Active transport mechanisms as determinants of drug absorption, distribution, and clearance have been the focus of considerable research effort over the past decade. Of the numerous transporter proteins recently investigated, the one for which the greatest amount of knowledge exists is P-glycoprotein (MDR1). Originally described as a transporter involved in imparting drug resistance to tumor cells, P-glycoprotein has been demonstrated to be important in reducing absorption of drugs from the intestinal lumen, in active secretion of drugs into urine and bile, and in extrusion of drugs from vital organs such as the brain and reproductive tissues (Troutman et al., 2002). As such, P-glycoprotein-mediated transport has become an important issue in the discovery and development of new drugs. For example, new compounds that are promising with regard to target receptor/enzyme activity can be severely hampered in their ability to elicit pharmacological effects in vivo should they be good substrates for P-glycoprotein, especially if the route of administration is intended to be oral or the target tissues is one rich in P-glycoprotein activity. Furthermore, the potential for drug-drug interactions arises in the event that the P-glycoprotein substrate is coadministered with another agent that can inhibit P-glycoprotein.

Several models have been developed to assess drugs as P-glycoprotein substrates. In vitro models have included the Caco-2 cell line, which expresses this transporter (among numerous others), and cell lines stably transfected with P-glycoprotein, such as Madin-Darby canine kidney cells (MDCK). Such cell lines can be used with P-glycoprotein inhibitors, or substrate transport can be studied in transfected versus control cells to measure P-glycoprotein activity. Additionally, animal models have also been used to assess the impact of P-glycoprotein on substrate pharmacokinetics and/or effect. Some specific breeds of dogs have been shown to lack functional P-glycoprotein, and the effects of drugs that are P-glycoprotein substrates can be markedly different in such animals (e.g., ivermectin toxicity in collie dogs; Mealey et al., 2003; Roulet et al., 2003). A very powerful tool has been developed by Schinkel and coworkers: the mdr1a/1b knockout mouse model (Schinkel et al., 1996). These are genetically modified animals in which both genes that are homologous to the human MDR1 gene have been disrupted, resulting in a viable line of animals. These animals have seen widespread use in investigating the function of P-glycoprotein in drug disposition. Additionally, a variant subpopulation of CF-1 mice lacking functional mdr1a has been identified and characterized (Umbenhauer et al., 1997), and used in studies to address the role of this transporter in drug disposition (Kwei et al., 1999; Yamazaki et al., 2001).

In our ongoing efforts to identify new agents for disorders of the central nervous system, we have frequently applied the mdr1a/1b knockout mouse model. Comparison of the brain/plasma (B/P) AUC ratio in knockout animals versus wild-type animals has become a standard experimental approach to determine whether P-glycoprotein-mediated efflux poses a potential threat to the activity of CNS agents in vivo, or a confounding factor in attempts to relate circulating free drug concentrations to effect in animal pharmacology models. Such in vivo data can be used in concert with data obtained using in vitro models, to assess the potential liability of P-glycoprotein-mediated efflux from the brain.

However, it has been a common observation of ours that the conclusion for various compounds from in vitro P-glycoprotein data have not been consistent with the data we have obtained in the mdr1a/1b knockout mouse model. When discord has been observed, it has been that the in vivo mouse model has suggested that P-glycoprotein is an important determinant in the CNS penetration of the compound, whereas the in vitro data have supported the opposite conclusion. A variety of factors could be responsible for such disagreement, including interspecies differences in intrinsic activity of P-glycoprotein (which are not identical between human and mouse and could have different expression levels at the blood-brain barrier between the species) or failure of the in vitro models to demonstrate adequate representation of P-glycoprotein activity relative to other phenomena, such as passive diffusion rates, that can occur in vitro.

