Aldehyde Oxidase Rapidly Metabolizes a P38 Kinase Inhibitor

In silico and in vitro pharmacogenetics: aldehyde oxidase rapidly metabolizes a p38 kinase inhibitor

X Zhang1, H-H Liu2, P Weller3, The clinical development of a candidate p38 kinase inhibitor was terminated 2 3 1 because of its unexpectedly rapid clearance in human subjects. Its short half- M Zheng , W Tao , J Wang , life and metabolic profile in human beings were vastly different from that in 1 3 G Liao , M Monshouwer and rats, dogs, and monkeys characterized during routine pre-clinical studies. G Peltz2 Mice generated the predominant drug (4-hydroxylated) metabolite produced in human beings, which was not found in other species. The data from 1Department of Genetics and Genomics, Roche a murine in vitro drug biotransformation assay that used liver extracts from 2 Palo Alto, Palo Alto, CA, USA; Department of 14 inbred mouse strains were analyzed by haplotype-based computational Anesthesia Stanford University, Stanford, CA, USA and 3Department of Drug Metabolism and genetic analysis. This led to the identification of aldehyde oxidase-1 (AOX1) Pharmacokinetics, Roche Palo Alto, Palo Alto, as the enzyme responsible for the rapid metabolism of this drug. Specific CA, USA enzyme inhibitors and expressed recombinant enzymes were used to confirm that AOX catalyzed the formation of the 4-hydroxylated drug metabolite in Correspondence: Dr G Peltz, Department of Anesthesia, Stanford mouse and man. Genetic variation within Aox1 regulated the level of hepatic University, 800 Welch Road Room 213, Aox1 mRNA, AOX1 protein, and enzyme activity among the inbred strains. Stanford, CA 94305-5796, USA. Thus, computational murine pharmacogenetic analysis can facilitate the E-mail: [email protected] identification and characterization of drug metabolism pathways that are differentially utilized by humans and other species. The Pharmacogenomics Journal (2011) 11, 15–24; doi:10.1038/tpj.2010.8; published online 23 February 2010

Keywords: pharmacogenetics; aldehyde oxidase; inbred mouse strains

Introduction

Drug development is an extremely complicated, costly, and time-consuming process. Poor bioavailability or metabolic stability of a candidate drug has contributed to failures in drug development. Therefore, the pharmacokinetic profile of a drug candidate is comprehensively assessed during pre-clinical development using in vitro systems and in several different animal species before it is administered to human subjects (reviewed in Singh1). The data obtained from animal studies are then extrapolated to predict the human pharmacokinetic profile through allometric modeling, which assumes that anatomical, physio- logical, and biochemical variables can be scaled across mammalian species as a power function of the body weight.2 However, allometric predictions do not adjust for inter-species differences within drug metabolizing enzymes, which can cause substantial inter-species differences in drug metabolism, as has been observed for the cytochrome P450 family.3–5 Despite the existence of multiple Received 26 May 2009; revised 20 November 2009; accepted 20 January 2010; published different strains within many of the species (rat, mouse, or dog) used for online 23 February 2010 pharmacokinetic testing, data are usually obtained from only a single strain. Computational in vitro pharmacogenetics X Zhang et al 16

As there is significant variability in drug metabolism among Pre-clinical pharmacokinetics and projection of human dose different strains of rats and mice,6–8 pharmacokinetic data All animal experiments were implemented under protocols generated from one strain may be atypical and may not even approved by the Roche Palo Alto Institutional Animal represent the pharmacokinetic profile of most strains of that Care and Use Committee. For the pharmacokinetics studies, species. Hanover–Wistar rats were obtained from Charles Rivers Therefore, it is critical to develop tools that provide Laboratories; beagle dogs and cynomolgus monkeys were information about the genetic basis for inter-individual or obtained from other commercial suppliers. All animals were inter-species differences in drug metabolism. We recently housed in a pathogen-free environment and provided food described how in vivo pharmacokinetic studies9 or a hepatic and water ad libitum with a 12 h:12 h light/dark cycle. All in vitro drug biotransformation system10 could be used in animals were acclimatized for an additional week before conjunction with haplotype-based computational genetic study. The animals were administered a single intravenous analysis11–14 to quickly identify genetic factors affecting the dose of RO1 (5 mg kg–1) in a simple aqueous formulation and metabolism of commonly prescribed drugs in mice. To do blood samples were collected into heparinized tubes from a this, a test drug is administered to multiple inbred strains, or minimum of three animals per study at prescribed times. incubated with hepatic extracts in vitro, and then individual For allometric scaling, the dose-normalized plasma drug metabolites are measured at multiple time points. The concentration in each animal species is adjusted for causal genetic factors are identified by correlating inter- clearance by dividing the time by BW0.25, where BW is body strain differences in drug metabolism with the pattern of weight (kg). The concentration–time curves obtained from genetic variation within genomic regions among the inbred these animals were converted into a human concentration– strains. Multiple pathways mediate the metabolism of most time curve using the elementary Dedrick plot with the drugs, and distinct enzymes may catalyze individual steps in following equation:16–19 each pathway. Therefore, analysis of the rate of formation of 1Àx individual metabolites across multiple inbred murine strains Thuman ¼ Tanimal ðBWhuman=BWanimalÞ ; reduced the complexity of this biological process, which enabled the genetic factors that contribute to the overall where x ¼ 0.75 variability in drug metabolism to be identified (reviewed in Liao et al.15). Here, the in vitro and in silico genetic analysis Chemicals method was used to analyze the metabolism of a novel p38 RO1, RO2, and its metabolites (RO1: 6-(2,4-difluoro-phenoxy)- kinase inhibitor, whose clinical development for treatment 2-((R)-2-hydroxy-1-methyl-ethylamino)-8-((S)-2-hydroxy-propyl)- of rheumatoid arthritis was terminated because of rapid 8H-pyrido[2,3-d]pyrimidin-7-one; M1: 6-(2,4-difluoro-phenoxy)- drug metabolism in human subjects. Its short half-life and 4-hydroxy-2-((R)-2-hydroxy-1-methyl-ethylamino)-8-((S)-2-hydroxy- metabolic profile were not predicted by its pharmaco- propyl)-8H-pyrido[2,3-d]pyrimidin-7-one) were synthesized kinetics in several animal species analyzed during pre-clinical by Chemical Services at Roche Palo Alto. Camptothecin evaluation. However, murine in vitro computational genetic (Sigma, St Louis, MO, USA) was used as the internal standard analysis identified the cause of its unusual pharmacokinetic (IS) for the liquid chromatography/tandem mass spectro- and metabolic profile in human subjects. metry (LC/MS/MS). Methanol and acetonitrile were purchased from Burdick & Jackson (Muskegon, MI, USA); formic acid, 1-aminobenzotriazole (ABT), and isovanillin (3-hydroxy-4-methoxybenzaldehyde) were purchased from Materials and methods Sigma-Aldrich (St Louis, MO, USA). All solvents used for HPLC and LC/MS/MS analysis were of chromatographic Phase I clinical trial grade. A single-center, ascending single-dose, randomized, observer-blinded, parallel-design phase I study was performed Preparation and analysis of human liver cytosol according to a protocol that was approved by the Institu- Segments of three frozen postmortem livers were obtained tional Review Board at Biokinetic Europe (Belfast, UK) that from Science Care (Phoenix, Arizona, USA). Approximately followed the Helsinki guidelines. Six healthy male subjects 2 g of liver tissues from donors 1 (male), 2 (male), and (18–43 years of age) received RO1 (50 mg), and 2 received 3 (female) were each homogenized in three volumes of placebo in an oral solution. Peripheral blood samples were 20 mM Tris buffer (pH 7.4) containing 1 mM EDTA, 1 mM collected from all subjects 0.25, 0.5, 1, 1.5, 2, 3, 4, 5, 6, 8, 12, dithiothreitol, and a complete protease inhibitors cocktail 24 h after dosing, and then daily for 7 days. This initial drug tablet (Roche Applied Sciences, Indianapolis, IN, USA). The dose was selected based on allometric projection of human homogenates were centrifuged at 100 000 g for 65 min. exposures from animal pK data. It was anticipated that this Fat was removed from the resultant supernatants by filtra- dose would produce a blood level above the measured IC50 tion through gauze tissues and protein concentrations of this drug for inhibition of TNFa-induced-IL1b production were determined from aliquots of the supernatants by BCA in human whole blood. The plasma concentration of RO1 protein assay (Pierce Biotechnology, Rockford, IL, USA). and its metabolites were analyzed by LC/MS/MS analysis as Individual donors activities were initially determined. For described below. screening assays, a pool was made by mixing equal amounts

