Improving cross-species extrapolation of chemical sensitivity

The aim of this thesis was to improve the development of models describing species sensitivity towards chemical stressors. This included unraveling decisions that might be of importance in the modeling process, obtaining a better mechanistic understanding of differences in species sensitivity, and providing recommendations for applying such modeling approaches within or across different taxonomic groups in freshwater ecosystems all over the globe. To accomplish this global applicability, we primarily focus on traits-based instead of taxonomy-based approaches on the development of predictive models. A trait is a phenotypic or ecological character of an organism at individual or population level, and describes the physical characteristics, ecological niche and functional role of a species within the ecosystem. We make use of traits, because they are transferable across geographies, add mechanistic and diagnostic knowledge and can be easily translated from taxonomic analyses priory preformed.We start this thesis with presenting a new predictive modelling approach for potential use in environmental risk assessment (ERA). This new approach constructs macroinvertebrate sensitivity rankings, and subsequently, predictive trait models for a set of pre-defined modes of action (MOA). Each model reveals different taxonomic patterns of species sensitivity, demonstrating that there is not one (set of) species which is most sensitive to all chemicals. The traits included in the models additionally provide information on the mechanisms underlying differences in species sensitivity. We identified, for instance, traits like life cycle duration and respiration mode as important in explaining species sensitivity, and which are hypothesized to be indicators of respectively metabolism and uptake rates. Additionally, we provide an accurate picture of which species-chemical combinations still lack sufficient data, and found that no or incomplete traits data were available for 71% of the species, making the lack of trait data the main obstacle in model construction.Although these models delivered a new understanding of species sensitivity, they also demonstrate that using traits-data alone will never results in explaining all differences in species sensitivity. Therefore, we develop the models further by adding taxonomic relatedness as an additional predictor of species sensitivity. We indeed find that model performance increases, although only marginally. Besides improving the models further, we also develop the tool further, so that it is now able to predict the sensitivity of species never tested before. We apply the developed models and tool to community assemblages at two different spatial scales (continental and national) and for two different MOAs (narcosis and acetylcholinesterase (AChE) inhibition). The community composition of European freshwater ecoregions (ERs) is used for the application of our models at the continental scale, while the reference database of the RIVPACS (River InVertebrate Prediction And Classification System) tool is used for river-type scale within the United Kingdom. We found that on a relative scale, 46% and 33% of European species were ranked as more sensitive towards narcosis and AChE inhibition, respectively. These more sensitive species were distributed with higher occurrences in the south and north-eastern regions, reflecting known continental patterns of endemic macroinvertebrate biodiversity. At the UK level, we found contradicting sensitivity patterns depending on the MOA, with more species displaying relative sensitivity to narcotic MOA in north and north-western regions, and more species with relative sensitivity to AChE inhibition MOA in south and south-western regions.Next, instead of looking for alternative predictors of species sensitivity, we look for alternative descriptors of species sensitivity. Recently, mechanistic effect models have been suggested as an alternative to the statistical summaries (e.g. LC50s) that are currently being used to assess the potential risk of chemicals to the diversity of all living species. An example of such quantitative mechanistic effect models are the toxicokinetic- toxicodynamic models of the General Unified Threshold models of Survival (GUTS) framework, which link external exposure and survival effects by describing the processes of uptake, biotransformation, elimination, damage and internal recovery. Parameterization of these models by means of traits enables prediction of standard sensitivity endpoints for a wide range of species and for multiple exposure patterns. We find that GUTS models parameterized on traits are indeed able to approximate the sensitivity of freshwater arthropods, and additionally allow us to understand the underlying processes of species sensitivity better.Since so many decisions have to be made regarding the predictors or descriptors that can be used to describe species sensitivity, we provide a review on currently existing cross-species extrapolation methodologies. We argue that risk assessment can benefit most from modelling approaches when species sensitivity is described based on effects that are ecologically relevant and robust. Additionally, specific attention should be paid to the heterogeneity of the training data, since this strongly influences the reliability of the resulting models. Regarding which predictors are useful for explaining differences in species sensitivity, we develop an integrated framework, combining the strengths of interspecies-correlation, relatedness-based, traits-based, and genomic-based extrapolation methods. Finally, we discuss that regardless of the statistical method used, statistically significant models can be found. However, the usefulness and applicability of these models varies considerably according to modelling choices made. We therefore recommend the publication of scientific code along with scientific studies to simultaneously clarify modelling choices, and enable continuation and elaboration on existing work.We conclude this thesis by explaining the three ways in which cross-species extrapolation methods can be used in the prospective risk assessment for chemicals: i) to support priority setting procedures to narrow down further assessment work, ii) to supplement the use of experimental data in weight-of-evidence approaches, and iii) to replace or completely substitute the need for experimental data. Through this development, our approach can help reduce animal testing and contribute towards a new predictive ecotoxicology framework