It has long been recognized that drug distribution is a function of relative tissue and plasma protein binding (Gillette, 1971; Wilkinson and Shand, 1975; Gibaldi and McNamara, 1978; Kurz and Fichtl, 1983). Albumin, α-1-acid glycoprotein, and lipoprotein are thought to account for the plasma binding of most drugs; however, relatively less is known about the determinants of tissue binding. Several reports indicate a generally important role for phosphatidylserine binding (Nishiura et al., 1986, 1987, 1988; Yata et al., 1990; Hanada et al., 1998; Suzuki et al., 2002) and lysosomal accumulation (Honegger et al., 1983; MacIntyre and Cutler, 1988; Xie et al., 1991; Ishizaki et al., 1998a,b,c) in determining the tissue distribution of basic drugs. For certain other drugs, binding to actin and myosin (Kurz and Fichtl, 1983), nuclei (Ishizaki et al., 1998a), melanin (Larsson and Tjalve, 1979), monoamine oxidase (Yoshida et al., 1989, 1990), and tubulin (Wierzba et al., 1987) has also been implicated. In drug discovery, knowledge of the particular tissue components governing the distribution of drug candidates is uncommon. However, many of the critical pharmacokinetic implications can be derived from information regarding the overall extent of tissue binding estimated nonspecifically. With relatively few assumptions, qualitative insight into the extent of tissue binding is easily obtained by pairing distribution measures with unbound plasma fraction. For example, it is known that tissue binding in excess of plasma binding can result in apparent volumes of distribution that greatly exceed physiologic fluid volumes (Gillette, 1971; Wilkinson and Shand, 1975; Gibaldi and McNamara, 1978). Another related concept is that, under freely diffusible conditions, the tissue-to-plasma concentration ratio ([tissue]/[plasma]) of a drug at equilibrium is expected to be equivalent to unbound plasma-to-tissue fraction (fu_plasma/fu_tissue) (eq. 1) (Fichtl et al., 1991).

Following this expectation, attempts have been made to predict tissue partitioning from in vitro estimates of unbound plasma fraction and unbound tissue fraction with mixed success (Bickel and Gerny, 1980; Bickel et al., 1987; Schuhmann et al., 1987; Clausen and Bickel, 1993). Previous reports suggest that the dilution, homogenization, and incubation process necessary to determine unbound tissue fraction by equilibrium dialysis may disrupt or destroy intracellular components that contribute to distribution in vivo. For example, it has been suggested that disruption of the postnuclear fractions containing the acidic organelles (e.g., lysosomes) may explain reported under-predictions in the distribution of the basic lipophilic drugs imipramine, desipramine, chlorpromazine, and methadone from liver, lung, and kidney homogenates (Clausen and Bickel, 1993). Further supporting this hypothesis, accurate predictions have been reported for acidic and neutral drugs in these same tissues and also for the same basic lipophilic drugs in tissues with less abundant lysosomes (e.g., brain, muscle, and adipose tissue).

We have recently extended this approach to provide an indirect determination of brain penetration (Kalvass and Maurer, 2002). In the event that a drug freely diffuses across the blood-brain and blood-CSF barriers, it is reasonable to expect that CSF concentrations will be reflective of those in the brain extracellular fluid (de Lange and Danhof, 2002). As such, it is also reasonable to expect that the following relationships should be demonstrable for such drugs (A, B and C, below). In these relationships, B/P, CSF/P, and CSF/B represent the in vivo brain-to-plasma, CSF-to-plasma, and CSF-to-brain concentration ratios at equilibrium. Unbound fractions in plasma and brain are represented by fu_plasma and fu_brain, respectively.

Consistent with these expectations, we have previously reported such predictive correlations for nine proprietary CNS compounds for which independent evidence indicated that the prerequisite assumptions regarding blood-brain and blood-CSF barrier diffusion were met (Kalvass and Maurer, 2002). In that work, it was further demonstrated that, as expected, fu_plasma/fu_brain over-predicts the B/P of P-glycoprotein efflux substrates. However, no consistent pattern was observed with regard to the magnitude of disparity in relationships A, B, and C. This latter finding is not surprising given the complex factors governing drug distribution between CSF and extracellular fluid for such compounds (de Lange and Danhof, 2002). However, as suggested previously, patterns of disparity among these relationships for any particular compound may provide additional insight into the nature of the active processes influencing the CNS distribution process. In this work, we sought to demonstrate these relationships among 33 marketed CNS drugs.

