Abstract
There is a considerable need to rationalize the membrane permeability and mechanism of transport for potential nutraceuticals. The aim of this investigation was to develop a theoretical permeability equation, based on a reported descriptive absorption model, enabling calculation of the transcellular component of absorption across Caco-2 monolayers. Published data for Caco-2 permeability of 30 drugs transported by the transcellular route were correlated with the descriptors 1-octanol/water distribution coefficient (log D, pH 7.4) and size, based on molecular mass. Nonlinear regression analysis was used to derive a set of model parameters a′, β′, and b′ with an integrated molecular mass function. The new theoretical transcellular permeability (TTP) model obtained a good fit of the published data (R2 = 0.93) and predicted reasonably well (R2 = 0.86) the experimental apparent permeability coefficient (Papp) for nine non-training set compounds reportedly transported by the transcellular route. For the first time, the TTP model was used to predict the absorption characteristics of six phenolic acids, and this original investigation was supported by in vitro Caco-2 cell mechanistic studies, which suggested that deviation of the Papp value from the predicted transcellular permeability (Papptrans) may be attributed to involvement of active uptake, efflux transporters, or paracellular flux.
Introduction
A growing number of epidemiological studies indicate an association between habitual coffee consumption and reduced risk of type 2 diabetes (van Dam and Hu, 2005). Coffee is a predominant source of phenolic acids in the form of hydroxycinnamic acids (HCAs) such as caffeic acids and ferulic acids, typically linked to a quinic acid moiety to form a subgroup known as chlorogenic acids. Chlorogenic acids and HCA derivatives may play an important role in biological mechanisms (Wang et al., 2008; Henry-Vitrac et al., 2010) with potential health benefits. Despite these reported pharmacological effects, the mechanism of intestinal absorption has been elucidated for only a small number of phenolic acids (Konishi and Kobayashi, 2004; Poquet et al., 2008). There is a considerable need for rationalizing intestinal permeability of phenolic acids because it is a critical step toward understanding their potential bioactivity. The Caco-2 cell line has been widely used as a model of intestinal absorption and estimate of oral bioavailability (Artursson et al., 2001). However, when the complex mixture of compounds present in foods is investigated, the number of permeability studies required creates several financial and technical challenges, and clarification of absorption pathways remains a labor-intensive and difficult task. The ability to make in silico predictions of the permeability of dietary components based on physicochemical properties would streamline the high-throughput screening of potential bioactive agents and provide an estimation of the major pathway of permeation and bioavailability.
Theoretical models of permeation through in vitro biological membranes such as the human intestine (Camenisch et al., 1998) have been developed within the pharmaceutical sciences. In principle, these models correlate passive intestinal permeability with physicochemical attributes (descriptors) to estimate the absorption of similar compounds. Most models assume that transport can occur through two main pathways: via the restricted aqueous spaces between cells (paracellular) and through the cell membrane (transcellular). Passive diffusion can be considered as the predominant mechanism of permeation for ingested compounds. Diffusion processes are largely governed by interrelated physicochemical descriptors: lipophilicity, polarity (charge, hydrogen bonding, and polar surface area) (van de Waterbeemd, 1998) and molecular mass (Camenisch et al., 1998) as described by Fick's first law of diffusion. Involvement of metabolism by cytochrome P450 enzymes and efflux processes and active transport may limit or improve membrane uptake, respectively, causing deviation from the predicted permeability.
Theoretical models are a useful tool for high-throughput screening of potentially bioavailable compounds but have never been used to investigate absorption characteristics of polyphenols. In the current study, we have refined a previously reported descriptive absorption model, which related permeation across a biological membrane to lipophilicity measured as the distribution coefficient (log D) and molecular mass (Camenisch et al., 1998). For the first time, the new theoretical transcellular permeability model was used to estimate the transcellular passive diffusion component for an external validation set consisting of 22 compounds, including a range of phenolic acids of diverse molecular mass.
Materials and Methods
Chemicals.
Compounds used in this study were obtained from Sigma-Aldrich (Dorset, UK) unless stated otherwise. All water refers to deionized Millipore water (Millipore UK Ltd., Hertfordshire, UK). Culture flasks and Transwell plates fitted with polycarbonate semipermeable membranes of 0.4-μm pore size and 4.67-cm2 area were manufactured by Corning Life Sciences (Lowell, MA) and supplied by Sigma-Aldrich. 5-O-Caffeoylquinic acid (5-O-CQA) and 3,5-di-O-caffeoylquinic acid (3,5-diCQA) were purchased from Biopurify Pharmaceuticals Ltd. (Sichuan, China). Standard 5-O-feruloylquinic acid (5-O-FQA) was synthesized at the Nestlé Research Center (Lausanne, Switzerland) and kindly donated by D. Barron (Nestlé Research Center, Lausanne, Switzerland).
Selection of the Training and Validation Set Compounds.
