Human P-glycoprotein (P-gp, MDR1) is known to be a determinant of drug absorption, distribution, and excretion of a number of clinically important drugs (Ambudkar et al., 1999; Fromm, 2000). P-gp is widely expressed in major organs, and, more specifically, P-gp is highly expressed in the capillaries of the blood brain barrier (BBB) and poses a barrier to brain penetration of its substrates (Schinkel, 1999). Given that P-gp efflux liability can be a major hurdle for CNS therapeutic drugs to cross the BBB and reach the target, the interactions of CNS compounds with P-gp may lead to the lack of CNS activity as a result of the decreased brain penetration. Thus, the prediction and understanding of the relevance of P-gp-mediated efflux transport have become important activities in the discovery and development of CNS drugs. In attempts to predict the effects of P-gp in vivo, a variety of in vitro P-gp assays have been developed to classify compounds as P-gp substrates. For instance, transwell-based assays using polarized cell lines such as the Madin-Darby canine kidney (MDCK) cell line. The MDCK cell line can be stably transfected with human MDR1 or mouse Mdr1a (MDR1-MDCK or Mdr1a-MDCK, respectively). Comparison of the efflux ratios between MDR1-MDCK and MDCK transwell assays can provide a measure of the specific human P-gp-mediated efflux activity (Polli et al., 2001). Another widely used P-gp in vitro assay is the P-gp ATPase assay for assessing drugs with P-gp interactions as substrates (Sarkadi et al., 1992; Ramachandra et al., 1998). The principal reagent of the ATPase assay is a membrane preparation from insect cells highly expressing human P-gp, and functional human P-gp will transport P-gp substrates across the membrane resulting in the release of inorganic phosphates (Scarborough, 1995; Litman et al., 1997). Finally, the in vitro calcein AM P-gp inhibition assay can be used to detect compounds that inhibit P-gp-mediated efflux of the fluorescent P-gp substrate, calcein. This assay can differentiate P-gp inhibitors from noninhibitors by measuring the fluorescence of calcein (Liminga et al., 1994; Tiberghien and Loor, 1996). However, it is important to note that the calcein AM assay is a P-gp inhibition assay and not a substrate assay and that P-gp substrates do not necessarily correlate with P-gp inhibitors.

In addition to P-gp in vitro assays, animal models have been used to assess the impact of P-gp on substrate pharmacokinetics in humans. Comparison of the brain/plasma (B/P) exposure ratios in Mdr1a/Mdr1b KO mice versus WT mice has become a standard experimental approach to determine whether P-gp-mediated efflux poses a potential threat to the activity of CNS agents in vivo (Schinkel et al., 1996). Such in vivo data can be used in concert with data obtained using in vitro models to assess the potential liability of P-gp-mediated efflux on brain exposure of compounds in humans.

Despite all efforts to develop in vitro assays to predict P-gp effects in vivo, clear guidance and rationale around the selection of models to aid in CNS drug discovery and development have been limited. Toward the pursuit of selecting appropriate in vitro P-gp assays to support drug discovery efforts, we have identified successful CNS drugs to provide a relevant context for the interpretation of preclinical P-gp models. As such, the P-gp interactions of a set of CNS and non-CNS drugs were evaluated in a battery of in vitro P-gp screening assays that were readily available and easy to perform. The data from the in vitro studies were compared with brain exposure determined in vivo using P-gp KO and WT mice as described by Doran et al. (2005) to assess the concordance of the in vitro data to the in vivo data. The objectives of drug selection were to identify a representative and diverse sampling of the most commonly used CNS therapeutic agents. Thirty-one CNS drugs and one active metabolite of a CNS drug were selected according to sales, number of prescriptions, structural diversity, availability, pharmacology, and analytical feasibility (Doran et al., 2005). The CNS drugs selected represented a number of different therapeutic indications including antidepressants, sedatives, anxiolytics, tranquilizers, anticonvulsants, and others, of which 23 were bases, 7 were neutral, and 2 were acids. The detailed rationale for compound selection for the study can be found in Doran et al. (2005). Seven non-CNS drugs were selected as controls for their consideration as P-gp modulators, although not necessarily as substrates for P-gp. Additionally, because the key concern in using these mouse models as tools to predict relevance to humans is the potential species differences that may be seen between human MDR1 and mouse Mdr1a/1b, we have performed a correlation analysis of efflux results determined in transwell assays using MDR1-MDCK and Mdr1a-MDCK cells for 3300 Pfizer compounds.

