Model validation in aquatic toxicity testing: implications for regulatory practice

Toxicity test validity is contingent on whether models and assumptions are appropriate and sufficient. A quality control evaluation of the acute toxicity testing protocol using the US. EPA fathead minnow database focused around three key assumptions that ensure results represent valid toxicological metrics: 1, it must be possible to estimate steady-state LC50s; 2, LC50s should occur at equivalent exposure durations; 3, all substantive toxicity modifying factors should be adequately controlled. About 8% of the tests failed the first assumption and are invalid and unusable. Examination of remaining data indicated variance from unquantified effects of toxicity modifying factors remained in LC50s, thereby failing assumption three. Such flaws in toxicity data generated via recommended LC50 testing protocols means resultant data do not represent consistent, comparable measures of relative toxicity. Current regulations employing LC50 testing data are acceptable due to the use of semiquantitative, policy-driven development guidance that considers such data uncertainty. Quantitative applications such as QSARs, mixture toxicity, and regulatory chemical grouping can be compromised. These validation failures justify a formal quality control review of the LC50 toxicity testing protocol. Interim improvements in the design, execution, interpretation, and regulatory applications of LC50 and related protocols using exposure-based dose surrogates are warranted.