Because of these uncertainties, we have undertaken a large study in which we have examined over 30 well established CNS agents of a variety of chemical and therapeutic classes in the mdr1a/1b knockout mouse model. Additionally, several non-CNS agents, some known to be P-glycoprotein substrates, were assessed in this model for comparison. The objective of this study was to determine how CNS-active agents, as a group, behave in the mdr1a/1b knockout mouse model with regard to brain-plasma pharmacokinetics. The following questions were kept in mind in pursuing this objective. 1) Are there CNS-active agents that demonstrate a large difference in B/P ratio in knockout versus wild-type mice? 2) Is the ratio of B/P AUC ratios between knockout and wild-type typically unity for CNS-active agents, or are there many also subject to some action by P-glycoprotein? If not unity, is there a reasonable cutoff value under which most CNS-active agents reside? 3) Do CNS-active agents consistently demonstrate B/P ratios in excess of unity? 4) Does cerebrospinal fluid (CSF) represent a viable sampling matrix that can adequately reflect the potential impact of P-glycoprotein on CNS activity? This study, in concert with the accompanying report (Maurer et al., 2005) attempts to provide insight into the application of P-glycoprotein models in the discovery of new CNS-active drugs.

Materials and Methods

Chemicals. Supplies of cyclobenzaprine, fluoxetine, methylphenidate, prazosin, and sertraline were obtained from Pfizer Global Material Management (Groton, CT). Citalopram, paroxetine, ritonavir, and loratadine were purchased from Sequoia Research Products (Oxford, UK). The following reagents were purchased from the respective vendors: amiodarone (Spectrum Chemical, Gardenia, CA), clozapine (MP Biomedicals, Irvine, CA), fluvoxamine (Tocris Cookson Inc., Ellisville, MO), 9-hydroxyrisperidone (SynFine Research, Richmond Hill, ON, Canada), midazolam (Cerilliant, Round Rock, TX), propoxyphene (U.S. Pharamcopoeia, Rockville, MD), and venlafaxine (Alchemie USA, Plantsville, CT). All other drugs included in the study were purchased from Sigma-Aldrich (St. Louis, MO).

Animals. Female FVB (wild-type) and mdr1a/1b (-/-, -/-) mice of approximately 9 weeks of age, weighing 25 to 30 g, were obtained from Taconic Farms (Germantown, NY). For quinidine, loperamide, and caffeine, male FVB (wild-type) and mdr1a (-/-) mice were used. Upon arrival, the mice were maintained for at least 5 days on a 12-h light/dark cycle in a temperature- and humidity-controlled environment with free access to food and water. The mice were housed in clear polycarbonate boxes (n = 5 per box) containing sawdust and nesting pads. The study was conducted in accordance with approved Pfizer Animal Care and Use Procedures.

Drug Selection. The process of drug selection for the study was approached with the objective of identifying a representative and diverse sampling of the most commonly used CNS therapeutic agents. To identify the top-selling CNS drugs, a market assessment was conducted for total dollar sales and number of worldwide and U.S. prescriptions filled for the year 2001. To ensure chemical diversity of the data set, a structure cluster analysis was conducted. The UNITY fingerprints of CNS drugs were calculated using the software package SYBYL 6.7.1 (Tripos Inc., St. Louis, MO) and then grouped based on a hierarchical clustering algorithm. Drugs for the study were then selected based on a combination of factors including availability, market prevalence, pharmacology, chemical structure (based on clustering algorithm), and analytical feasibility.

The exercise resulted in the identification of 32 CNS drugs representing 31 distinct structure clusters. In addition, the active metabolites of risperidone and carisoprodol (9-hydroxyrisperidone and meprobamate, respectively) were added to the study to evaluate the potential contribution of any metabolite interactions with P-gp on the biological activity of parent drug. Excluded from consideration as CNS agents were the anti-inflammatory drugs indicated for pain, and anti-migraine agents due to the peripheral location of their targeted receptors and the potential for disruption of the blood-brain barrier during migraine episodes (Goadsby, 2000).

The CNS drugs selected represent a number of different therapeutic indications including antidepressants, sedatives, anxiolytics, tranquilizers, anticonvulsants, etc. The molecular properties of the drug set are consistent with those expected for lipophilic membrane-permeable compounds. The average molecular weight for the drug set was 297 and ranged from 141 to 426. A majority of the drugs (n = 24) were basic compounds with calculated pK_a values ranging from 6.9 to 10.6. Several examples of weakly acidic or neutral drugs were represented in the data set and include: caffeine, carbamazepine, carisoprodol, diazepam, ethosuximide, meprobamate, midazolam, phenytoin, thiopental, and zolpidem. The mean calculated log D value for the CNS drug set was 1.95 and ranged from -1.12 to 4.42. The pK_a and log D values were calculated using ACD labs, version 6.0 (Advanced Chemistry Development Inc., Toronto, ON, Canada). Molecular weight and polar surface areas were calculated using MOE 2002 (Chemical Computing Group, Montreal, QB, Canada). Overall physicochemical characteristics of the study drugs are listed in Table 1.

