In Silico Identification of Potential Cholestasis-Inducing Agents Via Modeling of Na+-Dependent Taurocholate Cotransporting Polypeptide Substrate Specificity

toxicological sciences 129(1), 35–48 (2012) doi:10.1093/toxsci/kfs188 Advance Access publication May 28, 2012

In Silico Identification of Potential Cholestasis-Inducing Agents via Modeling of Na+-Dependent Taurocholate Cotransporting Polypeptide Substrate Specificity

Rick Greupink,*,1 Sander B. Nabuurs,†,2 Barbara Zarzycka,† Vivienne Verweij,* Mario Monshouwer,‡

Maarten T. Huisman,‡ and Frans G. M. Russel* Downloaded from https://academic.oup.com/toxsci/article/129/1/35/1629030 by guest on 29 September 2021

*Department of Pharmacology and Toxicology, Radboud University Nijmegen Medical Centre, Nijmegen Centre for Molecular Life Sciences, Nijmegen, The Netherlands; †Centre for Molecular and Biomolecular Informatics, Radboud University Nijmegen, Nijmegen Centre for Molecular Life Sciences, Nijmegen, The Netherlands; and ‡Department of Drug Metabolism and Pharmacokinetics, Janssen Pharmaceutical Companies of Johnson and Johnson, Beerse, Belgium

1To whom correspondence should be addressed at Department of Pharmacology and Toxicology 149, Radboud University Nijmegen Medical Centre, PO Box 9101, 6500 HB Nijmegen, The Netherlands. Fax: +(31) 24 36 14214. E-mail: [email protected]. 2Present address: Lead Pharma Medicine, Nijmegen, The Netherlands.

Received March 23, 2012; accepted May 16, 2012

The liver plays an important role in bile acid homeostasis. At Na+-dependent taurocholate cotransporting polypeptide (NTCP, physiological pH, bile acids do not effectively cross cell mem- SLC10A1) is the main transporter facilitating the hepatic uptake branes by passive diffusion because they exist for an important of bile acids from the circulation. Consequently, the interaction of part in the anionic form. Hepatocellular uptake of bile acids xenobiotics, including therapeutic drugs, with the bile acid binding pocket of NTCP could lead to impairment of hepatic bile acid from portal blood and subsequent excretion into bile, therefore, uptake. We pursued a 3D-pharmacophore approach to model the completely depends on a coordinated function of transport NTCP substrate and inhibitor specificity and investigated whether proteins located in the basolateral and canalicular membranes it is possible to identify compounds with intrinsic NTCP inhibi- of liver cells (Kosters and Karpen, 2008). As a consequence, tory properties. Based on known endogenous NTCP substrates, a any (xenobiotic-mediated) functional disruption of these trans- 3D-pharmacophore model was built, which was subsequently used port processes may lead to cholestasis (Pauli-Magnus and to screen two virtual libraries together containing the structures of Meier, 2006). 10 million compounds. Studies with Chinese hamster ovary cells The Na+-dependent taurocholate cotransporting polypeptide overexpressing human NTCP, human hepatocytes, ex vivo per- (NTCP, SLC10A1) is the major transporter that takes up bile fused rat livers, and bile duct–cannulated rats were conducted to acids from the portal blood, which represents the first step in validate the activity of the virtual screening hits. Modeling yielded hepatocellular bile acid handling. Hence, intact NTCP functioning a 3D-pharmacophore, consisting of two hydrogen bond acceptors and three hydrophobic features. Six out of 10 structurally diverse is considered important for the maintenance of the enterohepatic compounds selected in the first virtual screening procedure signifi- circulation of bile acids and hepatocyte homeostasis (Ho et al., cantly inhibited taurocholate uptake in the NTCP overexpressing 2004). In humans, it has been estimated that approximately cells. For the most potent inhibitor identified, an anthraquinone 80% of the hepatic uptake of conjugated bile acids occurs in a derivative, this finding was confirmed in human hepatocytes and sodium-dependent manner, whereas the remainder is taken up perfused rat livers. Subsequent structure and activity relationship through sodium-independent pathways, e.g., via organic anion studies with analogs of this derivative indicated that an appropri- transporting polypeptide 1B1 (OATP1B1, SLCO1B1) (Kosters ate distance between hydrogen bond acceptor features and pres- and Karpen, 2008). Under physiological conditions, bile ence of one or two negative charges appear critical for a successful acids are mainly confined to the enterohepatic circulation, but NTCP interaction. In conclusion, pharmacophore modeling was impaired uptake of bile acids from the portal blood will lead to an successfully used to identify compounds that inhibit NTCP. Our increase in bile acid concentrations in the systemic circulation. approach represents an important first step toward the in silico flagging of potential cholestasis-inducing molecules. During cholestasis, a 20- to 100-fold increase in systemic bile Key Words: NTCP; cholestasis; pharmacophore modeling; acid concentrations have been reported, which is associated with drug-induced toxicity. morbidity (Makino et al., 1969). For instance, hypercholanemia has been associated with extrahepatic adverse effects, a

