Evaluation of liquid and supercritical fluid chromatographic systems in modelling the skin permeability of pharmaceutical and cosmetic compounds

Oral administration of drugs is still the most commonly applied route. However, besides its numerous advantages, there are also drawbacks, such as adverse effects or a pronounced first pass effect. As an alternative, topical dosage forms can be used in which active substances remain superficially on the skin, to target local skin tissue or appendages, or even to enter into the systemic circulation. It is thus important to know the rate at which a chemical component passes through the skin (represented by the skin permeability coefficient, log Kp), not only for drug discovery and development purposes, but also when studying dermal exposure or risk assessments. To reduce or possibly even replace the currently used in-vivo and in-vitro skin permeability tests, quantitative structure-activity relationship (QSAR) models can be used. With these models, the skin permeability coefficient can be estimated using theoretical descriptors (linked to physicochemical properties), as well as experimentally determined descriptors, such as chromatographic retention. The aim of this thesis was to investigate multiple chromatographic methods, which are preferably fast and equally predictive as the currently applied methods, toestimate the skin permeability of pharmaceutical and cosmetic compounds.
A representative test set of various pharmaceutical and cosmetic components was used throughout this thesis. After analysis of this test set with the various chromatographic methods, some modelling approaches, i.e. (stepwise) multiple linear regression (MLR) and partial least squares (PLS) regression, were applied to model skin permeability. To further improve the models, two sets of theoretical descriptors obtained with Vega ZZ and E-Dragon software, respectively, were applied.
In the first part of this thesis, skin permeability models were built consisting only of theoretical descriptors. In this way, for every tested chromatographic method, the added value of the experimental descriptor relative to the purely theoretical models could be assessed. The stepwise MLR approach on the E-Dragon descriptors provided the best performing model. Furthermore, the first chromatographic methods were tested, applying an octadecyl silica column. The limited improvement of combining the chromatographic and theoretical descriptors, relative to the use of only theoretical descriptors, did not outweigh the extent of the experimental work. It was therefore decided to further investigate chromatographic methods, which show more resemblance to the skin structure and/or are faster.
A second part of the thesis focused on the use of micellar liquid chromatography (MLC), becausethe presence of micelles in the mobile phase mimics the lipid bilayer structure of membranes and changes the stationary phase properties accordingly. The results of an MLC method on a particle based and a monolithic column were compared. The latter column type offered the advantage of fast analyses, by the ability to increase the flow rate, without a loss of information. Both column types provided similar skin permeability models, which showed an improvement to the models containing only theoretical descriptors. Consequently, application of the monolithic column provides the preferred MLC method to estimate skin permeability, because of its fast characteristics.
In the third part of this thesis, stationary phases mimicking the skin structure, i.e. an immobilized artificial membrane (IAM) and a cholesterol-bonded column, were implemented to model skin permeability. Moreover, another high-throughput method applying a sub-2 µm particle octadecyl silica column in UHPLC was investigated. The best model was obtained with a stepwise MLR approach, including the UHPLC and E-Dragon descriptors. An improvement was noticed relativeto the best models obtained within the previous two parts. Because this model was also quite complex, it increased the risk of overfitting. Therefore, as an alternative, simplified MLR models, which still presented good performance parameters, were considered. Finally, UHPLC retention was preferred, based on its fast determination, while it resulted in the best performing skin
permeability model.
In the last part of the thesis, supercritical fluid chromatography (SFC) was applied as an alternative for the previously tested methods in liquid chromatography. SFC presents the advantage of being ‘green’ and high-throughput. Nine dissimilar stationary phases were tested, of which the retention on a phenyl column together with E-Dragon descriptors were included in the best stepwise MLR model. This model showed better performance parameters than the model
built with only theoretical descriptors, and performed comparably to the best UHPLC model.
Similarly to the observations in the third part, the high complexity of the best SFC model increased the possibility of overfitting. As a compromise, less complex models with E-Dragon and SFCdescriptors were determined, for which the model containing the retention on a cyanopropyl column excelled. Thus, to model skin permeability SFC seems a promising technique, for which certain parameters even can be further optimized in the future (e.g. temperature, backpressure, type of organic modifier and use of additives in the mobile phase).
Overall, it could be noticed that the best model in each part of the thesis was obtained with the stepwise MLR approach and E-Dragon descriptors. The MLR models with Vega ZZ descriptors provided simpler models of average quality, in which the selected theoretical descriptors could be easier related to skin permeability processes. The PLS models built with E-Dragon descriptors
were generally better than with Vega ZZ descriptors. However, their performance did not surpass those of the E-Dragon MLR models.
Summarized, the overall best performing model was obtained with some E-Dragon descriptors and the SFC retention on a phenyl column, closely followed by the equivalent model built with aUHPLC descriptor. Both the SFC and UHPLC methods provided high-throughput analyses, thus offering fast skin permeability estimations compared to the currently used in-vivo and in-vitromethods. However, as a general requirement for their application in real life conditions, these models should be validated using a new external data set