A smaller set of eight non-CNS drugs was selected for evaluation in the study as controls. The drugs were selected for their consideration as P-gp modulators, although not necessarily as substrates for transport, and to represent a range in the degree of interaction spanning from weak or moderate to strong. Included as positive controls for substrates transported by P-gp were quinidine, an antiarrhythmic agent; loperamide, an antidiarrheal; verapamil, an antihypertensive calcium channel blocker; loratadine, a nonsedating antihistamine; and ritonavir, an HIV protease inhibitor. The antiarrhythmic agent, amiodarone, was selected as a known P-gp inhibitor of very high lipophilicity. Also included were prazosin, the antihypertensive agent commonly used for photoaffinity labeling of the drug-binding site on P-gp, and prednisone, a glucocorticoid.

Dose Administration and Sample Collection. Mice were administered a single 3 mg/kg subcutaneous dose (n = 5 per genotype per time point) of drug for all drugs except quinidine (10 mg/kg), loperamide (1 mg/kg), and caffeine (5 mg/kg). Dosing solutions of each drug were prepared in phosphate-buffered saline, with or without 5% dimethyl sulfoxide or 10% glycerol formal, when necessary; or 20% hydroxypropyl β-cyclodextrin, such that doses were administered in a volume of 10 ml/kg. Dosing solutions were quinidine (10 mg/kg), loperamide (1 mg/kg), and caffeine (5 mg/kg). Mice were euthanized in a CO₂ chamber at 0.5, 1, 2.5, or 5 h postdose. CSF was collected via puncture of the cisterna magna using a 25 gauge needle attached to polyethylene tubing (internal diameter of 2.0 mm) and a syringe. Whole blood was collected by cardiac puncture into Microtainer tubes containing heparin and stored on ice until centrifuged for the preparation of plasma. Whole brains were collected by decapitation, rinsed in phosphate-buffered saline, and weighed. CSF and whole brains were immediately frozen on dry ice upon collection. Note that the data reported for 9-hydroxyrisperidone resulted from the separate and direct administration of the metabolite to FVB and mdr1a/1b mice, whereas concentrations of meprobamate were determined as the metabolite formed following carisoprodol administration.

Analysis of Drugs in Mouse Plasma, Brain, and CSF. Plasma, CSF, and brain samples for all drugs were analyzed by HPLC/mass spectrometry. Summaries of the analytical methods for plasma are available as Supplemental Information. In many cases, drugs were processed by liquid-liquid extraction, whereas others were subjected to solid phase extraction or deproteination using CH₃CN. Brains were homogenized in 4 volumes of water, and homogenates were processed and analyzed in a manner similar to plasma samples. CSF samples were analyzed using the same HPLC/mass spectrometry conditions and were either directly injected on HPLC or were first evaporated and reconstituted in mobile phase before HPLC analysis. Settings and potentials for the mass spectrometers were adjusted to optimize the signal for each analyte. All samples were analyzed for concentrations of intact parent drug with the exception of meprobamate, which was monitored as a metabolite formed following administration of carisoprodol.

Data and Statistical Analysis. The experiments were set up as a randomized complete block design with genotype as the main effect and time as a secondary effect. The times were 0.5, 1.0, 2.5, and 5 h postdose. The genotypes were wild type (WT) and knock out (KO). The WT animals were randomly assigned to five cages corresponding to each replicate, so that there were four animals per cage, with each representing one time point. The same procedure was repeated with the KO mice using an additional five cages. The caging was done in this manner to prevent errors in genotype identification posed by a randomized housing scheme. The final AUC calculations considered the effects of genotype-specific caging. The sample size was determined assuming a small percentage of dropouts for a significance level of 0.05 and a power of 0.8 at each time point.

The AUCs were calculated using the trapezoid rule and the variances of the AUCs were determined from the weighted sum of the concentration variance at each time point. The weights were determined from the time separation weights found in the trapezoid rule. The AUC ratios, B/P, CSF/brain, and CSF/plasma were calculated and the respective variances determined using the first order Taylor series expansion of the ratios. The correlation between brain, plasma, and CSF concentrations was incorporated into the variance estimate.