© The Author 2012. Published by Oxford University Press on behalf of the Society of Toxicology. All rights reserved. For permissions, please email: [email protected] 36 GREUPINK ET AL. well-known example being pruritus (Kuiper et al., 2010). Also, database, Accelrys, San Diego, CA) by screening the databases against the asymptomatic hypercholanemia during pregnancy, which is pharmacophore model. Once identified in silico, the selected chemicals were characterized by specifically elevated maternal serum levels of purchased from the registered chemical suppliers indicated in these databases. Both database resources contain only commercially available molecules bile acids > 40µM, has been associated with a higher incidence and can be found through http://cococo.unimore.it and http://accelrys.com. of preterm labor, fetal asphyxia, and third trimester intrauterine Chemical structures of the compounds are provided in this article. In addition, death (Abu-Hayyeh et al., 2010; Glantz et al., 2004; Williamson full chemical names are listed in Supplementary table 1. Prior to conducting in et al., 2004). Studies suggest bile acid–induced damage of vitro experiments, the purchased chemicals were dissolved in dimethylsulfox- 3 cell membranes as a potential biochemical mechanism of the ide (DMSO), unless indicated otherwise. H-taurocholic acid (specific activity 4.6 Ci/mmol) was obtained from Perkin Elmer (Boston, MA). L-Glutamine and observed toxicity during cholestasis, whereas, additionally, bile sodium pyruvate were obtained from Invitrogen, whereas DMSO and all other acids have also been shown to cause mitochondrial dysfunction reagents were of the highest obtainable grade available and were purchased and modulate second messenger pathways involved in the from local suppliers. regulation of cellular metabolism (Bouscarel et al., 1995; Experimental animals. Male Sprague Dawley rats (Harlan, Horst, The Downloaded from https://academic.oup.com/toxsci/article/129/1/35/1629030 by guest on 29 September 2021 Schölmerich et al., 1984; Spivey et al., 1993). Netherlands) weighing 220–240 g (in vivo experiments) or 240–300 g (liver Recently, some drugs and other xenobiotics have been iden- perfusions) were housed under a 12-h dark/light cycle at constant humidity tified as NTCP substrates, based on data obtained in isolated and temperature. Animals had free access to tap water and standard lab chow. hepatocytes and NTCP recombinant overexpression systems. Prior to conducting experiments, animals were allowed to acclimatize for 1 week after arrival. All experiments were approved by the local committees for For example, various statins, the liver function diagnostics bro- care and use of experimental animals and were performed according to strict mosulphophtalein and indocyanine green, as well as the antifun- governmental and international guidelines for the use of experimental animals. gal drug micafungin are transported by NTCP (Greupink et al., Generation of a common feature pharmacophore and database screen- 2011; Ho et al., 2006; Yanni et al., 2010). Hilgendorf et al. inves- ing. Molecular modeling studies were performed with the Discovery Studio tigated the hepatic mRNA concentration of basolateral and cana- Software package version 2.5.5 (Accelrys). A common feature, ligand-based licular transporters and found that NTCP expression is similar to pharmacophore model was built based on the molecular structure of five estab- that of the well-known and clinically relevant basolateral drug lished NTCP substrates, namely the unconjugated bile acid cholic acid, the uptake transporter OATP1B1, which is among the most abun- conjugated bile acids taurocholic acid, tauroursodeoxycholic acid and gly- cocholic acid, and the non–bile acid estrone-3-sulphate. The 3D chemical dantly expressed transporters in human liver (Hilgendorf et al., structures of the molecules were obtained from the PubChem online database 2007). Studies in human hepatocytes demonstrated that NTCP (National Center for Biotechnology Information, Bethesda, MD). Due to the can indeed contribute substantially to the overall uptake of drugs conformational freedom of the studied molecules, functional groups within in these cells (Ho et al., 2006; Yanni et al., 2010). This suggests molecules will not have fixed positions in 3D space. To account for the effect that next to cholestasis, xenobiotic-induced inhibition of NTCP of rotational freedom, for each molecule up to 255 conformers were generated, probing low-energy 3D molecular conformations. To this end, the “best” could also contribute to clinically relevant transporter-mediated conformer generation method of the Discovery Studio software package was drug-drug interactions and associated drug-induced toxicities. used, allowing for a maximum energy difference of 20 kcal/mol between con- In this article, we describe a pharmacophore modeling formers. Ten pharmacophore hypotheses were generated, assigning hydrogen approach, which can assist in identifying chemotypes that bond acceptors (HBAs), hydrogen bond donors, and hydrophobic features. The may inhibit NTCP. During hit-to-lead and lead-optimization hypothesis that fitted the structures of the training set best was selected for subsequent database screening. A total of three in silico–screening and in vitro– of new candidate drugs, the proposed in silico approach helps testing cycles were performed. In the first screening round, the pharmacophore to flag unwanted chemotypes with intrinsic NTCP inhibitory model was used to screen the Accelrys CAP database, containing the 3D struc- activity during the early stages of drug discovery. To develop tures of over 2.5 million commercially available compounds. The second and these models, we employed a computer-assisted ligand-based third screening rounds were performed in the context of model refinement and 3D-pharmacophore modeling approach to describe NTCP inhib- are described in a dedicated paragraph below. itor specificity. The pharmacophore models were subsequently Cell culturing of CHO-Parent and CHO-NTCP cells. CHO-NTCP cells used to identify potential NTCP inhibitors from two virtual com- stably transfected with the gene encoding human NTCP were obtained from pound libraries, containing the chemical structures of close to Solvo Biotechnologies (Budapest, Hungary). CHO-parent cells, which were 10,000,000 commercially available compounds. Subsequently, not transfected with the transporters of interest (CHO-P), were used as control cells to estimate non-NTCP cellular uptake. The cell lines were cultured the candidate molecules that were identified in silico were evalu- in a humidified atmosphere at 37°C in the presence of 5% CO2. The culture ated experimentally for their activity against NTCP. To this end, medium consisted of Dulbecco’s modified Eagle’s medium (Invitrogen, Breda, we employed in vitro studies with Chinese hamster ovary cells The Netherlands) supplemented with 10% fetal calf serum, 4mM l-glutamine, expressing human NTCP (CHO-NTCP) and human hepatocytes, and 1mM sodium pyruvate without antibiotics. ex vivo rat liver perfusions, and in vivo experiments in rats. Inhibition studies with CHO cells and human hepatocytes. As a first step to investigate whether molecules identified in silico could indeed interact with NTCP, it was studied whether the compounds could inhibit MATERIALS AND METHODS NTCP-mediated taurocholic acid transport. For inhibition studies, CHO-NTCP cells were incubated with 1µM 3H-taurocholic acid. Stock solutions of the Chemicals. New chemical structures were identified from the CoCoCo various inhibitors were made in DMSO, and cells were coincubated with database (maintained by the BioChemoInformatics Laboratory, University test compounds, the final DMSO concentration being 1% in all wells. The of Bologna, Italy) and the CAP database (Chemicals Available for Purchase incubation period with 3H-taurocholic acid was 2 min because previous studies IN SILICO IDENTIFICATION OF NTCP INHIBITORS 37 have shown that uptake was linear over this time period (Greupink et al., anesthesia for the duration of the experiment. A tracer dose of 3H-labeled tauro- 2011). Uptake of 3H-taurocholic acid in CHO-P cells served as a control to cholic acid was administered via the penis vein (75 µCi/kg), with or without test correct for uptake that was not related to NTCP-mediated transport. Data were compound (5 mg/kg). During the experiment, bile flow was spontaneous and the expressed as the percentage uptake relative to CHO-NTCP cells incubated with outflow of the cannula was collected in 5-min intervals. At the end of the experi- vehicle and were plotted against the nominal inhibitor concentration. In the ment, animals were euthanized via cardiac puncture. During experiments, the case of dose-response curves, a one-site binding model with variable slope was body temperature of the animals was maintained at 37°C by placing the animals fitted to the data in order to estimate IC50 values. In case of uptake studies on thermostatic pads and covering them with isolating tissues. Bile samples aiming to assess the apparent Km and Vmax of taurocholic acid uptake in the were immediately stored at −20°C. For analysis, 10 µl of homogenized bile was presence or absence of inhibitors, the data were analyzed using the standard mixed with 4 ml of scintillation fluid and analyzed for3 H-taurocholic acid con- Michaelis-Menten model, as described earlier (Greupink et al., 2011). tent. Injection solutions were also analyzed, and the excretion of 3H-taurocholic Studies in human hepatocytes were conducted with cryopreserved hepato- acid was expressed as a percentage of the injected dose. cytes purchased from Life Technologies (Breda, The Netherlands). Hepatocytes were suspended in Krebs-Henseleit buffer to a final concentration of 1 mil- Statistical analysis. In all graphs, data are expressed as mean ± SEM. Subsequent statistical analysis and curve fitting were performed with the lion cells per ml. Cells were incubated at 37°C with 1µM of taurocholic acid, in the absence or presence of sodium for various time points, together with GraphPad Prism 5 software package (version 5.02, GraphPad Software Inc.). Downloaded from https://academic.oup.com/toxsci/article/129/1/35/1629030 by guest on 29 September 2021 post hoc increasing concentration of the test compound. Incubations were stopped using One-way ANOVA in conjunction with Dunnett’s test was performed to the oil centrifugation method as described earlier (Yabe et al., 2011). Next assess statistical significance of differences when multiple groups were compared. to the incubations at 37°C, parallel incubations were performed at 4°C. The When two groups were compared, data were subjected to the two-sided Student’s t p transporter-mediated uptake of radiolabeled taurocholic acid into hepatocytes -test. Differences were considered statistically significant when < 0.05. was calculated by subtracting uptake at 4°C from the values observed at 37°C.

