Raffaele Longhi, Silvia Corbioli, Stefano Fontana, Federicia Vinco, Simone Braggio, Lydia Helmdach, Jürgen Schiller, and Hinnerk Boriss

Transcription

1 /11/ $20.00 DRUG METABOLISM AND DISPOSITION Vol. 39, No. 2 Copyright 2011 by The American Society for Pharmacology and Experimental Therapeutics 36095/ DMD 39: , 2011 Printed in U.S.A. Brain Tissue Binding of Drugs: Evaluation and Validation of Solid Supported Porcine Brain Membrane Vesicles (TRANSIL) as a Novel High-Throughput Method Raffaele Longhi, Silvia Corbioli, Stefano Fontana, Federicia Vinco, Simone Braggio, Lydia Helmdach, Jürgen Schiller, and Hinnerk Boriss GlaxoSmithKline Neurosciences Centre of Excellence for Drug Discovery, Verona, Italy (R.L., S.C., S.F., F.V., S.B.); Sovicell GmbH, Leipzig, Germany (L.H., H.B.); and Universität Leipzig, Medizinische Fakultät, Institut für Medizinische Physik und Biophysik, Leipzig, Germany (J.S.) Received August 25, 2010; accepted November 11, 2010 ABSTRACT: Estimating the unbound fraction of drugs in brain has become essential for the evaluation and interpretation of the pharmacokinetics and pharmacodynamics of new central nervous system drug candidates. Dialysis-based methods are considered to be accurate for estimating the fraction unbound in brain; however, these techniques are hampered by a low throughput. In this study, we present a novel, matrix-free, high-throughput method for estimating the unbound fraction, based on a sample pooling approach combining the TRANSIL brain absorption assay with liquid chromatography-mass spectrometry. The base measurement of the TRANSIL approach is the affinity to brain membranes, and this method is used directly to predict the free fraction in brain. The method was evaluated by comparing the free fraction of drugs in brain [f u,brain (%)] obtained using the TRANSIL brain absorption assay and equilibrium dialysis methods for a test set of 65 drugs (27 marketed and 38 proprietary drugs). A good correlation (r 2 > 0.93) of f u,brain (%) between the TRANSIL brain absorption assay and equilibrium dialysis was observed. Moreover, we compared the lipid composition of rat and porcine brain and analyzed the influence of the brain albumin content on brain tissue binding measurement. The comparison of the lipid composition indicated only minor differences between rat and porcine brain, and albumin appears to have a low impact on brain tissue binding measurements. The TRANSIL brain absorption assay with sample pooling methodology not only significantly reduces the biological matrix required but also increases the throughput, compared with the conventional dialysis methods. Introduction Estimation of the unbound fraction of drug in brain provides a valuable tool for identifying new therapies for central nervous system (CNS) diseases. For many years, emphasis has been placed on the magnitude of the brain/plasma ratio (Kp); although in many cases, brain total concentration (C brain ) has no or, at best, poor correlation with a mechanistic pharmacodynamic response. Today, C brain, measured in in vivo experiments, is corrected for the fraction of drug unbound determined by in vitro experiments to obtain an estimate of the brain unbound concentration (C u,brain ). This method has been successfully demonstrated across a range of CNS targets to yield a Article, publication date, and citation information can be found at doi: /dmd much better correspondence with receptor occupancy and pharmacodynamic endpoints (Boriss, 2010; Read and Braggio, 2010). A variety of in vivo, in situ, and in vitro techniques are currently available for estimating C u,brain. Brain microdialysis is the most realistic and accurate measure of free concentration in situ; however, it requires extensive resources, cannot be easily applied to lipophilic compounds, and has a very low throughput. The brain slice method is the second most realistic approach because it maintains the tissue integrity, unlike the brain homogenate methods (Becker and Liu, 2006; Friden et al., 2007). However, this approach is time-consuming and costly and is not amenable to a high-throughput format. A more rapid alternative is 96-well equilibrium dialysis with rat brain homogenate (Kalvass and Maurer, 2002; Maurer et al., 2005; Becker and Liu, 2006; Liu et al., 2006, 2009; Summerfield et al., 2006, 2007, 2008; Friden et al., 2007; Kalvass et al., 2007). At present, the dialysis-based assay is considered to be an accurate method for the ABBREVIATIONS: CNS, central nervous system; Kp, brain/plasma ratio; C brain, brain total concentration; C u,brain, brain unbound concentration; f u,brain, free fraction of drugs in brain; GSK, GlaxoSmithKline; HPLC, high-performance liquid chromatography; PBS, phosphate-buffered saline; CSF, cerebrospinal fluid; LC-MS/MS, liquid chromatography-tandem mass spectrometry; DMSO, dimethyl sulfoxide; UPLC/MS/MS, ultraperformance liquid chromatography-tandem mass spectrometry; RSA, rat serum albumin; HPTLC, high-performance thin-layer chromatography; MALDI-TOF-MS, matrix-assisted laser desorption and ionization time-of-flight mass spectrometry; SM, sphingomyelin; PC, phosphatidylcholine; PS, phosphatidylserine; PA, phosphatidic acid; TLC, thin-layer chromatography; PI, phosphatidylinositol; HBA, number of H-bond acceptors; HBD, number of H-bond donors; PSA, polar surface area; clogp, predicted log (octanol/water) partition coefficient; f u, free fraction of drugs; CV, coefficient of variation. 312

