Incorporating physiologically relevant mobile phases in micellar liquid chromatography for the prediction of human intestinal absorption

Abstract Micellar liquid chromatography is a popular method used in the determination of a compound's lipophilicity. This study describes the use of the obtained micelle–water partition coefficient (log P mw) by such a method in the prediction of human intestinal absorption (HIA). As a result of the close resemblance of the novel composition of the micellar mobile phase to that of physiological intestinal fluid, prediction was deemed to be highly successful. The unique micellar mobile phase consisted of a mixed micellar mixture of lecithin and six bile salts, i.e. a composition matching that found in the human intestinal environment, prepared in ratios resembling those in the intestine. This is considered to be the first method to use a physiological mixture of biosurfactants in the prediction of HIA. As a result, a mathematical model with high predictive ability (R 2 PRED = 81%) was obtained using multiple linear regression. The micelle–water partition coefficient (log P mw) obtained from micellar liquid chromatography was found to be a successful tool for prediction where the final optimum model included log P mw and polar surface area as key descriptors with high statistical significance for the prediction of HIA. This can be attributed to the nature of the mobile phase used in this study which contains the lecithin–bile salt complex, thus forming a bilayer system and therefore mimicking absorption across the intestinal membrane.

a greater possibility for improvement when compared with those with low intestinal permeability as drug solubility can be enhanced by choosing a more suitable formulation option. As a consequence of this, focussed synthesis of compounds with structures of reasonably high permeability during the early stages of drug development is preferential. Since drug lipophilicity is considered a key descriptor that dictates permeation across biological membranes (Rutkowska, PajIk, & Jóźwiak, 2012), the evaluation or determination of the lipophilicity of a drug is important for its characterization to ensure its potential to penetrate lipid barriers and subsequently be absorbed (Lipinski, Lombardo, Dominy, & Feeney, 2001;Scott & Clymer, 2002). Therefore, determining the lipophilicity of a compound can help in the prediction of human intestinal absorption (HIA).
Having the ability to explore the effects of micelles on the behaviour of a compound, micellar liquid chromatography (MLC) has been developed over the past 30 years to yield information on a wide variety of compounds where a surfactant aqueous solution is used above its critical micellar concentration (CMC; Berthod & Garcia-Alvarez-Coque, 2000) (Ruiz-Ángel, García-Álvarez-Coque, & Berthod, 2009). A very important physicochemical property indicating lipophilicity, log P mw , can be obtained using MLC in the presence of different surfactants as the micellar mobile phase to help characterize compounds (Kawczak et al., 2010;Marina & Garcia, 2000). For example, MLC has been used with a simple surfactant solution in previously published work for the prediction of HIA for a series of compounds using multiple linear regression analysis (Waters, Shokry, & Parkes, 2016). The novelty of the work presented in this study lies in the use of a very unique mobile phase mimicking the in vivo intestinal environment of humans. The composition of this mobile phase was closely related to that of the intestinal fluid through a combination of lecithin and bile salts which are normally found in the human intestine, used in ratios matching those found physiologically (Wiedmann, Liang, & Kamel, 2002

| Preparation of a mixed micellar solution for dilution
Different concentrations of the micellar mixture were prepared over the range of 5-17 mM by diluting the stock solution using a 2 mM mixture solution. The 2 mM mixture solution contained the same six bile salts and lecithin used in the preparation of the stock mixture solution in the same molar ratios. The 2 mM diluting mixture was prepared by transferring accurately weighed amounts equivalent to 0.32, 0.25, 0.24, 0.24, 0.55 and 0.4 mM of NaTC, NaTDC, NaDC, NaC, NaGC and NaGDC bile salts, respectively, and 0.75 mM of egg phosphatidylcholine to a 250 mL volumetric flask with buffer as detailed previously. The resultant solution was sonicated for 30 min then stored for 12 h before use. Dilution was carried out in this way as the 2 mM mixture is considered to be the monomer bile salt concentration that is required to be kept constant in each solution in order to keep the size of the micelle constant while its concentration is being changed (Wiedmann et al., 2002).