The Pharmacogenomics Journal Computational in vitro pharmacogenetics X Zhang et al 17

of protein from each of the three cytosol fractions. Aliquots as the ABT assay except for the substitution of isovanillin of the cytosol pool were apportioned into vials, and then (100 mM to 1 mM) for ABT. The inhibitor reactions were the vials were stored at À70 1C until use. Human liver performed in duplicates at 37 1C for 60 min. cytosols (1 mg ml–1) were incubated with substrate RO1 (100 mM), þ 100 mM isonanillin, in Tris (0.05 M, pH 7.4), LC/MS/MS analysis MgCl (5 mM), and 1.15% KCl for 1 h at 37 1C. All incubations The LC/MS/MS system consisted of a Shimadzu LC system, were performed in triplicate. At the end of the incubation, a vacuum degasser, an autoinjector, and a Sciex API 4000 100 ml of IS were added into each reaction, mixed, and then triple quadruple mass spectrometer (MS/MS) (Applied centrifuged at 10 000 rpm for 10 min to produce clear Biosystems, Foster City, CA, USA), and was equipped with supernatants. After centrifugation, aliquots (100 ml) of the Analyst software v. 1.4.1 (Applied Biosystems) for data acqui- resultant supernatants were transferred to the 96-deep-well sition and calibration. RO2 and its metabolites were sepa- plate and assayed for the hydroxylated metabolites M1 by rated on a Thermo BDS Hypersil-C18 column (50 Â 2 mm, the LC–MS/MS method described below. 5 mm) with HPLC guard column (Thermo Electron Corp., Bellefonte, PA, USA) at a flow rate of 0.4 ml min–1. The Preparation of murine liver cytosol mobile phases were 5 mM ammonium acetate in 0.1% Sixteen inbred mouse strains (MRL/MpJ, LG/J, NZW/LaC, formic acid (A) and 50% acetonitrile/50% methanol with 129S1/SvImJ, SM/J, NZB/B1nJ, LP/J, DBA/2, C57BL/6, A/J, 0.1% formic acid (B). The gradient of B was as follows: 10% C3H/HeJ, BALB/c, B10.D2-H2dH2-T18c Hc0/oSnJ, BALB/ (0–0.5 min), 10–90% (0.5–2.8 min), 90–10% (2.8–3 min), and cByJ, AKR/J, A/HeJ), ages 9–10 weeks, were obtained from 10% (3–3.5 min). The injection volume was 10 ml and the the Jackson Laboratory and housed at Roche Palo Alto in total run time was 3.5 min. M1 and the IS both eluted at a pathogen-free facility, and provided food and water 2.4 min. The MS was operated under positive ionization in ad libitum with a 12 h: 12 h light/dark cycle for at least the multiple reaction monitoring mode for quantitative 2 weeks before experimentation. Three male mice of each analysis and synchronized with the HPLC. The parent (m/z), strain were euthanized by cervical dislocation, and livers were daughter (m/z), collision energy, declustering potential, and harvested and snap frozen in liquid nitrogen. Equal amounts cell exit potential for each compound are RO1 (407.2, 330.9, of liver from each mouse were pooled and homogenized in 29, 81, 14), RO2 (407.2, 389.2, 29, 81, 14), M1 (423.1, 405.1, three volumes of 20 mM Tris buffer (pH 7.4) supplemented 31, 86, 14), and camptothecin (IS) (349.2, 305.2, 31, 86, 16). with 1 mM EDTA, 1 mM dithiothreitol, and a complete Calibration curves were developed by spiking blank CD1 protease inhibitors cocktail tablet (Roche Applied Sciences). mouse liver S9 fraction (BD Gentest) with different amounts In preliminary experiments, we were unable to detect of RO2 ranging from 1 to 50 000 ng ml–1. The peak area ratio production of the M1 metabolite after incubation of RO1 or of metabolites relative to that of the IS (camptothecin) was RO2 with human liver or human intestinal microsomal quantified, and correlation coefficients were calculated by extracts, murine liver S9 fractions (contain both microsomal least-square regression analysis. and cytosolic fractions), or with cytosolic fractions prepared by conventional methods. Therefore, cytosolic extracts were Computational genetic mapping rapidly isolated using a modified method with a single ultra- The computational genetic analysis of the inbred strain centrifugation step that minimizes the processing time.20 To data was performed as described earlier.11,12 In brief, a do this, the homogenates were centrifuged at 100 000 g for haplotype block map of the mouse genome was constructed, 65 min at 4 1C. Fat was removed from the supernatants and and single-nucleotide polymorphisms (SNPs) were orga- protein concentrations were determined as above. Liver nized into haplotype blocks, which typically consisted of cytosol aliquots were stored at À80 1Cuntiluse. 2–4 haplotypes.13,14 The haplotype-based computational analysis identified haplotype blocks in which the haplo- In vitro drug biotransformation assays typic strain grouping within the block correlates with the Mouse liver cytosol or cytosolic extracts of insect cells distribution of phenotypic data among the inbred strains expressing recombinant aldehyde oxidase (AOX) proteins analyzed. The program calculates a P-value, which assesses (1 mg ml–1) were pre-incubated with PBS buffer (pH 7.2) at a the likelihood that genetic variation within each block could final volume of 100 mlat371C for 5 min. The reactions were underlie the observed distribution of phenotypes among initiated by addition of 10 mM RO2. At designated time the inbred strains.11,13 The haplotype blocks are ranked periods, the reactions (100 ml) were terminated by adding a based on the calculated P-value. The genomic regions within 100 ml stop solution (40 ng ml–1 camptothecin in acetoni- haplotype blocks that strongly correlated with the pheno- trile). All samples were then centrifuged at 4000 g for 10 min typic data are then analyzed. The Roche SNP database to precipitate the proteins. The supernatant was collected (http://mousesnp.roche.com) contains 250 837 SNPs gene- for metabolite analysis by LC/MS/MS. CYP450 activity was rated from 21 inbred mouse strains covering 3346 genes, assessed by incubating the liver cytosol with 10 mM RO2 and including 60 Cyp450 enzymes, 8 aldehyde dehydrogenases, ABT (100 mM) in the presence of Tris (0.05 M, pH 7.4), MgCl 5 flavin monooxygenases, 88 transporters (Abc or Slc gene (5 mM), and 1.15% KCl with or without NADPH-P450 families), 3 AOXs, almost all UDP glucuronosyltransferase, oxidoreductase (BD Gentest, Bedford, MA, USA). The AOX glutathione S-transferase enzymes, sulfotransferases, and inhibition assay was carried out under the same conditions 41 nuclear hormone receptors. The P-values for numerical