This study used an existing data set with a previously reported passive permeability-lipophilicity relationship (Artursson and Karlsson, 1991; Camenisch et al., 1998). However, our aim was to develop a theoretical permeability equation to predict absorption exclusively via passive transcellular diffusion, referred to as the theoretical transcellular permeability (TTP) model. Thus, a literature search was performed to filter compounds reported to have extensive active or efflux transport, metabolism, or uptake by the paracellular route. Because transport of a single compound across the intestinal barrier may involve absorption via several routes (Avdeef and Tam, 2010; Sugano et al., 2010), it is a difficult task to identify compounds with transport wholly by the transcellular route. However, on the basis of the following evidence, the compounds present in the training set are mainly transported via the passive transcellular pathway. This conclusion was based on reported insensitivity to membrane junction tightness (Hilgendorf et al., 2000), in vitro mechanistic studies, which correct for the contribution of paracellular component of transport (Avdeef and Tam, 2010), ex vivo tissue transport studies (Bernards and Hill, 1992; Kern and Bernards, 1997), theoretical modeling based on pH partition theory (Obata et al., 2005), and human bioavailability data (Artursson and Karlsson, 1991; Palm et al., 1997; Yamashita et al., 2000) indicating rapid and significant absorption (greater than 50% of oral dose). The following compounds were selected as model compounds transported by mainly transcellular passive diffusion (n = 30): alprenolol, antipyrine, ceftriaxone, corticosterone, coumarin, dexamethasone, diazepam, epinephrine, felodipine, guanabenz, guanoxan, hydrocortisone, imipramine, ketoprofen, lidocaine, metoprolol, mibefradil, morphine, nitrendipine, phenytoin, piroxicam, practolol, propranolol, remikiren, sulpiride, testosterone, theophylline, tiacrilast, verapamil, and warfarin.
To evaluate the capacity of the TTP model to predict the transcellular component of permeability across Caco-2 monolayers for compounds outside of the training set, Papptrans values were calculated for a total of 22 non-training set compounds of known transport mechanism consisting of 6 phenolic acids and 9 transcellular compounds: acetylsalicylic acid, aminopyrine, 5-O-caffeoylquinic acid, caffeic acid, caffeine, carbamazepine, ferulic acid, 3,5-di-O-caffeoylquinic acid, 3,4-dimethoxycinnamic acid, 5-O-feruloylquinic acid, fluvastatin, griseofulvin, naproxen, nevirapine, and salicylic acid. Mannitol was used as the classic paracellular marker. d-Glucose (glucose transporter 2), Gly-Pro (dipeptide transporter) and l-dopa (amino acid transporter) were chosen as model compounds taken up by specific transporters. Cimetidine, rhodamine 123, and talinolol have been reported as substrates for the efflux transporter P-glycoprotein (Hilgendorf et al., 2000; Troutman and Thakker, 2003).
Interlaboratory Variability.
In brief, Caco-2 permeability studies and log D measurements were performed as described below using acetylsalicylic acid, alprenolol, atenolol, hydrocortisone, metoprolol, and salicylic acid to determine the extent of variability between our findings and the original published data for the training set (Artursson and Karlsson, 1991). Data are presented as mean ± S.D.
Determination of Distribution Coefficient.
Log D values were determined for acetylsalicylic acid, alprenolol, atenolol, caffeic acid, 5-O-CQA, 3,4-dimethoxycinnamic acid, 3,5-diCQA, ferulic acid, 5-O-FQA, hydrocortisone, metoprolol, and salicylic acid by the 1-octanol/water shake flask method as described in detail previously (Farrell et al., 2011). Each compound was identified by retention time and UV spectra relative to authentic standards, and the log D was determined as the mean of 10 replicates.
Cell Culture.
The human colon adenocarcinoma cell line, Caco-2 (HTB-37) was obtained from the American Type Culture Collection at passage 25 (LGC Promochem, Middlesex, UK). Permeability studies used Caco-2 cells between passages 30 and 50. Caco-2 cells were added to Transwell inserts (24-mm diameter, 4.67-cm2 growth area) at a density of 6 × 104 cells/cm2 and cultured for 21 days at 37°C under a humidified atmosphere of 95% air-5% CO2. The culture medium, Dulbecco's modified Eagle's medium, supplemented with 10% fetal bovine serum, 584 mg/l l-glutamine, 100 U/ml penicillin-streptomycin, 1% (v/v) minimum essential medium, and 0.25 μg/ml amphotericin B was replaced every other day.
Permeability and Mechanistic Studies.
On or after 22 days, permeability studies were initiated by careful aspiration of the culture medium from apical and basal compartments and 2 ml of modified HBSS, consisting of HBSS containing 1.8 mM calcium chloride (pH 7.4), was added to each compartment. Plates were incubated at 37°C in a humidified 5% CO2 atmosphere for 15 min to allow equilibration of tight junction integrity. Apical and basal solutions were carefully aspirated, and 2 ml of apical transport solution (pH 7.4) was added to the apical compartment; all basal solutions were modified HBSS (pH 7.4). Thus, both donor and receiver compartments used the same pH, which was a preassumption of the permeability-lipophilicity model described by Camenisch et al. (1998). The transport solution consisted of modified HBSS (pH 7.4) containing 500 μM test compound, which was similar to the amount used to determine Papp values for the training set compounds. A concentrated stock of test compound, which was either atenolol, alprenolol, 5-O-CQA, caffeic acid, 3,5-diCQA, 3,4-dimethoxycinnamic acid, ferulic acid, 5-O-FQA, hydrocortisone, metoprolol, or salicylic acid, was prepared in DMSO and then diluted in modified HBSS by 500-fold so that the percentage of DMSO was maintained at 0.2% in each transport experiment. The transepithelial electric resistance (TEER) was measured with a Millicell ERS volt-ohm meter fitted with a chopstick probe (Millipore Ltd., Watford, UK). Plates containing differentiated monolayers and a reference sample (2 ml) of transport solution in a cell-free compartment were incubated at 37°C in a humidified 5% CO2 atmosphere for 1 h without stirring. Permeation was monitored by removal of basal solution (100 μl) at 30, 60, 90, and 120 min, and the volume was replaced with modified HBSS. At the end of the incubation period, TEER measurements were repeated, an aliquot (1 ml) was removed from the apical and basal compartments to which acetic acid was added to obtain a final concentration of 10 mM, and samples were stored at −80°C. In all transport experiments, the pH measurements were recorded on the initial transport solution and repeated for all apical and basal compartment solutions at the end of the incubation period.