2 DETERMINING BRAIN TISSUE BINDING OF DRUGS IN DRUG DISCOVERY 313 rapid estimation of the free fraction of drugs in brain (f u,brain ). However, the equilibrium dialysis method is still a time-consuming assay with limited sample throughput capacity, and although improvements have been achieved by sample pooling approaches (Fung et al., 2003; Wan and Rehngren, 2006; Wan et al., 2007), the throughputs of dialysis-based methods are still limited by long equilibration times (4 6 h) and the readily available supply of brain homogenate. As a result, the development of a matrix-free, high-throughput method for assessing f u,brain would be highly desirable. This study describes a novel method, based on brain lipid membrane vesicles stabilized on silica beads (TRANSIL brain absorption), which enables rapid and convenient separation of the test compound from the biological phase, facilitating the following quantification step and thereby reducing time and costs. The TRANSIL brain absorption kit (Sovicell GmbH, Leipzig, Germany) measures the affinity of test substances to the porcine lipid membrane by quantifying the remaining free concentration in the buffer phase after incubation with TRANSIL beads. The brain membrane affinity is then used directly to predict the free fraction in brain. A schematic representation of the assay steps is shown in Fig. 1. The main advantage of the TRANSIL method is its 2-min equilibration time in comparison with the 4 to 6 h required by dialysis-based methods. Rapid equilibration is achieved through a very large TRANSIL bead surface area of more than 1 m 2, whereas dialysis systems typically equilibrate across a membrane area of less than 1 cm 2. Brain tissues are known to contain substantial amounts of albumin (Glees, 1988), whereas this newly developed brain membrane affinity assay uses only lipids. Because the affinity of many drugs to albumin can be substantial, we analyzed the albumin content of rat brain and compared the contribution of albumin binding to total brain tissue binding. Another factor that could influence the binding of drugs to brain tissue is the lipid composition of membranes. Di et al. (2008) reported only minor species differences in the brain lipid composition. However, considering that rat brain homogenate is commonly used with dialysis-based methods, whereas the TRANSIL assay is based on porcine brain lipid membranes, the lipid composition of rat and porcine brain was analyzed in detail to verify the comparability of the two brain tissue binding assays. The TRANSIL method was evaluated by a comparison of measured f u,brain (%) obtained using the TRANSIL brain absorption assay and equilibrium dialysis methods for a test set of 65 drugs (27 marketed and 38 proprietary drugs). The set of marketed drugs and GlaxoSmith- Kline (GSK) proprietary compounds was chosen to have CNS target activity, to be structurally diverse and to represent a wide range of physicochemical properties. Most of the compounds were basic or neutral, one was acidic, and two were amphoteric compounds. In addition, single compound measurements and a cocktail approach (pooling four compounds in one sample) have been evaluated and discussed for additional improvement of the throughput of the assay. Materials and Methods Materials. The marketed drugs and all other chemicals were of analytical grade and obtained from Sigma (Milan, Italy). The other reagents and solvents used were of analytical or high-performance liquid chromatography (HPLC) grade. Preparation of Rat Brain Extracts. Brains were isolated from male Wistar rats weighing between 220 and 250 g. For the preparation of homogenates without capillary blood, brains were perfused with Dulbecco s phosphatebuffered saline (PBS) before surgery. Brain tissue was homogenized in PBS or cerebrospinal fluid (CSF) surrogate, yielding a final dilution of 1:3, using a SONOPLUS HD 2070 (Bandelin; Bandelin GmbH, Germany) sonicator or a Covaris SX2 AFA homogenizer (Covaris Inc., Woburn, MA). Determination of Albumin and Protein Content in Rat Brain. The albumin content in rat brain homogenate was determined using an enzymelinked immunosorbent assay with anti-rat-albumin antibodies (Bethyl Laboratories, Montgomery, TX), according to the manufacturer s instructions. Five biological replicates of brain homogenate (dilution 1:6000) were used, and each replicate was measured in duplicate. A modified Bradford assay (Bradford, 1976) (Pierce Chemical, Rockford, IL) was used to determine the total protein content of rat brain homogenates. Four biological replicates of brain homogenate were used at a tissue dilution of 1:100 to ensure that quantification is within the linear range of the assay. Equilibrium Dialysis. Brain fractions unbound were determined using equilibrium dialysis in a 96-well format as described previously (Kalvass and Maurer, 2002). Brain homogenates were prepared using 2 ml/g CSF surrogate (7.30 g/l NaCl, mg/l KCl, mg/l MgCl 2, mg/l CaCl 2, and mg/l Na 2 HPO 4 7H 2 O, ph 7.4). Dialysis membranes (12 14-kDA cutoff) obtained from Spectrum Laboratories Inc. (Rancho Dominguez, CA) were conditioned in water (HPLC grade) for 60 min, followed by 20 min in 20% ethanol, and 15 min in CSF surrogate. Assay compound concentrations were 5 M. Diluted brain homogenate was spiked with the compound of interest, and 150- l aliquots were loaded into the 96-well equilibrium dialysis apparatus and dialyzed against an equal volume of CSF surrogate. Equilibrium was achieved by incubating the 96-well equilibrium dialysis apparatus in a temperature-controlled incubator at 37 C for 5 h, using an orbital shaker at 125 rpm. At the end of the incubation period, 50- l FIG. 1. The TRANSIL brain absorption assay.