The Pharmacogenomics Journal Computational in vitro pharmacogenetics X Zhang et al 18

phenotype evaluated in this experiment were calculated as 50 mM Tris (pH 8), 5 mM DTT, 250 mM NaCl, 0.5 mM described using ANOVA.11,12 As ANOVA-based computation EDTA with complete protease inhibitors cocktail tablet requires homogeneity of variance, the measured rate of M1 (Roche Applied Sciences) and homogenized by sonication. metabolite formation for each strain was logarithmically After centrifugation for 20 min at 410 000 rpm at 4 1C, the transformed for the analysis. supernatants were collected for measurement of total A relatively large number of genomic regions had a protein concentration and then used in the biotransforma- genetic pattern that correlated with the rate of drug tion assays at a final concentration of 1 mg ml–1. metabolism. As computational genetic mapping is based on comparison of phenotypic data with a large number of Immunoblot and protein quantitation haplotypic blocks (n ¼ 5255 here), a haplotype distribution In all, 40–50 mg of total cellular protein was separated pattern can correlate with trait values purely by chance, or by 10% SDS–PAGE and transferred to PVDF membranes because of genetic relatedness among the inbred strains. (Bio-Rad, Hercules, CA, USA). After 10 min of blocking in When the genetic differences within these identified regions 5% milk in TBS Tween 20 (milk/TBST), the membranes were do not contribute to the phenotypic differences observed, blotted with primary antibodies (1:1000 or 1:500 dilution in the association is a ‘false-positive’ prediction. One can milk/TBST) for 1 h at room temperature or overnight at 4 1C. attempt to eliminate false-positive predictions by setting a After washing in TBST for 10 min, the membranes were very high bar for the statistical significance of an identified incubated with alkaline phosphatase-conjugated a-mouse genomic region. However, this raises the risk of a false or rabbit immunoglobulin (Promega, Madison, WI, USA) negative—spuriously rejecting a true causative genetic factor secondary antibodies for 1 h at room temperature (1:1000 from consideration. As discussed elsewhere,21 as we cannot in 5% milk/TBST). After a 30 min wash in TBST, the bands compensate for a false negative, we have chosen to set our were developed with one-step NBT/BCIP (Pierce, Rockford, statistical significance at a threshold that minimizes the IL, USA). The primary antibodies were a-AOX1 (Clone 7 chance of a false negative. from BD Biosciences Pharmingen, San Diego, CA, USA), or anti-b-actin, rabbit monoclonal (Cell Signaling, Danvers, Microarray and RT–PCR gene expression analysis MA, USA). After developing, the blots were scanned on a For analysis of global gene expression patterns in liver, total flatbed scanner (Hewlett-Packard scanjet 8300, Palo Alto, RNA was isolated from liver obtained from female mice (age CA, USA). The signal density for each band was measured by 11 weeks) of 10 different inbred strains; mRNA was then using Image Analysis and Measurement Lab software (http:// isolated from three individual mice for each strain, and brneurosci.org/imal.html) and normalized to the highest 39 000 transcripts present on Affymetrix microarrays (Mouse density of the same protein. The relative ratio was compared Genome 430 2.0) array were analyzed in each sample as between normalized signal densities. A Spearman’s correla- described earlier.10 RT–PCR analysis was used to analyze the tion coefficient was calculated using SAS 9 (Cary, NC, USA) pattern of expression of AOX genes in liver. Total liver RNA software to assess the degree of correlation between AOX was prepared from female mice as described earlier.11 The activity and the normalized protein values. Aox mRNAs were analyzed by Taqman as described by the manufacturer (Applied Biosystems) using the following Results primers: Aox1 forward: 50-TTT TTGC CAT CCA TGA TGC A-30, Clinical pharmacokinetics of RO1 metabolism Aox1 reverse: 50-TTG AGT TTC CAT GGT CCA GAG A-30, The safety, tolerability, and pharmacokinetics of a novel p38 probe: 50-CTC CTG CCG CGC TGC CTT CA-30; MAP kinase inhibitor (RO1) for treatment of rheumatoid Aox3 forward: 50- AGG CTT GTC CCC AAT TTG G-30; arthritis (Figure 1a) were evaluated in six healthy Caucasian Aox3 reverse: 50-CAG GCC ATG CGG ATC ACT-30, male subjects in a phase I clinical study. RO1 (50 mg) or probe: 50-CCA TAA ACA GCC CTG CCA CAG CA-30 placebo was administered in a single ascending oral dose, randomized, observer-blinded trial, and the plasma concen- Recombinant AOX1 expression tration of RO1 and its metabolites were measured. Unexpec- 1 Aox1 cDNAs were isolated from AHeJ and MRL liver, and tedly, its observed terminal mean half-life (T2) was only 0.7 h human AOX1 from Huh-7 cells by RT–PCR using oligo-dT (Figure 1a), and the drug was extensively converted into six priming; and cloned into pVL 1392 univ ccdB vector. measurable metabolites. The drug was metabolized to such a Recombinant baculovirus was produced using the Rapid degree that it would not be possible to reach therapeutic Suspension Transfection method with an infection time levels even at higher doses, which led to discontinuation of of 72 h. For protein production, log-phase Sf9 cells (1.2 Â 106 clinical development of this compound. The allometric per ml) were infected with recombinant virus and grown at projections were based on pharmacokinetic data obtained 1 1 1 27 C in SF-900 II medium (Invitrogen, Carlsbad, CA, USA) from rats (T2 6.2 h), dogs, and monkeys (T2 4 h), which supplemented 10% heat-inactivated FBS. One hour after predicted a significantly longer half-life (5.9 h) (Figure 1b) infection, a supplement of 0.5 mM NaMoO4,10mM ribo- and total exposure level (B20-fold higher) in human flavin, 10 mM guanosine, and 80 mM FeSO4 was added to the subjects. Moreover, the major drug metabolite in human growth medium. On harvest, Sf9 cells expressing the recom- sera was a 4-hydroxylated metabolite (M1) (220% of AUC binant cDNAs were re-suspended in a solution containing relative to parent drug) that was rapidly glucuronidated

