The ACuteTox project has aimed to optimise and prevalidate an in vitro testing strategy for predicting human acute toxicity. Ninety-seven reference substances were selected and an in vivo acute toxicity database was compiled. Comprehensive statistical analyses of the in vivo LD50 data to evaluate variability and reliability, interspecies correlation, predictive capacities with regard to EU and GHS toxicity categories, and deduction of performance criteria for in vitro methods is presented. For the majority of substances variability among rodent data followed a log normal distribution where good reproducibility was found. Rat and mouse interspecies comparison of LD50 studies by ordinary regression showed high correlation, with coefficients of determination, ranging between 0.8 and 0.9. Substance specific differences were only significant for warfarin and cycloheximide. No correlation of compound LD50 range with presumed study quality rank (by assigning Klimisch reliability scores) was found. Modelling based on LD50 variability showed that with at least 90% probability ∼54% of the substances would fall into only one GHS category and ∼44% would fall within two adjacent categories. These results could form the basis for deriving a predictive capacity that should be expected from alternative approaches to the conventional in vivo acute oral toxicity test.
Copyright © 2010 Elsevier Inc. All rights reserved.