To assess the time-dependent transport of 5-O-CQA, 5-O-FQA, and 3,5-diCQA, permeability studies were performed in the apical-to-basal (A→B) and basal-to-apical (B→A) directions (bidirectional). In brief, transport solution (pH 7.4) consisting of test compound (500 μM) prepared from a concentrated DMSO stock and diluted in modified HBSS containing 1.8 mM calcium chloride was added (2 ml) to the donor compartment, and all receiver solutions (2 ml) were modified HBSS (pH 7.4). Samples were removed at 30, 60, 90, and 120 min and stored as described above.
Concentration dependence of membrane permeability was investigated for 5-O-CQA, 5-O-FQA, and 3,5-diCQA by bidirectional permeability studies using test compounds at a final concentration of 50, 100, 200, and 500 μM prepared from a concentrated DMSO stock as described above. In brief, transport solution (pH 7.4) consisting of test compound diluted in HBSS containing 1.8 mM calcium chloride was added (2 ml) to the donor compartment; modified HBSS (pH 7.4) was used as the receiver solution (2 ml). Transwell plates containing differentiated monolayers were incubated for 1 h, and thereafter samples (1 ml) were removed from the receiver compartment and stored as described.
Mechanistic studies were performed to elucidate the role of tight junction integrity on phenolic acid permeability across the Caco-2 monolayer. In brief, inserts containing differentiated monolayers (n = 3 per condition) were aspirated to remove apical and basal medium and replaced (2 ml) with HBSS (pH 7.4) containing either 18 μM or 3.6 mM calcium chloride and incubated at 37°C in a humidified 5% CO2 atmosphere for 15 min to vary the monolayer resistance. After 15 min, both compartments were carefully aspirated and transport solution (2 ml) consisting of 500 μM test compound dissolved in HBSS (pH 7.4) at the necessary calcium chloride concentration was added to the apical compartment, and HBSS (pH 7.4) of the same calcium chloride concentration was added (2 ml) to the basal compartment. TEER measurements were recorded, and the plates were incubated at 37°C in a humidified 5% CO2 atmosphere for 1 h without stirring. After 1 h, TEER measurements were repeated, and samples (1 ml) were collected from both compartments and stored as described above. TEER values were corrected for the intrinsic cell-free resistance of the insert in all cell culture studies.
To investigate the involvement of passive diffusion in the absorption process, the bidirectional permeation was evaluated. In brief, differentiated monolayers were incubated with transport solution consisting of HBSS containing 500 μM test compound and 1.8 mM calcium chloride added either to the apical or basal compartment, depending on the direction of transport; modified HBSS (pH 7.4) was the receiver solution. Plates were incubated for 1 h, and the amount permeating to the receiver compartment was determined by HPLC using a diode array detector (DAD). The efflux ratio for permeability was calculated to identify nonpassive transport mechanisms.
To study the role of ABC transporters on absorption, differentiated Caco-2 monolayers were exposed to 3,5-diCQA (500 μM) in the basal compartment and a 25 or 10 μM concentration of an inhibitor in the apical compartment. Both test compound and inhibitor were prepared from a concentrated DMSO stock and diluted in modified HBSS containing 1.8 mM calcium chloride; the final percentage of DMSO in all the experiments was maintained at 0.2% basal and 0.05% apical. The amount of 3,5-diCQA detected in the apical compartment after incubation (1 h) at 37°C and 5% CO2 was compared with the amount detected under control conditions, which was the absence of inhibitor. Papp values were determined for control and inhibitor conditions, and the percentage inhibition was calculated as the difference between the Papp for control and inhibitor, divided by the Papp for control, multiplied by 100. Inhibitors were used at concentrations similar to those often effective in inhibition studies and have been demonstrated to potently inhibit ABC transporter proteins: the role of breast cancer resistance protein (BCRP, ABCG2) was investigated using apigenin; P-glycoprotein (P-gp; ABCB1)-mediated transport was studied with cyclosporine; involvement of multidrug resistance-associated protein (MRP2, ABCC2) was analyzed using (E)-3-[[[3-[2-(7-chloro-2-quinolinyl)ethenyl]phenyl]-[[3-dimethylamino)-3-oxopropyl]thio]methyl]thio]-propanoic acid (MK571); and quercetin was used as a combined inhibitor of all three ABC transporters.
Deproteination of Culture Media.
Transport solutions were thawed and extracted as described previously (Farrell et al., 2011). In brief, transport solutions (46 μl) were combined with 6.4 μl of 50% aqueous formic acid, ascorbic acid (final concentration 1 mM), and sinapic acid (final concentration 50 μM) as an internal standard. To the mixture, 112.5 μl of acetonitrile was added dropwise to precipitate proteins, and the samples were vortexed for 1 min and allowed to stand (1 min); this procedure was repeated 2 times before centrifugation at 17,000g for 5 min. The supernatant (150 μl) was removed and dried under centrifugal evaporation (EZ-2 Plus; Genevac, Suffolk, UK). Dried samples were reconstituted in 50 μl of water containing acetonitrile (v/v, 5:95) and analyzed by HPLC-DAD.
