a global drug inhibition pattern for the human atp-binding...

12
A Global Drug Inhibition Pattern for the Human ATP-Binding Cassette Transporter Breast Cancer Resistance Protein (ABCG2) S Pa ¨ r Matsson, Gunilla Englund, Gustav Ahlin, Christel A. S. Bergstro ¨ m, Ulf Norinder, and Per Artursson Pharmaceutical Screening and Informatics, Department of Pharmacy, Uppsala University, Sweden (P.M., G.E., G.A., C.A.S.B., U.N., P.A.); and AstraZeneca R&D, So ¨ derta ¨ lje, Sweden (U.N.) Received May 9, 2007; accepted July 5, 2007 ABSTRACT In this article, we explore the entire structural space of regis- tered drugs to obtain a global model for the inhibition of the drug efflux transporter breast cancer resistance protein (BCRP; ABCG2). For this purpose, the inhibitory effect of 123 structur- ally diverse drugs and drug-like compounds on mitoxantrone efflux was studied in Saos-2 cells transfected with human wild-type (Arg482) BCRP. The search for BCRP inhibitors throughout the drug-like chemical space resulted in the identi- fication of 29 previously unknown inhibitors. The frequency of BCRP inhibition was 3 times higher for compounds reported to interact with other ATP-binding cassette (ABC) transporters than for compounds without reported ABC transporter affinity. An easily interpreted computational model capable of discrim- inating inhibitors from noninhibitors using only two molecular descriptors, octanol-water partition coefficient at pH 7.4 and molecular polarizability, was constructed. The discriminating power of this two-descriptor model was 93% for the training set and 79% for the test set, respectively. The results were sup- ported by a global pharmacophore model and are in agreement with a two-step mechanism for the inhibition of BCRP, where both the drug’s capacity to insert into the cell membrane and to interact with the inhibitory binding site of the transporter are important. The ATP-binding cassette (ABC) transporter breast cancer resistance protein (BCRP; ABCG2) has received much attention for its role in resistance to various cytotoxic agents (Doyle et al., 1998; Krishnamurthy and Schuetz, 2006) and has recently been shown to also influence the disposition of structurally unrelated drugs from other therapeutic classes (Gupta et al., 2004; Jonker et al., 2005; Zhang et al., 2005). BCRP is expressed in many tissue barriers throughout the body, including the intestine, the blood-brain barrier, the blood-placenta barrier, and the liver canalicular membrane (Maliepaard et al., 2001; Fetsch et al., 2006). A picture is emerging that, similar to the most well studied ABC transporter, P-glycoprotein (ABCB1), BCRP inter- acts with a wide variety of compounds, and it is one of the major ABC transporters affecting drug disposition throughout the body. The key role of BCRP in drug disposition was recently exemplified by a 111 times higher systemic exposure to the antiinflammatory drug sulfasalazine after oral administration to Bcrp1-knockout mice compared with wild-type mice (Zaher et al., 2006). Furthermore, the human oral bioavailability of the BCRP substrate topotecan was more than doubled after coad- ministration with the potent inhibitor GF120918 (Elacridar) (Kruijtzer et al., 2002), highlighting the risk of significantly altered drug exposure due to inhibition of BCRP. It would therefore be of great interest to develop models that can predict drug-mediated BCRP inhibition, but so far, studies have been limited to structurally homologous series of compounds (Gupta et al., 2004), and the only published computational model was not validated with an external test set (Saito et al., 2006). Most studies of BCRP-mediated drug transport have been This work was supported by the Swedish Research Council (Grant 9478), by the Knut and Alice Wallenberg Foundation, by the Swedish Fund for Research without Animal Experiments, and by the Swedish Animal Welfare Agency. Article, publication date, and citation information can be found at http://jpet.aspetjournals.org. doi:10.1124/jpet.107.124768. S The online version of this article (available at http://jpet.aspetjournals.org) contains supplemental material. ABBREVIATIONS: ABC, ATP-binding cassette; BCRP, breast cancer resistance protein; ABCG2, ATP-binding cassette transporter member G2; P-gp, P-glycoprotein; PCR, polymerase chain reaction; PBS, phosphate-buffered saline; HBSS, Hanks’ balanced salt solution; DMSO, dimethyl sulfoxide; 3D, three-dimensional; OPLS-DA, orthogonal partial least-squares projection to latent structures discriminant analysis; SLC, solute carrier; MRP, multidrug resistance-associated protein; logD 7.4 , octanol-water partition coefficient at pH 7.4; Ko143, 3-(6-isobutyl-9-methoxy-1,4- dioxo-1,2,3,4,6,7,12,12-octahydropyrazino[1,2:1,6]pryrido[3,4-b]indol-3-yl)-propionic acid tert-butyl ester. 0022-3565/07/3231-19–30$20.00 THE JOURNAL OF PHARMACOLOGY AND EXPERIMENTAL THERAPEUTICS Vol. 323, No. 1 Copyright © 2007 by The American Society for Pharmacology and Experimental Therapeutics 124768/3253326 JPET 323:19–30, 2007 Printed in U.S.A. 19 http://jpet.aspetjournals.org/content/suppl/2007/07/09/jpet.107.124768.DC1 Supplemental material to this article can be found at: at ASPET Journals on December 4, 2018 jpet.aspetjournals.org Downloaded from

Upload: vannhi

Post on 05-Dec-2018

215 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: A Global Drug Inhibition Pattern for the Human ATP-Binding ...jpet.aspetjournals.org/content/jpet/323/1/19.full.pdf · scribe RTase (50 U/ l), and 21 l of nuclease-free water. The

A Global Drug Inhibition Pattern for the Human ATP-BindingCassette Transporter Breast Cancer Resistance Protein(ABCG2)□S

Par Matsson, Gunilla Englund, Gustav Ahlin, Christel A. S. Bergstrom, Ulf Norinder, andPer ArturssonPharmaceutical Screening and Informatics, Department of Pharmacy, Uppsala University, Sweden (P.M., G.E., G.A., C.A.S.B.,U.N., P.A.); and AstraZeneca R&D, Sodertalje, Sweden (U.N.)

Received May 9, 2007; accepted July 5, 2007

ABSTRACTIn this article, we explore the entire structural space of regis-tered drugs to obtain a global model for the inhibition of thedrug efflux transporter breast cancer resistance protein (BCRP;ABCG2). For this purpose, the inhibitory effect of 123 structur-ally diverse drugs and drug-like compounds on mitoxantroneefflux was studied in Saos-2 cells transfected with humanwild-type (Arg482) BCRP. The search for BCRP inhibitorsthroughout the drug-like chemical space resulted in the identi-fication of 29 previously unknown inhibitors. The frequency ofBCRP inhibition was 3 times higher for compounds reported tointeract with other ATP-binding cassette (ABC) transportersthan for compounds without reported ABC transporter affinity.

An easily interpreted computational model capable of discrim-inating inhibitors from noninhibitors using only two moleculardescriptors, octanol-water partition coefficient at pH 7.4 andmolecular polarizability, was constructed. The discriminatingpower of this two-descriptor model was 93% for the training setand 79% for the test set, respectively. The results were sup-ported by a global pharmacophore model and are in agreementwith a two-step mechanism for the inhibition of BCRP, whereboth the drug’s capacity to insert into the cell membrane and tointeract with the inhibitory binding site of the transporter areimportant.

The ATP-binding cassette (ABC) transporter breast cancerresistance protein (BCRP; ABCG2) has received much attentionfor its role in resistance to various cytotoxic agents (Doyle et al.,1998; Krishnamurthy and Schuetz, 2006) and has recently beenshown to also influence the disposition of structurally unrelateddrugs from other therapeutic classes (Gupta et al., 2004; Jonkeret al., 2005; Zhang et al., 2005). BCRP is expressed in manytissue barriers throughout the body, including the intestine, theblood-brain barrier, the blood-placenta barrier, and the livercanalicular membrane (Maliepaard et al., 2001; Fetsch et al.,2006). A picture is emerging that, similar to the most well

studied ABC transporter, P-glycoprotein (ABCB1), BCRP inter-acts with a wide variety of compounds, and it is one of the majorABC transporters affecting drug disposition throughout thebody. The key role of BCRP in drug disposition was recentlyexemplified by a 111 times higher systemic exposure to theantiinflammatory drug sulfasalazine after oral administrationto Bcrp1-knockout mice compared with wild-type mice (Zaher etal., 2006). Furthermore, the human oral bioavailability of theBCRP substrate topotecan was more than doubled after coad-ministration with the potent inhibitor GF120918 (Elacridar)(Kruijtzer et al., 2002), highlighting the risk of significantlyaltered drug exposure due to inhibition of BCRP. It wouldtherefore be of great interest to develop models that can predictdrug-mediated BCRP inhibition, but so far, studies have beenlimited to structurally homologous series of compounds (Guptaet al., 2004), and the only published computational model wasnot validated with an external test set (Saito et al., 2006).

Most studies of BCRP-mediated drug transport have been

This work was supported by the Swedish Research Council (Grant 9478), bythe Knut and Alice Wallenberg Foundation, by the Swedish Fund for Researchwithout Animal Experiments, and by the Swedish Animal Welfare Agency.

Article, publication date, and citation information can be found athttp://jpet.aspetjournals.org.

doi:10.1124/jpet.107.124768.□S The online version of this article (available at http://jpet.aspetjournals.org)

contains supplemental material.

