Multi-scale modeling : prediction of in vivo hepatotoxicity from in vitro data
Résumé
The March 2013 European Directive banned testing cosmetic ingredients on animals and new in vitro technologies have been developed to facilitate safety assessment. However, there is a gap in bridging in vitro to in vivo toxicity assessment. We thus propose a methodology for in vitro in vivo extrapolation based on mathematical modelling of in vitro data coupled to toxicokinetics modeling at whole body scale. We focused on hepatotoxicity since the liver is the first affected by toxicity after exposure to repeated doses. Impedance metrics is a recently developed technology which provides real time cell impedance data. Analyzing these new data requires new methodologies, moving from static to dynamic toxicity assessment of acute and chronic exposures. We thus built TK/TD (toxicokinetic/toxicodynamic) models to describe and analyze impedance in vitro hepatotoxicity data obtained for HepaRG cells after single and repeated exposures to three hepatotoxic cosmetic ingredients (coumarin, isoeugenol and benzophenone-2). These models account for cyto-morphological and cell viability change over time. We integrated in the models: i) cosmetic uptake and elimination by the cells, ii) decrease of exposure concentration (due to evaporation, binding to well walls, or metabolism by cells), iii) cell morphological modifications due to exposure to cosmetics, iv) possible heterogeneity of sensitivity between cells. The model which described at best the acute data was based on a quasi-instantaneous equilibrium between cells and medium, a decrease of compound due to metabolism, and equal sensitivity between cells. We challenged the acute toxicity model to predict chronic data obtained for the same substances at 4-week exposure. Although the model predicted quite well chronic data of coumarin, it underestimated cell viability at intermediate concentration for isoeugenol and benzophenone-2. Therefore, a calibration of the model with chronic data has been performed for these compounds, which provided a better fit of these data. Once the TK/TD models were calibrated, they were coupled with human physiologicallybased pharmacokinetics (PBPK) models to predict a scaled-up liver toxicity. PBPK models consist in a series of compartments which represent an organ or groups of similar organs governed by blood flows. They permit to relate unique or repeated doses of exposure with the concentration time-profile in different organs. All the parameters of PBPK models have sound biological or toxicological basis and can be estimated based on QSAR models and in vitro data, as we did for the three compounds investigated. Another advantage of these models is that the consequences of inter-individual variability (in terms of age, sex, bodyweight or metabolism capacity) can be investigated, which we performed for coumarin, predicting long-term differences of sensitivity between low and fast metabolizers.