3 314 LONGHI ET AL. aliquots of brain homogenate and buffer were transferred to a 96-well plate, and the composition in each tube was balanced with control fluid to equalize the brain and buffer volumes. Sample extraction was performed by addition of 400 l of acetonitrile containing internal standard (rolipram). The samples were then vortexmixed and centrifuged, and 100 l of supernatants was transferred into a 96-well plate, diluted with 200 l of 16% acetonitrile/water, and analyzed by liquid chromatography-tandem mass spectrometry (LC-MS/MS). Brain unbound fraction was calculated from the ratio of analyte/internal standard peak area ratios determined in buffer versus brain homogenate samples using eq. 1, which accounts for the effect of tissue dilution on unbound fraction (Kalvass and Maurer, 2002): 1/D f u-undiluted 1/f u-diluted 1 1/D where D is the dilution factor in brain homogenate and f u-diluted is the measured free fraction of diluted brain tissue. Drug Preparation. For the TRANSIL brain absorption assay, drugs were prepared in 10-fold concentrated stock solutions in dimethyl sulfoxide (DMSO) such that the final DMSO concentration was 1%. For discrete incubations, each compound was prepared as an individual stock solution at 5 M in 50 mm phosphate buffer (ph 7.4) containing 10% DMSO to obtain 0.5 M final assay concentration in TRANSIL wells. For cassette incubation, individual drugs were prepared as 5 M stock solutions in 50 mm phosphate buffer (ph 7.4) containing 10% DMSO. A set of four drugs was pooled for cassette incubation, and concentrations were adjusted to obtain 0.5 M as final assay concentrations for each compound in the cassette. Brain Membrane Affinity. Brain membrane affinity was determined using the TRANSIL brain absorption assay kit, which is based on reconstituted porcine brain membrane vesicles on solid support (Fig. 1). The membrane affinity of a test substance to TRANSIL brain absorption beads was determined using eight vertical wells in the assay plate. Wells 1 and 8 served as references to control unspecific binding and therefore contained only buffer. The remaining six wells in each vertical row contained decreasing defined quantities of lipid bound to beads ranging from 0.7 to 7 l of lipids per milliliter of buffer. Compounds were incubated for 2 min at room temperature in a concentration of 0.5 M for discrete incubation or 2 M for cocktail incubation (0.5 M each compound). The TRANSIL brain absorption assay was automated on a Hamilton Microlab STARlet instrument equipped with 96-head disposable tips (Hamilton Bonaduz AG, Bonaduz, Switzerland). The automation consisted of the following five steps: 1) compound addition, after thawing the TRANSIL plate at room temperature, 45 l of test compounds (discrete or cocktail) was added to each well of a column of 8 wells; 2) mixing, Hamilton 96-head tips mixed compounds with the bead suspension by 10 cycles of aspiration and dispensed 110 l of solution; 3) incubation, the beads and compounds suspensions were incubated (room temperature) for 2 min; 4) separation of beads and buffer, the plate was centrifuged for 10 min at 750g to eliminate the solid supported brain membrane beads from the buffer containing the unbound compounds; 5) sampling of supernatant, 50- l aliquots of the supernatant were sampled for analysis and added to 50 l of acetonitrile containing internal standard (rolipram). The remaining compound concentration in the supernatant of each well was determined as an analyte/internal standard peak area ratio by using ultraperformance liquid chromatographytandem mass spectrometry (UPLC/MS/MS). After quantification of the test substance in both the reference and samples, the brain membrane affinity logma brain, which is used to estimate the brain free fraction, was calculated. The affinity to brain membrane is defined as the concentration of drug in membrane, c l, over the concentration of drug in buffer, c b : MA brain c l (2) c b The brain membrane affinity is calculated using the mass balance equation: (1) n t c b V b c l V l (3) where n t is the total amount of compound in each well; and V b and V l are the buffer and lipid membrane volumes, respectively, present in each well. Equation 3 is rearranged such that the membrane affinity can be determined from the slope of plotting the ratio of total amount of drug (n t ) over remaining compound concentration in the supernatant (c b ) against the lipid membrane volume present in each well: n t c l V c b c l V b MA V l V b (4) b The brain free fraction was predicted according to eq. 5 f u,brain exp(b l log MA brain b 2 ) (5) where b 1 and b 2 are parameters supplied by Sovicell GmbH and calculated based on in-house validation. Estimation of Albumin Binding in Brain. A TRANSIL rat serum albumin (RSA) binding kit (Sovicell GmbH) was used to determine the affinity (K D )of the drugs to rat serum albumin. Compounds were incubated in 2 M concentrations with six different concentrations (ranging from 55 to 110 M) of randomly immobilized RSA in Dulbecco s PBS. The remaining compound concentration in the supernatant was quantified via LC-MS/MS and used to determine K D. The TRANSIL brain absorption assay kit was used to determine the K D of the tested drugs to the lipid membranes. The lipid content in each TRANSIL well was expressed in concentrations ranging from 0.87 to 8.7 mm. This affinity model is different from the one used to calculate brain tissue binding (eq. 5) and may yield slightly different binding values. However, this model provides a means for prediction of the free fraction combining affinities to different membrane components, such as lipids and proteins, using the modified equation of Kremer et al. (1988): f b [RSA] RSA [Lipid] Lipid K D K D where f b represents the fraction of test article bound to RSA and lipids, K D is the dissociation constant, [RSA] is the concentration of albumin in rat brain, and [Lipid] is the concentration of lipids in rat brain. Dissociation constants K D were calculated according to the definition: K D [A] [P] (7) [AP] where [AP] is the concentration of drug A bound to the protein P and where [A] denotes the free concentration of drug and [P] denotes the free concentration of protein. The free concentration of drug can also be expressed as follows: (6) A f u A AP (8) The outcome of eq. 8 is entered into that of eq. 7, then the resulting equation can be simplified by considering that the fraction bound is defined as follows: f b AP / A AP (9) Thus, a linear model is obtained that can be fitted to the free drug concentrations in the TRANSIL wells: f b 1 [P] (10) f u K D Note that this equation requires that the concentration of the protein-drug complex [AP] should be much smaller than the total protein concentration in each well. Sample Quantification. The samples were analyzed using an Acquity UPLC system equipped with a thermostated autosampler and column compartment (Waters, Milford, MA), which was coupled to a Quattro Premier triple quadrupole mass spectrometer (Waters) and equipped with an electrospray ionization source. Chromatographic separation was performed on a Waters BEH C18 column ( mm, 1.7- m particle size) at 60 C using a flow rate of 0.4 ml/min. The mobile phase consisted of the following: phase A, water containing 0.1% (v/v) formic acid; and phase B, acetonitrile containing 0.1% (v/v) formic acid. The HPLC gradient started at 95% mobile phase A and