ABBREVIATIONS: ABC, ATP-binding cassette; BCRP, breast cancer resistance protein; ABCG2, ATP-binding cassette transporter member G2;P-gp, P-glycoprotein; PCR, polymerase chain reaction; PBS, phosphate-buffered saline; HBSS, Hanks’ balanced salt solution; DMSO, dimethylsulfoxide; 3D, three-dimensional; OPLS-DA, orthogonal partial least-squares projection to latent structures discriminant analysis; SLC, solutecarrier; MRP, multidrug resistance-associated protein; logD7.4, octanol-water partition coefficient at pH 7.4; Ko143, 3-(6-isobutyl-9-methoxy-1,4-dioxo-1,2,3,4,6,7,12,12�-octahydropyrazino[1�,2�:1,6]pryrido[3,4-b]indol-3-yl)-propionic acid tert-butyl ester.

0022-3565/07/3231-19–30$20.00THE JOURNAL OF PHARMACOLOGY AND EXPERIMENTAL THERAPEUTICS Vol. 323, No. 1Copyright © 2007 by The American Society for Pharmacology and Experimental Therapeutics 124768/3253326JPET 323:19–30, 2007 Printed in U.S.A.

19

http://jpet.aspetjournals.org/content/suppl/2007/07/09/jpet.107.124768.DC1Supplemental material to this article can be found at:

at ASPE

T Journals on D

ecember 4, 2018

jpet.aspetjournals.orgD

ownloaded from

Page 2: A Global Drug Inhibition Pattern for the Human ATP-Binding ...jpet.aspetjournals.org/content/jpet/323/1/19.full.pdf · scribe RTase (50 U/ l), and 21 l of nuclease-free water. The

performed in inside-out membrane vesicles, in which BCRPsubstrates added to the extravesicular medium are activelytransported to the vesicle interior (Ozvegy et al., 2001; Saitoet al., 2006). Inside-out membrane vesicles can thereby pro-vide direct information about the binding of drugs to thetransporter from the intracellular compartment. However, inmany BCRP-expressing tissues, such as the intestine and theblood-brain barrier, the efflux transporter is located in theexternally facing membrane (Maliepaard et al., 2001; Fetschet al., 2006). BCRP, therefore, most probably binds its sub-strates directly on their entry into the cell; consequently, itcan be hypothesized that the plasma membrane plays a ma-jor role in the presentation of drugs to the transporter.

This hypothesis is supported by an increasing body of ev-idence indicating that the well studied ABC transporter P-gpbinds its substrates and inhibitors from within the innerleaflet of the plasma membrane (Homolya et al., 1993; Sha-piro and Ling, 1997; Gatlik-Landwojtowicz et al., 2006). Inthis model, the substrate first partitions into the plasmamembrane and, after lateral diffusion within the membrane,binds to the transporter from the lipid phase in a second step.The two-step model for drug-P-gp interaction was recentlycorroborated by Omote and Al-Shawi (2006), who presented amodel supported by molecular dynamics simulations, wheresurface active substrates partition into the lipid membraneand subsequently exchange their interactions with water orthe polar lipid head groups in the membrane-cytosol inter-face for interactions with polar and charged amino acids inthe lipid-water interface region of P-gp. This study clearlydemonstrates the importance of the plasma membrane fordetermining drug binding to P-gp, but the mechanisticknowledge of drug transport mediated by other importantABC drug transporters such as BCRP is still lacking.

In this study, we use a global approach to determine thephysicochemical properties necessary for drug binding toBCRP. A data set of 123 compounds spanning the entirestructural space of registered drugs was studied for the in-hibition of BCRP-mediated mitoxantrone efflux in Saos-2cells transfected with human wild-type (Arg482) BCRP(Wierdl et al., 2003), resulting in the discovery of 29 previ-ously unknown BCRP inhibitors. An easily interpreted com-putational model was developed that is capable of discrimi-nating inhibitors from noninhibitors based on structuralfeatures related both to the drug’s capacity to insert into thecell membrane and to interact with the inhibitory bindingsite of the transporter. The results were supported by aglobal pharmacophore model and are in agreement with atwo-step mechanism for inhibition of BCRP.

Materials and MethodsMaterials. Fumitremorgin C was kindly provided by Dr. Robert

Robey (National Institutes of Health, Bethesda, MD). Ko143 was akind gift from Dr. Gerrit-Jan Koomen (Van’t Hoff Institute for Mo-lecular Sciences, University of Amsterdam, The Netherlands). Cele-coxib was a gift from Pfizer (Kalamazoo, MI). Famciclovir and ima-tinib were gifts from Novartis (Basel, Switzerland). Ganciclovir andvalganciclovir were gifts from Roche (Palo Alto, CA), and erlotiniband saquinavir were from Roche (Basel, Switzerland). GF120918,valacyclovir, and zanamivir were gifts from GlaxoSmithKline (Steve-nage, UK). Ritonavir and lopinavir were kindly provided by Abbott(Abbott Park, IL). Nevirapine and tipranavir were gifts from Boehr-inger-Ingelheim (Ingelheim, Germany). Astemizole was purchased

from MP Biomedicals (Eschwege, Germany). All other compoundswere purchased from Sigma-Aldrich (St. Louis, MO) and were of atleast 95% purity. Structure representations and references for com-pounds lacking generic names are provided in Supplemental Fig. 1.

Cell Culture Procedure. Saos-2 cells transfected with wild-type(Arg482) human BCRP (Saos-2/wtABCG2) and control cells trans-fected with the parental transfection vector pcDNA3 (Saos-2/pcDNA3) were kindly provided by Dr. J. Schuetz (St. Jude Children’sResearch Hospital, Memphis, TN) (Wierdl et al., 2003). The cellswere cultured in Dulbecco’s modified Eagle’s medium (Invitrogen,Carlsbad, CA) containing 10% fetal calf serum (Sigma-Aldrich) and 1mg/ml Geneticin (G-418; Invitrogen) under an atmosphere of 5% CO2

at 37°C.Analysis of Transporter Expression. Total RNA was isolated

with the RNAeasy minikit (QIAGEN, Hilden, Germany), using theprotocol provided by the manufacturer with the addition of an extraon-column DNase step (QIAGEN). The RNA quality was measuredusing a Bioanalyser (Agilent, Palo Alto, CA), and RNA concentrationwas measured using a Nanodrop ND-1000 Spectrophotometer(Nanodrop, Wilmington, DE). cDNA was synthesized using the HighCapacity cDNA Archive kit (Applied Biosystems, Foster City, CA)according to the manufacturer’s protocol. Five hundred nanogramsof the total RNA samples was added to a master mixture containing10 �l of 10� reverse transcriptase buffer, 4 �l of 25� deoxynucleo-side-5�-triphosphates, 10 �l of 10� random primers, 5 �l of Multi-scribe RTase (50 U/�l), and 21 �l of nuclease-free water. The reversetranscriptase PCR mixture was incubated at 25°C for 10 min and at37°C for 120 min, and the resulting cDNA was stored at �70°Cpending quantitative PCR analysis.

The amount of cDNA in the samples was analyzed in an ABIPrism 7900 HT Sequence Detection System with custom-designed384-well cards loaded with Assay-on-Demand gene expression as-says (Applied Biosystems). The quantitative PCR analysis was per-formed using 1 �l of reaction mixture per gene, containing 1 ng ofcDNA, TaqMan Universal PCR Master Mix (consisting of AmpliTaqGold DNA polymerase, deoxynucleoside-5�-triphosphates withdUTP, passive reference, and optimized buffer), and the Assay-on-Demand gene expression product mixture containing the specificprimers and the probe preloaded on the plate. The cycling conditionsconsisted of 2 min at 50°C, 10 min of polymerase activation at 95°C,and 40 cycles alternating between 95°C for 15 s and 60°C for 1 min.Amplification curves were analyzed using the SDS version 2.1 soft-ware (Applied Biosystems), extracting the threshold concentration(Ct) value as the cycle time when fluorescence is above a definedthreshold level. The relative gene expression was determined as2��Ct, using cyclophilin A as the internal standard.

Mitoxantrone Efflux Kinetics. Saos-2/wtABCG2 or Saos-2/pcDNA3 cells were seeded in 96-well plates (Corning Life Sciences,Acton, MA) 48 h before the experiment. On the day of the experi-ment, the cells were washed twice with 37°C PBS (Invitrogen) andincubated for 60 min in Hanks’ balanced salt solution (HBSS; Sigma-Aldrich), buffered with 25 mM HEPES to pH 7.4 and containing 0.5to 50 �M mitoxantrone. After the uptake phase, the incubationsolution was removed, and the cells were washed twice with 37°CPBS. Fresh PBS was added, and the mitoxantrone efflux was fol-lowed for 60 min at 37°C using continuous fluorescence measure-ment in a Tecan Saphire2 plate reader (Tecan, Mannedorf, Switzer-land), with the excitation at 633 nm and detection at the 735-nmemission peak wavelength. Nonlinear regression (Prism version4.02; GraphPad, San Diego, CA) was used to determine Michaelis-Menten kinetic parameters from the initial efflux rates, according toeq. 1:

V �V max � �S�

Km � �S�(1)

20 Matsson et al.

at ASPE

T Journals on D

ecember 4, 2018

jpet.aspetjournals.orgD

ownloaded from

Page 3: A Global Drug Inhibition Pattern for the Human ATP-Binding ...jpet.aspetjournals.org/content/jpet/323/1/19.full.pdf · scribe RTase (50 U/ l), and 21 l of nuclease-free water. The

where V is the efflux rate, Vmax is the maximal efflux rate, [S] is themitoxantrone concentration, and Km is the mitoxantrone concentra-tion at which V � 0.5 � Vmax.