In addition to the specific interaction with P-gp, it is known that the CNS penetration of a compound also depends on its permeability. Although much effort has been devoted to understanding the complex relationship between the two parameters, the role that each plays in CNS penetration is not yet well established. The permeability of the 32 compounds was determined using two in vitro permeability assays: parallel artificial membrane permeation assay (PAMPA), and apical-to-basolateral (A>B) apparent permeability (P_app) in MDCK cells. PAMPA is a model specifically designed to measure passive membrane permeability, which is energy-independent (Kansy et al., 1998; Kerns, 2001; Sugano et al., 2001; Wohnsland and Faller, 2001; Avdeef and Testa, 2002). Transport determined using MDCK cell monolayers approximates passive biological permeability, as the MDCK cells typically show low endogenous transporter activity.

The main goal of these studies was to evaluate the in vitro P-gp and permeability assays to find appropriate strategies for profiling new chemical entities for P-gp efflux liability relating to the distribution of drugs in the CNS. Additional insight is given regarding the strategies for effective use of the in vitro tools available to assess P-gp efflux and the utility of various permeability assays. Finally, we report the important findings regarding the correlation of human MDR1-MDCK and mouse Mdr1a-MDCK assay results.

Materials and Methods

Chemicals and Drug Selections. The 31 CNS drugs, 1 active metabolite of a marketed CNS drug, and 7 non-CNS drugs that were tested in the in vitro P-gp transporter assays and permeability assays were part of the 34 CNS compounds and 8 non-CNS drugs used in the previous publication of Doran et al. (2005). The suppliers of chemicals were the same as those described in Doran et al. (2005).

Materials. The WT MDCK and human MDR1-transfected MDCK cell line were acquired externally (Piet Borst, Netherlands Cancer Institute, Amsterdam, The Netherlands) and the murine Mdr1a-transfected MDCK cell line was constructed internally (Pfizer Inc., Groton, CT). PAMPA lipid used for PAMPA studies (4% dioleoylphosphatidylcholine with 2% steric acid in Dodecane) was prepared by Avanti Polar Lipids (Alabaster, AL). A high-throughput screening 96-multiwell insert system, cell culture plates with a polyethylene terephthalate membrane, 0.0625 cm² growth area, and 1 μm pore size, a 96-well angled bottom collection plate without a lid (nonsterile in 23 packs), T-flasks (175 cm²), and velocity V11 peelable seals were purchased from Falcon/BD (Bedford, MA). PAMPA permeability insert plates with polyvinylidene difluoride membranes (0.45 μm pore size) and associated labware used for PAMPA assay were obtained from Millipore (Danvers, MA). Cell culture reagents, transport buffer used for PAMPA and transwell assays (Hanks' balanced salt solution with 10 mM HEPES and 25 mM d-glucose, 1.25 mM CaCl₂, and 0.5 mM MgCl₂) and cell culture reagents were purchased from Invitrogen (Carlsbad, CA).

Sample Handling. Each compound was weighed out in duplicate on 96-well master plates and solubilized in DMSO. Daughter plates were made for each screen, and compounds were run according to the typical assay protocols as described in the following text.

MDCK Cell Culture. The following procedures were used to culture MDCK and the transfected MDCK cells. The cells were grown in minimum essential medium α with supplements at 37°C, 5% CO₂, and 95% humidity. The cells were harvested using trypsin and plated at a density of 2 × 10⁶ cells/cm² in Falcon/BD 96-well insert plates with a 1 μm pore polyethylene terephthalate filter. Seeded inserts were then placed into prefilled Falcon/BD feeder trays containing 37 ml of complete growth medium. The plates were incubated at 37°C with 95% humidity and 5% CO₂ for 4 days and subsequently used for assays.

Transwell Assay Procedures. The following procedures were used for all studies performed to determine compound P_app values using MDCK, MDR1-MDCK, and Md1a-MDCK cells. All transwell assays were performed with transport buffer (Hanks' balanced salt solution with 10 mM HEPES and 25 mM d-glucose, 1.25 mM CaCl₂, and 0.5 mM MgCl₂) at pH 7.4 (buffered with 10 mM HEPES). Assays were performed with 2 μM compound (30 mM DMSO stock solution diluted in transport buffer) in duplicate. Transport studies were performed by adding compounds in transport buffer to donor wells and measuring appearance in receiver wells after a 5-h incubation at 37°C. For A>B transport, the donor is the A compartment and the receiver is the B compartment. For B>A transport, the donor is the B compartment and the receiver is the A compartment. After the 5-h incubation period, samples were collected from the receiver compartment and from the original compound in transport buffer solution and retained for analysis. For A>B transport studies, receiver compartments were additionally washed with acetonitrile, and this wash was added to the receiver sample to recover potentially nonspecifically bound compound.