Refinement of pharmacophore models and subsequent database screen- ings. The initial pharmacophore that was derived from known natural sub- RESULTS strates/competitive inhibitors was refined using compounds 1–16. Inhibitory activities for compounds 1–16 were defined, classifying compounds 2, 15, and Pharmacophore Modeling 16 as very active, compounds 1, 3, 5, 6, and 13 as moderately active, and com- To identify the features that are relevant for a successful pounds 4, 7–12, and 14 as poorly active/inactive. The original pharmacophore interaction with the bile acid binding site of NTCP, we built was refined by assigning excluded volumes to the model. The refined and the nonrefined pharmacophore were used to screen the CoCoCo database, con- a ligand-based common feature pharmacophore model based taining close to 7 million commercially available compounds. For each phar- on five known NTCP substrates (Fig. 1A). Note that these macophore, the 5000 best scoring compounds were retrieved from the screen compounds are also inhibitors of each other’s NTCP-mediated and were subjected to a first filtering (selecting compounds with a MW < 700 uptake by virtue of binding to the same binding pocket. The and a cLogD < 3.5), using Pipeline Pilot Professional version 7.5 (Accelrys). resulting pharmacophore model (Fig. 1B) consisted of five fea- Subsequently, the following three subsets were combined: the top 250 hits of the original pharmacophore, the top 250 hits of the refined pharmacophore, tures. Two of which are HBAs, the smaller sphere (green) indi- and all hits retrieved by both pharmacophores, regardless of their goodness of cating the point of origin of the H-bond and the larger sphere fit. All hits were mapped against the refined pharmacophore using flexible fit- indicating the vector of the H-bond. The two HBA features ting. After Lipinski filtering, the top 150 compounds were selected for further are separated by three hydrophobic features (cyan spheres). analysis and testing. In addition to this first refinement, a second refinement In Figure 1C, the mapping of the prototypical NTCP substrate of the initial pharmacophore using the exact same approach was conducted as described above but now using only the anionic compounds for model optimi- taurocholic acid to the pharmacophore can be seen. The sul- zation. Database screening was also performed in a similar way as described fonic acid group that terminates the aliphatic side chain on the for the first refined pharmacophore. Additionally, a criterion was added that D ring and the hydroxyl group in the 7-position on the steroid retrieved hit compounds with a single or double negative charge at pH = 7.4. skeleton map to the HBAs, whereas the carbon atoms of the Cholestatic effects of compound 2 in perfused rat liver. In brief, male steroid backbone correspond to the hydrophobic features. Sprague Dawley rats were anaesthetized with isoflurane and the portal vein, bile duct, and liver vein were cannulated. Liver perfusions were performed in Initial Database Screening a single-pass setting. The perfusate consisted of warm Krebs-Henseleit buffer, In the first virtual screening round, of the 2.5 million saturated with a 95%/5% mixture of O /CO and contained 10µM taurocholic 2 2 molecules present in the CAP database, approximately 10,000 acid, to maintain bile flow. In addition,3 H-taurocholic acid (1200 dpm/ml) was added to monitor the effect of the test compound on the clearance of taurocho- molecules fitted the pharmacophore model well (mapping to all late. Throughout the experiment, samples of the bile and venous outflow were five features of the pharmacophore). This corresponds to around collected in 5-min intervals. Livers were equilibrated for 20 min, and then the 0.4% of the total number of compounds present in the database. test compound was added to the perfusate. Radioactivity in the perfusate, bile, To assure that the molecules put forward for experimental and venous outflow were measured via scintillation counting. The amount of validation would be soluble enough in physiological media, radioactivity in the bile and venous outflows was expressed as percentage of the amount entering the cannulated portal vein (perfusate). To assess the effect of subsequent selection steps were performed to select only protein binding, incubations were performed in the presence or absence of 1% molecules with a clogD < 3.5 and MW < 700 Da. Next, the bovine serum albumin (BSA) in the perfusate. remaining molecules were clustered by chemical similarity into 70 clusters of 10 hits each. From these different clusters, Cholestatic effects of compound 2 in rats in vivo. Compound 2 was also studied for its in vivo cholestatic potential in rats. Animals were anaesthetized an initial set of 10 chemically diverse molecules was manually selected to evaluate whether the molecules could interact with with O2/N2O/isoflurane, and the common bile duct was cannulated to allow monitoring of the excretion of compounds into the bile. The rats were kept under NTCP. The 2D chemical structures of the selected molecules 38 GREUPINK ET AL. Downloaded from https://academic.oup.com/toxsci/article/129/1/35/1629030 by guest on 29 September 2021