Efflux Inhibition Assay. A single-concentration method able todetermine the affinity of both competitive and noncompetitive BCRPinhibitors was developed using Saos-2 cells transfected with humanwild-type (Arg482) BCRP. Saos-2/wtABCG2 or Saos-2/pcDNA3 cellswere seeded in 96-well plates 48 to 72h before the experiment.DMSO stock solutions of the compounds under study were diluted inHBSS to a final concentration of 50 �M, resulting in final DMSOlevels of no more than 1% (v/v). As a comparison, DMSO levels of upto 10% were shown not to influence the mitoxantrone accumulationin control experiments. Cells were washed twice in 37°C PBS andincubated for 60 min in HBSS buffered with HEPES to pH 7.4,containing the studied compound and 1 �M mitoxantrone. After theuptake phase, the incubation solution was removed, and the cellswere washed twice with ice-cold PBS. The cells were detached using25 �l of trypsin solution (PBS containing 0.25% trypsin and 0.03%EDTA) and were resuspended in 175 �l of ice-cold PBS containing2% fetal calf serum, 0.5% sodium azide, and 3.2 mM trisodiumcitrate. The cells were then placed on ice, and the intracellularmitoxantrone fluorescence was analyzed using a Beckman CoulterFC500 flow cytometer (Beckman Coulter, Fullerton, CA) with theexcitation at 633 nm and detection using the FL4 channel (�675nm). Flow cytometric detection was preferred over detection in flu-orescence plate reader because this resulted in a higher signal/noiseratio (8� background compared with 2� background, SupplementalFig. 2). The cells were gated based on forward and side scatter, andonly viable cells (typically �80% of all analyzed events) were in-cluded in the analysis. This method of determining cell viabilityrelies on the fact that nonviable cells differ from viable cells in sizeand shape. The procedure was validated in control experiments bycostaining with the fluorescent viability marker propidium iodide.

To verify that intrinsic fluorescence of the studied compounds didnot interfere with the mitoxantrone analysis, cells incubated with 50�M compound solutions were analyzed using the same cytometersettings as in the mitoxantrone analysis. Doxorubicin was excludedfrom the data set because its intrinsic fluorescence significantlyinfluenced the analysis. All other compounds in the data set hadnegligible intrinsic fluorescence at the selected wavelength (10%fluorescence increase compared with the background levels detectedin cells incubated with HBSS buffer only). To determine whether theobserved increases in mitoxantrone accumulation were caused byspecific inhibition of BCRP, a representative selection of the com-pounds (corresponding to 67% of the identified inhibitors) was alsostudied in mock vector-transfected cells. For all compounds, signifi-cantly lower effects on mitoxantrone accumulation were observed incontrol cells than in cells expressing BCRP (Supplemental Fig. 3).Therefore, we conclude that the observed increases in mitoxantroneaccumulation are BCRP-specific.

The -fold increase in intracellular accumulation of mitoxantroneon coincubation with the compounds under study was used as ameasure of the BCRP inhibition. The increase in mitoxantrone ac-cumulation was normalized to that obtained using 0.5 �M of thepotent BCRP inhibitor Ko143 (100% inhibition). Compounds thatincreased the intracellular accumulation of mitoxantrone by morethan a factor of 3 were classified as BCRP inhibitors. The statisticalvalidity of the chosen cut-off value was shown by the fact that itresulted in the maximal statistical significance when the Student’s ttest was used to compare the experimental activity of the compoundsabove and below the cut-off.

The stability of the intracellular mitoxantrone levels throughoutthe analysis was demonstrated by incubating a full 96-well plate ofSaos-2/wtABCG2 cells with 1 �M mitoxantrone solution. This re-sulted in a total coefficient of variation of 7% over the plate, and themitoxantrone levels in the 12 last analyzed wells were statisticallyindistinguishable from those in the first 12 wells (p � 0.4; Student’st test). The interday variability of the intracellular fluorescence in

Saos-2/wtABCG2 cells incubated with 1 �M mitoxantrone was 8%(n � 20). Likewise, the effect of the potent BCRP inhibitor Ko143 onintracellular mitoxantrone levels was determined on five separateoccasions, giving an interday variability of 9%.

Computational Modeling. Molecular structures obtained fromSciFinder Scholar 2006 (American Chemical Society, WashingtonDC) were used as the input for 3D structure generation using Corinaversion 3.0 (Molecular Networks, Erlangen, Germany). Octanol-wa-ter partition coefficients (SlogD7.4) were calculated from the 3Dstructures using ADMETPredictor version 1.2.4 (SimulationsPlus,Lancaster, CA). Two-dimensional molecular descriptors were calcu-lated using the software package SELMA (AstraZeneca R&D, Moln-dal, Sweden). SELMA calculates a collection of commonly used mo-lecular descriptors representing molecular size, flexibility,connectivity, polarity, charge, and hydrogen bonding potential.

The data set was divided into a training set used for model devel-opment and a test set used to validate the predictivity of the finalmodel. The data set division was performed in two steps. First, thedata set was divided into two groups based on their experimentallydetermined BCRP inhibitory effect. Eighteen representative com-pounds from the group of compounds inhibiting BCRP and 25 rep-resentative noninhibitors were then included in the test set, corre-sponding to approximately one third of the compounds in eachactivity group. The remaining 28 BCRP inhibitors and 52 noninhibi-tors were used as training set compounds. The test set selectionprocedure was based on ChemGPS descriptions of the molecules(Oprea and Gottfries, 2001). In ChemGPS, the position of a com-pound in the drug-like chemical space is determined using principalcomponents calculated from descriptors of their chemical structure.By selecting the compounds with the largest distance to their nearestneighbors, structural diversity is maximized, resulting in a test setwith structural features that are representative of the compounds inthe training set (Fig. 1).

Orthogonal partial least-squares projection to latent structuresdiscriminant analysis (OPLS-DA), as implemented in Simca-P ver-sion 11.5 (Umetrics, Umeå, Sweden), was used to derive multivariateclassification models for separating BCRP inhibitors from noninhibi-tors. To optimize the models, a variable selection procedure was usedin which groups of molecular descriptors that did not contain infor-mation relevant to the problem (i.e., noise) were removed in a step-wise manner. Descriptors were kept outside the model if removingthem resulted in a statistically improved model, based on the leave-n-out cross-validated coefficient of determination (Q2). In addition tothe cross-validation procedure, the statistical validity of the models

−100102030

−6−3036−8

−4

0

4

8

t1t3

t2

Fig. 1. The data set investigated is representative of registered oraldrugs. The positions of the compounds in the drug space are determinedby the first three ChemGPS principal components (t1, t2, and t3), whichare summarized from a large number of molecular descriptors and de-scribe mainly the size, polarity, and flexibility of the molecules, respec-tively. The large blue circles denote compounds in the training set,whereas the large yellow diamonds denote test set compounds. The smallblack circles denote a reference set of 150 registered drugs from thePhysician’s Desk Reference (2005). The test set was selected to be repre-sentative of the compounds in the training set.

Inhibition Pattern for the Human BCRP/ABCG2 Transporter 21

at ASPE

T Journals on D

ecember 4, 2018

jpet.aspetjournals.orgD

ownloaded from

Page 4: A Global Drug Inhibition Pattern for the Human ATP-Binding ...jpet.aspetjournals.org/content/jpet/323/1/19.full.pdf · scribe RTase (50 U/ l), and 21 l of nuclease-free water. The

was tested using a random permutation test, in which the order ofthe response variable was randomly changed 100 times.

Common Features Pharmacophore Modeling. The three-di-mensional molecular structures (obtained as described above) wereimported into MacroModel version 9 (Schrodinger, San Diego, CA).The structures were energy minimized in vacuum by a 2000-stepPolak-Ribiere conjugate gradient procedure using the MMFF94sforce field and a dielectric constant of 1. After performing a premi-nimization of the structures, a 1000-step low-mode conformationalanalysis was used to identify low-energy conformations. Unique con-formations with energy levels lower than 50 kJ/mol above the min-imal energy conformation were stored and used for generating thepharmacophore models.

Low-energy conformations for the BCRP inhibitors in the trainingset were imported into Catalyst version 4.9 (Accelrys, San Diego,CA). Ten common feature pharmacophore hypotheses were devel-oped using the common features algorithm (HipHop) with hydrogenbond acceptor, hydrogen bond donor, positively ionizable, negativelyionizable, hydrophobic, and ring aromatic features as possible phar-macophore features. The ability of the 10 hypotheses proposed byCatalyst to distinguish between BCRP inhibitors and noninhibitorsin the training set was examined, and the hypothesis with thehighest discriminatory power was selected for visualization of thepreferential orientation of the BCRP inhibitors.

ResultsEndogenous Transporter Expression in the BCRP

Efflux Assay. Previous studies of active drug transport intransfected cell lines have shown that endogenous transport-ers and drug-metabolizing enzymes can have a significantinfluence on the results, obscuring the effects of the trans-fected transporter (e.g., Goh et al., 2002). We therefore usedreal-time PCR to determine the mRNA expression of 10 ABCtransporters and 24 SLC transporters previously shown totransport drugs, as well as the expression of seven majordrug-metabolizing CYP enzymes (see Supplemental Fig. 4).

