Immobilized Keratin HPLC Stationary Phase—A Forgotten Model of Transdermal Absorption: To What Molecular and Biological Properties Is It Relevant?

Chromatographic retention data collected on immobilized keratin (KER) or immobilized artificial membrane (IAM) stationary phases were used to predict skin permeability coefficient (log Kp) and bioconcentration factor (log BCF) of structurally unrelated compounds. Models of both properties contained, apart from chromatographic descriptors, calculated physico-chemical parameters. The log Kp model, containing keratin-based retention factor, has slightly better statistical parameters and is in a better agreement with experimental log Kp data than the model derived from IAM chromatography; both models are applicable primarily to non-ionized compounds.Based on the multiple linear regression (MLR) analyses conducted in this study, it was concluded that immobilized keratin chromatographic support is a moderately useful tool for skin permeability assessment.However, chromatography on immobilized keratin may also be of use for a different purpose—in studies of compounds’ bioconcentration in aquatic organisms.


Introduction
Many chemicals enter the human body through the skin. Transdermal absorption is an important route of drugs' administration, and it is also very important in the context of environmental toxicology, since undesired xenobiotics are often absorbed transdermally. The skin permeability coefficient K p is defined according to Equation (1): where: K m -the partition coefficient between the stratum corneum and the vehicle; D-the effective compound's diffusion coefficient through the stratum corneum; h-the diffusional pathlength.
The experimental values of skin permeability coefficients are measured in vivo (on human volunteers), ex vivo (on excised human skin), or on animal models [1], but such data are difficult to obtain due to ethical and financial problems, and the results of experiments in this area are often inconsistent due to variations in properties of different skin samples, even taken from the same human or animal.
Apart from skin absorption, an important property of compounds of environmental concern is their bioconcentration factor in aquatic organisms (BCF). The bioconcentration factor is the ratio of the chemical concentration in the organism (C B ) and water (C w ), accounting for the absorption via the respiratory route (e.g., gills) and skin. It is used to assess the bioaccumulation potential of compounds [2], especially in the absence of their bioaccumulation factor (BAF),which accounts for dietary, dermal, and respiratory exposures. According to different regulatory agencies, different criteria of bioaccumulation More recently, the bioconcentration of compounds in aquatic organisms has been investigated using chromatography on IAM stationary phases, developed initially to mimic molecule-biomembrane interactions in ADME studies [31,52]. Earlier research pointed to the importance of additional parameters, incorporated alongside log k IAM : (i) a biodegradation estimate, BioWin5, calculated using the EPISuite TM software and (to a lesser extent) topological polar surface area (TPSA) [52]; (ii) TPSA-the fraction of sp 3 carbon atoms (F Csp3 ) and hydrogen bond donor count (#HD) [31].
Turowski and Kaliszan postulated that predicting skin permeability of compounds should be based on molecules' lipophilicity and interactions with keratin, which is an important constituent of the outmost layer of the epidermis [34]. An immobilized keratinbased stationary phase, developed by Turowski and Kaliszan, was initially proposed to be an in vitro tool in investigations of solutes skin permeability (log K p ) [34]. However, it was discovered that the retention factor obtained on this sorbent (log k KER ) is not a sufficiently good predictor of skin permeability coefficient, and it cannot be used as a sole descriptor in log K p models. Turowski and Kaliszan reported that this descriptor can be combined with the chromatographic retention factor obtained by immobilized artificial membrane chromatography (log k IAM ), and the results of log K p predictions using multiple linear regression (MLR) models satisfy (Equation (2)): log K p = −6.56 + 1.92 log k IAM − 1.04 log k KER (n = 17, R 2 = 0.86) Turowski and Kaliszan concluded that skin permeability increases with the lipophilicity of solutes (encoded primarily by log k IAM ) and decreases with their affinity for keratin (expressed as log k KER ). Unfortunately, the model they proposed (Equation (2)) requires two sets of chromatographic data, obtained on different stationary phases, this being the likely reason why the immobilized keratin stationary phase they proposed has never become widely popular and, to the best of our knowledge, it is not commercially available.
In this study, a novel application of immobilized keratin stationary phases developed by Turowski and Kaliszan is proposed, and chromatography on immobilized keratin sorbent is used to model compounds' bioconcentration in aquatic organisms.

