Abstract
Membrane permeability and P-glycoprotein (Pgp) can be limiting factors for blood-brain barrier penetration. The objectives of this study were to determine whether there are differences in the in vitro permeability, Pgp substrate profiles, and physicochemical properties of drugs for central nervous system (CNS) and non-CNS indications, and whether these differences are useful criteria in selecting compounds for drug development. Apparent permeability (P app) and Pgp substrate profiles for 93 CNS (n = 48) and non-CNS (n = 45) drugs were determined by monolayer efflux. Calcein-AM inhibition assays were used to supplement the efflux results. The CNS set (2 of 48, 4.2%) had a 7-fold lower incidence of passive permeability values <150 nm/s compared with the non-CNS set (13 of 45, 28.9%). The majority of drugs (72.0%, 67 of 93) were not Pgp substrates; however, 49.5% (46 of 93) were positive in the calcein-AM assay when tested at 100 μM. The CNS drug set (n = 7 of 48, 14.6%) had a 3-fold lower incidence of Pgp-mediated efflux than the non-CNS drug set (n = 19 of 45, 42.2%). Analysis of 18 physicochemical properties revealed that the CNS drug set had fewer hydrogen bond donors, fewer positive charges, greater lipophilicity, lower polar surface area, and reduced flexibility compared with the non-CNS group (p < 0.05), properties that enhance membrane permeability. This study on a large, diverse set of marketed compounds clearly demonstrates that permeability, Pgp-mediated efflux, and certain physicochemical properties are factors that differentiate CNS and non-CNS drugs. For CNS delivery, a drug should ideally have an in vitro passive permeability >150 nm/s and not be a good (B → A/A → B ratio <2.5) Pgp substrate.
The delivery of a new drug candidate to the central nervous system (CNS) can be a significant challenge during drug development. Often, the CNS distribution of a drug is poor because of exclusion at the blood-brain barrier (BBB) (Abbott and Romero, 1996; Pardridge, 1997). The BBB is composed of a single layer of endothelial cells connected by tight junctions. Brain microvascular endothelial cells lack fenestrations, have few pinocytotic vesicles, and express a variety of metabolic enzymes and membrane efflux transporters, such as P-glycoprotein (Pgp) (Rubin and Staddon, 1999; Kusuhara and Sugiyama, 2001a,b). These features make the BBB a formidable barrier that drugs must overcome to reach the brain parenchyma.
Early assessment of the ability of a drug candidate to penetrate the CNS is critical during the drug discovery selection process, especially for therapeutic indications that require delivery to a CNS site of action. Equally important is the ability to design drugs for non-CNS indications that have minimal brain penetration to avoid undesirable CNS side effects. Over the past several years, academia and industry have invested significant effort in the development and implementation of lead optimization screens, including in vitro assays and computational models to evaluate CNS penetration.
A number of in vitro BBB and membrane transport models are available to aid in the selection of compounds (Polli et al., 2000, 2001b;Garberg, 1998). These models use various cell types (primary and immortalized, brain- and non-brain-derived), cell combinations (single or cocultures), and formats (grown on plastic, filters, or in hollow fibers), each having advantages and disadvantages. The Madin Darby canine kidney (MDCK) cell is increasingly used as a substitute for more labor-intensive in vitro BBB models in passive permeability and membrane transport studies, and is our model of choice for these types of studies (Veronesi, 1996; Sawada et al., 1999a,b; Polli et al., 2000).
Along with in vitro membrane permeability models, there has been great interest in using physicochemical properties and computational modeling to predict BBB penetration. A variety of computational approaches and methods have been described (Basak et al., 1996; Fischer et al., 1998;van de Waterbeemd et al., 1998; Ajay et al., 1999; Clark, 2001), and key physicochemical properties of molecules that passively diffuse across the BBB have been identified. For example, van de Waterbeemd et al. (1998) concluded that to enhance CNS penetration, a compound should have a molecular weight <450 and a total polar surface area (PSA) <90 Å. Others have correlated brain uptake with lipophilicity, hydrogen bond donors/acceptors, and rotatable bonds (Clark, 2001). Of note, several authors cautioned that the processes governing brain entry are complex and are unlikely to be related solely to physicochemical properties, but also influenced by other biological processes such as efflux transporter mechanisms.