In agreement with previous results on mRNA and proteinexpression (Wierdl et al., 2003), the Saos-2/wtABCG2 cellsexpressed significantly higher levels of BCRP/ABCG2 than ofother ABC transporters (e.g., 17 times higher expressionthan of MRP1/ABCC1, which was the ABC transporter withthe highest endogenous expression in Saos-2/wtABCG2cells), whereas BCRP expression in cells transfected with thecontrol pcDNA3 vector was undetectable. Both cell lines ex-hibited low or undetectable expression levels of major effluxtransporter genes such as P-gp/ABCB1, MRP1/ABCC1, andMRP2/ABCC2, as well as of genes encoding major SLC drugtransporters and drug-metabolizing CYP enzymes. The onlySLC transporter with significant expression was SLC16A1(monocarboxylate transporter 1), which was equally ex-pressed in both Saos-2/pcDNA3 and Saos-2/wtABCG2 cells atlevels corresponding to one fifth of BCRP/ABCG2. Only onecompound in the data set, quercetin, has been reported to

Fig. 2. Characterization of the inhibition assay. A, mitoxantrone effluxkinetics in Saos-2 cells stably transfected with human wild-type (Arg482)BCRP (closed symbols) or in cells transfected with a control vector (opensymbols). After loading the cells with mitoxantrone for 60 min, the effluxwas monitored using continuous fluorescence detection. The data arepresented as the mean of the initial efflux rates � S.E. (n � 4). B,inhibition of mitoxantrone efflux after coincubation with increasing con-centrations of Ko143. The intracellular mitoxantrone accumulation

was detected using flow cytometry after an incubation time of 60 min. Theclosed symbols show the inhibition induced by Ko143 in BCRP-trans-fected cells (IC50 � 0.19 � 0.09 �M), and the open symbols show theabsence of inhibition in control vector transfected cells. The data arepresented as the mean mitoxantrone accumulation � S.E. (n � 4), nor-malized to the accumulation observed in control cells incubated with 1�M mitoxantrone (100% inhibition). C, confocal microscopy visualizationof intracellular mitoxantrone levels after incubation of BCRP-transfectedcells with 1 �M mitoxantrone and 0, 0.1, 0.25, or 0.5 �M Ko143. Therightmost panels show control cells incubated with 1 �M mitoxantronewith and without the addition of 0.25 �M Ko143.

22 Matsson et al.

at ASPE

T Journals on D

ecember 4, 2018

jpet.aspetjournals.orgD

ownloaded from

Page 5: A Global Drug Inhibition Pattern for the Human ATP-Binding ...jpet.aspetjournals.org/content/jpet/323/1/19.full.pdf · scribe RTase (50 U/ l), and 21 l of nuclease-free water. The

interact with monocarboxylate transporter 1 (Ozawa et al.,2004). The results indicate that active efflux in the BCRP-transfected Saos-2 cells is not confounded by other transport-ers or by oxidative metabolism, and we conclude that Saos-2/wtABCG2 cells constitute a close to ideal model forstudying the interplay between active efflux and passivemembrane permeability in isolation from confounding fac-tors.

Characterization of the BCRP Efflux Assay. Mitox-antrone was chosen as a model substrate to follow BCRP-mediated efflux from Saos-2/wtABCG2 cells because mitox-antrone levels are readily measured using fluorescencedetection (Doyle et al., 1998; Gupta et al., 2004). Figure 2Ashows the concentration dependence of the initial efflux rateof mitoxantrone from Saos-2/wtABCG2 cells at nontoxic con-centrations, using continuous detection in a fluorescenceplate reader. The efflux kinetics were determined from thisplot using nonlinear regression, resulting in an apparent Km

of 18 � 3 �M. The Km determination was in agreement withresults obtained using flow cytometric detection (Km � 17 �9 �M).

Coincubation of BCRP transfectants with 1 �M mitox-antrone and 0.5 �M of the specific BCRP inhibitor Ko143resulted in complete inhibition of the mitoxantrone efflux(Fig. 2B), whereas inhibitor concentrations up to 1 �M didnot significantly influence mitoxantrone efflux in cells trans-fected with the control vector. Confocal microscopy confirmedthat intracellular mitoxantrone levels were completely re-stored in BCRP transfectants treated with 0.5 �M Ko143(Fig. 2C).

Data Set Selection. At the onset of this study, the num-ber of compounds reported to interact with BCRP was quitelow compared with other major ABC transporters, amount-ing to around 30 compounds with an affinity for the wild-typeBCRP. Furthermore, the published compounds were struc-turally homogenous, making predictions of structurally di-verse drug-like molecules unreliable. We therefore startedour data set selection by removing structural analogs fromthe data set obtained from the literature, including 18 of the30 compounds with a published BCRP affinity in our dataset.

Several of the compounds reported to have affinity forBCRP are known to also interact with other ABC transport-ers (Ozawa et al., 2004). We hypothesized that previouslyunknown BCRP-interacting compounds could be foundamong substrates and inhibitors of other major ABC trans-porters, so we added a set of structurally diverse compoundsinteracting with P-gp, MRP1, and MRP2 to the data set. Forcomparison, we also included a set of compounds that had notbeen previously indicated in ABC transporter interactions,selected from an in-house database of drugs from the Physi-cian’s Desk Reference (2005). Structural diversity in both theABC-interacting and the noninteracting compound set wasmaintained by selecting compounds with the largest distanceto their nearest neighbor in the ChemGPS drug-like chemicalspace (Oprea and Gottfries, 2001). Thereby, the coverage ofthe structural space of drug-like compounds was maximized(Fig. 1).

In summary, 123 endogenous or drug-like compounds wereincluded in this study, divided into three groups: 1) com-pounds with a reported affinity for BCRP (n � 18), 2) com-pounds with an affinity for other major ABC transporters (n �

42), and 3) compounds with previously unknown ABC trans-porter affinity, selected to maximize the diversity of the data set(n � 63) (Table 1).

Inhibition of Mitoxantrone Efflux. Based on the kineticparameters for mitoxantrone, we selected a standard concen-tration of 50 �M for the BCRP inhibition studies, correspond-ing to approximately 2.5 times the mitoxantrone Km. At thisconcentration, 46 of the 123 compounds (37%) in this studyinhibited BCRP-mediated mitoxantrone efflux, including all(100%) of the previously reported BCRP inhibitors. It is note-worthy that as many as 29 (63%) of the hits had not previ-ously been reported to be BCRP inhibitors.1 A closer exami-nation of the results revealed a significant enrichment of hitsin the group of compounds that were selected because of theirreported affinity for other ABC transporters. Of the com-pounds in this group, 45% inhibited BCRP, compared with16% in the compounds lacking published ABC affinity (Fig.3). These results support the notion that a significant affinityoverlap exists among the major ABC transporters (Bates etal., 2001).

It is generally accepted that drug binding to P-gp takesplace from within the inner leaflet of the plasma membrane(Gottesman and Pastan, 1993; Gatlik-Landwojtowicz et al.,2006; Omote and Al-Shawi, 2006). In this model, the drugfirst needs to partition to the lipid bilayer, explaining thestrong correlations often observed between P-gp affinity andmeasures related to membrane partitioning (Seelig andLandwojtowicz, 2000). To study whether BCRP functions bya similar mechanism, we included compounds with a widerange in lipophilicity in this study.

Although inhibitory effects were seen for all of the previ-ously reported BCRP inhibitors, four of the BCRP substratesincluded (methotrexate, sulfasalazine, cimetidine, and nitro-furantoin) did not appear as hits in our cell-based experimen-tal assay.1 Because these four compounds also were the mosthydrophilic ones, we reasoned that low passive membranepermeability was the most likely reason for the lack of activ-ity of these compounds. Thus, higher extracellular concen-trations of the compounds should lead to higher intramem-braneous and intracellular levels because of an increasedmass flux. Indeed, significantly increased intracellular mi-toxantrone accumulation was observed at higher concentra-tions of cimetidine and nitrofurantoin, supporting the notionthat passive membrane partitioning limits the access to thetransporter for these compounds (Table 2). However, formethotrexate, the possible concentration range was limitedby significant cell toxicity at concentrations above 500 �M.This very hydrophilic drug is a known BCRP substrate withreported Km as high as 5700 �M (Mitomo et al., 2003), wellexceeding the concentrations used in the present study.Thus, a low intrinsic affinity for the transporter in combina-tion with poor membrane permeability probably explains theabsence of an inhibitory effect for these compounds in thiswork.

To further investigate the importance of plasma membranepartitioning for drug binding to BCRP, we examined the

1 During the course of this study, cimetidine, nitrofurantoin (Jonker et al.,2005), and sulfasalazine (Zaher et al., 2006) were reported to be BCRP sub-strates in independent studies. Because their affinity for BCRP had not beenreported at the onset of this study, they were included in groups B and C inTable 1 and in Fig. 3.

Inhibition Pattern for the Human BCRP/ABCG2 Transporter 23

at ASPE

T Journals on D

ecember 4, 2018

jpet.aspetjournals.orgD

ownloaded from

Page 6: A Global Drug Inhibition Pattern for the Human ATP-Binding ...jpet.aspetjournals.org/content/jpet/323/1/19.full.pdf · scribe RTase (50 U/ l), and 21 l of nuclease-free water. The

TABLE 1The ABC transporter affinities of the 123 compounds in this studyExperimentally determined inhibition of BCRP-mediated mitoxantrone efflux and important molecular descriptors are shown for the compounds in this study, along withdata from the literature on the affinity for P-gp, MRP1, and MRP2 obtained from the University of Tokyo transporter database (Ozawa et al., 2004).