The B/P, CSF/plasma, and CSF/brain ratios were then used to compare KO with WT ratios to determine a stepwise significant increase from 1, 2, 3, and 4 (although results for 3 and 4 were not different from 2 and are not reported). The KO/WT ratios were compared using a log transform of the ratios and an approximation to the variance. The comparisons of the log-transformed ratios were done using a t test with unequal variances. All statistical calculations and analyses were completed using SAS Version 8.0 (SAS Institute, Cary, NC).

Results

Thirty-two drugs (and two metabolites) were selected for evaluating differences in central exposures of CNS-active agents using the mdr1a/1b double knockout mice model. Each drug was administered as a 3 mg/kg subcutaneous dose, and concentrations were determined in plasma, brain, and CSF at 0.5, 1, 2.5, and 5 h postdose. The resulting AUC, C_max, and T_max values are presented in Table 2.

Special consideration was given to the size and diversity of the drug set in selecting the dosing and sampling regimen for this study. Based on the number and completeness of the concentration profiles obtained, the regimen used provided sufficient data for assessment of P-gp impact on brain exposures for all drugs tested. For 85% of the CNS drugs, concentrations in plasma and brain samples were reportable for at least three of the four time points collected. In a limited number of cases (n = 6), plasma, brain, or CSF concentrations were observed for only two of the four time points collected: carisoprodol, ethosuximide, hydrocodone, meprobamate, sertraline, and sulpiride. Sertraline was the only example of a CNS drug for which CSF concentrations were detected at only a single time point in each genotype. Plasma and brain concentrations for all the control drugs were also captured in at least three of the four time points, although the CSF concentrations were more sporadic. Drug concentrations of amiodarone, loratadine, and ritonavir were not measurable in CSF of KO or WT mice and were only observed in two of the four time points for prazosin (WT and KO) and lopermide (WT only).

Representative time courses are presented in Fig. 1 as reflective of the diversity in the pharmacokinetic profiles observed across drugs as well as across matrices for individual drugs. As represented by the concentration profiles for zolpidem, most drugs exhibited monoexponential declines in plasma concentration profiles, with brain and CSF concentration profiles tracking in parallel with plasma. In some cases, such as paroxetine, a lag in the brain concentration profile was observed relative to the plasma concentration profile. For all the drugs studied, plasma T_max values were observed at the 0.5-h time point and were consistent between the FVB and mdr1a/1b dose groups, with the exception of lamotrigine and amiodarone, for which maximal plasma concentrations were observed at the 1- and 2.5-h sampling times, respectively.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

Representative plasma, brain, and CSF concentration-time profiles for zoplidem (a), paroxetine (b), metoclopramide (c), and risperidone (d) in FVB and mdr1a/1b (-/-, -/-) mice following 3 mg/kg subcutaneous administration. Data points represent mean ± S.E.M. values. Note differences in scaling of ordinate axis.

For 28 of 42 drugs examined, there was no statistical difference in the plasma AUC(0→t_last) values between mdr1a/1b knockout and FVB mice. To account for any differences in the systemic exposures resulting from the P-gp genotype, brain and CSF AUC(0→t_last) values were normalized for plasma AUC(0→t_last) and are reported as B/P and CSF/plasma (CSF/P) ratios, respectively (Tables 2, 4, and 5). In FVB mice, the B/P AUC values for all CNS drugs ranged from 0.060 (9-OH risperidone) to 24 (sertraline), and the CSF/P values ranged from 0.015 (paroxetine) to 1.6 (ethosuximide). A large percentage of CNS drugs, 80% (20 of 25 basic drugs) and 65% (22 of 34 all CNS drugs) demonstrated B/P ratios ≥1 in the wild-type mice. With the exception of diazepam, which had a B/P ratio of 2.0, all the neutral CNS drugs had B/P ratios less than unity in FVB mice. The B/P ratios of the metabolites, meprobamate and 9-hydroxyrisperidone, were also determined to be <1 and were decreased compared with their respective parents, carisoprodol and risperidone.

