Target Fishing by Cross-Docking to Explain Polypharmacological Effects

Target Fishing by Cross-Docking to ExplainPolypharmacological EffectsHitesh Patel,[a] Xavier Lucas,[b] Igor Bendik,[c] Stefan Gìnther,*[b] and Irmgard Merfort*[a]

Introduction

It is well known that most drugs possess undesired side effectsalong with their intended therapeutic effects.[1] The reasons forthese so-called adverse drug reactions can often be explainedby the fact that drugs are not selective; instead, they bind tomany protein targets by a phenomenon known as polyphar-macology.[2] A prominent example is indomethacin, a potentnonsteroidal anti-inflammatory drug that inhibits not only cy-clooxygenase (COX)-2, but also its isoform, COX-1. This causessevere side effects such as gastrointestinal toxicity and in-creased bleeding time.[3] Therefore, knowledge of the bindingpartners of a drug is essential for predicting or explaining itsefficacy, side effects, and mode of action. Hunting for such ad-ditional potential physiological targets of a known drug iscalled target fishing. Importantly, adverse drug reactions area severe public health concern, are very costly for the com-munity, and are one of the main reasons behind drug attritionand ultimately drug withdrawal from the market. Many suc-cessful efforts have been made to predict side effects by

chemical systems biology.[4, 5] Yang et al. used a dockingmethod to explore off-targets and candidate genes responsiblefor adverse drug reactions. They identified the Hsp70 proteinas an off-target of clozapine. This finding was supported bymRNA expression data.[6, 7] Other approaches, such as sideeffect inference based on chemical similarity, fragment match-ing, and known drug–target pairs have also been used.[8–10] Allthese approaches rely on identifying probable off-target inter-actions and are based on previous knowledge of interactingdrug–target pairs and known side effects.

However, such off-target interactions can also be responsiblefor beneficial effects, which were not intended or recognizedduring the drug development phase. Another motivationbehind target fishing is to decrease costs in the pharmaceuti-cal industry: the amount of financial investment has increaseddramatically during the last decade, whereas the number ofapproved drugs has decreased. Finding new uses for olddrugs, i.e. , drug repositioning, repurposing, or reprofiling, hasbeen considered as an alternative strategy to tackle this prob-lem.[11, 12] Many repositioned drugs or promising candidates forrepositioning have been reviewed and published;[13–17] relativeto de novo drug discovery, drug repositioning has significantadvantages, including decreased overall risk and the capacityto bypass several stages of the drug discovery and develop-ment pipeline. Another outcome of target fishing is the ex-planation of a drug’s mode of action. In this regard, traditionalmedicines such as Chinese or Ayurvedic are known to exertbeneficial effects in many disease conditions, but their modeof action is commonly unknown.[18]

Computational approaches for drug repositioning or targetfishing, namely chemical similarity searching, data mining andmachine learning, network analysis, bioactivity spectra, andpanel docking have been reviewed elsewhere.[19–23] Thesemethods have been successfully applied to predict new inter-

Drugs may have polypharmacological phenomena, that is, inaddition to the desired target, they may also bind to many un-desired or unknown physiological targets. As a result, theyoften exert side effects. In some cases, off-target interactionsmay lead to drug repositioning or to explaining a drug’s modeof action. Herein we present an in silico approach for targetfishing by cross-docking as a method to identify new drug–protein interactions. As an example and proof of concept, thismethod predicted the peroxisome proliferator-activated recep-

tor (PPAR)-g as a target of ethacrynic acid, which may explainthe hyperglycemic effect brought on by this molecule. The an-tagonistic effect of ethacrynic acid on PPAR-g was validated ina transient transactivation assay using human HEK293 cells.The cross-docking approach also predicted the potential mech-anisms of many other drug side effects and discloses new drugrepositioning opportunities. These putative interactions are de-scribed herein, and can be readily used to discover therapeuti-cally relevant drug effects.

[a] Dr. H. Patel,+ Prof. Dr. I. MerfortPharmaceutical Biology and BiotechnologyInstitute of Pharmaceutical Sciences, Albert-Ludwigs UniversityStefan-Meier-Str. 19, 79104 Freiburg (Germany)E-mail : [email protected]

[b] Dr. X. Lucas, Prof. Dr. S. GìntherPharmaceutical BioinformaticsInstitute of Pharmaceutical Sciences, Albert-Ludwigs UniversityHermann-Herder-Str. 9, 79104 Freiburg (Germany)E-mail : [email protected]

[c] Dr. I. BendikDSM Nutritional Products Ltd. , Department of Human Nutrition and HealthP.O. Box 2676, 4002 Basel (Switzerland)

[++] Current address : Chemische Biologie, Technische Universit�t Dortmund,Otto-Hahn-Straße 6, 44227, Dortmund (Germany)

Supporting information for this article is available on the WWW underhttp ://dx.doi.org/10.1002/cmdc.201500123.