Substance MitoxantroneAccumulationa

MolecularPolarizability logD7.4

ABCB1P-gpb

ABCC1MRP1b

ABCC2MRP2b

-Fold increase

Compounds previously reported as BCRPsubstrates or inhibitorsChrysin*c 8.4 26 2.8 – Y –Gefitinib*c,d 8.0 47 4.4 Y – –Imatinib mesylatec,d 8.0 58 2.4 Y – –Ko143c 8.0 54 3.6 – – –Fumitremorgin C*c 7.9 44 2.2 – – –Diethylstilbestrolc 7.7 34 4.6 Y – –GF120918 (Elacridar)c 7.7 64 4.6 Y N NCyclosporine-A*c 6.4 143 4.5 Y Y YPrazosin*d 5.7 40 1.6 Y – –Saquinavirc 5.7 79 4.6 Y Y –Ritonavir*c 5.1 83 4.3 Y Y –�-Estradiolcd 5.1 35 3.6 Y – –Verapamil*c 4.1 56 3.7 Y Y YTamoxifend 3.8 49 5.0 Y – –Hoechst 33342d 3.7 53 3.1 Y – –Quercetin*c 3.3 27 1.0 Y Y –Omeprazolec 3.1 38 1.6 Y – –Methotrexate*d 1.5 44 �4.8 Y Y Y

Compounds with a previously unknown BCRPaffinity but with a reported affinity for othermajor ABC transportersErgocristine*e 9.1 70 4.4 Y – –Nicardipinee 8.1 53 4.4 Y – –Ethinylestradiole 8.0 38 3.7 Y – –Astemizolee 7.8 55 5.4 Y – –Felodipinee 7.5 39 4.9 Y – –Glibenclamidee 7.4 53 2.7 Y Y YKetoconazolee 6.6 56 4.1 Y – –Chlorprotixene*e 6.5 38 4.3 Y – –Nitrendipinee 6.3 37 3.4 Y – –Chlorpromazinee 5.4 38 3.9 Y – –Progesteronee 4.9 42 3.9 Y – –Mifepristonee 4.7 56 4.9 Y Y –Dipyridamole*c,d,e,f 4.6 57 1.5 Y – –Lopinavire 3.9 76 5.2 Y – –Amiodaronee 3.9 58 4.8 Y – –Simvastatine 3.9 52 4.6 Y – YLoperamidee 3.7 58 4.8 Y – –Terfenadine*e 3.0 62 5.9 Y – –Clotrimazole 3.0 41 5.0 Y – –Spironolactone* 2.6 51 2.9 Y – –Maprotiline* 2.3 38 2.3 Y – –Digoxin* 2.0 87 2.4 Y – –Quinine 2.0 40 2.4 Y – –Fexofenadine* 1.9 62 2.4 Y – –Diltiazem* 1.4 47 2.4 Y – –Erythromycin* 1.4 83 3.1 Y Y –Etoposide 1.3 58 0.3 Y Y YPrednisone* 1.2 41 1.6 Y – –Trimethoprim 1.2 31 1.3 Y – –Chlorzoxazone* 1.2 15 1.8 Y – –Folic acid 1.2 42 �5.4 – – YLansoprazol 1.2 35 2.0 Y – –Ranitidine 1.1 34 0.7 Y – –Cimetidine*d,f 1.1 28 0.2 Y – –Indomethacin 1.1 37 0.2 N Y YPrednisolone 1.1 42 1.5 Y – –Propranolol 1.0 32 1.2 Y – –Timolol 0.9 35 0.2 Y – –Desipramine 0.9 35 1.8 Y – –Pravastatin 0.8 49 0.4 Y – YHydrocortisone 0.8 43 1.6 Y – –Sulfinpyrazone 0.8 46 �1.2 N Y Y

Compounds with no previously reported ABCtransporter affinityFenofibrate*e 7.9 40 5.0 – – –Tipranavir*e 7.4 64 5.0 – –Erlotinib*e 7.2 43 3.3 – – –

24 Matsson et al.

at ASPE

T Journals on D

ecember 4, 2018

jpet.aspetjournals.orgD

ownloaded from

Page 7: A Global Drug Inhibition Pattern for the Human ATP-Binding ...jpet.aspetjournals.org/content/jpet/323/1/19.full.pdf · scribe RTase (50 U/ l), and 21 l of nuclease-free water. The

TABLE 1 continued

Substance MitoxantroneAccumulationa

MolecularPolarizability logD7.4

ABCB1P-gpb

ABCC1MRP1b

ABCC2MRP2b

-Fold increaseFlupentixol*e 6.1 47 3.7 – – –Celecoxibe 5.6 36 3.4 – – –Thioridazinee 4.5 47 4.5 – – –Isradipinee 4.1 39 3.3 – – –Fendiline*e 4.1 43 3.6 – – –Medroxyprogesterone*e 3.9 44 3.5 – – –Pramoxinee 3.0 36 2.7 – – –Piroxicam* 2.7 33 0.2 – – –Terazosin 2.4 42 1.6 – – –Diazoxide 2.0 22 �1.0 – – –Oxazepam* 1.9 30 2.3 – – –Propafenone* 1.9 42 1.6 – – –Tinidazole 1.9 24 �0.4 – – –Meclizine* 1.7 49 4.9 – – –Tetracycline* 1.6 45 �2.6 – – –Budesonide 1.5 50 2.3 – – –Desmethyldiazepam 1.5 29 2.5 N – –Nevirapine 1.5 30 1.4 – – –Diazepam 1.4 31 2.9 N – –Zanamivir* 1.4 30 �5.4 – – –Flurbiprofen* 1.4 27 0.8 – – –Neomycin sulfate 1.3 32 �7.6 – – –Nitrofurantoin*d,f 1.4 19 �0.6 – – –Valacyclovir 1.4 32 �3.1 – – –Carbamazepine 1.3 28 2.4 N – –Chenodeoxycholic acid 1.3 51 2.7 – – –Hydrochlorothiazide 1.3 26 �0.3 – – –Amantadine* 1.2 21 �0.4 N – –Amoxicillin 1.2 37 �4.6 – – –Phenytoin 1.2 28 2.1 – – –Antipyrine 1.2 22 1.3 – – –Bendroflumethiazide 1.1 38 1.7 – – –Ganciclovir 1.1 23 �4.2 – – –Metoclopramide* 1.1 33 0.9 – – –Pindolol 1.1 30 0.0 – – –Warfarin 1.1 34 1.1 – – –Amiloride 1.1 19 �3.3 – – –Bupivacaine 1.0 38 3.4 – – –Carisoprodol 1.1 29 1.7 – – –Nizatidine 1.1 36 0.7 – – –Orphenadrine* 1.1 35 2.6 – – –Procyclidine 1.1 39 2.6 – – –Acyclovir 1.0 20 �1.3 – – –Atropine 1.0 35 0.8 – – –Captopril* 1.0 23 �2.2 N – –Furosemide* 1.0 30 �0.8 – – –Hydralazine 1.0 17 0.6 – – –Levothyroxine 1.0 47 0.6 – – –Salicylic acid 1.0 13 �1.5 – – –Sotalol 1.0 31 �1.5 – – –Valganciclovir 1.0 35 �3.5 – – –Levodopa 0.9 19 �4.6 – – –Methimazole* 0.9 12 0.0 – – –Sulindac 1.0 39 0.3 – – –Metoprolol 0.8 33 0.2 N – –Zidovudine 0.9 24 �0.2 – – –Gliclazide 0.8 36 �0.2 – – –Mesalazine 0.6 14 �2.0 – – –Bupropion 0.5 29 2.1 – – –Sulfasalazine*d,f 0.4 39 0.0 – – –

* Test set compounds.a Expressed as the ratio of the mitoxantrone accumulation after coincubation with inhibitor to that observed in cells incubated with mitoxantrone only. A more than 3-fold

increase was used as a cut-off for significant BCRP inhibition. The accumulation ratios were normalized to that obtained with the potent inhibitor Ko143 (100% inhibition)to account for interday variability. Ko143 and Fumitremorgin C were tested at 0.5 �M and GF120918 at 10 �M. All other compounds were tested at 50 �M.

b Y, compounds with an affinity for the transporter, according to the literature; N, compounds for which a lack of affinity for the transporter has been reported in theliterature; –, compounds for which no data are available in the literature.

c This compound has previously been reported to be an inhibitor of BCRP.d This compound has previously been reported to be a BCRP substrate.e This compound has not previously been reported to be a BCRP inhibitor.f During the course of this study, it was reported that cimetidine, nitrofurantoin (Jonker et al., 2005), dipyridamole (Zhang et al., 2005), and sulfasalazine (Zaher et al.,

2006) were BCRP substrates in independent publications.

Inhibition Pattern for the Human BCRP/ABCG2 Transporter 25

at ASPE

T Journals on D

ecember 4, 2018

jpet.aspetjournals.orgD

ownloaded from

Page 8: A Global Drug Inhibition Pattern for the Human ATP-Binding ...jpet.aspetjournals.org/content/jpet/323/1/19.full.pdf · scribe RTase (50 U/ l), and 21 l of nuclease-free water. The

correlation between BCRP inhibition and compound lipophi-licity, expressed as the calculated octanol-water partitioncoefficient (logD7.4) (Fig. 4A). We reasoned that it would be

possible to define a lipophilicity cut-off, below which a com-pound would not reach the substrate binding site in adequateamounts to elicit an inhibitory effect. Indeed, no compoundswith logD7.4 below 0.5 appeared as hits in the cell assay. Themedian lipophilicity was 4.0 for the BCRP inhibitors and 0.7for the noninhibitors, which indicates that the inhibitorswould accumulate in the plasma membrane and is consistentwith a need for membrane partitioning to reach the bindingsite (Fig. 4B).