High passive membrane permeability and the absence of efflux would likely favor CNS exposure. Conversely, low permeability and high efflux would diminish CNS exposure. To date, there has not been a systematic investigation of in vitro permeability and Pgp-mediated efflux to determine whether these factors discriminate between successful CNS and non-CNS medicines. The objectives of this study were to determine whether the in vitro permeability, Pgp substrate profiles, and physicochemical properties differed among 93 structurally diverse marketed drugs grouped by CNS and non-CNS indication, and to establish in vitro selection criteria to aid in drug discovery compound selection and lead optimization.
Materials and Methods
Materials
GlaxoSmithKline Chemical Registry supplied all test drugs. [4-3H]Propranolol (15–30 Ci/mmol) andd-[1-14C]mannitol (50–63 mCi/mmol) were purchased from Amersham Biosciences Inc. (Piscataway, NJ). Cell culture reagents were purchased from Invitrogen (Carlsbad, CA). All other chemicals were purchased from Sigma-Aldrich (St Louis, MO). Transwells (12-well, 11-mm diameter, 0.4 μm pores) were purchased from Corning Costar (Cambridge, MA).
Drug Set Selection and Limitations
Drugs were selected from the literature (Gilman et al., 1993;van de Waterbeemd et al., 1998) with the objective of having a balanced number of CNS and non-CNS drugs. Criteria used for selection were availability, therapeutic indication, molecular weight (range 150–800), and chemical stability/state (e.g., drugs that exist as gases or liquids were not considered). Of the 156 drugs identified, experiments were completed on 106. However, data were only reported on 93 drugs, primarily due to poor mass balance (≪50%) for 13 drugs in efflux studies. One limitation of grouping drugs by CNS and non-CNS indication is the general premise that CNS drugs must enter the brain to elicit an effect, whereas non-CNS drugs do not. The assumption that CNS-indicated drugs must penetrate the BBB is reasonable. However, the assumption that non-CNS drugs do not cross the BBB is not as reliable because many of these drugs are associated with CNS side effects. Therefore, this must be considered in interpreting the data from the non-CNS group. A further limitation of the selected compound set is that it may represent only a portion of the chemical diversity of currently marketed drugs.
Monolayer Efflux Studies
Multidrug resistance-transfected MDCK type II (MDR1-MDCKII) cells were obtained from the Netherlands Cancer Institute (Amsterdam, Netherlands). Culturing of cells and transport studies were completed as previously described (Polli et al., 2001b). Briefly, cells were split twice weekly at a ratio of 1:10 and grown in the absence of any selection agent to maintain Pgp expression. For transport studies, cells were seeded onto polycarbonate Transwell filter membranes at a density of 300,000 cells/cm2, and monolayers were ready for studies 3 days later. Drugs were dissolved at 20 mM in 100% dimethyl sulfoxide (DMSO) and then diluted in transport buffer (8.1 mM Na2HPO4, 138 mM NaCl, 0.5 mM MgCl2, 1.47 mM KH2PO4, 2.67 mM KCl, 0.9 mM CaCl2, 5.6 mM glucose, and 0.33 mM sodium pyruvate, pH 7.4). Drugs were tested at 10 μM concentration and in two directions [apical-to-basolateral (A → B) and basolateral-to-apical (B → A)] in duplicate. Monolayer efflux studies were conducted at 37°C in a humidified incubator with shaking (90 rpm) for 60 min. Transendothelial electrical resistance was measured with an Endohm Meter (World Precision Instruments, New Haven, CT). Reference drugs for paracellular transport ([14C]mannitol), transcellular transport ([3H]propranolol), and Pgp efflux (amprenavir) were included in each experiment. Concentrations of [14C]mannitol and [3H]propranolol were measured by liquid scintillation counting with Ready Safe Liquid Scintillation Fluid (Beckman Coulter, Fullerton, CA) using a Beckman LS501 counter. Amprenavir was analyzed by cassette LC/MS/MS analysis along with the test drugs.
Bioanalysis by High Throughput Cassette LC/MS/MS Analysis.
