A framework for ecological risk assessment of metal mixtures in aquatic systems.

Although metal mixture toxicity has been studied relatively intensely, there is no general consensus yet on how to incorporate metal mixture toxicity into aquatic risk assessment. We combined existing data on chronic metal mixture toxicity at the species level with species sensitivity distribution (SSD)-based in silico metal mixture risk predictions at the community level for mixtures of Ni, Zn, Cu, Cd, and Pb, to develop a tiered risk assessment scheme for metal mixtures in freshwater. Generally, independent action (IA) predicts chronic metal mixture toxicity at the species level most accurately, whereas concentration addition (CA) is the most conservative model. Mixture effects are noninteractive in 69% (IA) and 44% (CA) and antagonistic in 15% (IA) and 51% (CA) of the experiments, whereas synergisms are only observed in 15% (IA) and 5% (CA) of the experiments. At low effect sizes (∼ 10% mixture effect), CA overestimates metal mixture toxicity at the species level by 1.2-fold (i.e., the mixture interaction factor [MIF]; median). Species, metal presence, or number of metals does not significantly affect the MIF. To predict metal mixture risk at the community level, bioavailability-normalization procedures were combined with CA or IA using SSD techniques in 4 different methods, which were compared using environmental monitoring data of a European river basin (the Dommel, The Netherlands). We found that the simplest method, in which CA is directly applied to the SSD (CA ), is also the most conservative method. The CA has median margins of safety (MoS) of 1.1 and 1.2 respectively for binary mixtures compared with the theoretically more consistent methods of applying CA or IA to the dose-response curve of each species individually prior to estimating the fraction of affected species (CA or IA ). The MoS increases linearly with an increasing number of metals, up to 1.4 and 1.7 for quinary mixtures (median) compared with CA and IA , respectively. When our methods were applied to a geochemical baseline database (Forum of European Geological Surveys [FOREGS]), we found that CA yielded a considerable number of mixture risk predictions, even when metals were at background levels (8% of the water samples). In contrast, metal mixture risks predicted with the theoretically more consistent methods (e.g., IA ) were very limited under natural background metal concentrations (<1% of the water samples). Based on the combined evidence of chronic mixture toxicity predictions at the species level and evidence of in silico risk predictions at the community level, a tiered risk assessment scheme for evaluating metal mixture risks is presented, with CA functioning as a first, simple conservative tier. The more complex, but theoretically more consistent and most accurate method, IA , can be used in higher tier assessments. Alternatively, the conservatism of CA can be accounted for deterministically by incorporating the MoS and MIF in the scheme. Finally, specific guidance is also given related to specific issues, such as how to deal with nondetect data and complex mixtures that include so-called data-poor metals. Environ Toxicol Chem 2018;37:623-642. © 2017 SETAC.