Materials and Methods

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

Drug Selection. Drug inclusion and the associated rationales are the same as those used in the accompanying article (Doran et al., 2005). Ethosuxamide was excluded from this analysis due to a lack of analytical sensitivity necessary to estimate unbound plasma and brain fractions.

Animal Exposure Data. B/P, CSF/P, and CSF/B ratios from in vivo studies were obtained from the accompanying article (see Doran et al., 2005). Ratios used in this analysis were derived from the area under the curve measured up to 5 h after a 3 mg/kg subcutaneous dose to male FVB mice (except caffeine, administered at 5 mg/kg).

Nonspecific Binding Studies. A 96-well equilibrium dialysis apparatus was used to determine the plasma and brain free fraction for each drug (Banker et al., 2003). Spectra-Por 2 membranes with molecular cutoff of 12 to 14 kDa, obtained from Spectrum Laboratories Inc. (Rancho Dominguez, CA), were used for the dialysis. The membranes were conditioned in deionized water for 15 min followed by 30% ethanol for 15 min and 0.10 M sodium phosphate pH 7.4 buffer for 15 min. Fresh female FVB (wild-type) mouse blood and brain were obtained the day of the experiment. Fresh FVB mouse plasma and brain tissue was obtained on the day of the study. Brain samples were diluted with 2 volumes of 0.10 M sodium phosphate pH 7.4 buffer and homogenized via ultrasonic probe. Plasma and 1/3 brain homogenate were spiked with the test drug (1000 ng/g) and 150-μl aliquots were loaded into the 96-well equilibrium dialysis plate and dialyzed versus 150 μl of 0.10 M sodium phosphate pH 7.4 buffer. Equilibrium was achieved by incubating the 96-well equilibrium dialysis apparatus for 4.5 h in a 37°C reciprocating water bath (set at 155 rpm). Incubation time was based on prior experience with this system indicating that equilibrium is consistently achieved within 4.5 h (data not shown). After reaching equilibrium, the indicated aliquots of each matrix (Table 1) were taken from the 96-well equilibrium dialysis apparatus and added to HPLC vials containing an equivalent volume of dimethyl sulfoxide and 100 μl of acetonitrile. The appropriate amount of control buffer was added to the plasma and 1/3 brain homogenate samples, and the appropriate amount of either control plasma or control brain homogenate was added to the buffer samples to yield identical matrix between buffer and nonbuffer samples. A standard curve was made up in dimethyl sulfoxide over the concentration range of 1 nM to 2000 nM. The appropriate amount of control buffer and either control plasma or control brain homogenate was added to the standard samples as well as 100 μl of acetonitrile to yield anidentical matrix between samples and standards. The samples and standards were than vortexed and centrifuged, and the supernatant was assayed by liquid chromatography/tandem mass spectrometry. For plasma, the unbound fraction was determined as the ratio of concentrations determined in buffer and plasma. As previously described, the unbound fraction in undiluted brain was calculated via eq. 3, where D and fu_meas represent the fold dilution of brain tissue and the associated free fraction determined as the ratio of concentrations in buffer versus diluted brain tissue (Kalvass and Maurer, 2002).