PAMPA Assay Procedures. PAMPA studies were conducted using the transport buffer at pH 6.5 (buffered with 10 mM MES). Assays were performed with 10 μM compound (1 mM DMSO stock solution diluted in pH 6.5 transport buffer) in duplicate. The PAMPA experiment was performed by adding 5 μl of PAMPA lipid to the PAMPA permeability insert plate to make the PAMPA membrane. Compound was added to the donor compartment, and the PAMPA insert and receiver plate were incubated at 30°C for 2 h. Samples were taken from the donor (initial and final time) and receiver (final time) and retained for analysis.

Transwell and PAMPA Assay Sample Analysis. Samples were analyzed using liquid chromatography with tandem mass spectrometry. Briefly, 25 μl of sample was injected via a Gilson 215 (Gilson, Middleton, WI) or a custom designed dual-arm autosampler into a high-performance liquid chromatography system consisting of a pair of OPTI-LYNX 1 × 15-mm cartridge style columns custom packed with a Showa-Denko ODP polymeric material with 20-μm particle size. The system used a 10-port two-position Valco switching valve (Valco Instrument Co. Inc., Houston, TX). The aqueous mobile phase was 98% 2 mM ammonium acetate/2% 50:50 (acetonitrile/methanol). The organic mobile phase was 10% 2 mM ammonium acetate/90% 50:50 (acetonitrile/methanol). Flow rates were maintained at 2 ml/min using Jasco pumps. Detection and analysis was performed on a SCIEX API 3000 triple quadrupole mass spectrometer (Applied Biosystems, Foster City, CA), operated in either positive or negative ion mode. The effluent from the high-performance liquid chromatography column was directly introduced into the TurboIonSpray source. Analytes and internal standard responses were measured using multiple reaction monitoring (MRM). The MRM for the internal standard was 687.3/319.7 for positive ion mode or 685.3/366.1 for negative ion mode, and MRMs for the screen compounds were as determined previously.

Transwell Data Analysis. The P_app was calculated using eq. 1: where Area is the surface area of the cell monolayer (0.0625 cm²), C_D(0) is the initial concentration of compound applied to the donor chamber, t is time, M_r is the mass of compound appearing in the receiver compartment as a function of time, and dM_r/dt is the flux of the compound across the cell monolayer. The efflux ratio (ER) was calculated using eq. 2: where A>B and B>A denote the transport direction in which P_app was determined. The ratio of ratios (RR) was calculated using eq. 3: where ER_transfected is the ER determined in the transfected MDCK cells (MDR1-MDCK or Mdr1a-MDCK) and ER_wt is the ER determined in WT MDCK cells. RR_MDR1 and RR_Mdr1a denote RR determined for ER in MDR1-MDCK transfected cells and Mdr1a-MDCK transfected cells, respectively, compared with ER in WT MDCK.

PAMPA Data Analysis.P_app and membrane retention (Rf) were determined using eqs. 4 and 5, respectively: where V_A (ml) and V_D (ml) are the volumes of the acceptor and donor compartments, respectively, τ is the time to steady state (typically assumed to be ∼1800 s for convention), t is the time scale of the experiment (7200 s) and Area is the area of the filter (in square centimeters; 0.24 cm² for the insert plate listed above). Mass is M, for acceptor (A) and donor (D).

P-gp ATPase Assay. The ATPase assay uses membrane preparations from insect SF9 cells expressing only human P-gp. In the ATPase assay, two ATP molecules are hydrolyzed with the transport of each substrate molecule by P-gp, resulting in the release of two inorganic phosphates into the reaction buffer. The level of inorganic phosphate was measured via a secondary colorimetric reaction in which the phosphate reacts with ammonium molybdate to produce a blue color (Drueckes et al., 1995). The reactions were performed using a 96-well format in reaction buffer [50 mM MES, 2 mM EGTA, 2 mM dithiothreitol, 50 mM potassium chloride, and 5 mM sodium azide (pH adjusted to 6.8 with Tris)]. The reaction mixture included test drugs in 5% DMSO vehicle, 8 μg of human P-gp membranes (BD Gentest, Woburn, MA) and 1000 μM ATP magnesium salt. The incubation volume was 0.06 ml. Baseline activity was determined by incubations of P-gp with inhibitor (100 μM sodium orthovanadate). The reactions were incubated at 37°C for 20 min and stopped by the addition of 0.03 ml of 10% SDS. Two volumes (0.18 ml) of detection reagent (7 mM ammonium molybdate, 3 mM zinc acetate, and 16% ascorbic acid, pH 5.0) were dispensed to each well. The color reaction proceeded for 20 min in a 37°C incubator. Ammonium molybdate reacted with inorganic phosphate to form a colorimetric complex. The plate was read immediately for endpoint absorbance at 850 nm using a Safire plate reader (Tecan, Durham, NC). A potassium phosphate standard curve was used to extrapolate nanomoles of phosphate from the absorbance readings. ATPase activities were calculated in terms of nanomoles of phosphate per milligram of total vesicle protein per minute of incubation time. The substrate saturation model, Michaelis-Menten kinetics, was used to fit the dose-response data by nonlinear regression to obtain estimated apparent affinity values (K_m) for ATPase enzyme.