| Analytical instrumentation and measurement
Experiments were carried out with a chromatographic system consisting of an Agilent 1100 series binary pump, a Rheodyne injector through which 20 μL samples were injected in to the system and a UV detector (Perseptive Biosystems UVIS-205), set at a wavelength appropriate for each drug producing a peak via Picolog software indicating the retention of the solute within the column as a function of time. The mobile phase was filtered through a 0.45 μm Nylon filter and degassed in an ultrasonic bath. Data were recorded and then analysed to obtain retention factors and each run was repeated three times to ensure that reasonable accuracy and precision were achieved.
Analytical separation was accomplished using a reversed phase cyanopropyl column (Spherisorb 5 μm, 15 cm × 4.6 mm i.d., Waters). The flow rate used was 1.34 mL/min with all assays carried out at 37°C.
The mobile phase was placed in a water bath at 37°C throughout the duration of all experiments.

| Determination of dead time t 0
The dead time (t 0 ) is defined as the time taken by the solvent front to reach the detector, measured by the injection of water (Pramauro, Minero, Saini, Graglia, & Pelizzetti, 1988) or an organic solvent, e.g. acetonitrile or methanol (Khaledi, 1988;Khaledi, Peuler, & Ngeh-Ngwainbi, 1987). In this work, dead time was determined by injecting distilled water or acetonitrile in to the system and recording the retention time of the first peak that appeared after injection (solvent front).
The same method was repeated for each of the bile salt concentrations used. A reliable value of dead time used in the calculation of retention factor (k′) for all experiments (using eqn (1)) was determined from an average of at least 10 recordings. From the obtained chromatograms, the retention time of each drug was recorded for each bile salt concentration ( Figure 1). The retention factor for each retention time was calculated using the following equation: The reciprocal of each retention factor was obtained (1/k ′ ) with the average value plotted against the micellar concentration (C M ) that was calculated according to the following equation: The partition coefficient (log P mw ) was obtained from the slope and intercept of the line obtained from the plot of (C M ) against (1/k ′ ):

| Mixed micellar system
Since bile salts and lecithin (phosphatidylcholine) are considered to be two of the most common biosurfactants present in bile and involved in the digestion process, it was important to study the effect of using a mixed micellar system consisting of six bile salts and lecithin phospholipid as a mobile phase in MLC. The mixed micellar system used in this method consisted of a mixture of six bile salts (NaDC, NaC, NaTDC, NaTC, NaGC and NaGDC) which included dihydroxy, trihydroxy, con- ratios similar to that present physiologically (Wiedmann et al., 2002).
Having both a positively charged choline head group and a negatively charged phosphate group, lecithin is considered to be a zwitterionic compound that tends to self-assemble in water, forming characteristic bilayer membrane-like structures (Cheng, Oh, Wang, Raghavan, & Tung, 2014 (Malik, 2016). It was reported in previous studies that short, rod-like micelles were formed upon combining both bile salts and lecithin in a mixture (Cheng et al., 2014). The lecithinbile salt complex is considered as a balanced system where the lecithin on its own in water forms unstable bilayer structures of low aqueous solubility because of its bulky hydrophobic tails inhibiting its solubility in water that is compensated for and balanced by the presence of the bile salts of much greater water solubility. These can, in small amounts, stabilize the lecithin self-assembled structures by intercalating into these structures and thus promoting their water solubility, which is one of the main physiological applications of bile salts.
Initially, it was suggested by Mazer, Benedek and Carey that the aqueous lecithin/bile salt micelles were disc-like in shape (Mazer, Benedek, & Carey, 1980) but later on, different techniques provided evidence that these micelles are cylindrical in shape and can further grow into long flexible cylindrical micellar chains termed "worms" (Madenci, Salonen, Schurtenberger, Pedersen, & Egelhaaf, 2011;Walter, Vinson, Kaplun, & Talmon, 1991) which are similar to polymer chains where they entangle in a transient network rendering the solution highly viscous (Dreiss, 2007;Schurtenberger, Scartazzini, & Luisi, 1989;Shchipunov, 2001). This transformation of short cylinders to worms depends on the molar ratio of the two species and the ionic strength where an almost equimolar ratio of bile salt-lecithin (with high background counterion concentration) would induce the growth of the cylindrical micelles to worms (Cheng et al., 2014). As a result, caution was taken to avoid the formation of a highly viscous solution since the prepared micellar mixture was to be pumped through the chromatographic system. For this reason, the bile salt-lecithin mixed micellar system was prepared in a molar ratio much higher than that using an optimum counterion concentration (0.15 M NaCl).
Lecithin prefers to be present in the form of low-curvature cylindrical-shaped bodies owing to its two tails. It is expected that the bile salts will stabilize the hemispherical end caps of these cylinders as bile salts are generally present in water as highly curved small micelles.
Since stable end caps prevent the formed cylindrical micelles from further growing into long chains, adding more bile salts will result in more end caps being formed and therefore shorter cylinders (Cheng et al., 2014). Figure 2 summarizes the mechanism of micellization in the bile salt-lecithin mixed micellar system where lecithin prefers to form  Table 1.
For improving the linear relationship found between reported %HIA and experimental log P mw values, logit (HIA) was used as reported in similar studies (Norinder, Österberg, & Artursson, 1999;Raevsky, Fetisov, Trepalina, McFarland, & Schaper, 2000;Zhao et al., 2002). As a result, transformation of human intestinal absorption values to logit was carried out by substituting in eqn (4): Removal of drugs with %HIA values of 100 or 0% from the training set was essential.