Analysis of Drugs in Protein Binding Samples. All samples were quantified using HPLC-MS/MS as summarized in Table 1. Drugs were analyzed using either a PE-Sciex API-3000 triple quadrupole (Turbo Ionspray source 500°C, 81 psi/min nitrogen; PerkinElmerSciex Instruments, Boston, MA) or a PE-Sciex API-4000 triple quadrupole (Turbo V Ionspray source 700°C, 75 psi of nitrogen) mass spectrometer. Samples were injected (3-10 μl) using a CTC Analytics HTS Pal (with a 100-μl syringe) autosampler (CTC Analytics, Zwingen, Switzerland) onto either a Phenomenex Synergi Max-RP 2.0 × 30 mm, 4-μm column (Phenomenex, Torrance, CA) or a Primesphere 5-μm HC-C18 2.0 × 30 mm column (Phenomenex) maintained at 50°C. Analytes were eluted with a high-pressure linear gradient program, produced by two Shimadzu LC-10ADVP binary pumps and a 10-μl static mixer, consisting of methanol (solvent A), 10 mM ammonium acetate (containing 1% 2-propanol), ∼6.8 pH (solvent B), and acetonitrile (solvent C). The total run time was 3.0 min for drugs run on the API 3000 and 1.5 min for drugs run on the API 4000. For drugs run on the API 3000, the initial mobile phase conditions were held for 0.5 min, ramped to an intermediate condition between 0.5 and 2 min, held there for 0.5 min, then returned to the initial condition in a single step and held there for 0.5 min to allow for re-equilibration prior to the next injection (Table 1). During mobile phase ramping, the flow rate was also ramped from 500 to 750 μl/min. For drugs run on the API 4000, the initial mobile phase conditions were held for 0.25 min, ramped to an intermediate condition between 0.25 and 1.0 min, held there for 0.25 min, then returned to the initial condition in a single step and held there for 0.25 min to allow for reequilibration prior to the next injection (Table 1). During mobile phase ramping, the flow rate was also ramped from 750 to 1500 μl/min. Source parameters were optimized for each drug and were measured using multiple reaction monitoring with either positive or negative ionization. Nonlinear standard curves (quadratic with 1/x² weighting) were used for analysis of the equilibrium dialysis.

Results

Estimates of fu_brain, fu_plasma, and fu_plasma/fu_brain ratio varied by 955-, 102-, and 26-fold, respectively, among the 33 drugs examined (Table 2; Fig. 1). The ranges of fu_plasma estimates were similar between the nine weakly acidic/neutral drugs (i.e., caffeine, carbamazepine, carisoprodol, diazepam, meprobamate, midazolam, phenytoin, thiopental, and zolpidem) and the remaining 24 basic drugs. The most extensive binding to brain tissue was observed among the basic drugs, with several displaying a fu_brain less than 0.01 (i.e., clozapine, fluvoxamine, cyclobenzaprine, haloperidol, nortriptyline, paroxetine, fluoxetine, chlorpromazine, and sertraline). As such, the range of fu_brain estimates was far greater among basic drugs (i.e., 0.00066-0.63) than that observed for the weakly acidic/neutral drugs (i.e., 0.0272-0.52). This, in turn, resulted in a greater range of fu_plasma/fu_brain ratios for basic drugs (0.64-16.7) than was observed for the weakly acidic/neutral drugs (1.00-2.79).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

Relationship between brain/plasma ratio and fu_plasma/fu_brain ratio (A), CSF/brain ratio and fu_brain (B), and CSF/plasma ratio and fu_plasma (C) for the 33 CNS drugs examined in FVB mice. Solid symbols (•) represent the mean data observed for the 24 basic CNS drugs. Open symbols (○) represent the mean data observed for the 9 weakly acidic/neutral CNS drugs. Solid line represents that of unity and dashed lines represent 3-fold boundaries.

Overall, 23 of the examined drugs (i.e., 70%) achieved B/P, CSF/B, and CSF/P exposure ratios that were all within 3-fold of fu_plasma/fu_brain ratio, fu_plasma, and fu_brain, respectively (Fig. 1; Table 3, class 1). Mouse fu_plasma/fu_brain ratios over-predicted B/P ratios by 16-, 6-, and 4-fold for sulpiride, thiopental, and risperidone, respectively (Table 3, class 2). For these drugs, fu_plasma over-predicted CSF/P to a similar degree, whereas fu_brain predicted CSF/B to within 3-fold. Similarly, fu_plasma/fu_brain ratios over-predicted B/P ratios of carbamazepine, midazolam, phenytoin, and zolpidem by 4-, 7-, 4-, and 4-fold, respectively (Table 3, class 3). However, in this case, fu_brain under-predicted CSF/B to a similar degree, whereas fu_plasma predicted CSF/P to within 2-fold. Although the B/P ratios of buspirone and caffeine were predicted to within 2-fold by fu_plasma/fu_brain ratios, fu_plasma and fu_brain over-predicted CSF/P and CSF/B between 5- and 14-fold (Table 3, class 4). The largest over-prediction of B/P by fu_plasma/fu_brain was observed for 9-hydroxyrisperidone (i.e., 64-fold; Table 3, class 4). In this case, fu_plasma over-predicted CSF/P and fu_brain under-predicted CSF/B to a similar extent.