Eight concentrations of each drug were evaluated in triplicate in two independent assays. Initially, test compounds were tested at concentrations ranging between 1.25 and 150 μM, but some compounds were retested using higher concentrations to saturate the ATPase enzyme. Estimated K_m values were determined for those compounds exhibiting ATPase activity. Kinetics were not determined if no significant ATPase activity was observed.

Calcein AM P-gp Inhibition Assay. Parental MDCK and MDR1-transfected MDCK cells were grown as described above. For the studies, each cell line (50,000 cells/well) was plated into Costar 3904 black 96-well plates (PerkinElmer, Waltham, MA) with 100 μl of medium supplemented with 1% fetal bovine serum and allowed to become confluent overnight. Test compounds were added to monolayers in 100 μl of culture medium containing 1% DMSO as solvent. Plates were incubated at 37°C for 30 min. Calcein AM (Invitrogen) was added in 100 μl of phosphate-buffered saline (PBS) to yield a final concentration of 2.5 μM. Plates were incubated for another 30 min. Cells were then washed three times with ice-cold PBS. PBS was added to the cells, and the cells were read with a Victor fluorometer (PerkinElmer) at excitation and emission wavelengths of 485 and 535 nm, respectively. P-glycoprotein inhibition was calculated using the following equation: where “The amount of efflux” was defined as the fluorescence from MDCK cells subtracted by that from the MDR1-MDCK cells, and “The amount of inhibition” was (The amount of efflux)_untreated – (The amount of efflux)_treated. The IC₅₀ values were determined by fitting the percentage of inhibition-concentration data into the Hill equation using the ordinary least square method of the LabStats Excel Add-in (Pfizer Inc.). When fitting the Hill equation, cytotoxic data points for some compounds at the highest test concentration were removed. The cytotoxic data points were apparent because of the loss of calcein fluorescence at the highest test concentration of some compounds. The Hill slope, the minimum and the maximum % inhibition were not fixed, and the % inhibition data were not weighted. The IC₅₀ was based on the concentration at which the apparent 50% inhibition was achieved (i.e., apparent IC₅₀). Vinblastine, a known inhibitor of P-gp, was used as the positive control.

Results

Transwell Assay Results. Transwell assays are used to measure drug movement through cell monolayers, and comparison of the ER determined in MDR1-MDCK versus MDCK cells (RR_MDR1) is a commonly used method to enable the assessment of whether a compound is a specific human P-gp substrate. Based on retrospective statistical analysis of interweek variability of Pfizer transwell assay control data determined over 1 year (n > 280), a cutoff for RR_MDR1 of 1.7 can be used to classify compounds as potential P-gp substrates (Pfizer data on file). A similar analysis was performed for transwell studies performed with Mdr1a-MDCK versus MDCK cells and yielded a cutoff for RR_Mdr1a of 1.5. At Pfizer, the use of this cutoff has been demonstrated to minimize the incidence of false-positive and false-negative results in this assay, as assessed by standard P-gp nonsubstrate, poor substrate, and substrate controls (triprolidine, prazosin, and quinidine, respectively).

Thirty-one CNS drugs and one active metabolite were identified for evaluation in transwell assays using human MDR1-MDCK, mouse Mdr1a-MDCK, and MDCK cells, and the data are presented in Table 1. Using 1.7 as a cutoff for the RR_MDR1 and 1.5 as a cutoff for the RR_Mdr1a, 30 of 32 compounds did not show positive efflux in both assays. Risperidone and its active metabolite, 9-hydroxyrisperidone, had RR_MDR1 2.0 and 3.3, and RR_Mdr1a 1.8 and 4.7, respectively. As such, these two compounds were identified as positive substrates for P-gp efflux for both human MDR1 and mouse Mdr1a using these assays.