FIGURE 2
Schematic of the self-assembled structures formed by lecithin with, and without, bile salt in water (Cheng et al., 2014)  Minitab 17 ® was used in the statistical analysis of data. Data was analysed using multiple linear regression where all of the previously mentioned molecular descriptors (data not shown) were included and regressed against the dependant variable %HIA and backward elimination modelling strategy was used. As a result, PSA along with log P mw were the only descriptors included in the final developed model. Variables with high variance inflation factors were removed to keep it within the acceptable limits. Finally, an optimum model was obtained that provided a good summary of data. Assessment of the variables remaining in the final model for significance and relative importance was carried out using standardized coefficients and the associated p-values.
The final model predictive ability was evaluated using adjusted-R 2 and R 2 for prediction (R 2 PRED ), which is able to indicate the predictive  The t-statistic and standardized coefficient of log P mw are −8.51 (p < 0.05) and − 0.964, respectively, suggesting the statistical significance of log P mw as a predictor. Also the F -ratio of the overall model is statistically significant, F = 38.12 and the p-value is 0.007 (p < 0.05). Three drugs (acetaminophen, ibuprofen and salicylic acid) were used to test the obtained model. The model was able to predict the %HIA for these compounds within a minimum of 0.61% and a maximum of 4.43% difference between predicted and published data for %HIA.
The statistical model developed from this study using a bile saltslecithin physiological mixture confirms an enhanced capability for prediction of HIA, with an R 2 PRED of 81%, compared with a previous study using simple micelles of one bile salt (R 2 PRED of 75%; Waters et al., 2016). The current model involves fewer predictors (log P mw and polar surface area) than those in the previous model (logP mw , molecular weight and solubility), which simplifies the model. Based on previous research, PSA has been reported to be a successful parameter in the prediction of intestinal absorption (Clark, 1999;Palm, Stenberg, Luthman, & Artursson, 1997;Stenberg et al., 1999). Furthermore, drug absorption-relevant information has been shown to be sufficiently encoded in lipophilicity, along with PSA, without explicit reference to molecular weight (Egan, Merz, & Baldwin, 2000). These findings further corroborate that the current model displays superiority over the previously developed one for predicting intestinal absorption.

| CONCLUSION
Development of an MLC method that used a physiologically resembling bile salt-lecithin mixed micellar system was successful for the prediction of HIA. This method had a significant impact on the elution of compounds and the type of interaction they experienced upon being injected into the MLC system. The bile salt-phospholipid combination had a higher solubilizing capacity for compounds than that of the individual bile salt systems used before (Waters et al., 2016), confirmed by the behaviour of all compounds into binding solutes favouring the formed micelles. This developed MLC method has a higher predictive ability for HIA (R 2 PRED = 81%) compared with previous models. Overall, it can be concluded that there is a close resemblance between the 'physiologically occurring' and 'synthetic bile salt-phospholipid micellar mixture' used in this MLC method. This helped the compounds to behave in a manner closer to how they permeate through the human intestine, therefore simulating the human intestinal absorption process to some extent and ultimately leading to the construction of a mathematical model with a high predictive ability for HIA.