ChemMedChem 2015, 10, 1209 – 1217 Ó 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim1209

Full PapersDOI: 10.1002/cmdc.201500123

actions,[24] drug reposition-ing,[25, 26] and to propose themodes of action of drugs.[27]

In contrast to the knowledge-based methods discussed above,the technique presented hereinconsiders chemical interactionaffinity to identify putative noveltargets and uses known interac-tions only to validate the predic-tion capabilities of the scoringsystem. Various structure-basedapproaches for target fishinghave been reviewed.[28] For ex-ample, the drug interaction pro-files generated by docking witha series of protein targets wereused to predict the biologicaleffect of 1200 FDA-approveddrugs,[29] and for experimentallyvalidated angiotensin-convertingenzyme (ACE) and COX inhibito-ry predictions.[30] An inversedocking procedure (INVDOCK)was used to identify the knowntargets of 4-hydroxytamoxifenand vitamin E.[31] Paul et al. re-ported a similar study using thedocking software GOLD and re-covered true targets for four li-gands out of 2100 protein–ligand binding sites.[32] TarFis-Dock is a web server that per-forms inverse docking to finda query molecule’s target withinthe potential drug target data-base (PDTD), which includes 841protein structures.[33] A docking study was performed on fiverepresentative molecules from a combinatorial library sharinga 1,3,5-triazepan-2,6-dione scaffold against 2150 druggableactive sites from the RCSB Protein Data Bank (PDB).[34] Experi-mental validation showed that a panel of 1,3,5-triazepan-2,6-diones inhibited secreted phospholipase A2.[35]

In the present study, target fishing was performed usinga high-throughput cross-docking approach, in which 947 li-gands were docked to 515 proteins with the aim of identifyingnew drug–protein interactions that may explain reported sideeffects of drugs, propose drug repositioning, or suggest novelmechanisms of action. The method was validated usinghuman peroxisome proliferator-activated receptor (PPAR)-g.Additionally, many other predicted off-target interactions areproposed.

Results and Discussion

The intersection of co-crystallized protein targets with their li-gands in the PDB[34] and DrugBank[36] was used as input data

set; 947 ligands were cross-docked with each of the 515 pro-teins by means of virtual docking (Figure 1). Cross-docking pre-dicted many novel drug–target interactions. Afterward, inter-esting yet unknown potential drug–target pairs were studiedin detail for some marketed drugs. A statistical analysis of theresults is presented and discussed along with putative newdrug–target interactions, which may explain known side ef-fects, propose repositioning, and predict alternative modes ofaction.

Assessment of cross-docking results

After the all-against-all docking, a statistical analysis was car-ried out to evaluate the results. For each protein, ranking ofthe known ligand was determined; 198 redundant proteinshad their corresponding reference compound in the top 20and were considered for further study, representing 138unique proteins. Many reasons may explain the loss of the restof the targets, including the selection of standard parametersused in high-throughput cross-docking, the target dependency

Figure 1. Workflow depicting the selection of data sets and the cross-docking protocol.

ChemMedChem 2015, 10, 1209 – 1217 www.chemmedchem.org Ó 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim1210

Full Papers

of the docking algorithm, the unclear effect of water moleculesin protein–substrate recognition, and the possible inaccurateprediction of missing loops or side chains in the region of theactive pocket during protein preparation. Altogether, those ef-fects decreased the resulting set to 2760 interactions generat-ed from the 138 unique proteins and their predicted interac-tions with the 20 top-ranked ligands. Remarkably, in 171 out of198 protein–(known ligand) interaction pairs, new ligands wereranked better than the co-crystallized reference compounds(Figure 2 A).

We also analyzed the cross-docking results by taking into ac-count the docking score of each drug–protein complex. Dock-ing scores range from ¢4.2 to ¢17.8 kcal mol¢1, and the mini-mum and maximum differences between the 1st and 20th li-gand’s docking score for each protein are 0.7 and 6.6 kcalmol¢1, respectively. The relatively small ranges indicate that thecross-docking protocol produced reasonably consistent intra-

target docking scores along the 138 selected proteins (Fig-ure 2 B).

To evaluate the overall quality of the predictions, we calcu-lated a receiver operating characteristic (ROC) curve (Figure 3).The reasonable accuracy of the method is indicated by an areaunder the curve (AUC) value of 0.68.

Structure similarity comparison of predicted and co-crystal-lized ligands

Cross-docking aims at identifying putative ligands for each in-spected target. We calculated the Tanimoto similarity between2245 ligand pairs, considering each co-crystallized drug andthe other ligands in the top 20. As shown in Figure 2 C, in 93 %of the cases the Tanimoto coefficient is lower than 0.4, indicat-ing a remarkably low structural similarity between the co-crys-tallized drug and any other hit. Thus, the presented method

allows the exploration of a rea-sonably diverse chemical space,which may be out of the scopeof other high-throughput tech-niques based on ligand similari-ty.

