The potential of computer-based quantitative structure activity approaches for predicting acute toxicity of chemicals

Within the EU, the management of the risks of chemicals currently falls under a new legislation called Registration, Evaluation, and Authorization of Chemicals (REACH). Within the next 10 years, existing (eco)toxicological data gaps for the more than 100 000 chemicals on the European Inventory of Existing Commercial Substances (EINECS) should be filled. The challenge is to provide this toxicity information in a fast, cost effective manner, avoiding the use of experimental animals as much as possible. In this regard, REACH has provisions to allow for the use of in vitro and/or in silico methods, e.g. those based on (Quantitative) Structure Activity Relationships [(Q)SARs], to provide toxicity information or identify hazards of chemicals. This information can subsequently be used to identify priority chemicals for further risk evaluation. A QSAR is based on the assumption that the biological activity of a new or untested chemical can be inferred from the molecular structure, or properties of similar compounds whose activities have already been assessed. Therefore, using the chemical structure of chemical compounds as the sole input, one can build a toxicity prediction model based on parameters that define the physico-chemical properties and relative reactivity of the compounds. The objective of this thesis was to apply OECD guidelines in the development of validated QSAR models that describe acute toxicity of selected groups of EINECS chemicals to various organisms. In addition, an estimate was made of the total number of EINECS chemicals that could be possibly evaluated using (Q)SAR approaches. Based on experimental toxicity data from literature and in silico calculated log Kow (a measure of hydrophobicity) values, a QSAR advisory tool was developed that directs users to the appropriate QSAR model to apply for predicting toxicity of substituted mononitrobenzenes to five types of organisms within specified log Kow ranges. In a next study, QSAR models were developed to predict in vivo acute toxicity of chlorinated alkanes to fish based on data from in vitro experiments, and even based on in silico log Kow data only. Furthermore, using toxicity data from acute immobilization experiments with daphnids, an interspecies QSAR model was developed to predict toxicity of organothiophosphate pesticides to fish based on these data for daphnids and in silico log Kow values. The QSAR models for the mononitrobenzenes, chlorinated alkanes, and organothiophosphates covered in total 0.7 % of the 100 196 EINECS chemicals. In a final step, using chemical classification software, 54 % of the EINECS chemicals were grouped into specific classes that can in theory be subject to QSAR modeling. The safety assessment of one group of compounds that could not be classified e.g. botanical extracts might be done by further development of a method recently reported for the safety assessment of natural flavour complexes used as ingredients in food. This would result in an additional 3 % of the EINECS chemicals that could potentially be covered by SAR approaches, bringing the total percentage of EINECS compounds that can be covered by (Q)SAR approaches to 57. In conclusion, the results of this thesis reveal that, (i) in vitro experiments and even in silico calculations can help to reduce or replace animals used for experimental toxicity testing and (ii) despite the fact that individual QSARs may often each cover only limited, i.e. less than 1%, of the EINECS compounds, (Q)SAR approaches have the potential to cover about 57 % of the EINECS compounds