Comparison of species sensitivity distribution modeling approaches for environmental risk assessment of nanomaterials – A case study for silver and titanium dioxide representative materials

Abstract

Species sensitivity distributions (SSDs) are used in chemical safety assessments to derive predicted-no-effect-concentrations (PNECs) for substances with a sufficient amount of relevant and reliable ecotoxicity data available. For engineered nanomaterials (ENMs), ecotoxicity data are often compromised by poor reproducibility and the high level of nano-specific characterization needed to uniquely identify an ENM under test exposure conditions. This may influence the outcome of SSD modelling and hence the regulatory decision-making. This study investigates how the outcome of SSD modelling is influenced by; 1) Selecting input data based on the nano-specific “nanoCRED” reliability criteria, 2) Direct SSD modelling avoiding extrapolation of data by including long-term/chronic NOECs only, and 3) Weighting data according to their nano-specific quality, the number of data available for each species, and the trophic level abundance in the ecosystem. Endpoints from freshwater ecotoxicity studies were collected for the representative nanomaterials NM-300 K (silver) and NM-105 (titanium dioxide), evaluated for regulatory reliability and scored according to the level of nano-specific characterization conducted. The compiled datasets are unique in exclusively dealing with representative ENMs showing minimal batch-to-batch variation. Thus, any variation between studies is driven by exposure features such as duration, pH, media, inherent species sensitivity and endpoint, rather than variation in the pristine ENM tested. The majority of studies were evaluated as regulatory reliable, while the degree of nano-specific characterization varied greatly. The datasets for NM-300 K and NM-105 were used as input to the nano-weighted n-SSWD model, the probabilistic PSSD+, and the conventional SSD Generator by the US EPA. The conventional SSD generally yielded the most conservative, but least precise HC5 values, with 95 % confidence intervals up to 100-fold wider than the other models. The inclusion of regulatory reliable data only, had little effect on the HC5 generated by the conventional SSD and the PSSD+, whereas the n-SSWD estimated different HC5 values based on data segregated according to reliability, especially for NM-105. The n-SSWD weighting of data significantly affected the estimated HC5 values, however in different ways for the sub-datasets of NM-300 K and NM-105. For NM-300 K, the inclusion of NOECs only in the weighted n-SSWD yielded the most conservative HC5 of all datasets and models (a HC5 based on NOECs only could not be estimated for NM-105, due to limited number of data). Overall, the estimated HC5 values of all models are within a relatively limited concentration range of 25−100 ng Ag/L for NM-300 K and 1−15 µgTiO2/L for NM-105.