Discussion

For purposes of discussion, drugs were grouped into one of four general patterns of behavior that were observed when comparing unbound fraction data with in vivo exposure ratios (Table 3). Unbound fraction data were considered to be predictive of in vivo exposure ratios if the two values were within 3-fold. This criterion was chosen to allow for differences due to random experimental error and for actual differences that would be considered to be of little pharmacologic or pharmacokinetic consequence. The drugs falling into the four general classes of behavior and the implications of that behavior are discussed below.

Class 1. The majority of the drugs studied (23 of 33) followed a pattern of behavior in which fu_brain, fu_plasma, and fu_plasma/fu_brain ratios were predictive of CSF/B, CSF/P, and B/P, respectively (Table 3). Since this pattern would be expected under conditions in which a drug freely diffuses across both the blood-brain and blood-CSF barriers, the fact that the majority of the studied CNS drugs fall into this classification is not surprising. These results indicate that despite the greater than 50-fold range in absolute brain-to-plasma exposure ratios, free drug exposures in plasma and brain are equivalent for the drugs in this class. In other words, the large differences in brain-to-plasma ratio observed among these drugs are determined by differences in relative nonspecific plasma and tissue binding, not blood-brain barrier penetration. These results highlight the potential pitfalls of approaches that are associated with the implicit assumption that larger B/P values are indicative of a greater degree of brain penetration (Lombardo et al., 1996; Salminen et al., 1997; Norinder et al., 1998; Clark, 1999; Luco, 1999; Feher et al., 2000; Kaznessis et al., 2001; Keseru and Molnar, 2001; Platts et al., 2001). Because B/P is determined largely by nonspecific binding, efforts to optimize this parameter may actually lead to an unproductive or counterproductive design of drugs that are unnecessarily basic, lipophilic, and simply have a greater degree of nonspecific partitioning into brain tissue.

Class 2. Risperidone, sulpiride, and thiopental followed a pattern of behavior in which fu_brain was predictive of CSF/B, whereas fu_plasma and fu_plasma/fu_brain over-predicted CSF/P and B/P to a similar extent (Table 3). This pattern would be expected in the presence of a similar degree of impairment in diffusion across the blood-brain and blood-CSF barriers. In such an event, CSF exposure would fall short of free plasma exposure but would be equivalent to free brain exposure. Consistent with this scenario, the polar surface areas of sulpiride (110 Ų) and thiopental (90 Ų) are outside the 95% confidence interval of polar surface areas determined for the 22 drugs in class 1 (i.e., 30-50 Ų) and class 3 (i.e., 29-61 Ų) (Ertl et al., 2000). These values are also above the upper limits for optimal CNS penetration that have been suggested from the examination of large numbers of CNS drugs that have reached phase II efficacy studies (van de Waterbeemd et al., 1998; Ertl et al., 2000). It is important to note, however, that if the disparity between unbound fraction and in vivo exposure is attributable to less than optimal tissue diffusion, then equilibrium in unbound drug concentrations may simply be delayed. As such, the truncated 5-h brain, plasma, and CSF area under the curve estimates used herein may not accurately reflect the concentration ratios that would be achieved for these drugs under equilibrium conditions. Such a delayed equilibrium has been previously reported from microdialysis studies on thiopental (Mather et al., 2000). The other drug in this class, risperidone, also has a reasonably high polar surface area (64 Ų) as compared with the drugs in the other classes and is also now known to be a P-glycoprotein efflux substrate (see Doran et al., 2005). As such, a combination of less than optimal transcellular diffusion and blood-brain barrier efflux may be responsible for its behavioral pattern being borderline between class 2 and class 3.