Computational Modeling of BCRP Inhibition. To de-termine which descriptors of molecular structure are impor-tant for the interaction between BCRP and its inhibitors, wedeveloped a computational model for the discrimination be-tween inhibitors and noninhibitors. In our search for themolecular requirements for BCRP inhibition, we primarilyincluded 152 descriptors of molecular structure. It is surpris-ing that the statistical analysis showed that the final modelcould be based on only two descriptors: logD7.4 and the mo-lecular polarizability. Both descriptors are highly correlatedto the passive membrane permeability (Stenberg et al.,2001). Interestingly, this simple two descriptor model classi-fied 93% of the BCRP active compounds and 92% of theinactive compounds in the training set correctly (Fig. 5),indicating that membrane partitioning is an important factorfor drug interaction with BCRP. The OPLS-DA model wasfurther evaluated using a structurally diverse test set, re-sulting in correct classifications for 83% of the BCRP activecompounds and for 76% of the inactive ones, which confirmedthe predictivity of the model. Inclusion of additional molecu-lar descriptors only marginally increased the statistical qual-ity of the model, whereas the interpretation of the model wasunaffected because all descriptors were either related to thepolarizability (molar refractivity) or to logD7.4 (the number ofnonpolar atoms, the number of carbon atoms, and the surfacearea of nonpolar atoms).

Modeling the Drug-Transporter Interaction. We rea-soned that the strong influence of lipophilicity probably re-flects a need for membrane partitioning to occur for the drugto reach the BCRP binding site, which would be in agreementwith a two-step interaction model similar to that proposed forP-gp (Gottesman and Pastan, 1993; Seelig and Landwojtow-icz, 2000; Omote and Al-Shawi, 2006). With the aim of mod-eling the second step, i.e., the binding of drug to the trans-porter, we selected a lipophilicity-independent subset of thecompounds examined in this study. This was done through apairwise selection of 11 BCRP inhibitors and 11 noninhibi-

Fig. 3. Inhibition of BCRP-mediated mitoxantrone efflux by the 123compounds in this study. Mitoxantrone accumulation in Saos-2 cellsstably transfected with human wild-type (Arg482) BCRP was measuredafter incubation with 1 �M mitoxantrone with or without 50 �M inhibi-tor. Of the 123 compounds, 46 (37%) inhibited BCRP-mediated mitox-antrone efflux at this concentration. The inhibition was normalized tothat obtained using 0.5 �M of the potent BCRP inhibitor Ko143 (completeinhibition), and a greater than 3-fold increase in the intracellular mitox-antrone accumulation was used as the cut-off for significant BCRP inhi-bition (shown as a dashed line). The compounds are presented in thesame order as in Table 1. Inhibition was confirmed for all previouslyreported BCRP inhibitors (A). For the hydrophilic BCRP substrate meth-otrexate (Km � 5700 �M), a concentration greater than 50 �M wasrequired for inhibition, which reduced the hit frequency in this group to94%. Almost 3 times as many BCRP inhibitors were found in the group ofcompounds selected because of their affinity for other ABC transporters(B; 45%) compared with the group of compounds that lacked reportedABC transporter affinity (C; 16%). The data are presented as means �S.E. (n � 3–15).

TABLE 2The inhibitory effect of hydrophilic BCRP substratesHigher extracellular concentrations than those used in Table 1 were tested todetermine whether increased mass flux would result in BCRP inhibition for hydro-philic BCRP substrates that lack an inhibitory effect at the standard concentrationof 50 �M.

MitoxantroneAccumulationa

InhibitorConcentration logD7.4

-Fold increase �M

Nitrofurantoin 4.1 500 �0.6Cimetidine 4.0 5000 0.2Methotrexate 2.2b 500 c �4.8a Expressed as the ratio of the mitoxantrone accumulation at the stated inhibitor

concentration to that observed in cells incubated with 1 �M mitoxantrone only. Amore than 3-fold increase was used as the cut-off for significant BCRP inhibition.

b No significant effect on mitoxantrone accumulation at the maximal concentra-tion tested.

c Cytotoxicity was observed at concentrations of �500 �M.

26 Matsson et al.

at ASPE

T Journals on D

ecember 4, 2018

jpet.aspetjournals.orgD

ownloaded from

Page 9: A Global Drug Inhibition Pattern for the Human ATP-Binding ...jpet.aspetjournals.org/content/jpet/323/1/19.full.pdf · scribe RTase (50 U/ l), and 21 l of nuclease-free water. The

tors, so that each of the selected inhibitors had a noninhib-iting sibling with a corresponding lipophilicity. In this way,other factors that are important for discriminating inhibitorsfrom noninhibitors were not obscured by the significant in-

fluence of the membrane-partitioning step. Variable selectionresulted in a final model that correctly classified 91% of theactive and 82% of the inactive compounds in the lipophilicityindependent data set. The four descriptors in the final modelare related to �-electron energies and the abundance of ni-trogen atoms (Table 3), suggesting that hydrogen bonds andinteractions involving �-electron systems, such as �-� and�-cation interactions, are involved in inhibitor binding toBCRP.

Using the same procedure, a polarizability-independentdata set was constructed by selecting 10 pairs of active andinactive compounds with corresponding molecular polariz-abilities. The most influential molecular descriptors in thepolarizability-independent model were logD7.4 and the sur-face area of nonpolar atoms, both of which are mainly relatedto compound lipophilicity, and the surface area of nitrogenatoms, which is related to the hydrogen bonding capacity andthe possibility of acquiring a positive charge. The final modelclassified 100% of both the active and the inactive compoundsin the polarizability-independent data set correctly. Polariz-ability is correlated to the surface activity and the size of themolecule; therefore, a correlation to membrane partitioningis possible. However, other descriptors highly related tomembrane partitioning were required to discriminate be-tween inhibitors and noninhibitors in the polarizability-in-dependent data set. Thus, we deduce that polarizability isnot primarily related to the membrane partitioning step butrather that this descriptor explains charge delocalizationsthat are important for �-� interactions and hydrogen bondsbetween the drugs and the transporter.

Pharmacophore Modeling. For the purpose of investi-gating the preferential three-dimensional orientation of thestructurally heterogeneous BCRP inhibitors, we determinedthe molecular interaction points common to the 28 BCRPinhibitors in the training set. The modeling procedure re-sulted in a three-point pharmacophore consisting of two hy-drophobic features and one hydrogen bond acceptor feature(Fig. 6). The requirement for at least one hydrogen bondacceptor function is consistent with the OPLS-DA models ofthe drug-transporter interaction and is also reminiscent ofpharmacophore models previously presented for P-gp (e.g.,Pajeva and Wiese, 2002; Chang et al., 2006).

DiscussionIn this article, we explore the entire structural space of

registered drugs to determine a global model for inhibition ofthe drug efflux transporter BCRP. Consistent with otherstudies of more limited data sets, a large overlap was seen inthe affinity of the compounds in this investigation for themajor ABC efflux transporters P-gp, BCRP, MRP1, andMRP2 (Bates et al., 2001). Almost three times as many BCRPinhibitors were found in the group of compounds selectedbecause of their affinity for other ABC transporters than inthe group of compounds for which no ABC transporter affin-ity had been reported. The search for BCRP inhibitorsthroughout the drug-like chemical space resulted in the iden-tification of 29 new inhibitors. Our results corroborate pre-vious indications that BCRP may accept a set of inhibitors asdiverse as that found for P-gp (Gupta et al., 2004; Jonker etal., 2005; Zhang et al., 2005; Saito et al., 2006).

The result that a logD7.4 of at least 0.5 is needed for BCRP

Fig. 4. A, relationship between mitoxantrone efflux inhibition andlogD7.4. The closed circles denote compounds that inhibited BCRP in thisstudy, the open circles denote noninhibitors, and the closed squaresdenote BCRP substrates that did not inhibit BCRP in this study: 1)methotrexate, 2) nitrofurantoin, 3) cimetidine, and 4) sulfasalazine. Theeffect of increasing the concentration of these hydrophilic compoundsabove 50 �M is discussed in the text and in Table 2. The solid line showsthe cut-off used to discriminate between BCRP-inhibitors and noninhibi-tors. B, frequency distribution of logD7.4 in BCRP inhibitors (shaded bars,solid line) and in noninhibitors (open bars, dashed line).

Inhibition Pattern for the Human BCRP/ABCG2 Transporter 27

at ASPE

T Journals on D

ecember 4, 2018

jpet.aspetjournals.orgD

ownloaded from

Page 10: A Global Drug Inhibition Pattern for the Human ATP-Binding ...jpet.aspetjournals.org/content/jpet/323/1/19.full.pdf · scribe RTase (50 U/ l), and 21 l of nuclease-free water. The

inhibition in the whole-cell assay used in this work indicatesthat membrane partitioning significantly influences inhibi-tor binding to BCRP. The limiting effect of the plasma mem-brane was supported by examining the hydrophilic BCRPsubstrates nitrofurantoin and cimetidine. Under the stan-dard assay conditions, these two compounds were not able to

inhibit BCRP, which is explained by their low membranepermeability and the absence of significant expression ofuptake transporters in the Saos-2 cells. However, on increas-ing the concentration, and hence increasing the mass flux,both compounds inhibited BCRP. These results show that ahigh lipophilicity is not necessary for the binding of drugs toBCRP but is merely a prerequisite for reaching the bindingsite in sufficient amounts to elicit an inhibitory effect.

