Narcosis and chemical reactivity QSARs for acute fish toxicity

Quantitative structure activity relationships (QSAR) that describe the acute fish toxicity have been published for many different groups of reactive organic chemicals. The structural similarity of chemicals within such groups, suggests that they share a common mode of action (MOA) which is based on their common chemical reactivity. Often, however, a descriptor for this reactivity alone can not explain the observed toxicity satisfactory but addition of a hydrophobicity parameter, like log KOW, is found to improve the relationship. In the present paper, an alternative strategy is proposed and tested with three different literature data sets. Instead of searching for better descriptors to establish a QSAR for the whole data set, the assumption that all compounds within the set act by the same MOA was critically reviewed. We tested the hypothesis that some of the compounds within the data sets acted by narcosis (general anesthesia), a second plausible mode of action in acute fish toxicity. Narcosis potency at observed lethal exposure levels was modeled with a baseline toxicity QSAR. The literature data sets were split in a narcosis and a reactive subset and for each of them a separate, one-parameter QSAR was established. For a set of OP-esters, nine out of 20 compounds were identified as possible narcotic compounds and their toxicity could be described with a narcosis QSAR. For the 11 compounds remaining in the reactive subset, a good correlation between acute toxicity and measured, in-vitro AChE inhibition rate was found (r2=0.68) which would have been overlooked if the whole data set was used. The use of two separate QSARs instead of one mixed QSAR was also tested for literature data sets of nitrobenzenes and α,β-unsaturated carboxylates. It was shown that for the description of toxicity data of all three groups of reactive compounds, a model which uses two separate modes of action was superior to a mixed model which uses a reactivity and a hydrophobicity parameter in a multiple linear regression.