Theoretical and probabilistic methods for toxicokinetic predictions in daphnia magna and their application to environmental risk assessment

Collins, Jacob-Joe ORCID: 0009-0007-6191-4278 (2024). Theoretical and probabilistic methods for toxicokinetic predictions in daphnia magna and their application to environmental risk assessment. University of Birmingham. Ph.D.

Collins2024PhD.pdf
Text - Accepted Version
Available under License Creative Commons Attribution.
Download (12MB)

Abstract

The use of in silico and in vitro methods, commonly referred to as new approach methodologies (NAMs), has been proposed to support environmental (and human) chemical safety decisions, ensuring enhanced environmental protection. Toxicokinetic (TK) models developed for environmentally relevant species are fundamental to the deployment of a NAMs-based safety strategy, enabling the conversion between external and internal chemical concentrations, although they require historical TK data and robust physical models to be considered a viable solution. Daphnia magna is a key model organism in ecotoxicology albeit with limited quantitative TK data, as for most invertebrates, resulting in a lack of robust TK models. Moreover, current D. magna models are chemical specific, which restricts their applicability domain.

The overall aim of this thesis was to advance theoretical and probabilistic methods for TK predictions in Daphnia magna for use in environmental risk assessment (ERA). Firstly, to address current data limitations a D. magna TK dataset was collated from the literature and developed into an R package named AquaTK. Subsequently, a proof-of-concept Bayesian framework was developed to predict steady-state concentration ratios from the data. The application of the Bayesian framework to ERA was illustrated with an atrazine case study that showed prediction improvements (uncertainty reductions) with increasing amounts of data availability.

A substantial fraction of chemicals in the AquaTK dataset are ionisable at environmentally relevant pHs, therefore, the effect of ionisation on TK predictions was investigated within the Bayesian framework. Inferred steady-state concentration ratios were compared with predictions from the Bayesian predictive model and the state-of-the-art non-lipid organic matter (NLOM) model. Predictions from the Bayesian model did not improve after accounting for ionisation but the prediction errors were lower relative to the NLOM model. Additionally, the largest prediction errors occurred primarily for neutral chemicals, indicating that factors beyond ionisation also should be considered.

A protein surface-binding (PSB) model was developed to integrate protein binding, which is a key pharmacokinetic parameter, as a function of D. magna protein fraction and external concentration. A theoretical upper bound for the PSB model was determined and evaluated against the AquaTK dataset, which highlighted that the bound holds for a range of external concentration scenarios. This will have positive implications for ERA where risk assessors can predict concentration ratios under any exposure scenario with minimum data requirements and without the need for in vivo data.

To integrate biotransformation into TK models requires quantitative data for biotransformation products (BTPs) that can only be obtained with standards. However, the lack of commercially available standards for BTPs presents a significant challenge. Semi-quantification methods that predict concentrations from ionisation efficiency values have been developed. Therefore, a random forest regression model to predict relative ionisation efficiency values was developed on experimental parent and BTP data and showed promising results compared to other studies, in addition to robust predictions on unseen data. This was an important first step in the semi-quantification of BTPs for TK modelling purposes with further work required to predict concentrations of BTPs without standards.

A consistent theme throughout this research is the use of in silico NAMs with an “open data” approach to sharing and generating data for ERA. Overall, this work provides the foundations of a modern ERA that does not require animal testing through the increased use of theoretical and probabilistic methods. Further work should endeavour to integrate all these methods into a predictive tool for ERA.

Type of Work:

Thesis (Doctorates > Ph.D.)

Award Type:

Doctorates > Ph.D.

Supervisor(s):