4 DETERMINING BRAIN TISSUE BINDING OF DRUGS IN DRUG DISCOVERY 315 was held for 0.4 min. Mobile phase B was increased linearly to 95% over 0.2 min and held at 95% for 0.2 min. The total LC-MS/MS run time was 1 min. Autosampler temperature was kept at 4 C. The injection volume was 5 l. The ion source temperature was set to 300 C with an ion-spray voltage of 2.1 kv. Multiple reaction monitoring was used to monitor the precursor-to-product ion transition for all analytes. To avoid isobaric clashes in the analysis of cocktail samples, compounds were not included in the same cassette group if they had molecular mass differences of 0, 1, 2, and 18 atomic mass unit. Rolipram solutions were used as internal standards during the analysis. Analysis of Brain Lipid Composition. Lipid extractions of the full brain samples were performed immediately after surgery according to a method described previously by Dyer and Bligh (1959). Lipids were characterized by high-performance thin-layer chromatography (HPTLC) and matrix-assisted laser desorption and ionization time-of-flight mass spectrometry (MALDI- TOF-MS) as described previously by Fuchs et al. (2007). We used lysophosphatidylcholine, sphingomyelin (SM), phosphatidylcholine (PC), phosphatidylserine (PS), phosphatidic acid (PA), phosphatidylethanolamine, and phosphatidylglycerol as lipid standards. PA serves as the TLC standard for phosphatidylinositol (PI) because it has similar properties and is more readily available. High-Performance Thin-Layer Chromatography. Selected brain lipid extracts were subjected to HPTLC before MALDI-TOF-MS. Samples of the extracts (0.5 l), corresponding to an overall amount of approximately 12.5 g of lipids, were applied onto HPTLC silica gel 60 plates (10 10 cm with aluminum backs; Merck, Darmstadt, Germany) and developed in vertical TLC chambers using CHCl 3 /ethanol/water/triethylamine [35:35:7:35 (v/v/v/v)] as the mobile phase for the separation of phospholipids. Lipids were visualized by spraying with a solution of primuline (Direct Yellow; Sigma Seelze, Germany) according to a method described previously by White et al. (1998). Upon excitation by UV light (366 nm), individual lipids became detectable as colored spots. These spots were assessed by using a digital image system in combination with the program Argus X1 delivered by BioStep (Jahnsdorf, Germany). MALDI-TOF Mass Spectrometry. Total brain lipid extracts and individual lipid fractions were investigated by MALDI-TOF mass spectrometry. A 0.5 M 2,5-dihydroxybenzoic acid solution in methanol (Schiller et al., 1999) was used in all cases. All samples were premixed with the matrix before deposition onto the MALDI target. All MALDI-TOF mass spectra were acquired on a Bruker Autoflex mass spectrometer (Bruker Daltonics, Bremen, Germany). The system uses a pulsed nitrogen laser, emitting at 337 nm. The extraction voltage was 20 kv, and gated matrix suppression was applied to prevent the saturation of the detector by matrix ions. One hundred twenty-eight single laser shots were averaged for each mass spectrum. The laser fluence was kept at approximately 10% above threshold to obtain optimum signal-to-noise ratios. To enhance the spectral resolution, all spectra were generated in the reflector mode using delayed extraction conditions. A more detailed methodological description is available from Fuchs and Schiller (2009). Statistical Analysis. For statistical comparison, we used a logit transform of the interval scaled binding data. This transform was of particular relevance because we analyzed compounds with tissue binding values that varied over more than 4 orders of magnitude. Moreover, without the use of a logit transform, only weakly bound drugs would have influenced the correlation, whereas the physiologically more relevant differences of strongly bound compounds would have been underestimated in the correlation analysis. Results Correlation of Brain Fraction Unbound between Equilibrium Dialysis and TRANSIL Assay. The test set used for this study was composed of 38 GlaxoSmithKline proprietary compounds with CNS activity and 27 marketed compounds (23 CNS acting and 4 belonging to other therapeutic areas). Table 1 lists the 65 compounds used, along with the number of H-bond acceptors (HBA), number of H-bond donors (HBD), polar surface area (PSA), clogp [predicted log (octanol/water) partition coefficient] and acid/base character, which was calculated using ACD/PhysChem Suite (version 11; Advanced Chemistry Development Inc., Toronto, Canada); moreover, we reported f u,brain (%) measured using either equilibrium dialysis or the TRAN- SIL system. The brain tissue binding data for GSK compounds and for 17 marketed compounds have been generated in-house, whereas, for the remaining 10 marketed compounds, f u,brain (%) values were collected from published data (Maurer et al., 2005; Summerfield et al., 2007). A wide range of physicochemical properties were represented, and these varied as follows: molecular mass from 194 to 624 Da, lipophilicity (clogpacd) from 0.6 to 6.2, PSA from 3 to 26, HBA from 0 to 5, and HBD from 0 to 4. Most of the marketed compounds were bases (18 of 27) and neutral (9 of 27). GSK compounds were neutral (23 of 38), bases (8 of 38), and zwitterionic (2 of 38), and one was an acid (for four compounds, the ionized form was not calculated). Equilibrium dialysis f u,brain (%) values were obtained in triplicate in an intraday experiment. The TRANSIL brain tissue binding assay was designed such that one compound was incubated in six wells