To determine the effects of the lack of P-gp expression on the brain penetration of CNS drugs in mdr1a/1b knockout mice relative to their WT counterparts, the ratio of KO/WT values for B/P or CSF/P were compared for statistical significance against values of unity or 2-fold (Tables 3 and 5). Surprisingly, a majority of the CNS drugs (27 of 34) demonstrated a significant increase in the KO/WT B/P ratios when compared against unity (Fig. 2). In most cases, these increases were marginal and, in fact, only four drugs demonstrated a significant difference in the KO/WT ratio of B/P values when evaluated against a 2-fold increase: fluvoxamine, metoclopramide, propoxyphene, and risperidone (as well as the active metabolite 9-hydroxyrisperidone). The CNS drugs for which brain concentrations were most dramatically increased in the absence of P-gp were metoclopramide (6.6-fold), risperidone (10-fold), and 9-hydroxyrisperidone (17-fold). CSF concentrations for these drugs also showed marked alterations in P-gp KO mice.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

Effects of P-glycoprotein expression on the B/P ratios of CNS versus non-CNS drugs in FVB and mdr1a/1b (-/-, -/-) mice.

Results for the non-CNS drugs included in the study as controls are presented in Tables 3 and 5. The B/P ratios for all the non-CNS drugs examined were less than unity. Quinidine demonstrated the greatest increase in B/P ratio of any drug in the study when comparing KO to WT mice, 36-fold. Loperamide also demonstrated a significant increase in KO B/P relative to WT mice, 9.3-fold. The brain to plasma ratios of loratadine and verapamil increased 1.9- and 17-fold, respectively, in KO versus WT mice and were consistent with data reported in the literature from similar brain penetration studies conducted in P-gp genetically modified mice (Hendrikse et al., 1998; Chen et al., 2003). Results with amiodarone demonstrate a 21-fold increase in the B/P ratio of KO mice relative to WT mice, indicating that it is a substrate for P-gp. Moderate increases in KO/WT B/P ratios were also observed for prazosin and prednisone, 2.0- and 2.3-fold, respectively. Given the limited CSF data available for some of the non-CNS drugs, CSF/P values are only reported for loperamide, prazosin, prednisone, quinidine, and verapamil.

Discussion

The potential for P-gp-mediated transport poses a specific challenge to the discovery of new CNS agents by virtue of its function as an efflux transporter at the blood-brain barrier. For any agent intended to act in the brain, the success of eliciting a desired response is dependent on passive diffusion (related to the physicochemical properties of the compound) and the relative balance between active uptake and efflux transporters at the blood-brain barrier. Development of the mdr1a/1b knockout mouse model has provided a powerful tool for evaluating the effects of P-gp transport on brain and CSF pharmacokinetics for CNS-active agents in intact animals. To understand the prevalence and magnitude of P-gp interactions occurring within this class of therapeutic agents, we compared the central exposures of 32 clinically useful CNS drugs (and two active metabolites) across a 5-h time course in wild-type and mdr1a/1b (-/-, -/-) mice following subcutaneous administration. The results offer insight not only to the relevance of P-gp transport for CNS drugs, but also to the characteristics of CNS agents with regard to brain to plasma ratios and CSF to plasma ratios.

Brain-to-Plasma and CSF-to-Plasma Ratios.General Considerations. The CNS-active agents evaluated in the mouse did not demonstrate any particular trend regarding B/P ratios. Some B/P ratios for established CNS agents were as low as 0.060 to 0.46, whereas others were greater than 20. Thus, there is no requirement that CNS drugs achieve some threshold in total B/P ratios, although none exhibited B/P values below that which represents vascular volume in the brain (0.6-1.3 ml/100 g brain; Dagenais et al., 2000).

Our results demonstrate a difference in the brain pharmacokinetics of neutral or weakly acidic CNS drugs compared with basic CNS drugs. Neutral or acidic drugs tended to have lower B/P ratios in wild-type mice. Of the nine CNS drugs in the CNS drug set considered to be “nonbasic,” eight of them demonstrated B/P ratios equal to or less than 1, lower than those observed for the majority of basic CNS drugs. In contrast, only seven of the basic CNS drugs had B/P ratios <1 in the wild-type mice, and only two of those could be attributed to large P-gp efflux. These results indicate that lower brain partitioning is common among nonbasic CNS drugs and may not necessarily be associated with decreased potential for central activity. Lipophilic, weakly basic drugs, being cationic at physiological pH, may associate with phospholipid membranes, which could explain the difference in brain partitioning between weak bases and other molecules.