Fig. 1. (A) Chemical structures of estrone sulfate and four bile acid substrates of NTCP that were used to build the pharmacophore model. (B) 3D ligand-based common feature pharmacophore model derived from the structures depicted in panel A. Green spheres indicate HBA feature; cyan spheres indicate hydrophobic features. (C) Mapping of taurocholic acid to the common feature pharmacophore. Color version available online. are depicted in Figure 2A, full chemical names and goodness phthalamic acid) (compound 3), reduced taurocholic acid of fit to the pharmacophore model are listed in Supplementary uptake by 69 ± 3, 96.3 ± 0.6, and 54 ± 3%, respectively table 1. (p < 0.05). Interestingly, 2′,2′′-(hexamethylenedioxy)-dia cetanilide (compound 4) and 1-[1-(2,4-dimethoxybenzyl)- Inhibition Studies With CHO-P and CHO-NTCP Cells 4-piperidinyl]-4-(2-methoxyphenyl)piperazine (compound As a first step to study whether the selected molecules 9) stimulated taurocholate uptake to 144 ± 11 and 137 ± 19% of could interact with NTCP, we investigated their inhibitory control cells, respectively. effect on NTCP-mediated transport of 3H-labeled taurocholic For compounds 1–3, we investigated the mode of inhibition. acid. In Figure 2B, results are displayed that were obtained To this end, the concentration dependency of 3H-taurocholic in the CHO-NTCP cell model, and it can be seen that 6 out acid uptake by NTCP was studied in the absence or presence of of 10 molecules significantly inhibited taurocholic acid compound 1, 2, or 3. Data revealed that in the presence of all uptake at a concentration of 100µM. Compared with con- three compounds the uptake curve of taurocholic acid shifted to trol cells, the three most potent inhibitors, 2-(4-{[4-(4-methy the right (Figs. 3A–C). Coincubation with 100µM of compound l-3-sulfamoylphenyl)phthalazinyl]amino}phenyl)acetamide 1 increased the apparent Km for taurocholic acid from 13 ± 4µM (compound 1), 5-methyl-2-[[5-(4-methyl-2-sulfonatoanilino)- (absence of test compound) to 37 ± 7µM. Coincubation with

9,10-dioxoanthracen-1-yl]amino]benzenesulfonate 3 or 10µM of compound 2 increased the apparent Km for the (compound 2), and N,N′-dodeca-methylene-bis(hexahydro- taurocholate-NTCP interaction from 14 ± 4 to 35 ± 2 and 88 ± 19. IN SILICO IDENTIFICATION OF NTCP INHIBITORS 39 Downloaded from https://academic.oup.com/toxsci/article/129/1/35/1629030 by guest on 29 September 2021

Fig. 2. (A) Chemical structures of the 10 molecules selected for first round of in vitro testing. (B) Effect of test compounds 1–10 (100µM) on NTCP-mediated 3H-taurocholic acid uptake (1µM for 2 min) by CHO-NTCP cells. Open bars indicate absence of inhibition or stimulation of NTCP-mediated taurocholic acid transport; closed bars indicate a significant inhibition of taurocholic acid transport. Data are expressed as a percentage of control and represent the mean ± SEM of three separate experiments each performed in duplicate or triplicate. * indicates p < 0.05 compared with control. TC, taurocholic acid. 40 GREUPINK ET AL.

Finally, in the presence of 100µM of compound 3, the apparent

Km of NTCP for taurocholic acid increased to 42 ± 6µM. In all cases, the observed maximum taurocholic acid transport rates were not decreased, which indicates that the molecules inhibit NTCP in a competitive manner.

Structure Activity Relationship of Anthraquinone Derivatives To validate the pharmacophore, we next investigated whether analogues of compound 2 that failed to hit all features of the pharmacophore were also less-potent inhibitors of NTCP-mediated taurocholic acid uptake. Therefore, we studied Downloaded from https://academic.oup.com/toxsci/article/129/1/35/1629030 by guest on 29 September 2021 whether successful mapping to both HBA features is required for molecules to be active. In Figure 4A, the chemical structures of the selected molecules are displayed. Figure 4B depicts a graphical representation of how the compounds map to the pharmacophore model. For compound 2, the IC50 for inhibition of NTCP-mediated taurocholic acid uptake was 3.2 ± 1.0µM (Fig. 5). The two analogues 5-methyl-2-[[4-(4-methyl-2-sulfoanilino)-9,10-dioxoanthracen-1-yl]amino]benzenesulfonic acid (compound 15) and 2,2′-[(9,10-dihydro-5,8-dihydroxy-9,10-dioxo-1,4-an- thracenediyl)diimino]bis[5-methyl-]benzenesulfonic acid (compound 16) that had slightly different structures than compound 2 but did match with all features of the pharmacophore displayed a similar potency as compound 2. The IC50 values of compound 15 and compound 16 were determined to be 4.3 ± 1.7µM and 5.5 ± 1.7µM, respectively. In contrast, 1,4-bis(4-methylanilino)anthracene-9,10-dione (compound 11), which is very similar to compound 15, but is lacking the sulfonic acid substituents to cover the two HBA features of the pharmacophore model, was not active up to the highest concentration tested (100µM). 9,10-dioxoanthracene-1-sulfonic acid (compound 12), containing only one sulfonic acid group conjugated directly to the anthraquinone skeleton, was inactive, and also 1,5-Anthraquinonedisulfonic acid (compound 14), which does possess two sulfonic acid groups but not at an appropriate distance to cover both HBA features, is inactive up to 100µM. Also 5-methyl-2-[[4-(methylamino)-9,10-dioxoanthracen- 1-yl]amino]benzenesulfonic acid (compound 13), which contains only one sulfonic acid group conjugated to a single benzyl substituent of the anthraquinone skeleton, was found to be less potent than the compounds that hit all features of the pharmacophore. The IC50 of compound 13 was 12.2 ± 1.4µM, which is significantly higher (p < 0.05) than the IC50 values of compound 2, compound 15, and compound 16.