Study of newly predicted drug–target interactions

To evaluate newly predicteddrug–target interactions, we se-lected some commonly usedmarketed drugs and studied re-lated scientific literature in thecontext of described targets anddrug effects. Predictions sup-ported by published data aresummarized in Table 1. Exampleswere found for molecular mech-anisms of drug side effects, re-positioning opportunities, andundiscovered modes of action.For instance, mitochondrial argi-nine:glycine amidinotransferaseis predicted as a target of gaba-pentin, which could explain themechanism of its side effects, in-cluding memory problems ormyalgia.[37] Clofarabine is sug-gested as a novel antimalarialagent, as it might be an inhibitorof deoxyuridine 5’-triphosphatenucleotidohydrolase from Plas-modium falciparum. As an exam-ple for explaining a drug’s mech-anism of action, we propose thatthe potentiating effect of flavo-piridol on the efficacy of doxoru-bicin is related to an additional

Figure 2. A) Number of drug–target interactions with the co-crystallized drug ranked within the top 20. B) Dock-ing score of the 1st ranked ligand (orange), the 20th ranked ligand (blue), average score of the top-20 ligands(yellow), and difference between the 1st and 20th ligands’ score (green) for each of the 138 selected proteins.C) Binned Tanimoto coefficients (Tc) between the co-crystallized and new putative ligand pairs ; in 93 % of thecases, the new ligands are structurally unrelated (Tc<0.4).

ChemMedChem 2015, 10, 1209 – 1217 www.chemmedchem.org Ó 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim1211

Full Papers

interaction with carbonyl reduc-tase 1, which is known to metab-olize doxorubicin.[38] A completelist of all top-ranked predictionsis provided as Supporting Infor-mation. To further evaluate thein silico approach, we investigat-ed the predicted interaction ofethacrynic acid with PPAR-g ex-perimentally.

Interaction of ethacrynic acidwith PPAR-g

Ethacrynic acid (Edecrin, SodiumEdecrin) is a loop diuretic usedto treat high blood pressure andswelling.[39, 40] It acts by inhibiting(Na/K)Cl co-transport in the as-cending loop of Henle and isknown to cause ototoxicity,[41]

liver damage,[42] and hyperglyce-mia in a reversible manner.[43]

From the cross-docking analy-sis, human PPAR-g was identifiedas a target of ethacrynic acid.PPAR-g belongs to the nuclearreceptor superfamily and is also known as the glitazone recep-tor or nuclear receptor subfamily 1, group C, member 3(NR1C3).[44] Like other subtypes of PPARs, PPAR-g regulatesgene transcription by forming heterodimers with retinoid X re-ceptors (RXRs). This protein is mainly involved in the regulation

of lipid metabolism, insulin sensitivity, and blood glucose ho-meostasis.[45] Glitazones are PPAR-g agonists that decrease insu-lin resistance and have been used in the treatment of hyper-glycemia. It has been reported that PPAR-g activation increasesbasal and insulin-induced glucose uptake in 3T3L1 adipocytes,and conversely, PPAR-g suppression decreases insulin-inducedglucose uptake in 3T3L1.[46] Indeglitazar (3-[5-methoxy-1-(4-me-thoxyphenyl)sulfonylindol-3-yl]propanoic acid) is a knownligand of PPAR-g and has been co-crystallized with the pro-tein.[47] The crystal structure reveals that indeglitazar interactsby forming hydrogen bonds with the side chains of His323,His449, and Tyr473. Molecular modeling could reproduce thisbinding mode. Similarly, the carboxyl group of ethacrynic acidinteracts with the same residues (Figure 4). Presuming thatethacrynic acid acts as an antagonist of PPAR-g, its inhibitionshould decrease the uptake of blood glucose and subsequent-ly increase the blood glucose level. This new predicted interac-tion could therefore explain the observed hyperglycemic activi-ty of ethacrynic acid.[43]

To validate the proposed hypothesis, the antagonistic activi-ty of ethacrynic acid was assessed by using a PPAR-g transienttransactivation assay. The effect of ethacrynic acid and the pos-itive control, rosiglitazone, on PPAR-g is shown in Figure 5.PPAR-g transactivation is monitored by the relative lumines-

cence of firefly luciferase, normalized to the individual transfec-tion efficiencies using Renilla expression vector co-transfection.A solution of rosiglitazone (200 nm), a potent PPAR-g agonist,increased PPAR-g-mediated relative luminescence from 0.8 to6.3. Increasing concentrations of ethacrynic acid decreased the

Figure 3. ROC curve of generated predictions. The true positive rate wascomputed as the ratio of co-crystallized ligands identified by docking to thetotal number of predictions in each rank, whereas predicted but not co-crys-tallized interactions were treated as negatives.

Figure 4. Drugs and their predicted or co-crystallized interacting mode with PPAR-g : A1) Ethacrynic acid, A2) etha-crynic acid docked to PPAR-g, B1) indeglitazar, and B2) indeglitazar co-crystallized with PPAR-g (PDB code 3ET3).Both compounds show analogous polar interactions with the recognition site.

ChemMedChem 2015, 10, 1209 – 1217 www.chemmedchem.org Ó 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim1212

Full Papers

Table 1. List of putative new targets of marketed drugs and predicted outcomes.