All analyses were performed by dual high-performance liquid chromatography with tandem mass spectrometry (LC/MS/MS) and cassette analysis (Wring et al., 2000). Simultaneous assay of analytical standards or test samples was performed using cassette analysis, where samples or standards containing test drugs were pooled prior to injection (three per cassette). High-performance liquid chromatography was conducted on a Hewlett Packard 1100 (Hewlett Packard, Palo Alto, CA) equipped with a column-switching valve. The sample (injection volume, 10 μl or 20 μl) was loaded onto the column by means of a Gilson 215 autosampler (Gilson Medical Electronics, Middleton, WI) using a proprietary software add-in to HP Chemstation. Chromatography was performed on 30 × 2 mm (i.d.), 3 μm, Phenomenex Aqua C18 columns (Phenomenex, Torrance, CA) at a flow rate of 0.6 ml/min. The mobile phase consisted of two solvents: A, 10 mM ammonium formate, pH 3.5 with 1.5% (v/v) methanol; and B, 100% acetonitrile. The gradient profile was 0 to 2.0 min 1% (v/v) B; 2.0 to 3.0 min linear gradient to 95% (v/v) B; 3.0 to 3.9 min 95% (v/v) B; 3.9 to 4.0 min linear gradient to 1% (v/v) B; and 4.0 to 4.6 min 1% (v/v) B. Mass spectrometry was performed on a PerkinElmer Sciex API2000 (PerkinElmer Sciex, Toronto, ON, Canada) equipped with either a turbo ion spray source for electrospray ionization or a heated nebulizer source for atmospheric pressure chemical ionization. Detection by tandem mass spectrometry was based on precursor ion transitions to the strongest intensity product ions. Key instrumental conditions were optimized to yield best sensitivity. Selected ion monitoring was employed if tandem MS afforded inadequate response. The calibration range was typically 1.0 nM to 1.5 μM (n = 5) for each drug. Dose and donor solutions were diluted in transport medium/acetonitrile (1:1, v/v) as required to bring their concentrations into this range. Concentration of drug in the samples was calculated from the chromatographic peak area via proprietary software developed at GlaxoSmithKline. Sample analysis was typically completed within 6 h of each permeability study.
Calculations.
The apparent permeability (P
app) was calculated with the equation:
where P
app = apparent permeability; S = membrane surface area,C
0 = donor concentration at time 0, and dQ/dt = amount of drug transported per time. Data are presented as the averageP
app (nanometers per second) ± S.D. from two monolayers. A ratio of the B → A/A → BP
app values was calculated. Involvement of a Pgp-mediated efflux mechanism was concluded if the B → A/A → B ratio was >1.5. To confirm that drugs were Pgp substrates, drugs were also tested in the presence of 2 μM GF120918, a potent, specific Pgp inhibitor (Polli et al., 2001b). Inclusion of GF120918 reduces the B → A/A → B ratio to ∼1 for Pgp substrates.
Mass balance is the percentage of original drug mass accounted for at the end of the experiment (sum of the amount in the A and B chambers). Mass balance is calculated with the following equation.
where C
At andC
Bt are the drug concentrations in the apical (A) and basolateral (B) chambers at time (t),C
0 is the concentration of the donor at time 0, V
A andV
B are the volumes of the apical and basolateral chambers, and V
D is the volume of the donor solution added to the appropriate chamber. Mass balance was >70% except where noted in the tables.
Calcein Inhibition Assay
The calcein-AM assay was performed using the Vybrant Multidrug Resistance Kit (Molecular Probes, Eugene, OR) and MDR1-MDCKII cells. Cells were seeded at 70,000 cells per well (200 μl of culture medium) in 96-well black plates with clear bottoms (PerkinElmer Life Sciences, Boston, MA). The medium was changed 24 h after seeding, and the assay was performed 48 h later. On the day of the study, the medium was aspirated and monolayers were washed three times with transport buffer. Test drugs were added to monolayers in 50 μl of transport buffer containing 1% DMSO. Test concentrations of each drug (final concentrations of 1, 10, and 100 μM) were selected based on previous work with this assay (Polli et al., 2001b). DMSO concentration (1%) was constant in test and control wells (each n = 2). Plates were preincubated at 37°C for 10 min. Calcein-AM was added and plates were immediately placed in a SpectraMax Gemini cytofluorimeter (Molecular Devices Corp., Sunnyvale, CA) for 60 min and read at 15-min intervals at excitation and emission wavelengths of 485 and 530 nm, respectively. Pgp inhibition was quantified by use of the following equation:
where RFUcomp = fluorescence in the presence of 100 μM test compound (comp), RFUGF120918 = fluorescence in the presence of 1 μM GF120918, and RFUbackground = fluorescence in the absence of the drug (typically 45–65 RFU). A drug was determined to inhibit Pgp when percentage of maximum was >10%. A concentration of 100 μM was used for this calculation because only 1 of 93 drugs (bromocriptine) produced >10% response when tested at 1 or 10 μM.