Seven non-CNS drugs were also measured in transwell assays using human MDR1-MDCK, mouse Mdr1a-MDCK, and MDCK cells (Table 2). For the six reported P-gp substrates of the seven non-CNS compounds (loperamide, prazosin, prednisone, quinidine, ritonavir, and verapamil), RR_MDR1 values were all >1.7 and RR_Mdr1a values were all >1.5, except for verapamil with an RR_Mdr1a value of 1.4. RR_MDR1 and RR_Mdr1a values for loratadine, which has been reported as a weak P-gp substrate in literature (Polli et al., 2001), were 0.5 and 0.6, respectively. Thus, loratadine was not identified as a P-gp substrate in both human and mouse P-gp assays.

P-gp ATPase Assay Results. The same set of compounds was examined for ATPase activity. A summary of estimated K_m values of the CNS and non-CNS compounds is reported in Tables 1 and 2, respectively. Kinetics were not determined if no significant ATPase activity was observed. Affinity of P-gp substrates are ranked as follows: high affinity (K_m < 10 μM), moderate affinity (10 μM ≤ K_m < 50 μM), and low affinity (K_m ≥ 50 μM). Of the 32 CNS compounds, one compound, sertraline, had high affinity for human P-gp with a K_m of 9.4 μM, and four compounds (chlorpromazine, midazolam, paroxetine, and risperidone) had moderate P-gp affinity with K_m values of 41.0, 44.4, 26.2, and 36.1 μM, respectively. Of the seven non-CNS compounds, loratadine and verapamil had high affinity for P-gp, loperamide, quinidine, and ritonavir had moderate affinity for P-gp, and the remaining two compounds were identified as low affinity or negative P-gp substrates.

Calcein AM P-gp Inhibition Assay. The calcein AM fluorescent assay can be used as a P-gp inhibition assay, and the inhibitors were ranked as the following: potent inhibitors (IC₅₀ < 10 μM), moderate inhibitors (10 μM ≤ IC₅₀ < 50 μM), and weak inhibitors (IC₅₀ ≥ 50 μM). Of the 32 CNS compounds the calcein AM assay indicated that haloperidol was a P-gp inhibitor with high potency (IC₅₀ = 7.2 μM), buspirone, chlorpromazine, cyclobenzaprine, fluoxetine, hydroxyzine, paroxetine, and sertraline were identified as P-gp inhibitors with moderate potency, and the remaining compounds, including risperidone and 9-hydroxyrisperidone, were weak P-gp inhibitors or did not exhibit P-gp inhibition (Table 1). Of the seven non-CNS compounds, loperamide and loratadine were P-gp inhibitors with high potency, quinidine and verapamil were P-gp inhibitors with moderate potency, and the remaining compounds were identified as weak P-gp inhibitors or did not exhibit P-gp inhibition (Table 2).

Permeability Measurements: PAMPA and MDCK A>BP_appDeterminations. PAMPA is a high-throughput passive permeability assay used to assess membrane permeability characteristics for compounds. On the basis of historical PAMPA data at Pfizer, compounds with PAMPA P_app > 5 × 10^–6 cm/s showed good membrane permeability characteristics in biological models such as Caco-2 and/or MDCK, whereas those with PAMPA P_app < 5 × 10^–6 cm/s may show membrane permeability liability. It should be noted that the cutoff of 5 × 10^–6 cm/s is a general value selected from empirical analysis of PAMPA versus biology passive permeability data. Furthermore, we have observed that whereas 5 × 10^–6 cm/s is a general rule of thumb for predicting good membrane permeability, PAMPA P_app results can often be chemotype-dependent, thus producing specific membrane permeability cutoffs for different chemotypes will be ideal. The membrane permeability of compounds was also tested in the WT MDCK cells as an alternate method to gauge compound passive P_app using a biological model. As mentioned previously, endogenous expression of transporters in MDCK cells is low and therefore transmonolayer permeability is likely to be primarily mediated by passive processes. Indeed, for the compounds tested in the MDCK assay, the ER of unity was obtained for the majority and this lack of polarity confirms that transport across MDCK cells was passive. The magnitude of P_app (× 10^–6 cm/s) in the MDCK assay can be qualitatively grouped by the following: high permeability (P_app > 10), moderate permeability (1 ≤ P_app ≤ 10), and low permeability (P_app < 1).

The permeability of the 32 CNS compounds was tested using PAMPA and MDCK assays (Table 3). PAMPA data suggested that fluoxetine and sulpiride, had very low permeability, and two compounds, morphine and sertraline, could be ranked as low-permeability compounds as well, as judged by the empirical cutoff, whereas the remainder showed good permeability. In the MDCK assay, only one compound, sulpiride, had low permeability, and other CNS compounds had moderate to high permeability. Passive permeability data for the seven non-CNS compounds is shown in Table 4. Data from the two permeability assays indicated that all non-CNS compounds had moderate to high permeability.