Drug Putative new target Rationale for the putative new drug–target interaction

Explaining side effects

Aminocaproicacid

Mitochondrial arginine:glycine amidinotransferase (AGAT) * Aminocaproic acid and gabapentin have side effects such as memoryproblems, mental retardation, language or speech problems, and myalgia.

Gabapentin * AGAT catalyzes an intermediate step in the biosynthesis of creatine,[54]

which is involved in the ATP–PCr energy system and is necessary to pro-vide energy to brain and muscle cells.* It is known that inborn AGAT deficiency leads to severe decrease or ab-sence of creatine in the central nervous system, which is characterized bymental retardation, memory problems, language or speech problems, andmyalgia.[37, 55–58]

* Inhibition of AGAT by aminocaproic acid or gabapentin may lead to de-creased synthesis of creatine, which may explain the mechanism of theside effects.

Erlotinib Adenosine kinase (ADK) * ADK deficiency disrupts the methionine cycle and causes encephalop-athy.[59]

* Explains erlotinib-induced encephalopathy.[60]

Proposing putative drug repositioning

Clofarabine Adenosylhomocysteinase [Plasmodium falciparum(isolate 3D7)]

* Treatment of malaria[61–64]

Quinohemoprotein alcohol dehydrogenase ADH IIB[Pseudomonas putida (Arthrobacter siderocapsulatus)]

* Antibacterial[65]

Deoxyuridine 5’-triphosphate nucleotidohydrolase[Plasmodium falciparum (isolate 3D7)]

* Treatment of malaria[66]

Erlotinib Adenosine kinase (ADK) * ADK converts adenosine to adenosine monophosphate (AMP) and regu-lates adenosine concentration.[67]

* Inhibition of ADK may increase extracellular adenosine, which is anticon-vulsive and neuroprotective.[68–71]

* As an ADK inhibitor, erlotinib can be used as anticonvulsive, neuropro-tective,[72] or anti-hyperalgesic agent.

Enoyl-[acyl carrier protein] reductase [NADH] FabI [Helicobact-er pylori (strain ATCC 700392/26695) (Campylobacter pylori)]

* Key enzyme of the type II fatty acid biosynthesis pathway in prokaryotesand plants,[73] thus it may act as an antibacterial.

Flufenamicacid

Mitochondrial dihydroorotate dehydrogenase (quinone) * Treatment of rheumatoid arthritis and other autoimmune diseases.[74]

Enoyl-[acyl carrier protein] reductase [NADH] FabI [Helicobact-er pylori (strain ATCC 700392/26695) (Campylobacter pylori)]

* Key enzyme of the type II fatty acid biosynthesis pathway in prokaryotesand plants,[73] thus it may act as an antibacterial.

Fluorouracil Deoxyuridine 5’-triphosphate nucleotidohydrolase[Plasmodium falciparum (isolate 3D7)]

* Treatment of malaria.[66]

Gemcitabine 5’-Methylthioadenosine/S-adenosylhomocysteine nucleosidase[Vibrio cholera serotype O1 (strain ATCC 39315/El Tor InabaN16961)]

* Treatment of cholera.[75]

Imatinib Prolyl endopeptidase or prolyl oligopeptidase (POP) * Because POP inhibitors have neuroprotective, antiamnestic, and cogni-tion-enhancing properties,[76] imatinib could be used for the same purpose.

Carbonyl reductase [NADPH] 1 (CBR1) * CBR1 metabolizes doxorubicin to doxorubicinol, which is cardiotoxic.[77]

* CBR1 inhibitors were tested to improve treatment by anthracyclines.[78]

* Putative CBR1 inhibitor for protection from doxorubicin-induced toxici-ty.

Niflumic acid Mitochondrial dihydroorotate dehydrogenase (quinone) * Treatment of rheumatoid arthritis and other autoimmune diseases.[74]

Sulfasalazine 4,5:9,10-Diseco-3-hydroxy-5,9,17-trioxoandrosta-1(10),2-diene-4-oate hydrolase [Mycobacterium tuberculosis] (hsaD)

* Inhibition of hsaD has been patented for the treatment of tuberculo-sis.[79]

* Putative treatment of tuberculosis.

Proposing alternate or new modes of action for known effects of drugs

ChemMedChem 2015, 10, 1209 – 1217 www.chemmedchem.org Ó 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim1213

Full Papers

transactivation mediated by 200 nm rosiglitazone in a dose-de-pendent manner. This result can be explained by competitivebinding of ethacrynic acid and rosiglitazone to the same activesite of the PPAR-g receptor. We repeated the experiment withincreasing concentrations of ethacrynic acid without rosiglita-zone and observed a dose-dependent decrease in basal PPAR-g activity (Figure 5). At a concentration of 156 nm, ethacrynicacid decreased the relative luminescence by ~50 % relative tocontrol DMSO, and at 1.25 mm it was decreased to an evengreater extent. Thus, the PPAR-g reporter gene assay provided

evidence for the antagonisticeffect of ethacrynic acid onPPAR-g transactivation.