with increasing immobilized brain membrane surface area. Therefore, the assay provides a 6-fold determination of the membrane affinity. Mean f u,brain (%), S.D., and coefficient of variation (CV%) are reported in Table 1. f u,brain (%) values, ranging from 0.05 to 63%, varied by more than 4 orders of magnitude. For the equilibrium dialysis assay, recovery was determined by comparing the sum of peak areas found in the buffer and brain homogenate compartment after the dialysis period with the peak area of a nondialyzed spiked brain homogenate sample. For the compounds reported in this study, recovery was always in the range of 80 to 120%, which is generally accepted for this type of study. For the TRANSIL assay, compounds displaying low solubility in pure buffer solution or high nonspecific binding might result in lower concentrations in the reference wells (1 and 8) than in lipid-based wells (2 7). In this case, the reference value was determined from the first lipidbased well. A strong correlation (r , n 65) was observed between the brain free fraction obtained with equilibrium dialysis and brain free fraction determined using the TRANSIL kit (Fig. 2). Overall, 56 of the tested compounds in TRANSIL (i.e., 86%) gave f u,brain (%) values that were within 2-fold of the f u,brain (%) measured by using equilibrium dialysis. Of the nine compounds with discrepancies higher than 2-fold, five compounds (i.e., 8%) were underpredicted and four (i.e., 6%) were overpredicted by TRANSIL; however, it should be pointed out that the largest discrepancy observed never exceeded 3.3-fold (Table 1). Considering the wide range of f u,brain (%) measured, both assays gave good precision in the intraday experiments, with a CV% always within 20%, except for two compounds. Even for a subset of highly brain bound compounds [f u,brain 2% (Fig. 3)], the assay precision is still acceptable, with a CV% less than 20% for both assays, a value that is generally accepted for screening purposes (Fig. 3). Rat and Porcine Brain Lipid Composition. To investigate the difference in lipid composition between rat and porcine brain, the lipids of brain extracts from both species were separated using TLC. The corresponding thin-layer chromatograms of a known reference mixture of different phospholipids and of the brain extracts from rat and porcine are shown in Fig. 4. After TLC separation, phospholipids were individually characterized by MALDI-TOF, essentially as described previously by Fuchs et al. (2007). The results of MALDI- TOF-MS are summarized in Table 2, and, based on these data, it can be seen that no relevant differences in lipid composition were detected between rat and the pig brain. Hence, major species differences in binding to brain lipids can be considered unlikely. Brain Albumin and Lipid Affinity. To analyze the affinity of test compounds for brain lipids and brain albumin, the lipid and albumin concentrations in brain were determined. From six brains, an average

5 316 LONGHI ET AL. TABLE 1 List of proprietary CNS compounds and marketed drugs with their physicochemical properties and fraction unbound f u,brain (%) in rat brain tissue homogenate (equilibrium dialysis) and TRANSIL assay Acid/base was determined by ACD/PhysChem Suite, version 11. f u,brain (%) tissue homogenate (n 3) and f u,brain (%) TRANSIL (n 6) are shown as means with S.D. and CV%. Fold difference is the degree by which f u,brain (%) by equilibrium dialysis is overestimated or underestimated (in parentheses), with respect to data obtained using TRANSIL. Drug Name Molecular Mass HBA/HDB Acid/Base clogp PSA f u,brain (%) S.D. Tissue Homogenate CV% f u,brain (%) S.D. TRANSIL CV% Fold Change GSK /2 Base Trifluoperazine /0 Base (2.6) Thioridazine /0 Base GSK /2 Neutral (1.7) GSK /1 Base GSK /0 Neutral (2.3) GSK /2 Base (1.4) GSK /2 Base (1.1) Chlorpromazine /0 Base (S) (1.1) Paroxetine /1 Base (M) GSK /2 Base Fluoxetine /1 Base GSK /0 Neutral Nortriptyline /1 Base (M) GSK /2 Base Cyclobenzaprine /0 Base (M) (1.4) Fluvoxamine /1 Base (M) Amitriptylline /0 Base GSK /2 Neutral GSK /2 Neutral GSK /2 Neutral Hydroxyzine 375 2/1 Base (M) (2.1) Clozapine /1 Neutral Haloperidol /1 Base (1.3) GSK /2 Neutral GSK /1 Neutral GSK /1 Neutral (1.3) GSK /1 Neutral GSK /0 Base Midazolam /0 Neutral (1.8) GSK /1 Neutral Propranolol /2 Base GSK /1 Neutral (1.3) Diazepam /0 Neutral (2.3) GSK /2 Neutral GW /0 Zwitterion (1.2) GSK /4 Neutral (1.4) GSK /1 Acid Trazodone /0 Base GSK /2 Base GSK /0 Base GSK /0 n/c GSK /2 Base Phenytoin /2 Neutral GSK /0 n/c (1.4) Risperidone /0 Base GSK /0 Neutral GSK /3 Neutral (1.3) GSK /1 Neutral Carbamazepine /1 Neutral (1.1) GSK /0 n/c (1.2) Buproprion /1 Neutral Carisoprodol /2 Neutral (M) (1.5) Venlafaxine /1 Base Buspirone /0 Neutral (M) (1.1) GSK /0 n/c (1.7) Bupivacaine /1 Base GSK /3 Neutral (1.5) GSK /1 Neutral GSK /3 Neutral (1.1) GSK /3 Neutral Metoclopramide /2 Base GSK /0 Zwitterion Caffeine /0 Neutral (M) Sulpiride /2 Base (M) n/c, not calculated; (M), Maurer et al. (2005); (S), Summerfield et al. (2007). weight of 1.86 g and a dry weight of % was determined. The albumin content was determined by enzyme-linked immunosorbent assay at a 1:6000 brain extract dilution and was observed at g/ml in undiluted brain. Considering the albumin molecular mass of 66,500 Da, a rat brain albumin concentration of 0.8 M was calculated. From the Bradford assay for total protein, an estimate of

6 DETERMINING BRAIN TISSUE BINDING OF DRUGS IN DRUG DISCOVERY 317 FIG. 2. Correlation of f u,brain (%) between equilibrium dialysis and TRANSIL assay measurements (r ). The black line represents unit correlation, and dashed lines represent 2-fold boundaries. Gray circles represent proprietary compounds, and black circles represent marketed compounds mg/ml protein was obtained. The total lipid content of rat brain was estimated as 8.5% of the brain dry mass based on total brain dry mass measurements and brain total protein content analysis. This estimation results in a rat brain lipid concentration of 0.09 mg/ml. Considering an average phospholipid molecular mass of 750 g/mol (Harris and Gambal, 1963), a total lipid concentration of 120 M is estimated. For a test set of 16 marketed compounds, listed in Table 3, binding affinities to brain lipids and rat albumin were determined by TRAN- FIG. 3. Bar chart showing 19 highly brain tissue bound compounds [f u,brain (%) 2] with experimental variability (n 3 for equilibrium dialysis, n 6 for TRANSIL brain absorption assay).

