Identifying nanodescriptors to predict the toxicity of nanomaterials: a case study on titanium dioxide

Since the evaluation of nanomaterial (NM) hazards by animal testing is expensive, time-consuming and critical from an ethical point of view, much interest is being given to the development of alternative testing strategies such as computational (predictive) models based on in vitro testing. However, the variations in in vitro experimental conditions can influence the outcome of computational modelling. In this study, we aim to identify nanodescriptor(s) and biological endpoint(s) capable of predicting the toxicity of titanium-di-oxide (TiO2) NMs, and demonstrate how experimental variations determine the outcome of modelling using three case studies. We used TiO2in vitro data from our previously published study as case study 1 and two other external case studies (case study 2 and 3) performed under different exposure conditions (presence and/or absence of serum). Firstly, we identified the nanodescriptor(s) closely associated to biological endpoints. Secondly, we determined the strength of association of the identified nanodescriptor(s) with the respective biological endpoint. The results indicate that the experimental conditions influence the outcome of the computational modelling. Agglomerate size as a nanodescriptor was well associated with biological endpoints such as DNA damage and/or cytotoxicity. We conclude that, agglomerate size is an important nanodescriptor to assess the toxicological effects of TiO2 NMs in vitro. However, the agglomeration state of NMs can be potentially influenced by in vitro exposure conditions and such influences could be just a confounder in broader contexts such as safety-by-design approaches, which require linking of material specific properties to the toxicological outcome.