Conclusions

The presented cross-docking ap-proach can be applied to predictside effects of marketed drugsand candidates in the develop-mental phase, and to help de-crease drug candidate attrition.Knowledge of the target respon-sible for adverse drug reactionscould be useful for chemicalmodification of the potentialdrug in order to minimize unde-sired drug–target interactions,while retaining efficacy towardthe intended target. In additionto showing promise in targetfishing, this method can be used

to propose drug repositioning and can reveal the modes ofaction for known effects of drugs.

Experimental Section

A Scheme representing the applied workflow is depicted inFigure 1.

Collection of protein structures : Protein structures for dockingwere obtained from the PDB.[34] The data set was decreased by ob-taining the subset of proteins targeted by drugs listed in Drug-

Table 1. (Continued)

Drug Putative new target Rationale for the putative new drug–target interaction

Flavopiridol CBR1 * Flavopiridol is known to potentiate doxorubicin efficacy in vitro, in vivo,and in phase I clinical trials in patients with advanced sarcomas.[38]

* Flavopiridol as a CBR1 inhibitor may inhibit doxorubicin metabolism todoxorubicinol, thus potentiating its efficacy.

Lovastatin Retinoic acid receptor-g (RAR-g) * Statins are already known for their beneficial effects in chronic obstruc-tive pulmonary disease (COPD).[80–82]

* May act as an agonist of RAR-g (like retinoic acid[83]), thus inducing re-generation of alveoli and could be used in the treatment of emphysema(COPD).[84, 85]

Oxysterol receptor LXR-b * LXRs are known to play an important role in atherosclerosis and inflam-mation.[86]

* LXR ligands attenuate plaque formation[87] and inhibit the developmentof atherosclerosis in mice.[88]

* Another mechanism of anti-atherosclerotic activity.[89]

Resveratrol Neurophil collagenase (MMP-8) * MMP-8 plays important roles in inflammatory disorders and atheroscle-rosis.[90–92]

* Putative mechanism of anti-inflammatory and anti-atherosclerotic effectof resveratrol.[93–95]

3-Hydroxyacyl-[acyl carrier protein] dehydratase FabZ[Helicobacter pylori (Campylobacter pylori)] (FabZ)

* FabZ is a well-known target to inhibit H. pylori.[96, 97]

* Resveratrol is known to inhibit H. pylori.[98] We propose that its antibac-terial activity is due to targeting FabZ.

Figure 5. Antagonistic effect of ethacrynic acid on rosiglitazone-activated PPAR-g transactivation. Rosiglitazonetransactivated PPAR-g at a concentration of 200 nm, which is shown as the normalized relative luminescence ratioof firefly to Renilla luciferase enzymatic activity. Increasing concentrations of ethacrynic acid from 156 to25 000 nm antagonized the PPAR-g transactivation of rosiglitazone. Ethacrynic acid alone down-regulated thebasal PPAR-g transactivation in a dose-dependent manner. The basal PPAR-g activity in this transient transactiva-tion assay is given by vehicle (DMSO) control.

ChemMedChem 2015, 10, 1209 – 1217 www.chemmedchem.org Ó 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim1214

Full Papers

Bank.[36] This was achieved by mapping the list of UniProtKB men-tioned in DrugBank with the UniProtKB of each chain from thePDB. The resulting data set contained redundant structures forsingle proteins. All sequences of the resulting data set were clus-tered using CD-HIT[48] using a cutoff of 95 % in sequence identity.One protein from each cluster with at least one co-crystallizedligand from the DrugBank database and the highest atomic resolu-tion was selected as the representative protein structure. This pro-cedure yielded 515 different proteins.

Preparation of the ligand library : All ligands from the PDB werecollected using the LigandExpo set,[49] and their correspondingquantitative estimation of drug-likeness (QED) value[50] was calcu-lated using ChemicalToolBoX.[51] Compounds with QED>0.35 wereselected as having drug-like properties. This procedure yielded1682 compounds, which were subsequently mapped to the mole-cules in DrugBank and led to 947 unique ligands used for thecross-docking experiment.

Cross-docking experiment : The 947 collected ligands were pre-pared using LigPrep 2.5 (Schrçdinger LLC, New York, NY, USA), gen-erating 1572 ready-to-dock compound states with appropriate pro-tonation states. The final data set containing 515 proteins was pre-processed before performing docking studies. At first, PyWATER[52]

was used to identify putative conserved water molecules in theprotein structures, which may play a crucial role in ligand binding.For each protein, PyWATER generated a new file in .pdb formatwith putatively conserved water molecules, if any. Those files werepre-processed using Protein Preparation Wizard 2012 (SchrçdingerLLC). During protein preparation, hydrogen atoms were first addedto the protein, and bond orders and formal charges were adjusted.Missing side chains and loops were predicted using Prime 3.1(Schrçdinger LLC). If needed, the ionization and tautomerizationstates of protein residues and the ligand were also adjusted atpH 7�2 using Epik 2.3 (Schrçdinger LLC). The hydrogen bond net-work was optimized by rotating hydroxy and thiol groups, watermolecules, amide groups of asparagine and glutamine, and theimidazole rings of histidine residues. The protonation state of polaramino acids was predicted by PROPKA[53] at neutral pH. Subse-quently, full-atom protein structure minimization using theOPLS2005 force field was carried out. The prepared proteins weresubjected to the grid-generation protocol. From each preparedprotein file, one grid was generated for each co-crystallized ligandin the protein, considering the geometric center of the ligand asgrid centroid. Only drugs bound noncovalently to the protein wereconsidered for grid generation, yielding 515 grids. All 1572 ligandstates were cross-docked with each grid, requiring over 800 000single drug–protein complex evaluations. The high-throughputmolecular docking campaign was performed using the eXtra Preci-sion (XP) protocol from Glide 5.8 (Schrçdinger LLC) at the BlackForest Grid (Baden-Wìrttemberg, www.bfg.uni-freiburg.de).

