Using computational models to predict fish acute toxicity of pesticide metabolites

An analysis, published recently in Regulatory Toxicology and Pharmacology, has shown that Quantitative Structure–Activity Relationship (QSAR) models could potentially be used to predict whether plant protection product (PPP) metabolites are acutely toxic to fish. The work was led by the NC3Rs, in partnership with scientists from Dow AgroSciences, Syngenta and BASF.

PPPs are one of the most highly regulated groups of chemicals. The EU regulations require that not just PPPs but also their numerous metabolites are assessed for the risk they pose to the environment. This involves tests such as chronic and acute fish toxicity which can use a considerable number of animals, and acute fish toxicity tests in particular can be associated with a high level of suffering. In 2013 the European Food Safety Authority (EFSA) published a guidance document outlining opportunities where non-laboratory testing methods, for example in silico models, may be considered instead. Metabolite testing was one area that was highlighted. QSARs are computational models and well-established dry lab testing methods, which predict toxicity endpoints using a chemical's molecular structure and physico-chemical properties and are based on experimental toxicity data for related chemicals. The EFSA guidance provides an opportunity to establish the utility of QSARs in predicting acute fish toxicity of PPP metabolites.

Together with collaborators from the agrochemical industry, we carried out retrospective data analysis comparing experimental fish acute toxicity data for 150 pesticide metabolites (within a publically available database) with values predicted using QSARs. The team used ECOSAR (software from the US Environmental Protection Agency), a free QSARs tool which is easy to use and has been developed with the intention for use in a regulatory context.

For regulatory purposes it is important that the predictions are conservative, in other words that potential toxicity is not under-predicted. The results of the analysis showed a good correlation between the predicted and experimental data: 62% of the QSAR-predicted values were equal to or more conservative than the experimentally derived values. The correlation was better when only experimental data of highest quality was taken under consideration (71%). After applying a ‘tolerance factor’ to take into account potential experimental variability, the proportion of suitably predicted values increased further to 91%. Looking in more detail at other existing experimental data and the properties of the 9% of remaining compounds, values for only one chemical were not suitably predicted by the QSARs.

This analysis is a proof-of-principle showing that the QSAR approach has utility in the metabolite assessment scheme recommended by EFSA. The application of such a model by the agrochemical industry could ultimately lead to a decrease in the number of fish acute toxicity studies that are carried out to meet regulatory requirements, potentially saving approximately 1,800 fish annually across Europe. The next step is to work with regulators and the wider agrochemical industry to promote the use of the QSAR approach for modelling acute toxicity in fish.

Subscribe to our newsletter