A number of diverse in vitro model systems have been employed for the prediction of irritation potential of test articles. Monolayer systems have proven to be useful for preliminary screening but are not always capable of distinguishing mild effects or adaptable to fully formulated product. Three-dimensional reconstructed skin equivalents integrate cellular toxicity with the kinetics of exposure and absorption, serving as more realistic models; however, it is not obvious which of the three-dimensional models will give the most predictive response, and which biomarker should be used for an endpoint measurement for different groups of irritants. While evaluating these variables, we have shown that different irritants modulate various cytokine mRNA levels and secretion patterns in a time- and concentration-dependent manner that is unique to each product category. These profiles are also dependent on keratinocyte-fibroblast interactions. The most predictive combinations of model systems and biomarkers for each product category were identified following comparison to preclinical data and human in vivo skin responses. Using a panel of representative consumer products, we identified IL-1
, IL-lra, IL-8 and GM-CSF release from skin equivalents as being the best indicators of irritation.