The Pharmacogenomics Journal Computational in vitro pharmacogenetics X Zhang et al 19

F OH F previously described in vitro murine computational pharma- O O 10 N N cogenetic analysis method could provide information H H N N N O F N N N O F about how RO1 was biotransformed into M1. To avoid the OH OH regulatory requirements and complications associated with OH OH RO1 M1 working with a compound that was undergoing active F clinical testing at the time, RO2, a diastereomer of RO1 O N (Figure 1a) that had the same metabolic profile as RO1, was H N N N O F OH 1 used in some of the initial exploratory murine experiments. * RO1 OH M1 We found that cytosolic extracts that were rapidly isolated 0.1 M1 Gluc using modified method20 mediated the formation of a RO2 g/mL) significant amount of M1 (Figure 2a). As in human plasma, µ 0.01 a full-scale MS scan on the products formed after the in vitro 0.001 incubation indicated that the M1 was the predominant

Conc. ( metabolite formed (data not shown). Unlike human plasma, 0.0001 024681012 glucuronidated metabolites were not formed after any Time (hours) in vitro incubation with the murine extracts under any conditions tested. The rate of M1 formation after RO2 (10 mM) was incubated with liver cytosol fractions prepared 10.0000 Monkey from 14 inbred mouse strains was monitored by LC/MS/MS Dog 1.0000 Rat analysis (Figure 2a). For the computational haplotype-based genetic analysis, the amount of M1 formed after 15 min of 0.1000 incubation was analyzed because this time point was within

0.0100 the linear range for this enzymatic reaction and there were significant inter-strain differences in the rate of M1 forma- 0.0010 tion (Figure 2b). The computational genetic analysis of these 0.0001 data identified 126 haplotype blocks with a P-value o0.05 0481216 (Supplementary Table 1). A two-step filtering process used Dose Normalized Plasma Conc. (ug/mL) Time / BW 0.25 was applied to narrow the number of candidate genes that Figure 1 (a) The structure of RO1 and its metabolite (M1), which is could be responsible for the inter-strain differences. First, produced by the addition of a hydroxyl group at C4. RO2 is a a search of the functional annotations within the Genome diastereomer of RO1 at the indicated (*) position. The graph shows the Ontology Database (http://www.geneontology.org) indi- measured plasma concentration of RO1, M1, and M1-glucuronide at cated that only six of these haplotype blocks (shown in the indicated times after a single oral dose in human volunteers. The Figure 2c) encoded enzymes that could perform the ring B arrow indicates the half-life of RO1 ( 0.7 h). (b) The plasma oxidation reaction that generated the 4-hydroxy drug concentration of RO1 at designated times after a single bolus –1 metabolite. Second, we evaluated whether the remaining intravenous injection of RO1 (5 mg kg ) in monkey (’), dog (m), or rat (K). The x axis shows the time of exposure divided by body weight gene candidates were expressed in the liver. Although ^ 0.25. The y axis is the measured plasma RO1 concentration (mgml–1) Cyp19a1 and Cyp46a1 had correlated genetic patterns, divided by the administered dose (5 mg kg–1). The solid line indicates Cyp19a1 mRNA was not detected in any, and Cyp46a1 the predicted human pharmacokinetic profile using a Dedrick plot. mRNA was only detected (at a very low level) in 2 of the 12 inbred strains whose livers were examined. As these two enzymes were not expressed in liver of the inbred strains that mediated M1 formation, they were eliminated from (508% of parent drug AUC) and cleared through urinary consideration. This quickly narrowed the search for candi- excretion. Of importance, a negligible amount of this meta- date genes to two cytochrome P450 genes (Cyp2f2, Cyp2d10) bolite was formed in the rats, dogs, and monkeys tested, but and the AOX genes. Of note, the AOXs were the only the other drug metabolites in human plasma were also enzymes in the molybdenum hydroxylase family with a produced by these animal species (data not shown). This correlated genetic pattern. suggests that a metabolic pathway that was not used in rats, dogs, or monkeys was responsible for the unexpectedly Murine AOX1 mediates M1 generation rapid metabolism of this drug in human subjects. To determine whether Cyp450 or AOX enzymes covert RO2 to M1, in vitro biotransformation reactions were performed Computational in vitro pharmacogenetic analysis in the presence of 1-ABT22,23 or isovanillin,24 which are The ring oxidation reaction that introduced a hydroxyl specific inhibitors of Cyp450 or AOX enzymes, respectively. group at the C4 position, which converted RO1 to M1, could The inhibitors were evaluated using cytosolic extracts from be carried out by any enzyme within the following gene six strains with high, medium, or low rates of M1 formation. families: microsomal cytochrome P450 enzymes, alcohol As Cyp450 enzyme catalyzed reactions require NADPH as dehydrogenases, AOXs, or xanthine oxidase. Owing to the cofactors, all reactions were carried out in the presence and large number of possibilities, we investigated whether our absence of NADPH. ABT did not inhibit M1 formation,