IAM and Immobilized Keratin Chromatography
The chromatographic retention factors for the compounds analyzed in this study (Table 1) were taken from [34]. They were obtained on: (i) an IAM.PC.MG HPLC column purchased from Regis (150 × 4.6 mm, particle diameter 12 µm, pore diameter 300 Å) with a phosphate buffer (pH 6.0), including acetonitrile (95:5 v/v) mobile phase (flow rate-1 mL min −1 ); (ii) physically immobilized keratin sorbent with pH 4.2 phosphate buffer as a mobile phase (column dimensions-125 × 4 mm; flow rate-1 mL min −1 ). The mobile phase used in keratin chromatography (pH 4.2 buffer) was selected on the basis of QSRR studies as giving the "best" relationship between log k KER and structural descriptors (molecular weight and dipole moment) [34].

Reference Values of Skin Permeability Coefficient (log K p ) and Bioconcentration Factor (log BCF)
The experimentally determined values of log K p and log BCF are available for only some compounds within the studied group. For this reason, the models of skin permeability and bioconcentration factor, involving chromatographic and calculated descriptors, were generated and validated using log K p and log BCF values obtained in silico with the EpiSuite v. 4.1 software (log K p EPI -DERMWIN v. 2.02 and log BCF EPI -BCFBAF v. 3.02 modules, respectively), recommended by the US Environmental Protection Agency [54,55] and tested on sub-groups of solutes whose experimental log K p or log BCF values are known (log K p exp , log BCF exp ) [56,57]. The estimation methodology used by DERMWIN is based on an algorithm developed by Potts [58], and the estimations provided by BCFBAF are based on methodology developed by Meylan [59] and Arnot and Gobas [3]. The values of log K p EPI and log BCF EPI obtained using EpiSuite are given in Tables 2 and 3.

Statistical Tools
Multiple linear regression (MLR) models were generated using Statistica v. 13 by StatSoft Polska, Kraków, Poland, and this refers to the stepwise forward regression mode.
The models considered in this study were evaluated using the following procedures: • Cross-validation was performed, with n compounds from the initial training set split into 2 subsets, one of which was used to train a new model and the remaining one to test it. After cross-validation, the RMSEP (root mean squared error of prediction) for the particular N-compound test subset was calculated as follows (Equation (3)): • Comparison of the predicted log K p pred and log BCF pred values (calculated for the compounds, whose experimental log K p exp and log BCF exp data are available) was per-formed, and these data were analyzed using the squared coefficient of determination (R 2 exp ).

Keratin vs. IAM HPLC Skin Permeability Models
In this study, we compared the log K p models obtained using log k IAM and TPSA (Equation (4)) with the models including log k KER as a chromatographic parameter (Equation (5) It was observed that neither Equation (4), nor (5), gives satisfying results of log K p predictions for relatively strongly ionized solutes (compounds 14, 16, 23, and 27); when these compounds were excluded from the analysis, Equations (6) and (7) were obtained for a group of 28 neutral, basic, or weakly acidic compounds (Figures 1 and 2, Table 2).  The likely reason for such discrepancies between the predicted (Equations (4) an (5)) and reference values of log Kp for relatively strongly ionizable compounds is that th reference model has also its limitations: it overestimates the results for very hydrophili molecules, underestimates the values for non-hydrogen bonding solutes, and fails fo extremely lipophilic compounds or solutes having a very high tendency to hydroge The likely reason for such discrepancies between the predicted (Equations (4) and (5)) and reference values of log Kp for relatively strongly ionizable compounds is that the reference model has also its limitations: it overestimates the results for very hydrophilic molecules, underestimates the values for non-hydrogen bonding solutes, and fails for extremely lipophilic compounds or solutes having a very high tendency to hydrogen bonding [60][61][62]. At this point, the group of 28 studied compounds was divided into two subsets: a training set (1 to 20) and a test set (21 to 28). Equations (8) and (9) generated for the training set, and containing the same sets of independent variables as Equations (6) and (7), are as follows ( Table 2): The likely reason for such discrepancies between the predicted (Equations (4) and (5)) and reference values of log K p for relatively strongly ionizable compounds is that the reference model has also its limitations: it overestimates the results for very hydrophilic molecules, underestimates the values for non-hydrogen bonding solutes, and fails for extremely lipophilic compounds or solutes having a very high tendency to hydrogen bonding [60][61][62].
At this point, the group of 28 studied compounds was divided into two subsets: a training set (1 to 20) and a test set (21 to 28). Equations (8) and (9) generated for the training set, and containing the same sets of independent variables as Equations (6) and (7), are as follows ( Table 2)