To investigate the molecular descriptors that determinethe interaction between BCRP and its inhibitors, we usedOPLS-DA multivariate analysis to develop a computationalmodel for discrimination between inhibitors and noninhibi-tors. The two most influential molecular descriptors werelogD7.4 and the molecular polarizability, further demonstrat-ing the significant influence of membrane partitioning ondrug binding to BCRP. Interestingly, despite the fact that thefinal discriminant analysis model only contains two physico-chemical parameters and was developed solely on the basis ofqualitative data, it performs as well as, or better than, 3Dpharmacophore models for the P-gp transporter developedfrom quantitative binding affinity data. For example, if theIC50-based P-gp pharmacophores developed by Ekins et al.(2002) are used as classification models, with IC50 � 50 �Mas the cut-off for significant inhibition, on average, 60% of theclassifications obtained for the test sets in these studies arecorrect. Furthermore, many more false hits than false missesare found using the previously published models. Penzotti etal. (2002) used a selection of four-point pharmacophores toclassify compounds as either P-gp substrates or nonsub-strates, resulting in 53% correct classifications for the activecompounds and 79% for the inactive compounds in the testset. In comparison, the two-descriptor model developed andpresented here correctly classified 83% of the BCRP inhibi-tors and 76% of the noninhibitors in the test set. In additionto providing insight into the molecular mechanism of BCRPinhibition, the model will be useful in the development of newdrugs, where it can be used to focus more extensive experi-mental efforts on compounds with a high likelihood of exhib-iting BCRP interactions.

Because the importance of lipophilicity probably reflects aneed for membrane partitioning to occur for a drug to reach

Fig. 5. Prediction of BCRP inhibition from two molecular descriptors. OPLS multivariate discriminant analysis was used to develop a modeldiscriminating between BCRP inhibitors and noninhibitors. The final model was based on the two most influential molecular descriptors: logD7.4 andpolarizability. A, dashed line shows the division between BCRP inhibitors and noninhibitors, as determined from the training set compounds.Compounds in the shaded area are predicted to be BCRP inhibitors. The closed symbols denote compounds experimentally determined to inhibitBCRP, and the open symbols denote noninhibitors. Compounds in the training set are shown as squares, and test set compounds (that were withheldfrom the model development, see Materials and Methods) are shown as circles. B, percentage of correct predictions of BCRP inhibition made by theOPLS-DA model. Predictions for the experimentally determined BCRP inhibitors are presented in the lower level graphs, and predictions for thenoninhibitors are presented in the top-level graphs. White denotes true predictions, and black denotes false predictions. The models were evaluatedboth on the training set (left column) and on the test set (right column).

TABLE 3The most important molecular descriptors in the logD andpolarizability-independent data setsPolarizability- and logD-independent subsets of the data set in this study wereconstructed to determine the molecular descriptors that are important for drugbinding to BCRP from the membrane. The descriptors are listed in order of decreas-ing importance for the discrimination between BCRP inhibitors and noninhibitors inthe respective subsets.

logD-Independent Data Set Polarizability-Independent Data Set

�-System energy logD7.4�-System resonance energy Relative surface area of nonpolar atomsSurface area of N-sp2 Surface area of N-sp2

Lowest unoccupiedmolecular orbital energy

Fig. 6. Three-dimensional orientation of the BCRP inhibitors. A, a three-point pharmacophore model was developed based on the BCRP inhibitorsin the training set, using the common features algorithm (HipHop) inCatalyst (see Materials and Methods). The model consisted of two hydro-phobic centers (shown in blue in B) and one hydrogen bond acceptorfeature (shown in green in B). The arrow indicates the optimal directionof electron sharing in the hydrogen bond, and the hemisphere indicatesthe optimal placement of a hydrogen bond donor group in the BCRPbinding site. B, potent BCRP inhibitor Ko143 mapped to the pharma-cophore model with an excellent fit. The fit value describes how well thechemical features in the compound can be superimposed onto the phar-macophore interaction points, ranging from 0 (no fit) to 3 (perfect fit) fora three-point pharmacophore.

28 Matsson et al.

at ASPE

T Journals on D

ecember 4, 2018

jpet.aspetjournals.orgD

ownloaded from

Page 11: A Global Drug Inhibition Pattern for the Human ATP-Binding ...jpet.aspetjournals.org/content/jpet/323/1/19.full.pdf · scribe RTase (50 U/ l), and 21 l of nuclease-free water. The

the BCRP binding site, we reasoned that factors that areimportant for the binding of drugs to the transporter from thelipid bilayer could be revealed by studying a lipophilicity-independent subset of the compounds examined in this in-vestigation. This analysis showed that, in addition to lipophi-licity, descriptors related to �-electron energies and theabundance of nitrogen atoms are important for discriminat-ing inhibitors from noninhibitors. This suggests that hydro-gen bonds and interactions involving �-electron systems,such as �-� and �-cation interactions, are involved in inhib-itor binding to BCRP. The inhibitors in the training set couldall be aligned to a pharmacophore consisting of hydrophobicfeatures and a hydrogen bond acceptor feature, which is inaccordance with the results from the OPLS-DA models (Fig.6). It is noteworthy that the amino acids in the intracellularloops and the transmembrane domains of the BCRP proteincontain a large number of aromatic and hydrogen bond donorside chains (Doyle et al., 1998), complementary to the molec-ular features in the models developed in this study. The samesituation has been observed in the substrate binding regionsof P-gp (Seelig, 1998; Omote and Al-Shawi, 2006; Shilling etal., 2006), and this similarity could well contribute to theoverlapping substrate specificity of the two transporters.

So far, the low resolution of the crystal structures of hu-man ABC transporters precludes direct examination of theirdrug binding sites (Rosenberg et al., 2005; McDevitt et al.,2006). Biochemical data suggests as many as four distinctdrug binding sites for P-gp (Shapiro and Ling, 1997; Ambud-kar et al., 2006); likewise, the existence of two or threedistinct but symmetrical binding sites has recently been sug-gested for the wild-type Arg482 BCRP and the R482G mu-tant isoform, respectively (Ejendal and Hrycyna, 2005; Clarket al., 2006). The good prediction of BCRP inhibition obtainedwith the model presented here is not in agreement with suchcomplexity. As an alternative to models based on distinctbinding sites, large binding regions have been suggested forP-gp, in which several compounds can bind simultaneously topartially overlapping subregions (Sauna et al., 2004). Thisconcept is consistent with high-resolution X-ray crystallogra-phy data for structurally related bacterial multidrug trans-porters such as AcrB and Sav1866 (Dawson and Locher,2006; Murakami et al., 2006; Higgins, 2007). A model basedon a large binding region in BCRP is in better agreement

with the surprisingly good results obtained with the modeldeveloped here, where the two physicochemical propertieslipophilicity and molecular polarizability describe generalinteractions with different parts of the binding pocket. Ourresults are in line with the previously reported correlationbetween drug affinity for P-gp and the frequency of hydrogenbond acceptor patterns, where a modular binding conceptrather than a key lock-type pharmacophore was used (Seelig,1998; Gatlik-Landwojtowicz et al., 2006).

In conclusion, we investigated a data set representing theentire structural space of registered drugs to examine theinhibition of the drug efflux transporter BCRP. This resultedin the discovery of 29 previously unknown inhibitors. Aneasily interpretable computational model capable of discrim-inating inhibitors from noninhibitors was constructed basedon structural features related both to the drug’s capacity toinsert into the cell membrane and to interact with the inhib-itory binding site of the transporter. The discriminatingpower of this two-descriptor model was 93% for the trainingset and 79% for the test set, respectively. The results weresupported by a global pharmacophore model and are inagreement with a two-step mechanism for the inhibition ofBCRP (Fig. 7).

Acknowledgments

We thank Pia Brokhøj, Nina Ginman, Aki Heikkinen, and LuciaLazorova for skillful technical assistance and Constanze Hilgendorf,Johan Karlsson, and Anna-Lena Ungell for assistance with the geneexpression analysis. We also thank Daisuke Nakai for valuable com-ments on the manuscript.

ReferencesAmbudkar SV, Kim IW, and Sauna ZE (2006) The power of the pump: mechanisms

of action of P-glycoprotein (ABCB1). Eur J Pharm Sci 27:392–400.Bates SE, Robey R, Miyake K, Rao K, Ross DD, and Litman T (2001) The role of

half-transporters in multidrug resistance. J Bioenerg Biomembr 33:503–511.Chang C, Bahadduri PM, Polli JE, Swaan PW, and Ekins S (2006) Rapid identifi-

cation of P-glycoprotein substrates and inhibitors. Drug Metab Dispos 34:1976–1984.

Clark R, Kerr ID, and Callaghan R (2006) Multiple drug binding sites on the R482Gisoform of the ABCG2 transporter. Br J Pharmacol 149:506–515.

Dawson RJ and Locher KP (2006) Structure of a bacterial multidrug ABC trans-porter. Nature 443:180–185.