Discussion

In this study we assessed P-gp interactions with CNS and non-CNS drugs using different in vitro P-gp assay formats and passive permeability determination for these compounds. Our goal was to assess the ability of these in vitro studies in profiling P-gp efflux liability as it is associated with brain distribution. Thirty-one marketed CNS drugs, one active metabolite of a marketed CNS drug, plus seven non-CNS compounds, were tested in transwell assays using MDR1-MDCK, Mdr1a-MDCK, and MDCK cells, P-gp ATPase assay, and calcein AM P-gp inhibition assays, and data were compared with those generated in vivo from Mdr1a/Mdr1b (–/–/–/–) and WT mice (Doran et al., 2005). These studies were preformed provide guidance for the interpretation of preclinical in vitro and in vivo P-gp models.

Each of the three in vitro P-gp assays have strengths and weaknesses as drug discovery screens to identify P-gp substrates and to assess the relevance of P-gp efflux liability in vivo. The advantage of the transwell assay is that it is cell-based, mimicking in vivo conditions and that it measures transport representative of processes underlying disposition. The MDCK cell line has the essential property of spontaneously forming polarized monolayers on solid supports with tight junctions, such that the efflux transporters can insert into the apical membrane to impart polarity to the transport of substrates (Cho et al., 1989). Polarity can be readily assessed via comparison of transport performed in B>A and A>B directions by determining the ER. In our transwell studies we used three related cell lines (the parent MDCK, the human MDR1-transfected MDCK, and the murine Mdr1a-transfected MDCK). Both transfectants produce high level of their respective transfected gene product and are presumed to express all other MDCK-specific canine renal transporters as well, such as dog P-gp, but at a much lower level than the transfected gene product. Given the homogeneous nature of the screening panel, an MDCK background subtraction was intended to correct for the influence of transporters other than P-gp and underlies our rationale for performing studies in the WT MDCK cells in addition to the transporter-transfected MDCK cells. Thus, by comparing the ER of compounds in the MDR1- or Mdr1a-transfected MDCK cells with that of the parent MDCK cells, the respective RR values can be used to positively identify a compound as a specific human or mouse P-gp substrate.

The P-gp ATPase assay estimates apparent affinity of the test compounds by measuring the level of inorganic phosphate and can generate the apparent affinity (K_m) of a compound for P-gp. The ATPase assay uses P-gp membrane preparations from insect cells, which could function differently from mammalian cells. The third P-gp in vitro assay, the calcein AM P-gp inhibition assay, is able to identify P-gp inhibitors using a fluorescent probe substrate, but there is no necessary correlation between P-gp substrates and P-gp inhibitors, presumably due to multiple binding sites in P-gp. However, in comparison with the transwell assay, the P-gp ATPase assay and the calcein AM P-gp inhibition assay are more cost-effective and less labor-intensive and do not require a liquid chromatography with tandem mass spectrometry endpoint.