7 318 LONGHI ET AL. FIG. 4. Image of a developed TLC plate containing samples from rat and pig brain lipid extract subsequent to staining with primuline. Migration of a standard lipid mixture is shown in lane d, whereas differently concentrated samples from porcine brain were analyzed for lanes a to c and rat brain samples for lanes e to g. In the standard mixture, PA was used instead of PI. SIL assay, and free fractions were calculated according to eq. 6, based on either lipid affinity only [K D - (lipid)] or lipid affinity plus rat albumin affinity [K D - (RSA)]. Both values were compared to the free fraction determined by equilibrium dialysis, using perfused rat brains, to avoid bias through albumin from the capillary blood. The difference in free fraction based on binding to lipids-only and binding to lipids and brain albumin was denoted by RSA. As shown in Table 3, RSA was calculated for compounds with a wide interval of free fractions, ranging from 0.16 to 84.43%, to evaluate the contribution of albumin in compounds with low lipid affinity. RSA was always below 0.03, which means a percent difference always below 1%, except for ibuprofen, tolbutamide, and warfarin, which displayed RSA values of 4.44, 6.73, and 1.39, respectively, corresponding to a percent difference of 14.6, 8, and 3.4%. Correlation of Brain Fraction Unbound between Discrete and Cocktail Incubation Using the TRANSIL Kit. To further increase the sample throughput, the possibility of assaying multiple compounds was investigated. With highly selective LC-MS/MS methods, multiple compounds can be monitored in one sample as long as the compounds have distinctive parent and fragment ion masses. A fourin-one cocktail approach was investigated in this study. The percent unbound values of the new chemical entities and commercial compounds were comparable when assayed individually or as a cocktail (Table 4). The percent unbound values obtained individually and by the cocktail approach using the TRANSIL kit are shown in Fig. 5. In general, a good correlation (r ) was observed between single and pooled compounds, despite a wide range of f u (%) (percent free fraction of drugs) and diverse chemical structures. Overall, all 25 compounds tested displayed f u,brain (%) values that were within 2-fold of the f u,brain (%) determined when assayed individually (Table 4). Such variability should be acceptable for screening purposes. Discussion The importance of considering free drug concentration in the brain is well recognized. Indeed, assessing brain free fraction in early drug discovery has had a significant impact on establishing mechanistic pharmacokinetics/pharmacodynamics links and on in vitro prediction of brain penetration (Kp f u,blood /f u,brain, where f u,blood is the free fraction of drugs in blood) across different species. Because the mechanism regulating the fraction unbound in brain homogenate is mainly nonspecific, correlation between the unbound brain fraction and lipophilicity has been well described in the literature (Suzuki and Kudo, 1990). In general, the observed trend offers a tool to predict brain tissue binding and may serve to flag problems in a lead series in the early phase of drug discovery projects. However, an implicit error of few f u (%) units might pose problems, especially in the late optimization phase when more accurate predictions for compound prioritization are needed (Wan et al., 2007). Thus, a methodology capable of assessing the brain tissue binding more rapidly than equilibrium dialysis will be especially beneficial for the lead optimization phase. In this study, we compared a novel, matrix-free, high-throughput method for estimating the unbound fraction, with the standard approach based on equilibrium dialysis using rat brain homogenate. This method, referred to as the TRANSIL assay, determines brain tissue

8 DETERMINING BRAIN TISSUE BINDING OF DRUGS IN DRUG DISCOVERY 319 TABLE 2 Lipid composition of rat and porcine brain extracts Overview of the peaks detected in the positive ion MALDI-TOF mass spectra of the rat brain and the porcine brain extract and the corresponding assignments. Only the most prominent peaks were listed. Dihydroxybenzoic acid was used in all cases. For details, see Materials and Methods. m/z Rat Brain Extract m/z Porcine Brain Extract Lipid Class PC 14:0/16:0 ( H ) PC 14:0/16:0 ( Na ) SM 18:0 ( H ) PC 16:0/16:0 ( H ) PG (16:0/16:0) SM 18:0 ( Na ) PC 16:0/16:0 ( Na ) PC 16:0/18:1 ( H ) PC 16:0/18:1 ( Na ) 788, PC 18:0/18:1 ( H ) PE 18:0/20:4 ( Na ) PC 18:0/18:1 ( Na ) PS 18:0/18:1 ( H ) SM 24:1 ( H ) Galactosyl-Cer 24:1/18:1 ( Na ) PS 18:0/18:1 ( Na ) SM 24:1 ( Na ) SM 24:0 ( Na ) Hydroxylated Galactosyl-Cer 24:0/18:1 ( Na ) PS 18:0/18:1 (-H 2Na ) PI 18:0/20:4 ( H ) PI 18:0/20:4 ( Na ) Cer, ceramide; PE, phosphatidylethanolamine; m/z, mass charge ratio. binding by estimating the affinity of drugs to reconstituted solid supported porcine brain membranes. The TRANSIL method was evaluated by comparison with equilibrium dialysis, which is the standard for the determination of plasma protein and brain free fraction. A strong correlation (r ) was observed for f u,brain obtained from both methods for a set of 65 structurally diverse compounds and, indeed, 87% of compounds displayed f u,brain (%) values that were within 2-fold of the f u,brain (%) measured by using equilibrium dialysis. Higher discrepancies were observed for nine compounds (13%), even if the largest difference observed never exceeded 3.3-fold, which is expected to be still acceptable for screening purposes. Moreover, this variability appears not to be related to any physicochemical property, or to the extent of free fraction, or to a systematic difference between the two methods. Indeed, three compounds exhibited higher f u,brain (%) values with TRANSIL than with equilibrium dialysis, and six exhibited higher f u,brain (%) values with equilibrium dialysis than with TRANSIL (see Table 1). In this case, a 3-fold difference is likely related to differences in sensitivity of LC-MS/MS methods used for quantifying compounds. The TRANSIL methodology can therefore be considered as a tool comparable to the dialysis methodology for measuring the brain free fraction. The high correlation between the unbound fraction, estimated by dialysis and by using the