The Pharmacogenomics Journal Computational in vitro pharmacogenetics X Zhang et al 20

5 blank MRL C57 DBA 4 LPJ AJ NZW AKR 3 C3H BALB/cbyJ M1 Formation B10 NZB 2 AHeJ SMJ 129x S9_CD1 1 M1 Formation (ng/ml) 2 Control 1.5 Isovanillin 0 1 0 10203040506070 -0.5 Time of Incubation (min) -1

(M1 Formed) -1.5 10 -2

10 Log -2.5 8 MRL C57 DBA2 A/J B10.D2 SMJ

6 Figure 3 AOX catalyzes M1formation. RO2 (10 mM) was incubated for 4 60 min with hepatic cytosol fractions (1 mg ml–1) prepared from the indicated inbred mouse strains. The assays in the upper panel were 2 performed in the presence or absence of CYP450 enzyme inhibitor ABT

M1 Formation (ng/ml) (100 mM) or the CYP450 cofactor NADPH (20 mg ml–1). The assays in the lower panel are performed in the presence or absence of the AOX inhibitor isovanillin (100 mM). M1 formation was quantified by LC/MS/ MS and expressed in linear scale in the upper panel and in log scale in the lower panel. Isovanillin significantly inhibited M1 formation, whereas ABT did not.

Figure 2 Computational genetic analysis of the rate of M1 formation by an in vitro drug biotransformation system. (a) The amount of whereas isovanillin completely abolished the conversion of 4-hydroxy M1 metabolite formed after incubation of RO2 (10 mM) with RO2 to M1 in all strains (Figure 3). These results suggest that liver cytosol (1 mg ml–1 total protein) preparations obtained from the an AOX, and not a Cyp450 enzyme, mediated RO2 biotrans- indicated mouse strains was determined by LC/MS/MS analysis and is formation to M1. plotted as a function of time. Each data point represents the Three co-linear genes (Aox1, Aox3, and Aox4), which average±s.e.m. of three independent measurements performed on encode different AOXs of similar size (1333, 1335, and 1336 liver cytosol preparations for each strain. A blank sample containing amino acids, respectively), are within or immediately water and a liver S9 fraction from CD-1 mice were also run in parallel. adjacent to the computationally correlated haplotype block (b) The amount of M1 formed after 15 min of incubation with cytosol preparations from 14 strains. (c) The data in (b) was log transformed at 58 Mb on chromosome 1. Although multiple haplotype and analyzed using the haplotype-based computational genetic blocks with a slightly different haplotype pattern span the method. Listed in this table are the genes whose haplotype block three AOX genes, there is an extended haplotypic pattern pattern significantly correlated with the inter-strain differences in the across these three genes. The A/HeJ, A/J, AKR, B10.D2, rate of M1 formation (Po0.05) and have enzymatic activities that could C57BL6, SMJ, and C3H strains share one haplotype across mediate the M1 production. For each block, the chromosomal location, this region, although the BALB/b and MRL strains share the number of SNPs within a block, its gene symbol, and an indicator of other haplotype. The Aox1 and Aox3 mRNAs are abundantly gene expression in liver are shown. A colored block represents the expressed in mouse liver, whereas Aox4 mRNA was not haplotype for each strain. Strains with the same colored block share the same haplotype for the particular gene. The haplotype blocks are expressed in liver (Supplementary Figure 1). The two presented in the same order as indicated in the x axis of (b). The remaining candidates, Aox1 and Aox3, had very similar calculated P-value measures the probability that strain groupings within patterns of genetic variation (Supplementary Figure 2). As an an individual block would have the same degree of association with the SNP within Aox1 induced an amino-acid change at position phenotypic data by random chance. The hepatic expression level of 857 (D857E), we examined the ability of two different allelic each gene is also indicated: ‘Yes’ indicates the gene is expressed in all forms (D857 and E857) of expressed recombinant murine the strains, ‘No’ indicates the gene is expressed in none (or o3) of the AOX1 to convert RO2 to M1. Despite the amino-acid strains, whereas ‘unknown’ indicates no information is available. difference, both allelic forms of AOX1 efficiently catalyzed A complete list of all correlated haplotype blocks (P-valueo0.05) is shown in Supplementary Table 1. the biotransformation (Figure 4). We next investigated whether the genetic differences altered the level of hepatic Aox1 mRNA, protein expression or enzyme activity. AOX1 enzyme activity (rate of M1

The Pharmacogenomics Journal Computational in vitro pharmacogenetics X Zhang et al 21

1.20

1.00

0.80 mAox1 (D857) 0.60 mAox1 (E857) Control 0.40 M1 Formation

0.20

0.00 20 40 60 80 [Substrate] (uM)

Figure 4 Expressed recombinant murine AOX1 catalyzes the production of drug metabolite M1. cDNAs encoding two different allelic forms Figure 6 M1 metabolite formation mediated by human liver cytosol. of murine Aox1 were expressed in insect cells using the baculovirus The amount of M1 metabolite present 1 h after incubation of RO1 system. The indicated concentrations of RO2 were incubated with insect (100 mM) with human liver cytosol (1 mg ml–1) from three donors was cell lysates (protein concentration of 1 mg ml–1) for 60 min, and the determined by LC/MS/MS analysis. The incubations were performed in amount of M1 formed was determined by LC/MS/MS analysis. Both the absence (control) or presence of 100 mM isovanillin. Each bar recombinant AOXs generated significant amounts of M1, whereas represents the mean and the standard error of three independent control lysates did not. However, the two different allelic forms of measurements for each donor. murine AOX1 catalyzed the same rate of M1 formation.