Doyle LA, Yang W, Abruzzo LV, Krogmann T, Gao Y, Rishi AK, and Ross DD (1998)A multidrug resistance transporter from human MCF-7 breast cancer cells. ProcNatl Acad Sci U S A 95:15665–15670.

Ejendal KF and Hrycyna CA (2005) Differential sensitivities of the human ATP-binding cassette transporters ABCG2 and P-glycoprotein to cyclosporin A. MolPharmacol 67:902–911.

Ekins S, Kim RB, Leake BF, Dantzig AH, Schuetz EG, Lan LB, Yasuda K, ShepardRL, Winter MA, Schuetz JD, et al. (2002) Application of three-dimensional quan-titative structure-activity relationships of P-glycoprotein inhibitors and sub-strates. Mol Pharmacol 61:974–981.

Fetsch PA, Abati A, Litman T, Morisaki K, Honjo Y, Mittal K, and Bates SE (2006)Localization of the ABCG2 mitoxantrone resistance-associated protein in normaltissues. Cancer Lett 235:84–92.

Gatlik-Landwojtowicz E, Aanismaa P, and Seelig A (2006) Quantification and char-acterization of P-glycoprotein-substrate interactions. Biochemistry 45:3020–3032.

Goh LB, Spears KJ, Yao D, Ayrton A, Morgan P, Roland Wolf C, and Friedberg T(2002) Endogenous drug transporters in in vitro and in vivo models for theprediction of drug disposition in man. Biochem Pharmacol 64:1569–1578.

Gottesman MM and Pastan I (1993) Biochemistry of multidrug resistance mediatedby the multidrug transporter. Annu Rev Biochem 62:385–427.

Gupta A, Zhang Y, Unadkat JD, and Mao Q (2004) HIV protease inhibitors areinhibitors but not substrates of the human breast cancer resistance protein(BCRP/ABCG2). J Pharmacol Exp Ther 310:334–341.

Higgins CF (2007) Multiple molecular mechanisms for multidrug resistance trans-porters. Nature 446:749–757.

Homolya L, Hollo Z, Germann UA, Pastan I, Gottesman MM, and Sarkadi B (1993)Fluorescent cellular indicators are extruded by the multidrug resistance protein.J Biol Chem 268:21493–21496.

Jonker JW, Merino G, Musters S, van Herwaarden AE, Bolscher E, Wagenaar E,Mesman E, Dale TC, and Schinkel AH (2005) The breast cancer resistance proteinBCRP (ABCG2) concentrates drugs and carcinogenic xenotoxins into milk. NatMed 11:127–129.

Fig. 7. A proposed two-step model for drug binding to BCRP. The sche-matic illustration shows the efflux transporter BCRP inserted in theplasma membrane. An extracellularly applied compound needs to parti-tion to the plasma membrane (1). This step is described mainly by thelogD7.4 descriptor in the OPLS-DA model (Fig. 5). After flip-flop from theouter to the inner membrane leaflet (2) and lateral diffusion in themembrane, the compound can bind to the transporter (3). Hydrogenbonds and �–� interactions are probably involved in this step, which isdescribed mainly by the polarizability descriptor in Fig. 5.

Inhibition Pattern for the Human BCRP/ABCG2 Transporter 29

at ASPE

T Journals on D

ecember 4, 2018

jpet.aspetjournals.orgD

ownloaded from

Page 12: A Global Drug Inhibition Pattern for the Human ATP-Binding ...jpet.aspetjournals.org/content/jpet/323/1/19.full.pdf · scribe RTase (50 U/ l), and 21 l of nuclease-free water. The

Krishnamurthy P and Schuetz JD (2006) Role of ABCG2/BCRP in biology andmedicine. Annu Rev Pharmacol Toxicol 46:381–410.

Kruijtzer CM, Beijnen JH, Rosing H, ten Bokkel Huinink WW, Schot M, Jewell RC,Paul EM, and Schellens JH (2002) Increased oral bioavailability of topotecan incombination with the breast cancer resistance protein and P-glycoprotein inhibitorGF120918. J Clin Oncol 20:2943–2950.

Maliepaard M, Scheffer GL, Faneyte IF, van Gastelen MA, Pijnenborg AC, SchinkelAH, van De Vijver MJ, Scheper RJ, and Schellens JH (2001) Subcellular localiza-tion and distribution of the breast cancer resistance protein transporter in normalhuman tissues. Cancer Res 61:3458–3464.

McDevitt CA, Collins RF, Conway M, Modok S, Storm J, Kerr ID, Ford RC, andCallaghan R (2006) Purification and 3D structural analysis of oligomeric humanmultidrug transporter ABCG2. Structure 14:1623–1632.

Mitomo H, Kato R, Ito A, Kasamatsu S, Ikegami Y, Kii I, Kudo A, Kobatake E,Sumino Y, and Ishikawa T (2003) A functional study on polymorphism of theATP-binding cassette transporter ABCG2: critical role of arginine-482 in metho-trexate transport. Biochem J 373:767–774.

Murakami S, Nakashima R, Yamashita E, Matsumoto T, and Yamaguchi A (2006)Crystal structures of a multidrug transporter reveal a functionally rotating mech-anism. Nature 443:173–179.

Omote H and Al-Shawi MK (2006) Interaction of transported drugs with the lipidbilayer and P-glycoprotein through a solvation exchange mechanism. Biophys J90:4046–4059.

Oprea TI and Gottfries J (2001) Chemography: the art of navigating in chemicalspace. J Comb Chem 3:157–166.

Ozawa N, Shimizu T, Morita R, Yokono Y, Ochiai T, Munesada K, Ohashi A, Aida Y,Hama Y, Taki K, et al. (2004) Transporter database, TP-Search: a web-accessiblecomprehensive database for research in pharmacokinetics of drugs. Pharm Res21:2133–2134.

Ozvegy C, Litman T, Szakacs G, Nagy Z, Bates S, Varadi A, and Sarkadi B. (2001)Functional characterization of the human multidrug transporter, ABCG2, ex-pressed in insect cells. Biochem Biophys Res Commun 285:111–117.

Pajeva IK and Wiese M (2002) Pharmacophore model of drugs involved in P-glycoprotein multidrug resistance: explanation of structural variety (hypothesis).J Med Chem 45:5671–5686.

Penzotti JE, Lamb ML, Evensen E, and Grootenhuis PD (2002) A computationalensemble pharmacophore model for identifying substrates of P-glycoprotein.J Med Chem 45:1737–1740.

Physicians Desk Reference (2005) Physicians’ Desk Reference, Thomson Healthcare,Montvale, NJ.

Rosenberg MF, Callaghan R, Modok S, Higgins CF, and Ford RC (2005) Three-dimensional structure of P-glycoprotein: the transmembrane regions adopt anasymmetric configuration in the nucleotide-bound state. J Biol Chem 280:2857–2862.

Saito H, Hirano H, Nakagawa H, Fukami T, Oosumi K, Murakami K, Kimura H,Kouchi T, Konomi M, Tao E, et al. (2006) A new strategy of high-speed screeningand quantitative structure-activity relationship analysis to evaluate human ATP-binding cassette transporter ABCG2-drug interactions. J Pharmacol Exp Ther317:1114–1124.

Sauna ZE, Andrus MB, Turner TM, and Ambudkar SV (2004) Biochemical basis ofpolyvalency as a strategy for enhancing the efficacy of P-glycoprotein (ABCB1)modulators: stipiamide homodimers separated with defined-length spacers re-verse drug efflux with greater efficacy. Biochemistry 43:2262–2271.

Seelig A (1998) A general pattern for substrate recognition by P-glycoprotein. EurJ Biochem 251:252–261.

Seelig A and Landwojtowicz E (2000) Structure-activity relationship of P-glycoprotein substrates and modifiers. Eur J Pharm Sci 12:31–40.

Shapiro AB and Ling V (1997) Positively cooperative sites for drug transport byP-glycoprotein with distinct drug specificities. Eur J Biochem 250:130–137.

Shilling RA, Venter H, Velamakanni S, Bapna A, Woebking B, Shahi S, and vanVeen HW (2006) New light on multidrug binding by an ATP-binding-cassettetransporter. Trends Pharmacol Sci 27:195–203.

Stenberg P, Norinder U, Luthman K, and Artursson P (2001) Experimental andcomputational screening models for the prediction of intestinal drug absorption.J Med Chem 44:1927–1937.

Wierdl M, Wall A, Morton CL, Sampath J, Danks MK, Schuetz JD, and Potter PM(2003) Carboxylesterase-mediated sensitization of human tumor cells to CPT-11cannot override ABCG2-mediated drug resistance. Mol Pharmacol 64:279–288.

Zaher H, Khan AA, Palandra J, Brayman TG, Yu L, and Ware JA (2006) Breastcancer resistance protein (Bcrp/abcg2) is a major determinant of sulfasalazineabsorption and elimination in the mouse. Mol Pharm 3:55–61.

Zhang Y, Gupta A, Wang H, Zhou L, Vethanayagam RR, Unadkat JD, and Mao Q(2005) BCRP transports dipyridamole and is inhibited by calcium channel block-ers. Pharm Res 22:2023–2034.

Address correspondence to: Dr. Per Artursson, Pharmaceutical Screeningand Informatics, Department of Pharmacy, Uppsala University, P.O. Box 580,SE-751 23 Uppsala, Sweden. E-mail: [email protected]

30 Matsson et al.

at ASPE

T Journals on D

ecember 4, 2018

jpet.aspetjournals.orgD

ownloaded from