Data from transwell assays can be used to categorize compounds into three classes. The quantitative cutoffs have been established on the basis of retrospective analysis of historical performance of efflux substrates in the in vitro assay and comparison with B/P ratios from Mdr1a/1b KO and WT mice. The first class is composed of nontransported compounds, i.e., the compound has no significant efflux (ER < 2.5) in either MDCK or MDR1-MDCK cells. The second class of compounds has RR_MDR1 > 1.7, which is the hallmark for a positive human P-gp substrate. The third class of compounds has RR_MDR1 < 1.7, but there are individual effluxes (ER > 2.5) in MDR1-MDCK and MDCK cells, which are interpreted as having transporter activity but are not necessarily human MDR1-specific. The same criteria apply to results of studies performed with mouse Mdr1a-MDCK versus MDCK cells, with the exception of 1.5 instead of 1.7 as a cutoff. Of the 32 CNS drugs, the transwell study data indicated that 94% of compounds had mean RR_MDR1 and RR_Mdr1a values ≤ 1.7 and 1.5, respectively, and these were classified as non-P-gp substrates. Only one CNS drug, risperidone (RR_MDR1 of 2.0), and its metabolite, 9-hydroxyrisperidone (RR_MDR1 of 3.3), were identified as both human and mouse P-gp substrates using the RR_MDR1 and RR_Mdr1a data interpretation criteria. The P-gp ATPase assay identified risperidone and 9-hydroxyrisperidone as a moderate- and a low-affinity P-gp substrate with K_m values of 36.1 and 177 μM, respectively. Consistent with data from the in vitro transwell studies, risperidone and 9-hydroxyrisperidone showed marked differences in B/P AUC ratios between P-gp KO and WT mice of 10 and 17, respectively (Doran et al., 2005). In addition, the brain concentration for metoclopramide was also dramatically increased in the P-gp KO mice (6.6-fold), and the B/P values between P-gp KO and WT mice for the three compounds were significantly >4. A cutoff of 4 for B/P AUC ratios between P-gp KO and WT mice has been set to assess the P-gp interaction in vivo, after conducting a multivariate analysis of in vivo data for a population (n = 23) of Pfizer compounds with in vitro MDR1-MDCK data strongly indicative of nonsubstrates. The RR_MDR1 and RR_Mdr1a values for metoclopramide generated from transwell assays were both 1.4, and the compound was identified as a non-P-gp substrate in vitro. Possibilities for this discrepancy could include interspecies differences in the intrinsic activity of P-gp between human and mouse or different expression levels at the BBB between the species. There are reports in the literature indicating that a species difference in P-gp substrates between human and mouse exists (Yamazaki et al., 2001). However, in the present studies, transwell data from MDR1-MDCK cells correlated well with those from Mdr1a-MDCK cells. In addition, we have previously tested 3300 Pfizer compounds in MDR1-MDCK and Mdr1a-MDCK transwell assays, and there was high correlation (R² = 0.92) for efflux ratios between human MDR1-MDCK and mouse Mdr1a-MDCK cells (Fig. 1). Thus, species differences for P-gp between human and mouse are relatively rare, although they could be observed for a particular structural class of compounds. Another consideration is that there are two Mdr genes, Mdr1a and Mdr1b, in the mouse, although Mdr1a has been reported to be the major Mdr gene encoding mouse P-gp. In vitro, the compounds were only tested in the Mdr1a-MDCK assay, whereas in vivo studies were done in Mdr1a/Mdr1b double knockout mice. This discrepancy may explain the difference between the in vitro and in vivo observations with metoclopramide. Additionally, it is important to note that the transfected cell lines possess approximately 4-fold more transporter activity relative to endogenous canine P-gp activity in the parental MDCK cells. Consequently, the in vitro data may not necessarily reflect the magnitude of the in vivo effect. The definitive answer for the in vitro versus in vivo discrepancy with metoclopramide needs to be investigated further. Overall, compared with the in vivo B/P AUC ratios in P-gp KO and WT mice for the 32 CNS compounds, the transwell assay predicted no false-positive and few false-negative results (3%) and had a 97% correlation of in vitro versus in vivo data.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

Correlation of the efflux ratio in transwell assays using mouse Mdr1a-MDCK and human MDR1-MDCK cells for 3300 Pfizer compounds. The y-axis is the mouse Mdr1a-MDCK efflux ratio, and the x-axis is the human MDR1-MDCK efflux ratio. R² = 0.92. Solid line is the linear regression line, and the dotted line is the 95% prediction interval.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

Correlation of the ratio of ratios in transwell assays using MDR1-MDCK versus MDCK cells with brain/plasma AUC ratios in Mdr1a/1b KO mice and FVB wild-type mice for the 32 CNS (○) and 7 non-CNS drugs (•). The dotted lines indicate a cutoff of 4 for brain/plasma ratios between P-gp KO and WT mice and a cutoff of 1.7 for the ratio of ratios in MDR1-MDCK versus MDCK cells.

The seven non-CNS compounds tested were identified as human P-gp substrates using data from transwell studies, except for loratadine. Loratadine was not identified as a P-gp substrate in vivo on the basis of the B/P in P-gp KO and WT mice (Doran et al., 2005), although loratadine has been reported in the literature to be a weak P-gp substrate (Polli et al., 2001). Interestingly, the ATPase assay identified loratadine as a P-gp substrate with a high affinity (K_m = 1.4 μM). This disconnect could be due to the good passive permeability of loratadine; thus, efflux in the transwell assay does not have a large impact for this compound. But this supposition certainly requires further investigation. In the work by Doran et al. (2005), the B/P AUC ratios between P-gp KO and WT mice for the seven compounds, loperamide, prazosin, prednisone, quinidine, ritonavir, verapamil, and loratadine, were 9.3, 2, 2.3, 36, 1.2, 17, and 1.9, respectively. Only loperamide, quinidine, and verapamil had B/P ratios in KO and WT mice greater than the cutoff of 4, whereas the values for loratadine as well as for prazosin, prednisone, and ritonavir were not statistically significant in vivo.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