Model Refinement Based on the new activity data obtained with compounds Fig. 3. Concentration dependency of NTCP-mediated 3H-taurocholic acid 1–16, the initial pharmacophore was refined by adding 22 uptake by CHO-NTCP cells in the absence or presence of 100µM of compound excluded volumes (Fig. 6B). Both the old and the new phar- 1 (A), 3 or 10µM of compound 2 (B), or 100µM of compound 3 (C). The data represent the mean ± range of observations of a representative experiment per- macophore were used to screen the CoCoCo database, and the formed in duplicate. top 5000 compounds for each pharmacophore were retrieved. IN SILICO IDENTIFICATION OF NTCP INHIBITORS 41 Downloaded from https://academic.oup.com/toxsci/article/129/1/35/1629030 by guest on 29 September 2021

Fig. 4. (A) Chemical structures of analogues of the anthraquinone derivative compound 2. (B) Graphical representation of mapping the selected compounds to the pharmacophore model. Similar to compound 2, compounds 15 and 16 hit all features of the model. Compounds 11–14 failed to hit both HBA features.

Additional filtering for compounds with a molecular weight with Lipinski’s rule of five were filtered out from the top < 700 g/mol and a cLogD < 3.5 reduced the number of 150 compounds by manual selection. In this final step, com- hits by approximately 50%. Combination of the top 250 pound selection was mainly driven by vendor availability molecules obtained by screening with the original and the (Fig. 6A). Surprisingly, only a very small number of com- refined model, with the hits obtained with both pharmaco- pounds displayed inhibition of NTCP as tested using CHO- phores, yielded 616 hits. Compounds that did not comply NTCP cells. Moreover, the compounds that were active also 42 GREUPINK ET AL.

obtained from the in silico screening round with this pharmacophore are depicted in Figure 7A. Compound 31 did not dissolve in the assay buffer and could not be tested. However, 4 out of the 7 remaining molecules exhibited significant inhibitory activity (p < 0.05, compared with control) against taurocholic acid uptake in CHO-NTCP cells, displaying more than 50% up to complete inhibition (Fig. 7C). Effects of Compound 2 in Cryopreserved Human Hepatocytes For the most potent NTCP inhibitor, compound 2, we investigated whether the observations made in the recombinant Downloaded from https://academic.oup.com/toxsci/article/129/1/35/1629030 by guest on 29 September 2021 system expressing the transporter also translated to cryopreserved human hepatocytes. In line with the high NTCP affinity observed in the CHO recombinant expression model, we found that compound 2 also potently inhibited sodium-dependent taurocholic acid in human hepatocytes (Fig. 8). Incubation with 2µM of compound 2 reduced taurocholic acid uptake from 77 ± 5 to 31 ± 5 pmol taurocholate per million hepatocytes (p < 0.05), whereas 20 and 200µM of compound 2 reduced taurocholic acid uptake to 18.0 ± 0.4 and 13.0 ± 0.3 pmol taurocholate per million hepatocytes (p < 0.05), respectively. These are similar to the levels observed at 4°C or when incubated in the absence of sodium.

Cholestatic Effects of Compound 2 in Perfused Rat Liver and in Rats In Vivo To further extrapolate the in vitro findings to a more physiological setting, we conducted an ex vivo rat liver perfusion study and an in vivo rat study to investigate whether compound 2 could reduce the biliary clearance of radiolabeled taurocholic acid. In Figure 9, the results of the liver perfusion experiments are shown. The effect of compound 2 on 3H-labeled taurocholic acid clearance was measured in a single-pass perfusion setting. After a stabilization period of 20 min, compound 2 was added to the perfusion medium at end concentrations of 10 and Fig. 5. Inhibitory effect of compound 2 and structural analogues 11–16 100µM. In Figure 9A, it can be seen that addition of compound on NTCP-mediated taurocholic acid uptake by CHO-NTCP cells. Active mole- 2 to the perfusion medium did not influence venous outflow, cules are depicted in panel A, inactive molecules in panel B. The data represent the mean ± SEM of three separate experiments each performed in duplicate or but that the amount of taurocholate in the venous outflow relative to inflow increased dramatically in the presence of 100µM triplicate. In order to estimate IC50 values, a one-site binding model with variable slope was fitted to the data. Derived IC50 values are mentioned in the text. of compound 2 compared with control (Fig. 9B). Moreover, 100µM led to cholestasis as reflected by a reduced bile flow (Fig. 9C) and a corresponding reduction in 3H-taurocholic acid only showed a limited efficacy at the tested concentration of excretion (Fig. 9D). In the liver perfusion experiments, we 100µM (Fig. 6C). additionally found that addition of 1% BSA to the perfusion Upon analysis of the chemical structures obtained from medium totally abolished the cholestatic effect of 100µM of screening with the refined pharmacophore, we found uncharged compound 2. molecules to be over-represented compared with negatively Intravenous administration of 5 mg/kg of compound 2 to bile charged compounds. To explore whether the presence of 1 or 2 duct–cannulated rats did not result in a statistically significant negative charges may result in a better interaction with NTCP, effect on bile flow (Fig. 10A) nor did it affect the biliary tau- we conducted a second refinement of the original pharmacoph- rocholic acid excretion profile compared with control animals ore, now only taking into account the active molecules with a (Fig. 10B). Pilot studies showed that the plasma concentration net negative charge at pH = 7.4. This yielded a third pharma- of compound 2 approximated 100µM (5 min after an iv bolus cophore model, which is depicted in Figure 7B. The molecules injection, we measured a total plasma concentration of 115µM). IN SILICO IDENTIFICATION OF NTCP INHIBITORS 43 Downloaded from https://academic.oup.com/toxsci/article/129/1/35/1629030 by guest on 29 September 2021

Fig. 6. (A) Compounds that were retrieved from the CoCoCo database using the refined pharmacophore and were tested for NTCP inhibitory activity in CHO-NTCP cells. (B) Graphical representation of the refined pharmacophore used to retrieve the molecules depicted in (A). In the refined model, 22 excluded volumes were added, based on experimental data obtained with molecules 1–16. (C) Effect of test compounds 17–29 (100µM) on NTCP-mediated 3H-taurocholic acid uptake (1µM for 2 min) by CHO-NTCP cells. Compound 27 did not dissolve in the uptake buffer and could not be tested. Open bars indicate an absence of inhibition or a stimulation of NTCP-mediated taurocholic acid transport; closed bars indicate a significant inhibition of taurocholic acid transport. Data are expressed as a percentage of control and represent the mean ± SEM of three separate experiments each performed in duplicate or triplicate. * indicates p < 0.05 compared with control. TC, taurocholic acid. 44 GREUPINK ET AL. Downloaded from https://academic.oup.com/toxsci/article/129/1/35/1629030 by guest on 29 September 2021

