Prioritization of phytotoxins according to their threat to water quality

Natural toxins are pollutants of emerging concern. Despite being ubiquitous in the European environment, thus far little action has been taken to conduct screening-level assessment of their presence in drinking water. The need to assess and prioritize natural toxins is more acute in the context of climate change, since distributions and environmental behaviour of plants are expected to change. To address the need for screening assessment, estimated properties representing the persistence and mobility of natural toxins are needed, but experimentally obtained values to build predictive models are still scarce. Existing QSPR models can be applied to estimate Kow and half-lives, with an important caveat; compounds have to lie within the applicability domain of the QSPR model to assure a reliable prediction. This thesis reports the assessment of the applicability domain of two popular QSPR models from the US EPA’s EPI Suite™ software package that estimate Kow and biodegradability, and evaluates how many toxins in a database for Switzerland lie within domain of the models. Results demonstrate that nearly half of the plant toxins in the database lie outside the applicability domain of one or both models, and thus that screening predictions of the toxins’ persistence and mobility are subject to unquantifiable uncertainties. This work points to a need to measure property data for more natural toxins to improve the empirical basis for predictive QSPR modeling.