is complete, we will use those results to identify “contaminants of concern” in whole mixtures and evaluate the relative degree of health protection from “traditional” risk assessment as compared to those based on predicted or measured whole mixture bioactivity. Furthermore, we will work with Aim 2 to compare risk prioritization based on traditional methods with those based on interaction and mediation analyses. 

Journal Articles on this Report : 1 Displayed | Download in RIS Format

Publications Views

Other project views:	All 3 publications	3 publications in selected types	All 3 journal articles

Publications

Type	Citation	Project	Document Sources
Journal Article	Tsai H, Ford L, Chen Z, Dickey A, Wright F, Rusyn I. Risk-Based Prioritization of PFAS Using Phenotypic and Transcriptomic Data from Human Induced Pluripotent Stem Cell-Derived Hepatocytes and Cardiomyocytes. ALTEX-ALTERNATIVES TO ANIMAL EXPERIMENTATION 2024;41(3):363-381	R840450 (2024)	Full-text: NLM - Full Text HTML

Supplemental Keywords:

mixtures, risk assessment, analytical, toxicology, NAMs, UVCBs

Progress and Final Reports:

Original Abstract

2023 Progress Report

Final Report