(P-value ¼ 0.005). A Spearman’s correlation analysis indicated that Aox1 mRNA and AOX1 protein levels were significantly correlated (r ¼ 0.83, P ¼ 0.015) and that AOX1 enzyme activity was significantly correlated with the amount of AOX1 protein in the cytosol (r ¼ 0.95, P ¼ 0.001). Taken together, these results indicate that there are genetically determined differences in hepatic Aox1 mRNA, AOX1 protein, and enzyme activity levels among the inbred strains; and these differences affect the rate of biotransformation of RO2.

Human AOX1 mediates M1 formation This analysis was undertaken to identify the enzyme that converted the kinase inhibitor to its 4-hydroxy metabolite (M1) in human subjects. On the basis of the murine results, we investigated whether cytosolic preparations from human liver could generate M1. Conventional preparations of Figure 5 The amount of AOX1 protein in hepatic cytosol correlates human hepatic S-9 or cytosolic fractions lacked appreciable with the rate of M1 formation among the inbred strains. The amount of activity, which was probably because of proteolysis during AOX1 protein in hepatic cytosolic preparations from eight inbred mouse centrifugation. Therefore, a rapid procedure for preparing strains was determined by immunoblotting. b-actin was used as a loading and normalization control. The rate of M1 formation in the the cytosolic fraction was developed, which used protease eight hepatic cytosol preparations was determined as in Figure 3. The inhibitors and anti-oxidants to preserve enzyme activity. enzyme activities were plotted along with the normalized amount of Although there was some variability in the rate of conver- AOX1 protein in the cytosol fraction. The amount of protein was sion, human liver cytosol fractions that were prepared significantly correlated with enzyme activity (r2 ¼ 0.93, P ¼ 0.0009). from three different donors by this method were able to convert RO1 to M1. A kinetic analysis performed on pooled human liver cytosol indicated that RO1 was a low-affinity formation) and protein expression (immunoblotting and substrate for the human cytosolic enzyme (Km ¼ 112 mM) densitometry) were simultaneously measured in cytosolic (Supplementary Figure 3). However, virtually all (499%) of extracts prepared from eight inbred mouse strains represent- the M1 formation catalyzed by the liver cytosolic extract was ing three haplotypes (Figure 5). Consistent with LPJ having a inhibited by addition of an AOX inhibitor (isovanillin) to unique haplotype, the level of hepatic AOX1 protein was the reaction (Figure 6). lower in LPJ liver than in the strains with the BALB/b A human AOX1cDNA, which is the human homolog of haplotype. A Wilcoxon’s rank sum test indicated that the murine Aox1, was expressed in insect cells. Cytosolic amount of hepatic Aox1 mRNA was significantly different extracts prepared from these insect cells efficiently catalyzed between the strains with the SMJ and BALB/b haplotypes the biotransformation of either RO1 or its stereoisomer

The Pharmacogenomics Journal Computational in vitro pharmacogenetics X Zhang et al 22

3 a murine in vitro biotransformation system led to the 2.5 identification of AOX1 as the enzyme causing the rapid metabolism of this drug. Experiments using specific inhibi- 2 Enzyme Substrate hAOX1 RO1 tors and recombinant enzymes showed that AOX1 catalyzed 1.5 hAOX1 RO2 the formation of the predominant 4-hydroxylated drug Cyp2d10 RO1 1 Cyp2d10 RO2 metabolite produced in human subjects, which was not