HPLC-DAD Analysis.
Reconstituted samples were injected onto a Rapid Resolution HPLC column with a DAD fitted with a deuterium lamp, semimicro flow cell (6-mm pathlength, 1200 series; Agilent Technologies, Berkshire, UK). Elution was achieved at 30°C on an Eclipse Plus C18 column (2.1 × 100 mm, 1.8 μm; Agilent Technologies) using a 60-min gradient of (A) premixed water containing acetonitrile (v/v, 95:5) and (B) premixed acetonitrile containing water (v/v, 95:5), both containing 0.1% formic acid at a flow rate of 0.26 ml/min. Elution was initiated at 0% solvent B and maintained for 17 min; the percentage of solvent B was then increased to 16% over the next 21 min and increased to 100% for 5 min before initial starting conditions were resumed for a 15-min column reequilibration.
Recovery Efficiency.
Recovery efficiency experiments were performed for caffeic acid, 5-O-CQA, 3,5-diCQA, 3,4-dimethoxycinnamic acid, ferulic acid, and 5-O-FQA in the absence of cells and used as a correction factor in quantification. In brief, test phenolic acids, prepared from a concentrated stock in DMSO, were diluted with HBSS (pH 7.4) to 10 μM, and the final percentage of DMSO was 0.2%. Samples were incubated for 2 h at 37°C in the absence of cells in Transwell plates. After incubation, a portion (100 μl) was removed and deproteinated as described above, and recovery was determined by HPLC-DAD (n = 10).
Data Analysis.
Papp values were either taken from the literature or determined in this study by HPLC-DAD quantification of the concentration of compound in the receiver compartment after transport across the Caco-2 monolayer. The calculation was described by the equation:
where VD is apical donor volume (cm3), A is membrane surface area (cm2), dMR/dt is the change in the amount (moles) of compound in the basal receiver compartment over time (seconds), and MD is the initial amount of compound in the apical donor compartment at time 0, which was determined from the reference sample of the transport solution. Values of Papp are only valid under experimental conditions that provide a constant concentration gradient; that is, the basal receiver amount should not exceed 10% of the amount in the apical compartment (Hubatsch et al., 2007). Therefore, sampling intervals in this study (30, 60, 90, and 120 min) were selected to accommodate slow and rapid transport characteristics. For most compounds, the dMR was measured as the amount of compound accumulated at the end of the 60 min or for rapidly permeated compounds before more than 10% of the compound had permeated to the basal receiver compartment and thus sink conditions were preserved. The amount transported to the basal compartment was less than 10% for all compounds except for metoprolol and alprenolol for which transport was less than 15%.
Mass balance, defined as the sum of the amounts of compound recovered in the apical and basal compartments at the end of the transport experiment as a percentage of the initial apical donor amount, was calculated for each transport study using the equation:
where MD and MR are the amounts in the donor and receiver compartment (moles) at the start (0) and end (fin) of the experiment, and Ms(t) was the amount withdrawn in samples removed at time intervals (t). A mass balance of 80% is generally acceptable for an approximation of Papp (Hubatsch et al., 2007). Typical mass balance was found to be between 86 and 109%; thus, in the authors' opinion providing valid estimations of membrane permeability.
To establish the mechanistic nature of absorption for the phenolic acids, the efflux ratio for permeability was calculated as follows:
where Papp B→A is the permeability coefficient determined by the transport of compound from the basal to the apical compartment and Papp A→B is the permeation coefficient for transport in the apical to basal direction. An efflux ratio of 1 was indicative of passive diffusion; values less than 0.5 and greater than 2 were regarded as being indicative of active influx and active efflux, respectively.
The residual error, defined as the variation between the observed Papp and Papptrans in the logarithmic form was calculated as follows:
where a positive residual indicates a greater permeability than predicted and a negative residual indicates a lower permeability.
Data modeling of the training set was performed using the statistics package R (R Development Core Team, 2010). Linear regression and correlation relationships were investigated by analysis of the coefficient of determination (R2) and Pearson correlation coefficient, respectively. The statistical difference between samples under different cell transport conditions was investigated by analysis of variance, followed by a pairwise multiple comparison test to compare all conditions. Assumption of normality of the data sets and equality of variance was confirmed by Shapiro-Wilk and Levene's test statistics, respectively, and statistical significance was set at the 0.05 level (PASW Statistics 17). When an experimental log D value could not be located, an in silico value was calculated using MarvinSketch (version 5.3.1, 2010; ChemAxon, http://www.chemaxon.com).
Results
The Training Set.
The training set consisted of 30 compounds permeating mainly by the transcellular route and compounds used as examples are shown in Fig. 1. The physicochemical descriptors, summarized in Table 1, are an expanded set based on the works of Artursson and Karlsson (1991) and Camenisch et al. (1998). The compounds have a diversity of structure, a molecular mass range of 146 to 631 Da and sufficient scope of lipophilicity to permit statistically relevant assessments of molecular mass and log D correlations with Caco-2 permeability.
Example compounds investigated in the current study.