Calculated Physicochemical Properties
Solute McGowan volume (Vx), excess molar refractivity (R2), dipolarity/polarizability (π), summation of hydrogen bond acidity (αH), and summation of hydrogen bond basicity (βH) were calculated as described by Platts et al. (1999). Log octanol/water partition coefficient (clogP) and molar refraction (cmr) were calculated using Daylight Software v4.71, (Daylight Chemical Information Systems Inc., Irvine, CA). PSA was calculated as described by Clark (1999). All other physicochemical descriptors were calculated using GlaxoSmithKline proprietary software.
Statistical Analysis
The nonparametric Wilcoxon rank sums statistical test was used to determine statistical significance using JMP, version 4.0.5 (SAS Institute Inc., Cary, NC). The significance level wasp < 0.05.
Results
Results for CNS-Indicated Drugs.
TheP app A → Bvalues ranged from 2.52 to 748 nm/s and theP app B → Avalues ranged from 5.88 to 788 nm/s for the 48 CNS-indicated drugs (Tables 1 and2). The mean passive permeability value (P app B → A + GF120918) was 474 nm/s and the range was 2.17 to 847 nm/s, indicating that, on average, these drugs have good membrane flux. Forty-six of the 48 drugs (95.8%) had passive permeability values > 150 nm/s (Fig. 1), consistent with high passive permeability being a key feature of CNS-indicated drugs. The two drugs with permeability values <150 nm/s were both anti-migraine compounds (sumatriptan and zolmitriptan). The B → A/A → B efflux ratios ranged from 0.76 to 44.7 with only 7 of 48 (14.6%) of the CNS drug set undergoing efflux (B → A/A → B ratio > 1.5) across MDR1-MDCKII monolayers (Table 2 and Fig.2). Only one drug, eletriptan, had a B → A/A → B ratio > 5, suggesting that it is a good Pgp substrate. The B → A/A → B ratio for these seven drugs reduced to <1.3 in the presence of the specific Pgp inhibitor GF120918, confirming that they are Pgp substrates.
Summary of in vitro assay results
Transport and calcein-AM results for CNS-indicated drugs
Distribution of P appvalues for CNS- versus non-CNS-indicated drugs. Overall, both CNS and non-CNS drugs had good permeability, represented by the numbers of each drug type spanning the range of P app values. There was a lower number of CNS drugs (2 of 48) compared with the non-CNS drugs (13 of 45) that had P appvalues <150 nm/s.
Distribution of Pgp efflux ratios for CNS- versus non-CNS-indicated drugs. The majority (72.0%) of drugs were not Pgp substrates (efflux ratio <1.5). Of those that were substrates, more were from the non-CNS set (19 of 45) than from the CNS set (7 of 48), and 73.7% of the non-CNS drugs had B → A/A → B ratios >2.5 compared with 28.6% for the CNS drugs.
Of the 48 CNS drugs, 28 (58.3%) were positive (percentage of maximum response >10%) in the calcein-AM inhibition assay when tested at 100 μM (Tables 1 and 2). The percentage of maximum fluorescence ranged from −1.01 to 84.4. Eight drugs gave a calcein-AM response >40% of maximum at 100 μM (Table 2). Bromocriptine was the only drug that had a positive response (i.e., 34.7% of maximum) at 10 μM, and none of the CNS drugs elicited a response when tested at 1 μM (data not shown). This suggests that for all drugs except bromocriptine, the concentration used in efflux does not inhibit Pgp-mediated transport. There was agreement between the efflux and calcein-AM assay results for 19 of 48 (39.6%) drugs (Table 2). Of the remaining 29 drugs that did not show agreement between the two assays, 4 were positive in monolayer efflux and negative in the calcein-AM assay, and 25 were negative in monolayer efflux and positive in the calcein-AM assay. Of the latter 25 drugs that were negative in efflux and positive in calcein-AM, only 2 had passive permeability values <400 nm/s (bromocriptineP app = 182 nm/s and metergolineP app = 216 nm/s). This suggests that high passive permeability is a feature that explains the lack of concordance between efflux and calcein-AM assays for CNS drugs.
Results for Non-CNS-Indicated Drugs.
