Replacing Fish Early Life Stage Toxicity Test: Performance of Three QSAR Tools for Recently Submitted REACH Data

The fish early life stage toxicity (FELS) test is important for hazards classification under European regulation on Classification, Labelling, and Packaging of Chemicals (CLP). To replace and reduce vertebrate testing, alternative methods are being developed to predict the apical effects measured in FELS test. Still, their predictivity needs to be tested against experimental FELS outcomes, and the applicability domains of these methods need to be determined. We tested how well three quantitative structure-activity relationship ((Q)SAR) tools can predict the FELS test outcome: ECOSAR, VEGA, and iSafeRat. We compared the predicted outcome with no observed effect concentrations (NOECs) derived from 49 experimental FELS studies on fathead minnow (Pimephales promelas), zebrafish (Danio rerio) and rice fish (Oryzias latipes), which have been submitted for the purpose of the regulation concerning Registration, Evaluation, Authorisation, and Restriction of Chemicals (REACH). We observed that the QSAR models to predict chronic toxicity apply to a limited number of substances (ECOSAR, 25/49 predictions without uncertainty flags; VEGA, 2/49 predictions indicated as highly reliable; iSafeRat, 18/49). While no firm conclusions on the predictivity of the models can be made because of the limited number of reliable predictions produced, the current analysis suggests that the models could fairly well estimate the toxicity in a conservative direction with few exceptions.