Fig. 7. (A) Chemical structures of the compounds that were retrieved from the CoCoCo database and were tested for NTCP inhibitory activity in CHO-NTCP cells. (B) Graphical representation of the third pharmacophore used to retrieve molecules depicted in (A). The model results from the second refinement step, using only activity data of the anionic compounds. (C) Effect of test compounds 30–37 (100µM) on NTCP-mediated 3H-taurocholic acid uptake (1µM for 2 min) by CHO-NTCP cells. Compound 31 did not dissolve in the uptake buffer and was not tested. Open bars indicate absence of inhibition or stimulation of NTCP-mediated taurocholic acid transport; closed bars indicate a significant inhibition of taurocholic acid transport. Data are expressed as percentage of control and represent the mean ± SEM of an experiment performed in triplicate. * indicates p < 0.05 compared with control. TC, taurocholic acid. IN SILICO IDENTIFICATION OF NTCP INHIBITORS 45 Downloaded from https://academic.oup.com/toxsci/article/129/1/35/1629030 by guest on 29 September 2021

Fig. 8. Effect of compound 2 on 3H-taurocholic acid uptake by cryopreserved human hepatocytes. Data represent the mean ± SEM of an experiment performed in fourfold. * indicates p < 0.05 compared with control.

This should be a plasma concentration that results in Ntcp inhibition. Based on the observations from liver perfusion experiments, the lack of in vivo cholestatic activity is therefore likely to be the result of high protein binding, leading to unbound plasma concentrations of compound 2 that are substantially lower than the concentrations required for Ntcp inhibition.

DISCUSSION AND CONCLUSIONS

Here, we describe a pharmacophore model capable of identifying NTCP inhibitors from large virtual compound databases. Inhibition of NTCP function by xenobiotics will interfere with bile acid homeostasis and may lead to cholestasis. The flagging of chemical scaffolds with intrinsic NTCP inhibitory activity early during drug discovery is relevant because it can be used to identify chemotypes that may possess undesired off-target effects against this transporter. The substrate specificity of NTCP with regard to bile acid–like substrates has been studied, using a series of structurally similar bile acid analogs, but this did not culminate in the generation of a model that could identify non-bile acid–like structures (Kramer et al., 1999; Meier et al., 1997; Pauli-Magnus and Meier, 2006). To date, interaction studies with NTCP have therefore been mainly confined to in vitro screening of a large number of compounds (Kemp et al., Fig. 9. Results from ex vivo liver perfusion experiments. Effect of compound 3 2005; Kim et al., 1999; Kouzuki et al., 2000). To our knowl- 2 on venous outflow (A) and on the amount of H-taurocholic acid measured in the venous outflow (B). The effect of compound 2 on bile flow (C) and on the edge, our models are the first 3D ligand-based pharmacophore amount of 3H-taurocholic acid excreted in bile (D). All data represent the mean models to describe the substrate/inhibitor specificity of NTCP of two liver preparations per condition ± the range of the observations, except the of human origin. Also, we are the first to apply and validate data obtained in the presence of 1% BSA, which is derived from one liver. 46 GREUPINK ET AL.

to be competitive inhibitors. This was in line with the nature of the pharmacophore model. The model was based on bile acids, which bind to the substrate binding pocket of NTCP and are also competitive inhibitors of the transporter (Meier et al., 1997). The identified inhibitors from the virtual screen were therefore expected also to bind to the same binding pocket and hence also be competitive inhibitors, which was indeed confirmed. Further confidence in the model was obtained from the finding that structural analogues of compound 2, which did not hit all of the pharmacophore features, were inactive or less potent than compounds that hit all features of the model. The analogues that did not map simultaneously to the two HBA features of the model Downloaded from https://academic.oup.com/toxsci/article/129/1/35/1629030 by guest on 29 September 2021 were less active (compound 13) or inactive (compounds 11, 12, and 14). Moreover, compound 14 did possess two HBAs, but these were not of sufficient distance from each other in the molecule to map to the HBA features of the pharmacophore model. This suggests that an appropriate distance between HBAs in the molecule is important for a more successful interaction with the NTCP substrate binding pocket, and the presence of simply two HBAs is not sufficient for inhibitory activity. Also, from the model refinement studies, it appeared that negatively charged compounds were better inhibitors than uncharged compounds, which is in line with results obtained in earlier work (Kim et al., 1999; Kouzuki et al., 2000). Some of the compounds tested in the CHO-cell assay stimulated NTCP-mediated taurocholate transport. Stimulation has been observed for NTCP and other transporters before, but the underlying biochemical mechanism and clinical relevance of this phenomenon are currently unknown (Kim et al., 1999; Wittgen et al., 2011). To strengthen the argument that NTCP inhibition may have clinically relevant implications, we analyzed how well the model is able to match known competitive inhibitory drugs or endogenous compounds. Unfortunately, for many compounds, the type of inhibition was not reported, which hampered this effort (Kim et al., 1999; Mita et al., 2006; Sandhu et al., 2005). Fig. 10. Effect of a 5 mg/kg iv bolus injection of compound 2 on bile Nevertheless, we found that the progesterone metabolites flow (A) and biliary3 H-taurocholic acid excretion (B) in bile duct–cannulated allopregnanolone sulfate and epiallopregnanolone sulfate are Sprague Dawley rats. Bile flow was expressed as percentage of control, which competitive NTCP inhibitors, with K values of 8 and 6 M, was obtained during the stabilization period prior to injection of the test com- i µ pound or vehicle. 3H-taurocholic acid excretion is expressed as percentage of respectively (Abu-Hayyeh et al., 2010). Elevated plasma lev- the injected dose. All data represent the mean ± SEM of 3–5 rats per group. els of these metabolites in pregnancy are potentially linked to clinical symptoms of cholestasis such as pruritus. Indeed, the structures of these compounds also mapped well to our mod- such models by screening large virtual compound databases to els, covering both HBA features (see Supplementary figure 1). successfully identify currently unknown NTCP inhibitors that Bosentan is a potent, noncompetitive inhibitor of rat NTCP are not of bile acid origin. (IC50 = 0.7µM) and a competitive but less potent inhibitor