TheP app A → Bvalues ranged from 1.51 to 792 nm/s and theP app B → Avalues ranged from 3.17 to 834 nm/s for the 45 non-CNS-indicated drugs (Tables 1 and 3). The mean passive permeability value was 331 nm/s and the range was 2.49 to 674 nm/s (Fig. 1). Overall, most drugs had good membrane flux. The B → A/A → B efflux ratios ranged from 0.78 to 261 with 19 of 45 (42.2%) of the non-CNS drugs undergoing Pgp-mediated transport (Table 3 and Fig. 2). The B → A/A → B ratios were typically reduced to <1.3 in the presence of the specific Pgp inhibitor GF120918. However, the B → A/A → B ratios for several Pgp substrates (indinavir, saquinavir, nelfinavir, and pirenzepine) did not attenuate to ∼1 in the presence of GF120918. As well, one drug (sulfasalazine) with a B → A/A → B ratio > 1.5 was classified as a nonsubstrate because the ratio did not change in the presence of GF120918. It is possible that these drugs are substrates for other endogenous transporters present in MDCKII cells that are not inhibited by GF120918.
Transport and calcein-AM results for non-CNS-indicated drugs
Of the 45 non-CNS-indicated drugs, 18 (40.0%) were positive in the calcein-AM inhibition assay (Tables 1 and 3). The percentage of maximum fluorescence ranged from −3.45 to 100. Only four drugs (astemizole, loperamide, terfenedine, and verapamil) gave a calcein-AM response > 40% maximum at 100 μM (Table 3). None of the drugs in the non-CNS set were positive when tested at 1 or 10 μM (data not shown), suggesting that the concentration used in the efflux assay does not inhibit Pgp-mediated transport of drug. There was agreement between the efflux and calcein-AM assay results for 28 of 45 (62.2%) drugs (Table 3). Of the remaining 17 drugs, 9 were positive in monolayer efflux and negative in the calcein-AM assay, and 8 were negative in monolayer efflux and positive in the calcein-AM assay. The mean passive permeability of the nine drugs positive in the efflux assay and negative in the calcein-AM assay was 6.5-fold lower than the overall mean permeability for the non-CNS group (51.4 versus 331 nm/s; Table 1). In addition, none of these nine drugs had a passive permeability > 120 nm/s (Table 3, section 3). In contrast, the eight drugs that were negative in efflux but positive in the calcein-AM assay had a mean passive permeability of 497 nm/s, with the lowest rate being 387 nm/s. These eight drugs appear to have properties more similar to drugs in the CNS set. These observations suggest that passive permeability is an important characteristic (in addition to interaction with Pgp) for a drug to give a positive response in the calcein-AM assay.
Comparison of the Results from the CNS and Non-CNS Sets.
The mean passive permeability was 474 and 331 nm/s for the CNS and non-CNS drug sets, respectively (Table 1). Although the mean values were statistically different (p < 0.05), drugs in both sets displayed high passive flux and had similar overall permeability profiles (Fig. 1). However, there was a striking difference among the CNS and non-CNS groups in the number of drugs that had passive permeabilities in the 0- to 150-nm/s range. Thirteen of 45 (28.9%) non-CNS drugs had passive permeability values <150 nm/s, whereas only 2 of 48 (4.2%) CNS drugs had values <150 nm/s. Thus, the 7-fold lower incidence of passive permeability being <150 nm/s for the CNS drug set suggests that permeability is a discriminating factor among the two groups.
The majority of drugs in this study (72.0%, n = 67 of 93) were not Pgp substrates in the efflux assay. However, non-CNS drugs (19 of 45, 42.2%) had a 3-fold higher incidence of being Pgp substrates than CNS drugs (7 of 48, 14.6%) (Fig. 2). Of the 19 non-CNS drugs undergoing efflux, 14 (73.7%) had B → A/A → B ratios > 2.5. In contrast, only 2 of 7 (28.6%) CNS drugs had B → A/A → B ratios >2.5. These results suggest that the Pgp substrate profiles differed between the two groups, with CNS drugs having a 3-fold lower incidence in both the number of drugs being Pgp substrates and the magnitude of the efflux ratio (B → A/A → B ratios > 2.5).
Overall, 49.5% (46 of 93) of the drugs were positive in the calcein-AM assay when tested at 100 μM. Both sets had a similar range of response in the assay, and few compounds from each set were strong inhibitors (>40% of maximum at 100 μM). However, more CNS drugs (28 of 48, 58.3%) than non-CNS drugs (18 of 45, 40.0%) were positive in the assay, and the concordance between efflux and calcein-AM was lower for the CNS group (39.6%) compared with the non-CNS group (62.2%). The results suggest that Pgp inhibition is not a factor that discriminates between CNS and non-CNS drugs.