M1 Formation 0.5 produced in rats, dogs, or monkeys. The pattern of genetic variation within a number of haplotype blocks correlated 0 0 1020304050607080 with the rate of drug biotransformation in the panel of Substrate Concentration (µM) inbred strains analyzed. However, a very limited set of plausible candidate genes were selected by application of gene expression criteria and knowledge-based filtering21 to 2.5 identify the candidate genes that encoded enzymes that were expressed in liver. 2 AOXs are members of the molybdenum hydroxylase family of enzymes that are present in a wide variety of 1.5 species. All AOXs are homodimers of B150 kDa monomers 1 that contain a molybdopterin cofactor, an FAD, and two different two Fe–S centers. They are liver cytosolic enzymes 0.5 M1 Formation (nM) with a broad substrate activity; they metabolize endo- genous compounds (dopamine, serotonin, catecholamine 0 25 Control + isovanillin metabolites, and retinals) and many drugs, including methotrexate, zonisamide,26 zaleplon,27 ziprasidone,28 Figure 7 Expressed recombinant human AOX1 catalyzes the produc- pyrazinamide, and famciclovir.29,30 AOX enzyme activity is tion of drug metabolite M1. A cDNA encoding human AOX1 was known to vary among different mammalian species, which expressed in insect cells using the baculovirus system, and lysates (1 mg ml–1) were prepared from these cells. (a) The indicated explains why RO1 pharmacokinetics in human subjects concentrations of RO1 and RO2 were incubated with the lysates for differed from that in the other species evaluated. Rats and 60 min, and the amount of M1 formed was determined by LC/MS mice have low activity, and dogs have no detectable activity analysis. As a control, lysates were also prepared from insect cells (reviewed in Beedham31). There is also considerable inter- expressing recombinant Cyp2d10. Each bar indicates the average and individual variation among rat strains32 and human standard error of three independent determinations. (b) Lysates beings.33 A Gly110Ser polymorphism that may alter homo- expressing recombinant human AOX1 were incubated for 60 min with dimer formation contributes to the observed 20- to 40-fold 10 mM of RO1 in the absence (control) or presence of 100 mM isovanillin. variation in AOX activity among rat strains.34–36 An allelic The amount of M1 formed after 60 min of incubation was determined by LC/MS analysis. Each bar indicates the average and standard error of difference in human AOX1 (Asn1135Ser), when combined three independent determinations. with assessment of thiopurine methyl-transferase activity, has recently been shown to affect azathioprine response in inflammatory bowel disease.37 In mice, we found that genetically determined differences in hepatic Aox1 mRNA (RO2) to M1 (Figure 7a). Furthermore, this biotransforma- and AOX1 protein expression contributed to the more tion was completely inhibited by addition of 100 mM subtle differences observed in this enzyme’s activity across isovanillin (Figure 7b), confirming that this chemical inbred mouse strains. It is also noteworthy that there are inhibits human AOX1. In contrast, lysates prepared from different numbers of AOX genes in the genomes of various insect cells expressing recombinant Cyp2d10 were unable species. Human beings have only one gene (AOX1); mouse to mediate this drug biotransformation reaction. Taken has three (Aox1, Aox3, and Aox4); and rat has two (Aox1 and together, this data indicates that human AOX1 biotrans- Aox2). Human AOX1 has 82% amino-acid identity with formed this p38 kinase inhibitor to the principal drug both rat and mouse AOX1.38 metabolite formed in human subjects, which accounts for AOX activity led to the rapid clearance of a novel p38 its unexpectedly rapid elimination in human subjects. kinase inhibitor in human subjects. The pharmaceutical industry has substantial knowledge about the chemical structure of cytochrome P450 substrates and can conveni- Discussion ently assay cytochrome P450-mediated drug metabolism in vitro and in vivo. Although cytochrome P450 enzymes The clinical development of a candidate p38 kinase generally catalyze the oxidation of carbon atoms with a inhibitor was terminated because of its rapid clearance in high electron density, AOXs catalyze the oxidation of human subjects. The half-life and the metabolic profiles of electro-deficient double-bonded carbon atoms that are often this candidate drug in human subjects were vastly different present in nitrogen heterocycles. Although some drugs can in several pre-clinical species studied. Haplotype-based serve as substrates for both types of enzymes, xenobiotics computational genetic analysis of the data generated using that are good substrates for one are usually poor substrates

The Pharmacogenomics Journal Computational in vitro pharmacogenetics X Zhang et al 23

