Filling the concept with data: integrating data from different in vitro and in silico assays on skin sensitizers to explore the battery approach for animal-free skin sensitization testing.

Tests for skin sensitization are required prior to the market launch of new cosmetic ingredients. Significant efforts are made to replace the current animal tests. It is widely recognized that this cannot be accomplished with a single in vitro test, but that rather the integration of results from different in vitro and in silico assays will be needed for the prediction of the skin sensitization potential of chemicals. This has been proposed as a theoretical scheme so far, but no attempts have been made to use experimental data to prove the validity of this concept. Here we thus try for the first time to fill this widely cited concept with data. To this aim, we integrate and report both novel and literature data on 116 chemicals of known skin sensitization potential on the following parameters: (1) peptide reactivity as a surrogate for protein binding, (2) induction of antioxidant/electrophile responsive element dependent luciferase activity as a cell-based assay; (3) Tissue Metabolism Simulator skin sensitization model in silico prediction; and (4) calculated octanol-water partition coefficient. The results of the in vitro assays were scaled into five classes from 0 to 4 to give an in vitro score and compared to the local lymph node assay (LLNA) data, which were also scaled from 0 to 4 (nonsensitizer/weak/moderate/strong/extreme). Different ways of evaluating these data have been assessed to rate the hazard of chemicals (Cooper statistics) and to also scale their potency. With the optimized model an overall accuracy for predicting sensitizers of 87.9% was obtained. There is a linear correlation between the LLNA score and the in vitro score. However, the correlation needs further improvement as there is still a relatively high variation in the in vitro score between chemicals belonging to the same sensitization potency class.