A.T. Salminen, J. Willoughby, J.M. McKim, Jr.
LifeNet Health LifeSciences, Virginia Beach, VA
Background and Purpose:
Predicting the bioavailability of a chemical following topical application or exposure is essential for assessing the risk and efficacy of dermatological products. In vivo studies using preclinical species are often constrained by cost, complexity, and limited translational relevance. As a result, new approach methodologies (NAMs), including in vitro and in silico tools, are increasingly relied upon in toxicological evaluations. In vitro permeation testing (IVPT) with excised human skin can be combined with mechanistic skin permeation models to extrapolate these data toward in vivo predictions. However, refining the implementation and understanding the limitations of such models is necessary before they can be widely adopted for regulatory purposes. In this study, IVPT data were compared with simulated results generated using a skin permeation model implemented in MoBi, part of the Open Systems Pharmacology suite. Additionally, layer-deposition data from both experimental and simulated datasets were analyzed to extrapolate the in vivo bioavailability of the test compounds.
Methods:
IVPT was performed on four test articles, caffeine (CAS No. 58-08-2, MW 194.19 g/mol, log Kow -0.07, salicylic acid (CAS No. 69-72-7, MW 138.12 g/mol, log Kow 2.26), testosterone (CAS No. 58-22-0, MW 288.43 g/mol, log Kow 3.32), and mannitol (CAS No. 69-65-8, MW 182.17 g/mol, log Kow -3.10). Excised human skin (LifeNet Health; cryopreserved, cadaver, leg or back, 500 µm dermatomed) was mounted on Franz diffusion cells (PermeGear; 0.64 cm2 active area) connected to a heated water circulator system. Skin temperature and barrier integrity were measured by infrared thermometer and transepidermal water loss (Delfin VapoMeter), respectively. A finite dose (10 µL/cm2) of test article (1% w/v in buffer) was applied to the apical skin surface, and receptor fluid (DPBS) collections were performed at 0, 0.5, 1, 2, 3, 4, 6, and 24 hours post-dosing. Apical wash, stratum corneum tape strips (three per sample), and tissue were collected following the final (24 hour) receptor fluid sampling. All samples were extracted and analyzed by liquid chromatography tandem mass spectroscopy (LC-MS/MS). In silico predictions of chemical dermal absorption were collected in parallel. Test article physicochemical properties and test system parameters were input into the MoBi (version 12.0.434) implementation of the (Dancik, Miller et al. 2013) skin permeation model (https://github.com/Open-Systems-Pharmacology/Skin- permeation-model) and simulated cumulative permeation, flux, and layer deposition data were generated.
Results:
The Mean ± SEM 24-hour cumulative absorption observed in the in vitro permeation test ranked from lowest to highest as follows: mannitol (no detectable absorption), testosterone (2.53 ± 0.20 µg/cm2), salicylic acid (7.70 ± 3.93 µg/cm2), caffeine (14.84 ± 5.30 µg/cm2). The simulated mean 24-hour cumulative absorption, alternatively, ranked from lowest to highest as follows: mannitol (2.07x10-29 µg/cm2), caffeine (0.87 µg/cm2), testosterone (1.75 µg/cm2), salicylic acid (99.92 µg/cm2). For testosterone, the model layer deposition data predicted 96.69% of the applied dose would remain on the apical skin surface versus the experimentally observed 91.62 ± 3.40%. Simulated versus experimental comparisons of apical retention for the remaining test articles were as follows: caffeine = 41.94% vs 70.86% ± 5.32%, salicylic acid = 0.06% vs 3.02% ± 0.18%, mannitol = 57.98% vs 20.85% ± 10.24%.
Conclusions:
The mechanistic skin model was limited in its ability to accurately predict the cumulative absorption of the test compounds. The layer deposition data generated, however, was more accurate at correctly ranking the results observed experimentally. This study underscores the value of mechanistic skin permeation models in refining in vivo dermal absorption predictions. Integrating IVPT data with computational models can strengthen predictive accuracy and increase confidence in risk assessments. Conversely, computational models can contextualize experimental findings, offering deeper insights into observed results. In both scenarios, employing multiple NAMs within a weight-of-evidence framework enhances the robustness and regulatory relevance of the assessment.