Class 3. Similar to those in class 2, fu_plasma/fu_brain over-predicted B/P for drugs in this class. In contrast, fu_brain under-predicted CSF/B, whereas fu_plasma was predictive of CSF/P. In each case, the fold over-prediction of B/P was similar to the fold under-prediction of CSF/B. This pattern would be expected in the presence of a distributional impairment that is specific to the blood-brain barrier. In such an event, CSF may accurately reflect free plasma concentrations while providing an overestimation of drug concentrations in the extracellular fluid of the brain. Consistent with the blood-brain barrier impairment implied by these data, increases in the brain extracellular fluid levels of both carbamazepine and phenytoin have been reported in the presence of the MRP inhibitor probenecid and/or in MRP2-deficient TR^- rats (Potschka et al., 2001, 2003; Potschka and Loscher, 2001; Loscher and Potschka, 2002) despite equivalent free plasma and CSF concentrations of these drugs in rats (Chou and Levy, 1981; Sokomba et al., 1988). In both cases, the observed impairment attributed to MRP2 efflux at the blood-brain barrier was enough to account for the differences between the unbound fractions and in vivo exposures reported in this analysis.

In the case of midazolam, further application of this approach to rats suggests that the detected blood-brain barrier impairment may be unique to mice. For example, the unbound fractions of midazolam in mouse and Sprague-Dawley rat plasma are very similar at 0.046 ± 0.004 and 0.0357 ± 0.0010, respectively. The unbound fractions of midazolam in mouse and rat brain are also very similar at 0.0272 ± 0.0018 and 0.0219 ± 0.0014, respectively. As such, the mouse and rat brain-to-plasma ratios predicted by these unbound fractions are very similar at 1.69 ± 0.08 and 1.63 ± 0.05, respectively. However, the actual in vivo B/P estimates of midazolam in mice and rats are approximately 10-fold different at 0.23 (Table 2) and 2.3 (Arendt et al., 1987; Mandema et al., 1992). These data together indicate that the B/P in rats is governed primarily by nonspecific binding, whereas some additional and limiting factor limits the brain partitioning of this drug in mice.

Unlike midazolam, the B/P of zolpidem is very similar between mice (Table 2) and rats (Garrigou-Gadenne et al., 1991). Therefore, it seems unlikely that the impairment in blood-brain barrier penetration implicated by unbound fraction data is unique to the mouse. It is also unlike carbamazepine and phenytoin in that a blood-brain barrier efflux mechanism has not been identified for this drug. Further investigation will be required to understand the underlying mechanism responsible for its pattern of behavior matching the other drugs in this classification.

Class 4. Drugs in this class followed a pattern of behavior that suggests a complex equilibrium between drug concentrations in the CSF and extracellular fluid of the brain. For example, the B/P ratios of buspirone and caffeine were very accurately predicted by fu_plasma/fu_brain, whereas CSF/P and CSF/B were over-predicted to a similar degree by fu_plasma and fu_brain (Table 3). In the case of 9-hydroxyrisperidone, fu_plasma/fu_brain over-predicted B/P by 64-fold. This impairment is qualitatively consistent with the 17-fold difference in B/P ratios between FVB and MDR1a/1b (-/-, -/-) mice. Because the absolute B/P in FVB mice approximated the fractional blood volume of the brain (∼5%), the latter estimate of impairment may represent an underestimate and highlights one of the limitations of the MDR1a/1b (-/-, -/-) mouse model (see Doran et al., 2005). As with buspirone and caffeine, a simple relationship between the over-prediction of B/P ratio and the over- or under-predictions of CSF/B and CSF/P was not evident.

In summary, these findings illustrate that the B/P ratio of the examined CNS drugs is largely determined by unbound plasma and brain fraction, both of which can be accurately estimated from equilibrium dialysis studies. Several of the examined drugs with disparities between B/P and fu_plasma/fu_brain of greater than 3-fold are known to have additional properties that are expected to influence brain distribution (e.g., high polar surface area, and/or efflux). These findings are also consistent with previous reports that CSF exposures do not necessarily reflect free brain exposure (e.g., classes 3 and 4 in Table 3). Lastly, these results support our previous assertions that unbound plasma and tissue fractions can be used as a model for examining brain penetration. In contrast to transgenic models, this approach is neither mechanism- nor species-specific. As such, it is particularly useful in drug discovery programs pursuing CNS targets.