Physicochemical properties and Caco-2 monolayer permeability of training set compounds
The calculated log D value of ketoprofen was used in the current study because an experimental log D value based on 1-octantol/water was not available at the time.
Interlaboratory Variability.
It was our intention to develop a theoretical permeability equation based on a previously reported model (Camenisch et al., 1998) to calculate transport characteristics of the phenolic acids, which were examined in our laboratory using a Caco-2 cell system. Thus, it was necessary to investigate the interlaboratory reproducibility of the Papp and log D physicochemical parameters. Experimental determinations of Papp (Fig. 2A) and measurements of 1-octanol/water distribution coefficients (Fig. 2B) in the current study are in excellent agreement with the published data; results indicated coefficients of determination of R2 = 0.96 and R2 = 0.99, supported by a Pearson coefficient = 0.99, which was significant at the 1% level.
A, comparison of apparent permeability coefficients determined experimentally (log Pappexp) in differentiated Caco-2 monolayers (n = 3) with data published previously for the training set compounds (Artursson and Karlsson, 1991). B, comparison of experimentally determined 1-octanol/water distribution coefficients (log Dexp) at 37°C, pH 7.4, by the shake flask method (n = 10) with published measurements. ●, mean of the experimentally determined value ± S.D.
Training the Theoretical Transcellular Permeability Model.
Data previously published suggested that, for a given narrow range of molecular masses, Caco-2 permeability was reasonably well explained by a combination of log D and molecular mass with a sigmoidal relationship (Camenisch et al., 1998), referred to as the descriptive absorption model (eq. 5):
where Papp is the permeability across the Caco-2 monolayer (centimeters per second), Doct is the 1-octanol/water distribution coefficient, and a, β, and b are model parameters determined by a least-squares fit of model (eq. 5) to an experimental data set. In this current study, the model (eq. 5) has been expanded so that the model parameters depend on the molecular mass via a quadratic relationship. The choice of a quadratic relationship was driven by inspection of the model parameters obtained by fitting the model (eq. 5) to subsets of the training data with different ranges of molecular masses. In brief, the permeability-lipophilicity relationship of the 11 lowest molecular mass drugs (146–259 Da) in the training set (Table 1) was statistically analyzed by a least-squares nonlinear regression fit of the model (eq. 5) to that data subset, giving estimated model parameters a, β, and b specific to this range of molecular mass. Then the lowest molecular mass compound was removed and, in exchange, the 12th (262 Da) lowest molecular mass drug was substituted, and new model parameters were estimated. This process continued until the 11 largest molecular mass drugs (331–631 Da) were used. Thus, a series of model parameters were calculated to describe the change in the relationship between permeability and Doct as molecular mass increased. Each parameter was plotted against the corresponding mean molecular mass for that data subset, showing a quadratic relationship in each case. We therefore generated an expanded set of model parameters incorporating a quadratic dependence on molecular mass (MM). The new parameters were incorporated into the model (eq. 5) to form the TTP model (eq. 6):
where Papptrans was the predicted transcellular permeability, and the expanded model parameters are a′ = (a0 + (a1 · MM) + (a2 · MM2), β′ = (β0 + (β1 · MM) + (β2 · MM2), b′ = (b0 + (b1 · MM) + (b2 · MM2) as summarized in Table 2.
Summary of model parameters for the TTP model with integrated molecular mass function
Model parameters a′, β′, and b′ were derived by least-squares nonlinear regression fit of the descriptive absorption model (eq. 5).
Because of the integrated molecular mass function in the TTP model, it was possible for the first time to calculate a predicted value for the transcellular permeability of each compound in the training set on the basis of its specific molecular mass and log D. The TTP model successfully predicted the Caco-2 transcellular permeability of 30 training set compounds as concluded by an excellent agreement between Papp and Papptrans, which was supported by a Pearson coefficient of 0.96, significant at the 1% level (Fig. 3). The calculated residual errors were between 0.3 to −0.3 log unit, except for morphine and sulpiride (−0.4 and −0.5 log unit, respectively). Furthermore, the residual errors were normally distributed, as concluded by the Shapiro-Wilk test of normality (P > 0.05), and residual errors showed no obvious association with molecular mass or log D (Supplemental Fig. 1).
Relationship between apparent permeability (log Papp) and predicted transcellular permeability (log Papptrans) of training set compounds as predicted by the TTP model (eq. 6).
Shape and Molecular Mass Dependence.
A graphical representation of the TTP model is shown in Fig. 4. The 30 training set compounds were arranged in groups of increasing molecular mass and are intersected by contours, which represent the calculated transcellular permeability at specified molecular mass bands. The results indicated, as expected, that a bilinear relationship between permeability and log D can explain the absorption behavior of the training set compounds. An increase in molecular mass was associated with a gradual shift toward lower permeability because of weight-related restriction of membrane diffusion.
Comparison of the apparent permeability for the training set compounds and the predictions for transcellular permeability based on the log D and molecular mass (MM). Training set compounds: ▴, 146 to 194 Da; ○, 231 to 346 Da; ●, 360 to 496 Da; ■, 555 Da; □, 631 Da; dashed line, calculated transcellular permeability predictions for given molecular mass values.
Model Evaluation and Validation.