B/P and CSF/P ratios were not correlated. In the absence of active transport between blood and brain, CSF concentrations should approximate unbound plasma concentrations provided that the drug can penetrate the tissue. Thus, there should be no expectation that CSF and brain concentrations would be correlated, since total brain concentrations are largely driven by nonspecific binding of drugs to brain tissue (Maurer et al., 2005).

Comparison of P-Glycoprotein Knockout and Wild-Type Mice. Most of the drugs assessed in this study showed some difference between wild-type and P-gp knockout mice in brain-to plasma or CSF-to-plasma ratios (Fig. 2). Thus, it appears that successful CNS agents can still be substrates for P-gp. However, the relative impact of P-gp-mediated efflux needs to be balanced with rates of passive diffusion. For in vitro transport studies, Polli et al. (2001) categorized drugs with regard to penetrability and active transport. One category was referred to as “nontransported substrates” and included drugs such as chlorpromazine and midazolam. Such drugs appear to be substrates for P-gp but with no readily apparent impact of transport affecting disposition. In comparing B/P ratios in P-gp knockout versus wild-type mice, it appears that many CNS compounds fit into such a category. B/P ratios in knockout mouse are statistically greater than corresponding values in wild-type mice, but the values rarely exceed 2. Such drugs are probably not limited in their capability to elicit central effects despite being substrates of P-gp in vivo. It is noteworthy that B/P and CSF/P ratios in wild-type versus knockout animals were well correlated but lacked a one-to-one correspondence (Tables 4 and 5). This suggests that CSF represents a viable biofluid that can be used to assess the impact of P-gp on CNS exposure.

Consideration of Specific CNS-Active Drugs. The most interesting results obtained for a centrally active agent were for risperidone. Clearly, risperidone is a well established, clinically useful antipsychotic agent. Yet the data obtained regarding B/P ratios in knockout versus wild-type mice suggest a profound impact of P-gp on the central exposures for risperidone and its active metabolite 9-hydroxyrisperidone. The KO/WT ratio of B/P ratios was 10 for risperidone, with a corresponding value for CSF/plasma of 6.3. Should this same magnitude of efflux impact risperidone disposition in humans, it would be expected that lack of concordance would be apparent between efficacious free plasma concentrations, receptor affinity, and receptor occupancy. However, receptor occupancy, as measured using SPECT for the D₂ receptor, ranges, on average, between 50% and 79% for steady-state efficacious doses of risperidone of 2 to 4 mg/day (Lavalaye et al., 1999; Bressan et al., 2003). Concentrations associated with these doses range around 10 ng/ml for risperidone and 35 ng/ml for 9-hydroxyrisperidone; correcting for free fraction (0.11 and 0.23, respectively; Thummel and Shen, 2001) yields concentrations of approximately 1.1 and 8.1 ng/ml, respectively. The D₂ receptor K_i values for risperidone and 9-hydroxyrisperidone are 2.5 and 1.9 ng/ml, respectively (6 and 4.5 nM; Schotte et al., 1996). Thus, although P-gp could decrease the “receptor-available concentrations” of risperidone and 9-hydroxyrisperidone relative to the free plasma concentration, risperidone remains an effective antipsychotic agent, due to its high potency. This is also consistent with receptor occupancy data gathered in the rat for risperidone in which striatum versus pituitary D₂ receptor occupancy is markedly different (Kapur et al., 2002), presumably due to active efflux of risperidone from the brain. Thus, significant P-gp efflux does not preclude a new drug from being a successful CNS agent; however, it can alter the relationship between free plasma concentrations and potency.

Citalopram was previously reported to be a P-gp substrate in vivo, with a KO/WT B/P ratio of 3.3 (Uhr and Grauer 2003), versus our presently reported value of 1.9. In either case, the impact of P-gp on the action of citalopram appears to be minor and likely of little consequence. Other antidepressants (both selective serotonin reuptake inhibitors and tricyclic antidepressants) demonstrated statistically significant yet minor differences in brain/plasma ratios between wild-type and knockout animals, which is consistent with other data addressing brain penetration of this class of drugs (Uhr et al., 2000, 2003). Of the antidepressants examined, the impact of P-gp on brain to plasma ratios was less than 2.3-fold, but sertraline was the only antidepressant tested that demonstrated no statistically significant difference between wild-type and knockout strains.