for the other.39 To decrease the frequency of dosing, the 11 Liao G, Wang J, Guo J, Allard J, Cheng J, Ng A et al. In silico genetics: pharmaceutical industry is now producing novel chemical identification of a functional element regulating H2-Ea gene expression. structures with long half-lives that are not substrates for Science 2004; 306: 690–695. 12 Wang J, Liao G, Usuka J, Peltz G. Computational genetics: from mouse cytochrome P450 enzymes. Owing to this, it is likely that to man? Trends Genet 2005; 21: 526–532. enzymes such as AOX may contribute to the metabolism of 13 Wang J, Peltz G. Haplotype-Based Computational Genetic Analysis in Mice. these drugs. Although this is only a single example, the Computational Genetics and Genomics: New Tools for Understanding murine computational pharmacogenetic analysis method Disease. Humana Press Inc.: Totowa, New Jersey, 2005. pp 51–70. 14 Wang J, Liao G, Chang J, Nguyen A, Guo J, Chou S et al. Haplotypic provides a tool that could facilitate the rapid identification Structure of the Mouse Genome. Computational Genetics and Genomics: and characterization of these less commonly used drug New Tools for Understanding Disease. Humana Press Inc.: Totowa, metabolism pathways. As this example shows, after a New Jersey, 2005. pp 71–83. 00 00 00 00 candidate has been computationally identified, the effect 15 Liao G, Zhang X, Clark DJ, Peltz G. A genomic roadmap to better drugs. Drug Metab Rev 2008; 40: 225–239. of a specific enzyme inhibitor can be very rapidly experi- 16 Boxenbaum H, Ronfeld R. Interspecies pharmacokinetic scaling and the mentally assessed using this in vitro drug biotransformation Dedrick plots. Am J Physiol 1983; 245: R768–R775. system. 17 Dedrick RL. Animal scale-up. J Pharmacokinet Biopharmacol 1973; 1: 435–461. 18 Dedrick R, Bischoff KB, Zaharko DS. Interspecies correlation of plasma concentration history of methotrexate (NSC-740). Cancer Chemother Conflict of interest Rep 1970; 54: 95–101. 19 Boxenbaum H. Interspecies pharmacokinetic scaling and the evolu- The manuscript has not been published elsewhere, nor is it tionary-comparative paradigm. Drug Metab Rev 1984; 15: 1071–1121. being evaluated for publication elsewhere. Several authors 20 Kurosaki M, Demontis S, Barzago MM, Garattini E, Terao M. Molecular cloning of the cDNA coding for mouse aldehyde oxidase: tissue are employees of Roche Palo Alto, but there are no other distribution and regulation in vivo by testosterone. Biochem J 1999; conflicts of interest around this manuscript. 341(Part 1): 71–80. 21 Zheng M, Shafer SS, Liao G, Liu H-H, Peltz G. Computational genetic mapping in mice: ‘The Ship has Sailed’. Sci Transl Med 2009; 1: Acknowledgments 3ps4. 22 Capello S, Henderson L, DeGrazia F, Liberato D, Garland W, Town C. This work was partially supported by grant 756 GM068885-05 from The effect of the cytochrome P-450 suicide inactivator, 1-aminobenzo- the NIGMS awarded to GP. We thank Michael Soth and Adrian triazole, on the in vivo metabolism and pharmacologic activity of Fretland for valuable discussions and Zhaomei Zhang, Hoangdung flurazepam. Drug Metab Dispos 1990; 18: 190–196. Ho, Erin Farrell, Wendy Klopf, and Olga Vilenski for reviewing the 23 Mathews JM, Dostal LA, Bend JR. Inactivation of rabbit pulmonary manuscript. cytochrome P-450 in microsomes and isolated perfused lungs by the suicide substrate 1-aminobenzotriazole. J Pharmacol Exp Ther 1985; 235: 186–190. References 24 Panoutsopoulos GI, Beedham C. Enzymatic oxidation of vanillin, isovanillin and protocatechuic aldehyde with freshly prepared Guinea 1 Singh SS. Preclinical pharmacokinetics: an approach towards safer and pig liver slices. Cell Physiol Biochem 2005; 15: 89–98. efficacious drugs. Curr Drug Metab 2006; 7: 165–182. 25 Klaassen CD. Casarett & Doull’s Toxicology: The Basic Science of Poisons. 2 Poggesi I. Predicting human pharmacokinetics from preclinical data. 6th edn. The McGraw-Hill Companies, Inc.: Columbus, OH, 2001. Curr Opin Drug DiscovDevel 2004; 7: 100–111. 26 Sugihara K, Kitamura S, Tatsumi K. Involvement of mammalian liver 3 Nelson DR, Zeldin DC, Hoffman SM, Maltais LJ, Wain HM, Nebert DW. cytosols and aldehyde oxidase in reductive metabolism of zonisamide. Comparison of cytochrome P450 (CYP) genes from the mouse and Drug MetabDispos 1996; 24: 199–202. human genomes, including nomenclature recommendations for genes, 27 Kawashima K, Hosoi K, Naruke T, Shiba T, Kitamura M, Watabe T. pseudogenes and alternative-splice variants. Pharmacogenetics 2004; Aldehyde oxidase-dependent marked species difference in hepatic 14: 1–18. metabolism of the sedative-hypnotic, zaleplon, between monkeys and 4 Martignoni M, Groothuis GM, de Kanter R. Species differences between rats. Drug Metab Dispos 1999; 27: 422–428. mouse, rat, dog, monkey and human CYP-mediated drug metabolism, 28 Beedham C, Miceli JJ, Obach RS. Ziprasidone metabolism, aldehyde inhibition and induction. Expert Opin Drug Metab Toxicol 2006; 2: oxidase, and clinical implications. J Clin Psychopharmacol 2003; 23: 875–894. 229–232. 5 Lin JH. Species similarities and differences in pharmacokinetics. Drug 29 Clarke SE, Harrell AW, Chenery RJ. Role of aldehyde oxidase in the Metab Dispos 1995; 23: 1008–1021. in vitro conversion of famciclovir to penciclovir in human liver. Drug 6 Casley WL, Menzies JA, Girard M, Larocque L, Mousseau N, Whitehouse Metab Dispos 1995; 23: 251–254. LW et al. Differences in caffeine 3-demethylation activity among 30 Rashidi MR, Smith JA, Clarke SE, Beedham C. In vitro oxidation of inbred mouse strains: a comparison of hepatic Cyp1a2 gene expres- famciclovir and 6-deoxypenciclovir by aldehyde oxidase from human, sion between two inbred strains. Fundam Appl Toxicol 1997; 40: guinea pig, rabbit, and rat liver. Drug Metab Dispos 1997; 25: 805–813. 228–237. 31 Beedham C. Molybdenum hydroxylases as drug-metabolizing enzymes. 7 Bernard-Helary K, Ardourel MY, Hevor T, Cloix JF. In vivo and in vitro Drug Metab Rev 1985; 16: 119–156. glycogenic effects of methionine sulfoximine are different in two inbred 32 Gluecksohn-Waelsch S, Greengard P, Quinn GP, Teicher LS. Genetic strains of mice. Brain Res 2002; 929: 147–155. variations of an oxidase in mammals. J Biol Chem 1967; 242: 8 Koster AS, Nieuwenhuis L, Frankhuijzen-Sierevogel AC. Comparison of 1271–1273. microsomal drug-metabolizing enzymes in 14 rat inbred strains. 33 Kitamura S, Sugihara K, Ohta S. Drug-metabolizing ability of molyb- Biochem Pharmacol 1989; 38: 759–765. denum hydroxylases. Drug Metab Pharmacokinet 2006; 21: 83–98. 9 Guo YY, Weller PF, Farrell E, Cheung P, Fitch B, Clark D et al. In silico 34 Itoh K, Maruyama H, Adachi M, Hoshino K, Watanabe N, Tanaka Y. Lack pharmacogenetics: warfarin metabolism. Nat Biotechnol 2006; 24: of dimer formation ability in rat strains with low aldehyde oxidase 531–536. activity. Xenobiotica 2007; 37: 709–716. 10 Guo YY, Liu P, Zhang X, Weller PMM, Wang J, Liao G et al. In vitro and in 35 Itoh K, Maruyama H, Adachi M, Hoshino K, Watanabe N, Tanaka Y. silico pharmacogenetic analysis in mice. Proc Natl Acad Sci USA 2007; Lack of formation of aldehyde oxidase dimer possibly due to 377G4A 104: 17735–17740. nucleotide substitution. Drug Metab Dispos 2007; 35: 1860–1864.

The Pharmacogenomics Journal Computational in vitro pharmacogenetics X Zhang et al 24

36 Adachi M, Itoh K, Masubuchi A, Watanabe N, Tanaka Y. Construction 38 Wright RM, Clayton DA, Riley MG, McManaman JL, Repine JE. cDNA and expression of mutant cDNAs responsible for genetic polymorphism cloning, sequencing, and characterization of male and female rat in aldehyde oxidase in Donryu strain rats. J Biochem Mol Biol 2007; 40: liver aldehyde oxidase (rAOX1). Differences in redox status may 1021–1027. distinguish male and female forms of hepatic APX. J Biol Chem 1999; 37 Smith MA, Marinaki AM, Arenas M, Shobowale-Bakre M, Lewis CM, 274: 3878–3886. Ansari A et al. Novel pharmacogenetic markers for treatment outcome 39 Rettie AE, Fisher MB. Transformation enzymes: oxidative; non-P450. in azathioprine-treated inflammatory bowel disease. Aliment Pharmacol In: Woolf TF (ed). Handbook of Drug Metabolism. Marcel Dekker: Ther 2009; 30: 375–384. New York, New York, 1999. pp 131–151.

Supplementary Information accompanies the paper on the The Pharmacogenomics Journal website (http://www.nature.com/tpj)

The Pharmacogenomics Journal