Selection of putative new drug–target interactions : The resultingthousands of drug–target interactions from docking were subse-quently filtered. Assuming that the grids and docking calculationsare more reliable if the co-crystallized ligand is ranked higher, onlythose grids in which the co-crystallized ligand was ranked at posi-tion 20 or better were considered for further analysis. This yielded138 grids. For each of them, the other 19 highly ranked ligandswere investigated. Those interactions show new putative drug–target interactions, which were studied further by extensive litera-ture review.

Availability of test compounds : Rosiglitazone (5-((4-(2-(methyl-2-pyridinylamino)ethoxy)phenyl)methyl)-2,4-thiazolidinedione) with

a purity of 99.5 % was obtained from BIOTREND Chemicals AG(Zìrich, Switzerland), and ethacrynic acid was purchased from EnzoLife Sciences (Lçrrach, Germany) with a purity of �98 %.

PPAR-g transient transactivation assay : The transient transfec-tions were performed in HEK293 cells (ATCC, Molsheim, France)grown in minimum essential medium (Eagle) with Earle’s balancedsalt solution without l-glutamine and supplemented with 10 %fetal bovine serum (Sigma–Aldrich, St. Gallen, Switzerland), 2 mmGlutamax (Life Technologies AG, Basel, Switzerland), 0.1 mm non-essential amino acids (Life Technologies), and 1 mm sodium pyru-vate (Life Technologies) at 37 8C in 5 % CO2. For transfection, 7.5 Õ104 cells were plated per well (80 mL) in white clear-bottom 96-wellcell culture plates (Corning, Basel, Switzerland) in minimum essen-tial medium (Eagle) with Earle’s balanced salt solution without l-glutamine and without phenol red supplemented with 10 % char-coal-treated fetal bovine serum (HyClone Laboratories Inc. , Logan,UT, USA), 2 mm glutamax, 0.1 mm non-essential amino acids, and1 mm sodium pyruvate. The cells were transiently transfectedbefore stimulation the following day at >80 % confluence by poly-ethylene-imine-based transfection. Compound stocks were pre-pared in DMSO, pre-diluted in PBS (0.45 % final DMSO concentra-tion), and added in the respective dilution 5 h after the addition ofthe transfection mixture onto the cells. The cells were again incu-bated for an additional 16 h before firefly and Renilla luciferaseswere measured sequentially in the same cell extract using buffersaccording to established protocols (Promega AG, Dìbendorf, Swit-zerland). Transfection efficiency was normalized to the pRL-TK Re-nilla luciferase reporter. The ligand binding domain of PPAR-g wasexpressed from a GATEWAY (Invitrogen, Zug, Switzerland)-compati-ble version of pCMV-BD (Stratagene Corp., Santa Clara, CA, USA) asa fusion to the GAL4 DNA binding domain (residues 1–147); pFR-Luc (Stratagene) was used as a reporter plasmid.

Acknowledgements

The authors acknowledge the Baden-Wìrttemberg and BlackForest Grids. The authors also acknowledge Bjçrn A. Grìning (De-partment of Computer Science, Albert-Ludwigs University Frei-burg, Germany) for assistance in installing the Schrçdinger suiteon the grids. We thank Mrs. Pascale Fuchs (DSM Nutritional Prod-ucts) for excellent technical assistance and Dr. Joseph Schwager(DSM Nutritional Products) for fruitful discussions.

Keywords: ethacrynic acid · high-throughput virtualscreening · polypharmacology · PPAR-g · side effects

[1] A. L. Hopkins, Nat. Chem. Biol. 2008, 4, 682 – 690.[2] J. Scheiber, J. L. Jenkins, S. C. K. Sukuru, A. Bender, D. Mikhailov, M. Milik,

K. Azzaoui, S. Whitebread, J. Hamon, L. Urban, M. Glick, J. W. Davies, J.Med. Chem. 2009, 52, 3103 – 3107.

[3] A. L. Blobaum, M. J. Uddin, A. S. Felts, B. C. Crews, C. A. Rouzer, L. J. Mar-nett, ACS Med. Chem. Lett. 2013, 4, 486 – 490.

[4] N. P. Tatonetti, T. Liu, R. B. Altman, Genome Biol. 2009, 10, 238.[5] J. Scheiber, B. Chen, M. Milik, S. C. K. Sukuru, A. Bender, D. Mikhailov, S.