Correlation of P-gp ATPase K_m (A) and calcein AM P-gp inhibition IC₅₀ (B) with brain/plasma AUC ratios in Mdr1a/1b KO mice and FVB wild-type mice for the 32 CNS (○) and 7 non-CNS drugs (•). The dashed lines indicate a cutoff of 4 for brain/plasma ratios between P-gp KO and WT mice and a cutoff of 100 μM for P-gp ATPase K_m and calcein AM P-gp inhibition IC₅₀ to differentiate P-gp substrates versus non-P-gp substrates and P-gp inhibitors versus non-P-gp inhibitors. Compounds with negative K_m values in the ATPase assay were graphed in (A), with K_m values of 500 μM, and compounds with IC₅₀ > 100 μM were graphed in (B), with IC₅₀ values of 200 μM.

Collectively, the studies with 32 CNS and 7 non-CNS drugs suggested that RR values generated from transwell studies using MDR1-MDCK and MDCK cells show a good correlation with B/P AUC ratios in P-gp KO versus WT mice in vivo (Fig. 2). All CNS compounds, except metoclopramide, fell in the two expected quadrants using 1.7 as a cutoff for RR in the transwell assay and 4 as a cutoff for B/P in KO and WT mice. It is expected that RR values can classify positives and negatives but may not necessarily reflect the magnitude of the in vivo effect for these compounds. From these studies we conclude that there is a high utility of this in vitro approach as a means to profile P-gp liability that may be observed in vivo, and, as such, this in vitro approach provides a useful screen to cull compounds that can then be further assayed using the more resource-intense in vivo studies.

Compared with the transwell assay, the other two P-gp assays, ATPase and calcein AM, generated more false-positive and more false-negative results for the 32 CNS compounds (Fig. 3). Because the transwell assay is a cell-based assay, the efflux ratio will reflect the resultant interplay of the permeability of compounds and the P-gp interaction, which is more reflective of the in vivo situation. It is possible that compounds, such as loratadine, are positive in the ATPase assay but are negative in the transwell assay because of their cell permeability, whereas the third P-gp in vitro assay, the calcein AM P-gp inhibition assay, was used to identify P-gp inhibitors. Given the differences among the three P-gp in vitro assays, it is not a surprise that compared with the P-gp ATPase and the calcein AM P-gp inhibition assays, the transwell assay correlated better with the in vivo P-gp KO and WT mice studies. We found that the transwell assay is the most reliable for predicting the P-gp efflux liability in vivo and that it is the method of choice for evaluating drug candidates. However, the P-gp ATPase and calcein AM assays can serve as specialized assays to understand the kinetics and inhibition of CNS compounds and can complement the transwell assay to better characterize and predict P-gp modulators in vivo.

In the development of CNS compounds, P-gp efflux liability is a key factor to assess; however, the intrinsic permeability of a compound is also an important property to consider. Regardless of the level of P-gp efflux liability, a compound that is unable to cross cell membranes will not penetrate into the brain. Consequently, low passive permeability is a risk for brain penetration. For compounds that can cross membranes, the relationship between the efflux and permeability is complex (Lentz et al., 2000), and their relationship in CNS penetration is not well understood. However, understanding the contribution of permeability versus efflux transport of a compound across the BBB is critical to the understanding and translation of P-gp interactions from in vitro to in vivo settings. We have utilized two in vitro permeability assays, PAMPA and MDCK, to assess the passive permeability of compounds. The PAMPA assay assesses the passive permeability of compounds across an artificial membrane. The MDCK cells contain low endogenous transporters and the apparent permeability is typically governed primarily by passive processes. According to their PAMPA or MDCK A>B P_app values (Table 3), all of the 32 CNS compounds were ranked as having moderate-to-high permeability, except for sulpiride, which is consistent with its high polar surface area, as reported in Doran et al. (2005). In general, MDCK A>B P_app and PAMPA data correlated well despite the differences between the two systems.

On the basis of in vitro transwell data and the in vivo B/P in P-gp K/O and WT mice for the 32 marketed CNS compounds, the majority of the CNS compounds included in this study did not exhibit P-gp efflux liability and had moderate-to-high passive permeability. We conclude that the transwell assay using MDR1-MDCK and MDCK cells is a valuable in vitro assay to evaluate human P-gp interaction with compounds targeting CNS. Additionally, we have reported a high correlation between MDR1-MDCK and Mdr1a-MDCK assays giving further confidence that the mouse model is a useful tool to predict P-gp activity in humans. We hereby recommend a multifaceted approach using in vitro P-gp and permeability assays, in concert with in vivo P-gp KO and WT mice, for the successful development of CNS drugs.