The results (Table 3) demonstrate that the TTP model can be used to accurately predict the permeability of compounds transported via transcellular passive diffusion with an error between 0.3 and −0.3 log unit. A graphical representation (Fig. 5) showed that the Papp values of all nine transcellular compounds fit within the range of permeability predicted by their log D and molecular mass. Investigation of the permeation behavior of paracellular, active influx, or active efflux substrates (Table 3) confirmed that deviation from the predicted model may be explained by a nontranscellular component of uptake across the Caco-2 monolayer. Compounds with a paracellular or active component of uptake resulted in a greater Papp value than predicted by the TTP model, indicated by a positive residual error greater than 0.3 log unit. An active efflux component resulted in a lower Papp value than predicted, indicated by a negative residual error less than −0.3 log unit. The model did not predict the active uptake of l-dopa, which may have been complicated by involvement of multiple carrier-mediated processes.
Comparison of the physicochemical properties, percentage human oral absorption, Caco-2 permeability, and predicted transcellular permeability of nontraining set compounds
Pathway of absorption was based on percentage of human oral absorption and mechanistic studies (see footnotes). For compounds transported via transcellular diffusion the Caco-2 permeability (log Papp) can be estimated by the TTP model prediction with a residual between 0.3 and −0.3. Residual error >0.3 indicates a paracellular or active influx component; residual error <−0.3 indicates an active efflux component.
Graphical representation of the fit for nine non-training set compounds transported by the transcellular route. Inset, good linearity (R2 = 0.86) between log Papp and the log Papptrans calculated using the TTP model. Dashed line, calculated transcellular permeability predictions for given molecular mass values. 1, acetylsalicylic acid, molecular mass 180 Da; 2, salicylic acid, molecular mass 138 Da; 3, caffeine, molecular mass 194 Da; 4, aminopyrine, molecular mass 231 Da; 5, fluvastatin, molecular mass 411 Da; 6, nevirapine, molecular mass 266 Da; 7, carbamazepine, molecular mass 236 Da; 8, griseofulvin, molecular mass 353 Da; 9, naproxen, molecular mass 230 Da.
Theoretical Transcellular Permeability for Test Phenolic Acids.
To investigate the absorption characteristics of the test phenolic acids, their Papptrans values were calculated (Table 4) and compared with the Papp values determined in this current study. The results, graphically displayed in Fig. 6, indicated that the TTP model correctly predicted the Caco-2 permeability of 3,4-dimethoxycinnamic acid and ferulic acid. In contrast, the Papp values for 5-O-CQA and 5-O-FQA were greater than predicted by the model. The permeation of caffeic acid and 3,5-diCQA across the Caco-2 membrane were found to be slightly less than predicted by the model.
Summary of the mechanistic data for permeability of phenolic acids across Caco-2 monolayers and the calculated transcellular permeability predictions
Permeability in all mechanistic studies was investigated by measuring the transfer of compound (500 μM) from donor (pH 7.4) to receiver (pH 7.4) compartment after a 1-h incubation with confluent monolayers (n = 3) at 37°C in a humidified 5% CO2 atmosphere. Values are mean Papp ± S.D.
Comparison of the apparent permeability of six phenolic acids determined in the current study and the predicted transcellular permeability calculated using the TTP model. Values are mean log Papp ± S.D. Dashed lines, calculated transcellular permeability predictions. Inset, comparison of the apparent permeability (○) and the predicted transcellular permeability ± 0.3 log unit (residual error) (●). 1, 5-O-CQA; 2, 5-O-FQA; 3, caffeic acid; 4, 3,5-diCQA; 5, ferulic acid; 6, 3,4-dimethoxycinnamic acid.
Mechanistic Investigation of Phenolic Acid Absorption.
The relevance of the TTP model for the estimation of absorption pathways was investigated using Caco-2 mechanistic studies to elucidate the involvement of paracellular, active efflux, and active uptake components of transport. The results, summarized in Table 4 and Fig. 7, revealed that bidirectional transport was linear with respect to time and concentration with a corresponding efflux ratio approaching 1 for all phenolic acids tested, except for 3,5-diCQA. Investigation of the route of absorption was deduced by varying the diameter of the tight junction pore, as measured by TEER resistance to permeation. Ferulic acid was the only test compound with no statistical difference between the Papp value at a TEER of 1155 ± 70 Ω · cm2 and transport at low TEER (680 ± 40 Ω · cm2), and thus we concluded that ferulic acid transport was mainly passive transcellular. In contrast, the Papp values for 5-O-CQA and 5-O-FQA were significantly (P < 0.05) higher at low TEER. Caffeic acid was previously shown to be permeated mainly via the paracellular pathway (Konishi and Kobayashi, 2004). Thus, no further mechanistic investigations were performed in the current study for this test compound. Finally, the transport of 3,5-diCQA was more complex; mechanistic studies revealed involvement of an active efflux component, as concluded by an efflux ratio of 5.4.
Time- and concentration-dependent permeation of 5-O-CQA (A and B), 5-O-FQA (C and D), and 3,5-diCQA (E and F). ——, linear relationship (R2 greater than 0.85); – – –, nonlinear relationship; ○, apical-to-basal permeation; ●, basal-to-apical permeation.
Efflux of 3,5-diCQA was further investigated using inhibitors that have been shown previously to inhibit ABC transport proteins (Matsson et al., 2009). The results of the current study (Table 5) suggest that all inhibitors significantly (P < 0.05) reduced the B→A permeation (efflux) of 3,5-diCQA. Pairwise analysis of the statistical difference between mean Papp values within the inhibitor group confirmed that there was no significant difference between the potency of MK571, cyclosporine, and apigenin. However, the potency of quercetin was significantly (P < 0.05) greater than that for the other three inhibitors. These results suggest that ABC transport proteins are involved in the active efflux of 3,5-diCQA in a serosal-to-luminal direction, and the process may occur by a combination of BCRP, P-gp, and MRP2 proteins.