The means and ranges for 18 calculated physicochemical properties of the 93 drugs are listed in Table4. Overall, the two compound sets had similar physicochemical profiles that covered a broad range of values. This suggests that the selected compound sets are equally chemically diverse and cover similar chemical space. However, there were several properties that were significantly different between the groups. The CNS group had fewer hydrogen bond donors (donors, α, HBD), fewer positive charges, greater lipophilicity (clogP, clogD), lower PSA, and reduced flexibility compared with the non-CNS group (p< 0.05). Of particular note was the observation that even though size and bulk descriptors (molecular weight , cmr, and Vx) were not statistically different between the groups (p > 0.60), all non-CNS drugs (10 of 10) with a molecular weight >400 (cmr > 11.5, Vx > 3.0) were Pgp substrates. In contrast, only 1 of 6 CNS drugs with these attributes was a substrate. Taken together, these observations suggest that Pgp efflux is an important discriminator between marketed non-CNS and CNS drugs that are large (molecular weight >400) and bulky (cmr > 11.5, Vx > 3.0).
Mean (range) values of calculated physical chemical properties for CNS and non-CNS drugs
Discussion
Achieving adequate CNS delivery or exclusion can be a challenge to the development of a new drug. The primary purpose of this study was to measure to what extent CNS and non-CNS drugs differ in their in vitro permeability, Pgp substrate profiles, and physicochemical properties, and to establish in vitro selection criteria to aid in compound selection during drug discovery. Based on the frequency found in a large diverse set of marketed drugs (n = 93), passive membrane permeability, Pgp-mediated efflux, and certain physicochemical properties were distinguishing factors between the two drug sets.
Membrane permeability was a clear discriminating factor between these two groups, with only 2 of 48 CNS drugs having a passive permeability rate of <150 nm/s. In contrast, 13 of 45 non-CNS drugs had rates <150 nm/s. This difference in distribution frequency was reflected as well in the overall higher mean passive permeability for CNS drugs (481 versus 331 nm/s). Taken together, this suggests that high membrane permeability (>150 nm/s) is a prevalent characteristic of marketed CNS drugs. This finding is in agreement with the guidance proposed by our laboratory that a compound should have an in vitroP app > 200 nm/s to achieve good in vivo CNS penetration via passive permeability (Polli et al., 2000). The permeability guidance was based on the analysis of in vitro MDCK type I permeability data, rat brain unidirectional influx rates determined by in situ brain perfusion, and rat brain/plasma ratio results from 28 internal drug candidates.
The two CNS drugs with passive permeability values <150 nm/s are both used in the treatment of migraine. Interestingly, it has been suggested that the BBB breaks down during a migraine and/or that the target serotonin receptors are present on the BBB microvasculature and other peripheral sites (Goadsby, 2000). Therefore, anti-migraine drugs may not need to cross the intact BBB for clinical efficacy. This class of drugs represents a unique exception to the “broad” parenchymal delivery strategy typically used for the treatment of CNS diseases and resembles non-CNS drugs in its low passive membrane permeability and high Pgp-mediated efflux (see below).
Besides passive membrane permeability, membrane transporters such as Pgp can limit the CNS penetration of a drug (Polli et al., 1999;Schinkel, 1999). We found that Pgp-mediated efflux was a second discriminating factor between CNS and non-CNS drugs. Overall, 26 of 95 (27.4%) drugs in the combined set underwent Pgp efflux. The incidence of efflux across MDR1-MDCKII monolayers was 3-fold lower in the CNS drug set than in the non-CNS drug set (14.6% versus 42.2%). Furthermore, for drugs that were Pgp substrates, the incidence of the Pgp efflux ratio being greater than 2.5 was lower (2.6-fold) for CNS drugs when compared with non-CNS drugs. Interestingly, the two CNS drugs that had efflux ratios >2.5 (eletriptan and methysergide) are anti-migraine agents. As noted above, these agents may not need to cross an intact BBB for efficacy and resemble non-CNS drugs, which have larger efflux ratios.