Whitebread, J. Hamon, K. Azzaoui, L. Urban, M. Glick, J. W. Davies, J. L.Jenkins, J. Chem. Inf. Model. 2009, 49, 308 – 317.

[6] L. Yang, J. Chen, L. He, PLoS Comput. Biol. 2009, 5, DOI: 10.1371/jour-nal.pcbi.1000441.

[7] L. Yang, K. Wang, J. Chen, A. G. Jegga, H. Luo, L. Shi, C. Wan, X. Guo, S.Qin, G. He, G. Feng, L. He, PLoS Comput. Biol. 2011, 7, DOI: 10.1371/journal.pcbi.1002016.

[8] E. Pauwels, V. Stoven, Y. Yamanishi, BMC Bioinf. 2011, 12, 169.

ChemMedChem 2015, 10, 1209 – 1217 www.chemmedchem.org Ó 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim1215

Full Papers

[9] Y. Yamanishi, E. Pauwels, M. Kotera, J. Chem. Inf. Model. 2012, 52, 3284 –3292.

[10] F. Cheng, W. Li, Z. Wu, X. Wang, C. Zhang, J. Li, G. Liu, Y. Tang, J. Chem.Inf. Model. 2013, 53, 753 – 762.

[11] C. R. Chong, D. J. Sullivan, Nature 2007, 448, 645 – 646.[12] M. J. Keiser, V. Setola, J. J. Irwin, C. Laggner, A. I. Abbas, S. J. Hufeisen,

N. H. Jensen, M. B. Kuijer, R. C. Matos, T. B. Tran, R. Whaley, R. A. Glen-non, J. Hert, K. L. H. Thomas, D. D. Edwards, B. K. Shoichet, B. L. Roth,Nature 2009, 462, 175 – 181.

[13] T. T. Ashburn, K. B. Thor, Nat. Rev. Drug Discovery 2004, 3, 673 – 683.[14] A. Corbett, J. Pickett, A. Burns, Nat. Rev. Drug Discovery 2012, 11, 833 –

846.[15] A. Corbett, G. Williams, C. Ballard, Pharmaceuticals 2013, 6, 1304 – 1321.[16] M. Aldea, L. Craciun, C. Tomuleasa, I. Berindan-Neagoe, G. Kacso, I. S.

Florian, C. Crivii, Tumor Biol. . 2014, 35, 5101 – 5110.[17] J. S. Shim, J. O. Liu, Int. J. Biol. Sci. 2014, 10, 654 – 663.[18] F. Cheng, C. Liu, J. Jiang, W. Lu, W. Li, G. Liu, W. Zhou, J. Huang, Y. Tang,

PLoS Comput. Biol. 2012, 8, e1002503.[19] J. L. Jenkins, A. Bender, J. W. Davies, Drug Discovery Today Technol. 2006,

3, 413 – 421.[20] A. Bender, J. Scheiber, M. Glick, J. W. Davies, K. Azzaoui, J. Hamon, L.

Urban, S. Whitebread, J. L. Jenkins, ChemMedChem 2007, 2, 861 – 873.[21] A. Bender, D. Mikhailov, M. Glick, J. Scheiber, J. W. Davies, S. Cleaver, S.

Marshall, J. A. Tallarico, E. Harrington, I. Cornella-Taracido, J. L. Jenkins, J.Proteome Res. 2009, 8, 2575 – 2585.

[22] D.-L. Ma, D. S.-H. Chan, C.-H. Leung, Chem. Soc. Rev. 2013, 42, 2130 –2141.

[23] Z. Wu, Y. Wang, L. Chen, Mol. Biosyst. 2013, 9, 1268 – 1281.[24] A. F. Fliri, W. T. Loging, P. F. Thadeio, R. A. Volkmann, Proc. Natl. Acad. Sci.

USA 2005, 102, 261 – 266.[25] N. Wale, G. Karypis, J. Chem. Inf. Model. 2009, 49, 2190 – 2201.[26] F. Napolitano, Y. Zhao, V. M. Moreira, R. Tagliaferri, J. Kere, M. D’Amato,

D. Greco, J. Cheminf. 2013, 5, 30.[27] F. Iorio, A. Isacchi, D. di Bernardo, N. Brunetti-Pierri, Autophagy 2010, 6,

1204 – 1205.[28] D. Rognan, Mol. Inf. 2010, 29, 176 – 187.[29] Z. Simon, Ý. Peragovics, M. Vigh-Smeller, G. Csukly, L. Tombor, Z. Yang,

G. Zahor�nszky-Kûhalmi, L. V¦gner, B. Jelinek, P. H�ri, C. Het¦nyi, I.Bitter, P. Czobor, A. M�ln�si-Csizmadia, J. Chem. Inf. Model. 2012, 52,134 – 145.

[30] L. V¦gner, Ý. Peragovics, L. Tombor, B. Jelinek, P. Czobor, A. Bender, Z.Simon, A. M�ln�si-Csizmadia, J. Med. Chem. 2013, 56, 8377 – 8388.

