QSAR models for the (eco-)toxicological characterization and prioritization of emerging pollutants: case studies and potential applications within REACH.

Under the European REACH regulation (Registration, Evaluation, Authorisation and Restriction of Chemical substances - (EC) No 1907/2006), there is an urgent need to acquire a large amount of information necessary to assess and manage the potential risk of thousands of industrial chemicals. Meanwhile, REACH aims at reducing animal testing by promoting the intelligent and integrated use of alternative methods, such as in vitro testing and in silico techniques. Among these methods, models based on quantitative structure-activity relationships (QSAR) are useful tools to fill data gaps and to support the hazard and risk assessment of chemicals. The present thesis was performed in the context of the CADASTER Project (CAse studies on the Development and Application of in-Silico Techniques for Environmental hazard and Risk assessment), which aims to integrate in-silico models (e.g. QSARs) in risk assessment procedures, by showing how to increase the use of non-testing information for regulatory decision-making under REACH. The aim of this thesis was the development of QSAR/QSPR models for the characterization of the (eco-)toxicological profile and environmental behaviour of chemical substances of emerging concern. The attention was focused on four classes of compounds studied within the CADASTER project, i.e. brominated flame retardants (BFRs), fragrances, prefluorinated compounds (PFCs) and (benzo)-triazoles (B-TAZs), for which limited amount of experimental data is currently available, especially for the basic endpoints required in regulation for the hazard and risk assessment. Through several case-studies, the present thesis showed how QSAR models can be applied for the optimization of experimental testing as well as to provide useful information for the safety assessment of chemicals and support decision-making. In the first case-study, simple multiple linear regression (MLR) and classification models were developed ad hoc for BFRs and PFCs to predict specific endpoints related to endocrine disrupting (ED) potential (e.g. dioxin-like activity, estrogenic and androgenic receptor binding, interference with thyroxin transport and estradiol metabolism). The analysis of modelling molecular descriptors allowed to highlight some structural features and important structural alerts responsible for increasing specific ED activities. The developed models were applied to screen over 200 BFRs and 33 PFCs without experimental data, and to prioritize the most hazardous chemicals (on the basis of ED potency profile), which have been then suggested to other CADASTER partners in order to focus the experimental testing. In the second case-study, MLR models have been developed, specifically for B-TAZs, for the prediction of three key endpoints required in regulation to assess aquatic toxicity, i.e. acute toxicity in algae (EC50 72h Pseudokirchneriella subcapitata), daphnids (EC50 48h Daphnia magna) and fish (LC50 96h Onchorynchus mykiss). Also in this case, the developed QSARs were applied for screening purposes. Among over 350 B-TAZs lacking experimental data, 20 compounds, which were predicted as toxic (EC(LC)50 ≤ 10 mg/L) or very toxic (EC(LC)50 ≤ 1 mg/L) to the three aquatic species, were prioritized for further experimental testing. Finally, in the third case-study, classification QSPR models were developed for the prediction of ready biodegradability of fragrance materials. Ready biodegradation is among the basic endpoints required for the assessment of environmental persistence of chemicals. When compared with some existing models commonly used for predicting biodegradation, the here proposed QSPRs showed higher classification accuracy toward fragrance materials. This comparison highlighted the importance of using local models when dealing with specific classes of chemicals. All the proposed QSARs have been developed on the basis of the OECD principles for QSAR acceptability for regulatory purposes, paying particular attention to the external validation procedure and to the statistical definition of the applicability domain of the models. QSAR models based on molecular descriptors generated by both commercial (DRAGON) and freely-available (PaDELDescriptor, QSPR-Thesaurus) software have been proposed. The use of free tool allows for a wider applicability of the here proposed QSAR models. Concluding, the QSAR models developed within this thesis are useful tools to support hazard and risk assessment of specific classes of emerging pollutants, and show how non-testing information can be used for regulatory decisions, thus minimizing costs, time and saving animal lives. Beyond their use for regulatory purposes, the here proposed QSARs can find application in the rational design of new safer compounds that are potentially less hazardous for human health and environment.