Although the model only consists of two types of pharma- (IC50 = 24µM) of human NTCP (Leslie et al., 2007). The use of cophoric features, namely three hydrophobic and two HBA this endothelin antagonist has been associated with hepatotoxic features, we expected that in the first virtual screening quite a effects. The structure of bosentan could be mapped to our mod- large number of compounds would fit the pharmacophore. To els of human NTCP. Only one HBA feature was covered by our surprise, only 0.4% of 2.5 million molecules screened fitted the structure, which appears to be in line with its intermediate to the model. This low number is in line with only a handful of inhibitory potency against human NTCP (see Supplementary non–bile acid NTCP inhibitors currently known and suggests figure 1). that the model was NTCP specific. Also, the three most potent In the past, pharmacophore modeling studies have been per- compounds identified in the first screening round were found formed for the rabbit and human apical sodium-dependent bile IN SILICO IDENTIFICATION OF NTCP INHIBITORS 47 acid transporter (ASBT, SCL10A2) (Baringhaus et al., 1999; although our model can successfully identify compounds with Zheng et al., 2009). ASBT facilitates the reabsorption of bile intrinsic NTCP inhibitory activity, it is helpful to combine our acids from the gut lumen, displays 35% overlap in amino NTCP pharmacophore model with subsequent in silico selec- acids with NTCP, and possesses a similar membrane topology tion steps that address in vivo protein binding. (Kosters and Karpen, 2008). ASBT pharmacophore modeling In conclusion, we successfully applied pharmacophore mod- studies were born out of an interest as a potential target for eling to identify new chemotypes with intrinsic NTCP inhibi- cholesterol-lowering therapies. The pharmacophore model for tory activity. Our approach represents a first step toward the human ASBT described by Zheng et al. has two HBA and a in silico identification of potential cholestasis-inducing com- number of hydrophobic features. This is similar to our results pounds during the early stages of drug discovery. and in line with the substrate overlap between the two transporters. However, the spatial orientation of the pharmacophoric features in the ASBT model is slightly different from the NTCP SUPPLEMENTARY DATA Downloaded from https://academic.oup.com/toxsci/article/129/1/35/1629030 by guest on 29 September 2021 model, in which the two HBA features appear to be positioned closer together (Zheng et al., 2009). Instead of the two HBAs Supplementary data are available online at present in human NTCP (present data) and ASBT (Zheng et al., http://toxsci.oxfordjournals.org/. 2009), the rabbit ASBT model contains one acceptor and one donor feature, suggesting differences in substrate specificity Funding between the rabbit and human proteins. To study how the data obtained in silico and in the in vitro Janssen Pharmaceutical Companies of Johnson & Johnson; NTCP recombinant expression system translated to more physi- Dutch Organization for Scientific Research (NWO) (700.58.410 ologically relevant systems, we tested compound 2 in cryopre- to S.B.N.). served human hepatocytes and perfused rat livers and studied the effect on biliary taurocholic acid excretion in rats in vivo. Similar to the results obtained in CHO-NTCP cells, com- Acknowledgments pound 2 exhibited a strong inhibition of taurocholic acid uptake Mr Tom Meeuws and Mr Guy van de Perre are gratefully in hepatocytes, with 2µM reducing taurocholate uptake by acknowledged for executing the liver perfusion experiments. approximately 60%. This is similar to the inhibitory potency of Mr Jos van Houdt is thanked for skillful conduction of hepato- cyclosporin A, which exhibits an IC of 1µM and is one of the 50 cyte uptake studies. The authors thank Accelrys Limited for most potent NTCP inhibitors known to date (Kim et al., 1999). providing the Discovery Studio software package and Katalin Also in rat liver perfusion experiments, compound 2 reduced Nadassy for her kind support. 3H-taurocholic acid uptake from the perfusate and induced cholestasis, which is in line with NTCP inhibition; however, effects on other bile acid–transporting proteins cannot be excluded. References In perfused rat liver, the effect could be observed only at higher concentrations of compound 2, suggesting that the affinity for Baringhaus, K. H., Matter, H., Stengelin, S., and Kramer, W. (1999). Substrate rat Ntcp may be lower than for the human protein. Such differ- specificity of the ileal and the hepatic Na(+)/bile acid cotransporters of the rabbit. II. A reliable 3D QSAR pharmacophore model for the ileal Na(+)/bile ences have been described before for the interaction between acid cotransporter. J. Lipid Res. 40, 2158–2168. bile acids and NTCP/Ntcp and for non–bile acid compounds Bouscarel, B., Ceryak, S., Gettys, T. W., Fromm, H., and Noonan, F. (1995). that interact with the transporter. For instance, the Km of tauro- Alteration of cAMP-mediated hormonal responsiveness by bile acids in cells cholic acid is higher for rat Ntcp than for the human ortholog of nonhepatic origin. Am. J. Physiol. 268(6 Pt 1), G908–G916. (Meier et al., 1997). Another example is the more potent inhibi- Abu-Hayyeh, S., Martinez-Becerra, P., Sheikh Abdul Kadir, S. H., Selden, C., tion of rat Ntcp compared with human NTCP by the endothelin Romero, M. R., Rees, M., Marschall, H. U., Marin, J. J., and Williamson, C. receptor antagonist bosentan (Leslie et al., 2007). (2010). Inhibition of Na+-taurocholate co-transporting polypeptide-mediated bile acid transport by cholestatic sulfated progesterone metabolites. J. Biol. Interestingly, we found that compound 2 did not display any Chem. 285, 16504–16512. relevant effects on biliary taurocholic acid excretion in vivo, Glantz, A., Marschall, H. U., and Mattsson, L. A. (2004). Intrahepatic cholesta- despite the fact that sufficiently high total plasma concentra- sis of pregnancy: Relationships between bile acid levels and fetal complica- tions were obtained. This is likely due to a very high protein tion rates. Hepatology 40, 467–474. binding of this compound in vivo resulting in a low unbound Greupink, R., Dillen, L., Monshouwer, M., Huisman, M. T., and Russel, F. plasma concentration, as our results obtained in rat liver perfu- G. (2011). Interaction of fluvastatin with the liver-specific Na+ -dependent sion experiments showed that addition of 1% BSA to the perfu- taurocholate cotransporting polypeptide (NTCP). Eur. J. Pharm. Sci. 44, 487–496. sate protected the liver from the cholestatic effects of compound Hilgendorf, C., Ahlin, G., Seithel, A., Artursson, P., Ungell, A. L., and 2. Unfortunately, iv injection of higher concentrations in order Karlsson, J. (2007). Expression of thirty-six drug transporter genes in human to test this hypothesis further was precluded due to solubility intestine, liver, kidney, and organotypic cell lines. Drug Metab. Dispos. 35, reasons. Nevertheless, these experiments demonstrated that 1333–1340. 48 GREUPINK ET AL.