Basal-to-apical permeability (efflux) of 3,5-di-O-caffeoylquinic acid in the presence of ABC transporter inhibitors
Papp values in the basal-to-apical direction were determined by analysis of the apical solution after basal incubation (1 h) of 3,5-diCQA (500 μM) with differentiated monolayers (n = 6) and ABC transporter inhibitor (apical side) at 37°C and 5% CO2. Papp is expressed as ×10−6 ± S.D.
Discussion
For the first time, the absorption characteristics of a series of dietary phenolic acids were investigated using a theoretical permeability model. Lipophilicity is considered as an important determinant for absorption (van de Waterbeemd, 2009) and is commonly measured as the partition coefficient (log P) in an 1-octanol/aqueous system (van de Waterbeemd and Gifford, 2003). The relevance of log P to membrane permeability has been attributed to the ability of hydrated 1-octanol to simulate the structural homology and hydrogen bonding capacity of more complex lipid bilayers (Artursson et al., 2001). Furthermore, the distribution coefficient (log D) can be considered to provide a more meaningful description of partition processes in the body for ionizable compounds (van de Waterbeemd and Gifford, 2003) and encodes several physicochemical properties: hydrogen bonding capacity and influence of ionization on solubility and molecular mass (van de Waterbeemd, 1998; Liu et al., 2011). A good predictive relationship between Papp and log D, with a molecular mass as a third descriptor, has been observed in the Caco-2 model (Camenisch et al., 1998). Thus, the previously reported descriptive absorption model (eq. 5) based on a bilinear relationship between Papp and log D was thought to be suitable for further investigation of absorption across Caco-2 cell monolayers.
Statistical modeling performed on the training set revealed a quadratic relationship between the model parameters and molecular mass. When these novel expanded parameters were integrated into the new TTP model (eq. 6), a more useful relationship between Papp and log D could be obtained compared with the descriptive absorption model (eq. 5), which was only fitted by spline smoothing to create a model limited to the estimation of passive diffusion (Camenisch et al., 1998). An excellent fit was observed between the TTP model and training set data (R2 = 0.93). A bilinear relationship between permeability and lipophilicity was described with molecular mass dependence due to membrane partitioning and diffusion effects (Xiang and Anderson, 1994). The transport plateau often observed with very lipophilic compounds (high log D) is due to rate-limiting diffusion of lipophilic compounds through the stagnant aqueous boundary layer close to the membrane surface (Stehle and Higuchi, 1972).
A molecular mass dependence for permeability was observed in the TTP model (eq. 6) as indicated by the positioning of the predicted permeation curves (Fig. 4), which show a reduction in permeability as molecular mass increases as was reported previously (Camenisch et al., 1998). The importance of molecular mass in membrane permeability can be attributed to a role in lipophilicity as described by eq. 7 (van de Waterbeemd, 1998):
where log D is the distribution coefficient, a is a regression constant, V is the molar volume (molecular size component), and Λ is a polarity term (related to hydrogen bonding capacity). Thus, the affinity of a compound for a lipophilic environment can be considered as a product of its molecular size and hydrogen bonding capacity. Furthermore, molecular size was shown to be inversely related to the diffusion coefficient D as revealed by the Einstein-Stokes equation. Thus, molecular size, represented by molecular mass in this study, may be considered a governing factor for transcellular diffusion in biological membranes (Xiang and Anderson, 1994).
Using the TTP model, good estimations of the transcellular permeability of nine structurally diverse compounds in the validation set were obtained with a Pearson correlation coefficient of 0.97, significant at the 1% level. The scope of the TTP model to estimate transport mechanisms of absorption was evaluated using commonly used markers of paracellular, active influx, and active efflux processes in the Caco-2 model. The results suggested that deviation from the model may be used to reveal interference by nontranscellular transport processes; however, additional transport studies are required to indicate which mechanisms have a role. Divergences can be rationalized by considering the design of the TTP model, which predicts only the transcellular fraction of absorption; thus, any transport via the paracellular route would be observed as an additional flux. Likewise, any active transport would enhance (influx) or hinder (efflux) permeability and result in a positive or negative deviation from the theoretical value, respectively. To distinguish between paracellular and active processes, we recommend that an additional Caco-2 mechanistic study be performed. Transport of the test compound under conditions of variable TEER has become a common method to elucidate the role of tight junctions in membrane permeation (Konishi and Kobayashi, 2004). Compounds transported primarily via the paracellular route will be attenuated at high TEER if the pore size was sufficiently restrictive. In contrast, transport of the test compound under chilled conditions (on ice) would retard active uptake, thus revealing the involvement of active processes.
Of interest, the model overestimated the permeability of l-dopa. The poor uptake may be explained by differential expression of the amino acid transporter (Sun et al., 2002) in Caco-2 cells or the involvement of active efflux as suggested by others (Fraga and Soares-da-Silva, 2006). In reference to human bioavailability, comparative in vivo/Caco-2 in vitro permeability relationships have been previously characterized (Sun et al., 2002). It should be noted that a good correlation exists between Caco-2 permeability and human intestinal uptake. However, the predicted permeability for carrier-mediated or metabolized compounds by the Caco-2 model may be expected to differ from in vivo results because of significant differences in gene expression of specific transporters and metabolic enzymes (Sun et al., 2002).