Keratin HPLC Models of Bioconcentration Factor
According to our earlier research, the bioconcentration factor log BCF can be predicted using log k IAM and two additional parameters: F Csp3 and TPSA [31]. The predictive potential of Equation (10) (Figure 3) is compared to that of a model based on chromatographic retention factors obtained using immobilized keratine as a stationary phase (Equation (11), Figure 4).

Discussion
In our study, we investigated the possibility of using log kKER in skin permeability models, alongside additional descriptors that were either not considered or not available when the keratin stationary phase was originally developed. We studied correlations between log kKER and the key physico-chemical properties associated with compounds' ability to cross biological barriers ( Table 4) and discovered that log kKER encodes primarily lipophilicity (log Kow) and aqueous solubility (log S), which are important factors governing the ability of compounds to cross the skin barrier, but the correlations are moderate.  At this point, the group of 32 studied compounds was divided into two subsets: a training set (1 to 20) and a test set (21 to 32). Equations (12) and (13) generated for the training set, and containing the same sets of independent variables as Equations (10)

Discussion
In our study, we investigated the possibility of using log k KER in skin permeability models, alongside additional descriptors that were either not considered or not available when the keratin stationary phase was originally developed. We studied correlations between log k KER and the key physico-chemical properties associated with compounds' ability to cross biological barriers ( Table 4) and discovered that log k KER encodes primarily Pharmaceutics 2023, 15, 1172 9 of 12 lipophilicity (log K ow ) and aqueous solubility (log S), which are important factors governing the ability of compounds to cross the skin barrier, but the correlations are moderate. Predictive models of log K p , involving retention parameters obtained on immobilized keratin (Equations (7) and (9)), have similar (or, in fact, slightly better) statistical parameters compared to those reported for models based on IAM chromatography (Equations (6) and (8)). Log K p values predicted using Equation (7) are in a slightly closer agreement with experimental log K p exp data available for a subset of 18 compounds than those calculated using Equation (6). It must be noted, however, that, in the process of descriptors' selection by forward stepwise regression, chromatographic parameters log k KER and log k IAM behave differently. Log k IAM (Equation (6)) is selected first, and it accounts for ca. 66% of total log K p variability; log k KER (Equation (7)) is selected second (after TPSA), and it accounts for just 16% of total log K p variability.
The significance of log k KER as an independent variable is much higher in models of bioconcentration factor log BCF. In Equation (11), log k KER is the most important independent variable, accounting for 39% of total log K p variability; further variables (selected as follows: TPSA, MR, and #ArHvAt) account for 24, 18, and 7% of total log K p variability, respectively. In the IAM chromatography-based model of log BCF (Equation (10)), log k IAM accounts for 73%, and other independent variables (F Csp3 and TPSA) account for 12 and 2% of total log K p variability, respectively. The keratin chromatographic retention-based model (11) has statistical parameters similar to those of Equation (10), derived from IAM chromatography; however, Equation (11) seems to fit the experimental data (log BCF exp ) reported for a subset of 10 compounds better than Equation (10).

Conclusions
Immobilized keratine-based chromatographic stationary phase was developed in the late 1990s to help in in vitro investigations of compounds' transdermal absorption. A new model of a skin permeability coefficient was developed in the current study, which involves the chromatographic retention factor measured on the immobilized keratine sorbent (log k KER ) and four additional independent variables (Equation (7)). This model has slightly better statistical parameters and is in a better agreement with experimental log K p data than the model derived from IAM chromatography (Equation (6)); both models are applicable primarily to non-ionized compounds (with carboxylic acids removed from Equations (4) and (5)). Based on the MLR analyses conducted in this study, it was concluded that immobilized keratin chromatographic support is a moderately useful tool for skin permeability assessment. However, similarly to IAM chromatography in the past, chromatography on immobilized keratin may serve a different purpose; designed for applications in pharmacokinetic studies, it may also be of use in the realm of environmental science, in studies of compounds' bioconcentrations in aquatic organisms.