TRANSIL brain absorption assay, implicitly pointed out that there are no relevant differences between binding to rat and porcine brain. This result was also supported by our analysis of the lipid composition of rat and porcine brain, which was found to be consistent between the two species. This finding was a clear indication that the TRANSIL assay, which is based on porcine brain lipid membranes, can be used to predict not only the free fraction of drugs in rat brain but also the unbound fraction in humans, considering the excellent correlation noted between unbound brain fractions across species (Summerfield et al., 2008; Read and Braggio, 2010). Measurements of f u,brain (%) by using equilibrium dialysis methods are usually performed with homogenized unperfused brain. In contrast, the TRANSIL assay is based on the estimation of the membrane affinity considering only the partition into brain lipids as the main component to the tissue binding. However, because in brain tissue the weight of protein is comparable to that of lipid (protein 0.079; lipid 0.11) (Snyder et al., 1975), and because drug binding to albumin might be relevant, the contribution of albumin to brain binding was evaluated, especially for compounds with low lipid affinity, by estimating the binding affinities to brain lipids and RSA. Affinities to brain lipids and RSA were determined by TRANSIL assay, and the free fractions, based on lipid affinity only as well as the combined affinity to lipids and RSA, were calculated according to eq. 6 above. The free fractions determined via the K D approach yielded slightly different estimates than those via measuring membrane affinity. However, this approach, allowing for the combination of competing agents into a single binding model, might be used to predict the theoretical influence on brain free fraction of different binding agents that can act in concert. As shown in Table 3, most of the drugs had higher affinities for lipids than albumin, up to 2 orders of magnitude higher for amitriptylline, chlorpromazine, clozapine, fluoxetine, and TABLE 3 Influence of rat albumin on brain tissue binding Binding affinities to brain lipids K D (lipid) and RSA K D (RSA) were determined by TRANSIL experiments and the free fractions based on binding to lipids only f u (lipid). The values for binding to lipids and brain tissue albumin f u (lipid RSA) were calculated according to eq. 6. The free fraction determined by equilibrium dialysis f u (ED) was assessed using perfused rat brains. The difference in the unbound fraction between binding to lipids only and binding to lipids and brain tissue albumin is denoted by RSA. Compound f u (ED) K D (lipid) K D (RSA) f u (lipid) f u (lipid RSA) RSA % M M % % % Amitriptyline Carbamazepine Chlorpromazine Clozapine Diazepam Fluoxetine Haloperidol Ibuprofen Metoclopramid Risperidon Thioridazin Tolbutamid Trazodon Trifluorperazine Venlafaxin Warfarin

9 320 LONGHI ET AL. TABLE 4 Measured f u,brain (%) in TRANSIL, as discrete or as a cocktail (pooling four compounds in one sample) Drug Name f u,brain (%) S.D. TRANSIL Discrete f u,brain (%) S.D. TRANSIL Cocktail Fold Change GSK GSK GSK GSK GSK GSK Clozapine Amitriptyline Haloperidol GSK GSK Propranolol GSK Midazolam GSK GSK GSK GSK GSK GSK Carbamazepine Bupivacaine GSK Buproprion GSK trifluoperazine. Consequently, the influence of albumin binding was negligible for these drugs. These results are in line with the assumption that brain tissue binding is governed, in particular, by hydrophobicity because phospholipids are present in brain in great excess (Poulin and Theil, 2000). Drugs may primarily partition into phospholipids, whereas other macromolecules would have an insignificant contribution due to much lower concentrations. Ibuprofen, tolbutamide, and warfarin exhibited higher affinities to RSA and high free fractions in brain. The highest contribution of RSA binding to total brain tissue binding was observed for these compounds, with albumin (in theory) decreasing the fraction unbound from to 26.10% for ibuprofen, to for warfarin, and 84.3 to 77.6% for tolbutamide. Although high affinity to albumin for compounds that are loosely bound to brain lipids might theoretically cause an underestimation of binding by the TRANSIL method, practically, this appears not to affect the brain binding. For these three compounds, we observed that the free fractions values calculated based only on the lipid affinity [f u (lipid)] were closer to those measured by equilibrium dialysis with brain perfused homogenates [f u (ED)] than the values obtained based on the combined affinity to lipids and albumin [f u (lipid RSA)]. By using the TRANSIL 96-well assay format, we were able to process 12 samples at the same time. To further increase the sample throughput, we investigated the possibility of dosing multiple compounds. Previous work has already demonstrated the suitability of sample pooling for screening drug-plasma protein binding (Wan and Rehngren, 2006) and brain tissue binding (Wan et al., 2007). The main assumption here is that the equilibrium of a compound when it is pooled with others is expected to be the same as when it is on its own, provided that the pooled total drug concentration is still much lower than the lipid concentration in the homogenate, leading to a consistent f u independent of drug concentration. To further verify this theoretical model and to demonstrate that sample pooling is a viable approach for drug tissue binding screening with the TRANSIL assay, a four-in-one cocktail approach was investigated. The percent unbound values of 25 new chemical entities were compared for individual dosing and the cocktail approach. As shown in Table 4, the estimated brain free fractions displayed a strong correlation (r ) between single compound measurement and four pooled compounds measurement. This result indicates that the TRANSIL method is well suited for the sample pooling approach, thus increasing the throughput of the assay. When the cocktail approach with the 96-well FIG. 5. Correlation of f u,brain (%) between cocktail and discrete TRANSIL assay measurements (r ). The black line represents unit correlation, and dashed lines represent 2-fold boundaries. Gray circles represent proprietary compounds, and black circles represent marketed compounds.