Ho, R. H., Leake, B. F., Roberts, R. L., Lee, W., and Kim, R. B. (2004). Meier, P. J., Eckhardt, U., Schroeder, A., Hagenbuch, B., and Stieger, B. (1997). Ethnicity-dependent polymorphism in Na+-taurocholate cotransporting Substrate specificity of sinusoidal bile acid and organic anion uptake systems polypeptide (SLC10A1) reveals a domain critical for bile acid substrate rec- in rat and human liver. Hepatology 26, 1667–1677. ognition. J. Biol. Chem. 279, 7213–7222. Mita, S., Suzuki, H., Akita, H., Hayashi, H., Onuki, R., Hofmann, A. F., and Ho, R. H., Tirona, R. G., Leake, B. F., Glaeser, H., Lee, W., Lemke, C. J., Sugiyama, Y. (2006). Inhibition of bile acid transport across Na+/tauro- Wang, Y., and Kim, R. B. (2006). Drug and bile acid transporters in rosu- cholate cotransporting polypeptide (SLC10A1) and bile salt export pump vastatin hepatic uptake: Function, expression, and pharmacogenetics. (ABCB 11)-coexpressing LLC-PK1 cells by cholestasis-inducing drugs. Gastroenterology 130, 1793–1806. Drug Metab. Dispos. 34, 1575–1581. Kemp, D. C., Zamek-Gliszczynski, M. J., and Brouwer, K. L. (2005). Pauli-Magnus, C. and Meier, P. J. (2006). Hepatobiliary transporters and Xenobiotics inhibit hepatic uptake and biliary excretion of taurocholate in drug-induced cholestasis. Hepatology 44, 778–787. rat hepatocytes. Toxicol. Sci. 83, 207–214. Sandhu, P., Lee, W., Xu, X., Leake, B. F., Yamazaki, M., Stone, J. A., Lin, J. H., Kim, R. B., Leake, B., Cvetkovic, M., Roden, M. M., Nadeau, J., Walubo, Pearson, P. G., and Kim, R. B. (2005). Hepatic uptake of the novel antifungal A., and Wilkinson, G. R. (1999). Modulation by drugs of human hepatic agent caspofungin. Drug Metab. Dispos. 33, 676–682.

sodium-dependent bile acid transporter (sodium taurocholate cotransporting Schölmerich, J., Becher, M. S., Schmidt, K., Schubert, R., Kremer, B., Downloaded from https://academic.oup.com/toxsci/article/129/1/35/1629030 by guest on 29 September 2021 polypeptide) activity. J. Pharmacol. Exp. Ther. 291, 1204–1209. Feldhaus, S., and Gerok, W. (1984). Influence of hydroxylation and conjuga- Kosters, A. and Karpen, S. J. (2008). Bile acid transporters in health and dis- tion of bile salts on their membrane-damaging properties–studies on isolated ease. Xenobiotica 38, 1043–1071. hepatocytes and lipid membrane vesicles. Hepatology 4, 661–666. Kouzuki, H., Suzuki, H., Stieger, B., Meier, P. J., and Sugiyama, Y. (2000). Spivey, J. R., Bronk, S. F., and Gores, G. J. (1993). Glycochenodeoxycholate- Characterization of the transport properties of organic anion transporting induced lethal hepatocellular injury in rat hepatocytes. Role of ATP depletion polypeptide 1 (oatp1) and Na(+)/taurocholate cotransporting polypeptide and cytosolic free calcium. J. Clin. Invest. 92, 17–24. (Ntcp): Comparative studies on the inhibitory effect of their possible sub- Williamson, C., Hems, L. M., Goulis, D. G., Walker, I., Chambers, J., strates in hepatocytes and cDNA-transfected COS-7 cells. J. Pharmacol. Donaldson, O., Swiet, M., and Johnston, D. G. (2004). Clinical outcome in a Exp. Ther. 292, 505–511. series of cases of obstetric cholestasis identified via a patient support group. Kramer, W., Stengelin, S., Baringhaus, K. H., Enhsen, A., Heuer, H., Becker, BJOG 111, 676–681. W., Corsiero, D., Girbig, F., Noll, R., and Weyland, C. (1999). Substrate Wittgen, H. G., van den Heuvel, J. J., van den Broek, P. H., Dinter-Heidorn, specificity of the ileal and the hepatic Na(+)/bile acid cotransporters of the H., Koenderink, J. B., and Russel, F. G. (2011). Cannabinoid type 1 receptor rabbit. I. Transport studies with membrane vesicles and cell lines expressing antagonists modulate transport activity of multidrug resistance-associated the cloned transporters. J. Lipid Res. 40, 1604–1617. proteins MRP1, MRP2, MRP3, and MRP4. Drug Metab. Dispos. 39, Kuiper, E. M., van Erpecum, K. J., Beuers, U., Hansen, B. E., Thio, H. B., de 1294–1302. Man, R. A., Janssen, H. L., and van Buuren, H. R. (2010). The potent bile acid Yabe, Y., Galetin, A., and Houston, J. B. (2011). Kinetic characterization of rat sequestrant colesevelam is not effective in cholestatic pruritus: Results of a hepatic uptake of 16 actively transported drugs. Drug Metab. Dispos. 39, double-blind, randomized, placebo-controlled trial. Hepatology 52, 1334–1340. 1808–1814. Leslie, E. M., Watkins, P. B., Kim, R. B., and Brouwer, K. L. (2007). Differential Yanni, S. B., Augustijns, P. F., Benjamin, D. K., Jr, Brouwer, K. L., Thakker, inhibition of rat and human Na+-dependent taurocholate cotransporting pol- D. R., and Annaert, P. P. (2010). In vitro investigation of the hepatobiliary ypeptide (NTCP/SLC10A1)by bosentan: A mechanism for species differ- disposition mechanisms of the antifungal agent micafungin in humans and ences in hepatotoxicity. J. Pharmacol. Exp. Ther. 321, 1170–1178. rats. Drug Metab. Dispos. 38, 1848–1856. Makino, I., Nakagawa, S., and Mashimo, K. (1969). Conjugated and uncon- Zheng, X., Ekins, S., Raufman, J. P., and Polli, J. E. (2009). Computational jugated serum bile acid levels in patients with hepatobiliary diseases. models for drug inhibition of the human apical sodium-dependent bile acid Gastroenterology 56, 1033–1039. transporter. Mol. Pharm. 6, 1591–1603.