Of the two benzodiazepines examined, diazepam and midazolam, neither demonstrated a difference in B/P ratios in knockout versus wild-type mice. Midazolam showed a remarkably low B/P ratio, approximately 8-fold lower than diazepam. Midazolam has been claimed to be a P-gp substrate, but with a very high rate of passive diffusion, overall minimal differences were observed in in vitro permeability studied in MDR1-transfected versus nontransfected MDCK cells (Tolle-Sander et al., 2003). Our data suggest that P-gp has no impact on the brain penetration of midazolam.

Pollack and coworkers have demonstrated an impact of P-gp on the action of morphine in rats and mice using modulators of P-gp activity and genetically modified animals (Aquilante et al., 1999; Letrent et al., 1999; Zong and Pollack, 2000; Daganais et al., 2004). However, a corresponding impact of P-gp on the action of morphine in humans has not been shown (Drewe et al., 2000; Skarke et al., 2003). Our data show a minimal impact of P-gp on brain penetration of morphine and were very similar to previously reported findings (Zong and Pollack, 2000).

Metoclopramide showed a substantial difference in B/P ratio between wild-type and P-gp knockout animals. B/P ratios were greater than unity in both strains, but were nearly 7-fold greater in knockout animals. There is an unclear picture regarding the site of action of metoclopramide as an antiemetic agent, as to whether it is central or peripheral. The site of dopaminergic activity of metoclopramide has been claimed to be in the fourth ventricle, which is outside the blood-brain barrier (Gralla, 1983), and antiemetic activity has also been ascribed to peripheral cholinergic activity (Singamma and Prabhakara Rao, 1992). In context to this proposed site of action, it is interesting to note that a less pronounced increase in the CSF/plasma ratio versus brain/plasma ratio using the P-gp knockout genotype was observed for metoclopramide. In light of our observations, the potential impact of P-gp on metoclopramide potency and CNS side effects in humans warrants further investigation, and a determination of the effect of P-gp inhibition on metoclopramide pharmacokinetic/pharmacodynamic relationships in the clinic may provide valuable insight to the mechanism of action of this agent.

Summary

This report represents the first systematic evaluation of brain penetration of a large number of successful CNS-active agents in the P-gp knockout mouse model. It is clear that most drugs, even those targeting the CNS, demonstrate some difference between B/P and CSF/plasma ratios between knockout and wild-type animals. However, differences of 2- to 3-fold or less between brain to plasma ratios of KO and WT animals do not appear to have much impact on the success of drugs as CNS-active compounds. Some CNS-active drugs even demonstrated substantially greater B/P ratios in KO mice versus WT mice, yet are still successful therapeutic agents (i.e., risperidone and metoclopramide). It is likely that for such compounds the relationship between free plasma concentrations, target receptor affinity, and resulting receptor occupancy (and, hence, effect) demonstrate some discord. Total B/P ratios themselves do not appear to be important in distinguishing CNS drugs, as some agents had ratios of well below unity (e.g., sulpiride, midazolam, zolpidem, etc.) This is consistent with the well accepted “free drug hypothesis” and information in the accompanying report (Maurer et al., 2005), since B/P ratios are largely driven by nonspecific brain tissue binding, which represents drug that is not necessarily available for receptor binding. Finally, CSF appears to be a viable biofluid from which the impact of P-gp on CNS exposure can be addressed.

The extent to which differences in B/P ratios in P-gp knockout versus wild-type mice can be applied to estimating human concentration-response relationships remains undetermined and an area of future research. Limitations of this model likely exist. Since the brain is a complex, multicompartmented organ, there likely exist other distributional aspects of drugs within the brain that cannot be addressed with measurements of total brain concentrations. Also, it has recently been shown that knockout of P-gp in the mouse can result in changes in expression of other transporters (Cisternino et al., 2004). Finally, the potential for mouse-human differences in intrinsic activity of their respective P-gp transporters as well as species similarities/differences in expression levels at the blood-brain barrier needs to be explored.