The permeability of the phenolic acids has never been investigated using a theoretical permeability model before, and the results of this study represent valuable information that can streamline future polyphenol bioavailability investigations. Using the TTP model, we have calculated the theoretical transcellular permeability of six phenolic acids on the basis of their molecular mass and our experimentally determined log D values. Mechanistic studies were performed on Caco-2 monolayers to evaluate the accuracy of the TTP model when the contribution of specific pathways in phenolic acid absorption is measured (Table 4) and are discussed below.
In the current study, the efflux ratio of ferulic acid was equivalent to 1, suggesting that transport was passive, and permeation was independent of tight junction integrity; thus, we concluded that the model correctly identified the absorption as passive transcellular transfer, which was in agreement with previous investigations (Poquet et al., 2008). We infer that the transport of 3,4-dimethoxycinnamic acid was also passive transcellular, but no mechanistic data were available at this time. The transport of 5-O-CQA and 5-O-FQA was found to be linear with respect to time and concentration, bidirectional permeation was equivalent to 1, and transport was significantly increased (P < 0.05) at low TEER. Thus, we concluded that the transport of these compounds was passive paracellular, as correctly predicted by the TTP model. The Papp value for 3,5-diCQA was less than predicted, as concluded by a residual error less than −0.3 log unit, suggesting that transport was impeded by active efflux or cell metabolism. Deesterification was not thought to be a major factor in the absorption characteristics of 3,5-diCQA as concluded by the observation that 3,5-diCQA was a poor substrate for chlorogenate esterase (Guy et al., 2009). The relative resistance of the diCQAs to hydrolysis and negligible metabolism was apparent in the current study as shown by an excellent mass balance recovery of 98 ± 3%. Furthermore, mechanistic studies revealed that permeation was enhanced in the B→A direction (efflux ratio of 5.4); thus, the involvement of an active efflux transporter was confirmed. Further efflux inhibition studies have implicated the ABC transporter family as being important to the secretion of 3,5-diCQA in the serosal-to-luminal direction. This compound was the most lipophilic HCA investigated, and it is likely to partition readily in to the lipid bilayer and diffuse via the lipid envelope, suggesting that the partitioned compound would be ideally placed for rapid export to the lumen by P-gp, BCRP, and MRP2, which are colocalized to the apical membrane of intestinal epithelial cells and have notable affinity for lipophilic substrates (Matsson et al., 2009).
The permeability of caffeic acid was less than predicted by the TTP model, which indicated potential interference by either efflux or metabolic processes. We have previously reported that the Caco-2 metabolism of caffeic acid accounts for approximately 0.1% of the amount absorbed (Farrell et al., 2011); thus, extensive metabolism was ruled out as a reason for the reduced permeation. Investigation of the bidirectional transport of caffeic acid revealed that transport was passive as concluded by the efflux ratio of 1.2. This discrepancy may be attributed to small losses due to membrane binding because the mean mass balance for the transport study was 84%.
In conclusion, a refined theoretical model has been used to successfully predict the transcellular permeability of an external validation set and a good estimation of the major transport pathways was observed. The usefulness of theoretical models to predict Caco-2 permeability could be a benefit to in vitro absorption studies and may be of interest to human bioavailability studies because permeability coefficients for passive transcellular transport in Caco-2 cells are good indicators of oral absorption (Sun et al., 2002; Press and Di Grandi, 2008).
Authorship Contributions
Participated in research design: Farrell, Poquet, Barber, and Williamson.
Conducted experiments: Farrell and Barber.
Contributed new reagents or analytic tools: Barber and Dew.
Performed data analysis: Farrell and Barber.
Wrote or contributed to the writing of the manuscript: Farrell, Poquet, Barber, and Williamson.
Acknowledgments
We thank D. Barron for his generous gift of the feruloylquinic acid standard.
Footnotes
This research was supported by Nestlé Research Center, Lausanne, Switzerland.
Article, publication date, and citation information can be found at http://dmd.aspetjournals.org.
↵
The online version of this article (available at http://dmd.aspetjournals.org) contains supplemental material.
-
ABBREVIATIONS:
- HCA
- hydroxycinnamic acid
- 5-O-CQA
- 5-O-caffeoylquinic acid
- 3,5-diCQA
- 3,5-di-O-caffeoylquinic acid
- 5-O-FQA
- 5-O-feruloylquinic acid
- TTP
- theoretical transcellular permeability
- HBSS
- Hanks' balanced salt solution
- DMSO
- dimethyl sulfoxide
- TEER
- transepithelial electrical resistance
- A→B
- apical-to-basal
- B→A
- basal-to-apical
- HPLC
- high-performance liquid chromatography
- DAD
- diode array detector
- ABC
- ATP-binding cassette
- BCRP
- breast cancer resistance protein
- P-gp
- P glycoprotein
- MRP
- multidrug resistance-associated protein
- MK571
- (E)-3-[[[3-[2-(7-chloro-2-quinolinyl)ethenyl]phenyl]-[[3-dimethylamino)-3-oxopropyl]thio]methyl]thio]-propanoic acid.
- Received July 7, 2011.
- Accepted November 17, 2011.
- Copyright © 2012 by The American Society for Pharmacology and Experimental Therapeutics