Quantitative analysis of lab-to-lab variability in Caco-2 permeability assays.

In this study, Caco-2 permeability results from different laboratories were compared. Six different sets of apparent permeability coefficient (P) values reported in the literature were compared to experimental P obtained in our laboratory. The differences were assessed by determining the root mean square error (RMSE) values between the datasets, which reached levels as high as 0.581 for the training set compounds, i.e. ten compounds with known effective human permeability (P). The consequences of these differences in P for prediction of oral drug absorption were demonstrated by introducing the P into the absorption and pharmacokinetics simulation software application GastroPlus™ for prediction of the fraction absorbed (F) in humans using calibrated "user-defined permeability models". The RMSE were calculated to assess the differences between the simulated F and experimental values reported in the literature. The RMSE for F simulated with the permeability model calibrated using experimental P from our laboratory was 0.128. When the calibration was performed using P from literature datasets, the RMSE values for F were higher in all cases except one. This study shows quantitative lab-to-lab variability of Caco-2 permeability results and the potential consequences this can have in the use of these results for predicting intestinal absorption of drugs.