10 DETERMINING BRAIN TISSUE BINDING OF DRUGS IN DRUG DISCOVERY 321 assembly was adopted, the throughput per plate was increased to 48 compounds. In addition to dosing multiple compounds, sample preparation is another important area in high-throughput screening. As previously mentioned, equilibrium dialysis methods suffer from long equilibrium times (6 20 h), matrix stability, volume shifts, and extensive preparation of dialysis membranes before use, all of which affect the turnaround time of the assay. The TRANSIL assay can shorten the time drastically to 30 min for each 96-well plate sample, including 2 min for sample incubation and 10 min for centrifugation, by using automation with a robotic system. Furthermore, the TRANSIL assay is a matrix-free method; thus, compound stability in biological matrix is unlikely to be an issue. In addition, because no sample preparation is required, the supernatant can be injected directly into the LC-MS/MS system, thus further improving the throughput of the assay. This method is especially beneficial for the drug discovery process because it more rapidly assesses the brain tissue binding potential of drug candidates. This work presents a novel high-throughput method for in vitro screening of drug brain tissue binding based on TRANSIL technology in combination with UPLC/MS/MS. This approach provides an efficient way to obtain unbound brain fraction data in early drug discovery, yielding results comparable with conventional equilibrium dialysis using brain homogenate. In addition, the speed and simplicity of this new assay considerably increases the throughput of the brain tissue binding assay and efficiently eliminates matrix effects in mass spectrometry. Furthermore, the TRANSIL assay substantially reduces both the assay and downstream analytical time through its matrix-free and fully automatable set up. The assay also contributes to the reduction of laboratory animal usage in drug discovery, which is of increasing importance in pharmaceutical research. Acknowledgments We thank K.D. Read for reviewing the manuscript and helpful suggestions. Authorship Contributions Participated in research design: Boriss, Longhi, and Fontana. Conducted experiments: Helmdach, Longhi, and Vinco. Contributed new reagents or analytic tools: Schiller and Boriss. Performed data analysis: Longhi. Wrote or contributed to the writing of the manuscript: Helmdach, Boriss, Longhi, and Fontana. References Becker S and Liu X (2006) Evaluation of the utility of brain slice methods to study brain penetration. Drug Metab Dispos 34: Boriss H (2010) Brain availability is the key parameter for optimizing permeability of CNS drugs. Drug Discovery 7: Bradford MM (1976) A rapid and sensitive method for the quantitation of microgram quantities of protein utilizing the principle of protein-dye binding. Anal Biochem 72: Di L, Kerns EH, and Carter GT (2008) Strategies to assess blood-brain barrier penetration. Expert Opin Drug Discov 3: Dyer WJ and Bligh EG (1959) Extraction of lipids in solution by the method of Bligh and Dyer. Can J Biochem Physiol 37: Fridén M, Gupta A, Antonsson M, Bredberg U, and Hammarlund-Udenaes M (2007) In vitro methods for estimating unbound drug concentrations in the brain interstitial and intracellular fluids. Drug Metab Dispos 35: Fuchs B, Schiller J, Süß R, Schürenberg M, and Suckau D (2007) A direct and simple method of coupling matrix-assisted laser desorption and ionization time-of-flight mass spectrometry (MALDI-TOF MS) and thin-layer chromatography (TLC) for the analysis of phospholipids from egg yolk. Anal Bioanal Chem 389: Fuchs B and Schiller J (2009) Application of MALDI-TOF mass spectrometry in lipidomics. Eur J Lipid Sci Technol 111: Fung EN, Chen YH, and Lau YY (2003) Semi-automatic high-throughput determination of plasma protein binding using a 96-well plate filtrate assembly and fast liquid chromatographytandem mass spectrometry. J Chromatogr B Analyt Technol Biomed Life Sci 795: Glees P (1988) The Human Brain, Cambridge University Press, Cambridge, UK. Harris RA and Gambal D (1963) Fluorometric determination of total phospholipids in rat tissues. Anal Biochem 5: Kalvass JC and Maurer TS (2002) Influence of nonspecific brain and plasma binding on CNS exposure: implications for rational drug discovery. Biopharm Drug Dispos 23: Kalvass JC, Maurer TS, and Pollack GM (2007) Use of plasma and brain unbound fractions to assess the extent of brain distribution of 34 drugs: comparison of unbound concentration ratios to in vivo P-glycoprotein efflux ratios. Drug Metab Dispos 35: Kremer JM, Wilting J, and Janssen LH (1988) Drug binding to human alpha-1-acid glycoprotein in health and disease. Pharmacol Rev 40:1 47. Liu X, Smith BJ, Chen C, Callegari E, Becker SL, Chen X, Cianfrogna J, Doran AC, Doran SD, Gibbs JP, et al. (2006) Evaluation of cerebrospinal fluid concentration and plasma free concentration as a surrogate measurement for brain free concentration. Drug Metab Dispos 34: Liu X, Van Natta K, Yeo H, Vilenski O, Weller PE, Worboys PD, and Monshouwer M (2009) Unbound drug concentration in brain homogenate and cerebral spinal fluid at steady state as a surrogate for unbound concentration in brain interstitial fluid. Drug Metab Dispos 37: Maurer TS, Debartolo DB, Tess DA, and Scott DO (2005) Relationship between exposure and nonspecific binding of thirty-three central nervous system drugs in mice. Drug Metab Dispos 33: Poulin P and Theil FP (2000) A priori prediction of tissue:plasma partition coefficients of drugs to facilitate the use of physiologically-based pharmacokinetic models in drug discovery. J Pharm Sci 89: Read KD and Braggio S (2010) Assessing brain free fraction in early drug discovery. Expert Opin Drug Metab Toxicol 6: Schiller J, Arnhold J, Benard S, Müller M, Reichl S, and Arnold K (1999) Lipid analysis by matrix-assisted laser desorption and ionization mass spectrometry: a methodological approach. Anal Biochem 267: Snyder WS, Cook MJ, Nasset ES, Karhausen LR, Howells GP, and Tipton IH 1975 Report of the Task Group on Reference Man. Pergamon Press, New York. Summerfield SG, Stevens AJ, Cutler L, del Carmen Osuna M, Hammond B, Tang SP, Hersey A, Spalding DJ, and Jeffrey P (2006) Improving the in vitro prediction of in vivo central nervous system penetration: integrating permeability, P-glycoprotein efflux, and free fractions in blood and brain. J Pharmacol Exp Ther 316: Summerfield SG, Read K, Begley DJ, Obradovic T, Hidalgo IJ, Coggon S, Lewis AV, Porter RA, and Jeffrey P (2007) Central nervous system drug disposition: the relationship between in situ brain permeability and brain free fraction. J Pharmacol Exp Ther 322: Summerfield SG, Lucas AJ, Porter RA, Jeffrey P, Gunn RN, Read KR, Stevens AJ, Metcalf AC, Osuna MC, Kilford PJ, et al. (2008) Toward an improved prediction of human in vivo brain penetration. Xenobiotica 38: Suzuki T and Kudo Y (1990) Automatic log P estimation based on combined additive modeling methods. J Comput.-Aided Mol Des 4: Wan H and Rehngren M (2006) High-throughput screening of protein binding by equilibrium dialysis combined with liquid chromatography and mass spectrometry. J Chromatogr A 1102: Wan H, Rehngren M, Giordanetto F, Bergström F, and Tunek A (2007) High-throughput screening of drug-brain tissue binding and in silico prediction for assessment of central nervous system drug delivery. J Med Chem 50: White T, Bursten S, Federighi D, Lewis RA, and Nudelman E (1998) High-resolution separation and quantification of neutral lipid and phospholipid species in mammalian cells and sera by multi-one-dimensional thin-layer chromatography. Anal Biochem 258: Address correspondence to: Raffaele Longhi, Aptuit Verona, Via Fleming 4, Verona, Italy. raffaele.longhi@aptuit.com