Analysis of a variety of calculated physicochemical properties revealed that CNS drugs had fewer hydrogen bond donors (donors, α, HBD), fewer positive charges, greater lipophilicity (clogP, clogD), lower PSA, and reduced flexibility compared with the non-CNS group. These trends have been noted by others during the development of computational models to predict CNS penetration (van de Waterbeemd et al., 1998; Ajay et al., 1999; Clark, 2001). We have observed that these physicochemical properties also appear to be features in compounds that are Pgp substrates (Polli et al., 2001a). This was particularly evident within this drug set for large, bulky drugs (molecular weight >400 and cmr > 11.5) where 10 of 10 non-CNS drugs were Pgp substrates, whereas only 1 of 6 CNS drugs was a Pgp substrate. High-throughput discovery screens have led the drug industry toward the selection of highly potent molecules that are large and bulky (Lipinski, 2000). This screening approach may also unintentionally select drug candidates that are more likely to be substrates for Pgp because of increased size and lipophilicity.
The interaction with Pgp measured by efflux is a better discriminator between CNS and non-CNS drugs than when measured by calcein-AM. More of the CNS (58.3%) drugs were positive in the calcein-AM assay than in the efflux assay (14.6%). In contrast, non-CNS drugs had similar responses in the efflux and calcein-AM assays (42.2 versus 40.0%; concordance = 62.2%). Thus, the calcein-AM assay is not useful for discriminating CNS from non-CNS drugs. The lack of concordance may be explained by failure of the efflux assay due to saturation of Pgp, or it may be that these drugs partition across the cell membrane too rapidly to allow measurement of transport (Eytan et al., 1996;Pauli-Magnus et al., 2000). The physicochemical properties of the drugs positive in the calcein-AM assay are consistent with this possibility. These drugs have high lipophilicity (clogP > 3) and few (0 to 1) hydrogen bond donors (Table 4), features that are found in most CNS-indicated drugs and that enhance passive diffusion across cell membranes. It is also possible that drugs positive in the calcein-AM assay and negative in the efflux assay may be Pgp inhibitors, but not substrates. For this drug set, this appears only to be a possibility for bromocriptine because this was the only compound to inhibit calcein-AM efflux at 10 μM, the test concentration used in the efflux studies. Finally, we observed that most drugs positive in efflux but negative in the calcein-AM assay have lower passive membrane permeability, which may limit their membrane partitioning. This feature, along with a lower affinity for Pgp than for calcein-AM, may result in these drugs not being able to effectively compete with calcein-AM for efflux.
In conclusion, the large set of successful medicines examined here shows that between CNS and non-CNS drug sets, there is overlap in the passive permeability, Pgp efflux, and physicochemical properties. However, some features that favor CNS exposure have clearly emerged. For delivery to the CNS, a drug should ideally have an in vitro passive permeability > 150 nm/s and should not be a good Pgp substrate (B → A/A → B ratio < 2.5), especially if the drug has a molecular weight >400. In contrast, to exclude a drug from the CNS, it should have low passive permeability (<50 nm/s) and be a strong Pgp substrate (B → A/A → B ratio > 5).
Acknowledgments
We thank Dr. Michael Emptage for statistical analysis.
Footnotes
- Received May 21, 2002.
- Accepted July 15, 2002.
↵1 Current address: Preclinical Drug Metabolism and Pharmacokinetics, GlaxoSmithKline, Inc., Mail Stop-UW2720, 709 Swedeland Rd., King of Prussia, PA 19406.
K.M.M.D. was a GlaxoSmithKline postdoctoral fellow in Preclinical Drug Metabolism and Pharmacokinetics.
DOI: 10.1124/jpet.102.039255
Abbreviations
- CNS
- central nervous system
- BBB
- blood-brain barrier
- Pgp
- P-glycoprotein
- MDCK
- Madin Darby canine kidney cells
- PSA
- polar surface area
- MDR
- multidrug resistance protein
- DMSO
- dimethyl sulfoxide
- A → B
- apical to basolateral
- B → A
- basolateral to apical
- B → A/A → B ratio
- Papp B → A/Papp A → B
- LC
- liquid chromatography
- MS
- mass spectrometry
- Papp
- apparent permeability
- AM
- acetoxymethyl ester
- clogP
- calculated octagonal/water partition coefficient
- cmr
- calculated molar refraction
- HBD
- count of hydrogen bond donor groups
- RFU
- relative fluorescence unit
- The American Society for Pharmacology and Experimental Therapeutics