[31] Y. Z. Chen, D. G. Zhi, Proteins Struct. Funct. Genet. 2001, 43, 217 – 226.[32] N. Paul, E. Kellenberger, G. Bret, P. Mìller, D. Rognan, Proteins Struct.

Funct. Genet. 2004, 54, 671 – 680.[33] H. Li, Z. Gao, L. Kang, H. Zhang, K. Yang, K. Yul, X. Luo, W. Zhu, K. Chen,

J. Shen, X. Wang, H. Jiang, Nucleic Acids Res. 2006, 34, W219 – W224.[34] F. C. Bernstein, T. F. Koetzle, G. J. B. Williams, E. F. Meyer, M. D. Brice, J. R.

Rodgers, O. Kennard, T. Shimanouchi, M. Tasumi, Eur. J. Biochem. 1977,80, 319 – 324.

[35] P. Muller, G. Lena, E. Boilard, S. Bezzine, G. Lambeau, G. Guichard, D.Rognan, J. Med. Chem. 2006, 49, 6768 – 6778.

[36] D. S. Wishart, C. Knox, A. C. Guo, D. Cheng, S. Shrivastava, D. Tzur, B.Gautam, M. Hassanali, Nucleic Acids Res. 2008, 36, D901 – D906.

[37] V. Leuzzi, M. Mastrangelo, R. Battini, G. Cioni, Epilepsia 2013, 54, 217 –227.

[38] J. J. Luke, D. R. D’Adamo, M. A. Dickson, M. L. Keohan, R. D. Carvajal,R. G. Maki, E. De Stanchina, E. Musi, S. Singer, G. K. Schwartz, Clin.Cancer Res. 2012, 18, 2638 – 2647.

[39] P. J. Cannon, M. M. Kilcoyne, Prog. Cardiovasc. Dis. 1969, 12, 99 – 118.[40] J. C. Somberg, J. Molnar, Am. J. Ther. 2009, 16, 102 – 104.[41] S. K. Bosher, Acta Oto-Laryngol. 1980, 89, 407 – 418.[42] K. K. Datey, S. N. Deshmukh, C. P. Dalvi, N. M. Purandare, Br. Med. J.

1967, 3, 152 – 153.[43] A. J. Cowley, R. S. Elkeles, Lancet 1978, 311, 154.[44] R. M. Evans, D. J. Mangelsdorf, Cell 2014, 157, 255 – 266.[45] F. A. Monsalve, R. D. Pyarasani, F. Delgado-Lopez, R. Moore-Carrasco, Me-

diators Inflammation 2013, 2013, 549627.[46] W. Liao, M. T. A. Nguyen, T. Yoshizaki, S. Favelyukis, D. Patsouris, T. Ima-

mura, I. M. Verma, J. M. Olefsky, Am. J. Physiol. Endocrinol. Metab. 2007,293, E219 – E227.

[47] D. R. Artis, J. J. Lin, C. Zhang, W. Wang, U. Mehra, M. Perreault, D. Erbe,H. I. Krupka, B. P. England, J. Arnold, A. N. Plotnikov, A. Marimuthu, H.Nguyen, S. Will, M. Signaevsky, J. Kral, J. Cantwell, C. Settachatgull, D. S.Yan, D. Fong, A. Oh, S. Shi, P. Womack, B. Powell, G. Habets, B. L. West,K. Y. J. Zhang, M. V. Milburn, G. P. Vlasuk, K. P. Hirth, K. Nolop, G. Bollag,P. N. Ibrahim, J. F. Tobin, Proc. Natl. Acad. Sci. USA 2009, 106, 262 – 267.

[48] W. Li, A. Godzik, Bioinformatics 2006, 22, 1658 – 1659.[49] Z. Feng, L. Chen, H. Maddula, O. Akcan, R. Oughtred, H. M. Berman, J.

Westbrook, Bioinformatics 2004, 20, 2153 – 2155.[50] G. R. Bickerton, G. V. Paolini, J. Besnard, S. Muresan, A. L. Hopkins, Nat.

Chem. 2012, 4, 90 – 98.[51] B. A. Grìning, X. Lucas, G. Von Kuster, H. Patel, M. Chbeib, S. Bleher,

K. K. Telukunta, I. Merfort, S. Gìnther, ChemicalToolBoX :https://github.com/bgruening/galaxytools/tree/master/chemicaltool-box.

[52] H. Patel, B. A. Grìning, S. Gìnther, I. Merfort, Bioinformatics 2014, 30,2978 – 2980.

[53] C. R. Søndergaard, M. H. M. Olsson, M. Rostkowski, J. H. Jensen, J. Chem.Theory Comput. 2011, 7, 2284 – 2295.

[54] A. Humm, E. Fritsche, S. Steinbacher, R. Huber, EMBO J. 1997, 16, 3373 –3385.

[55] E. B¦ard, O. Braissant, J. Neurochem. 2010, 115, 297 – 313.[56] R. Battini, V. Leuzzi, C. Carducci, M. Tosetti, M. C. Bianchi, C. B. Item, S.

Stçckler-Ipsiroglu, G. Cioni, Mol. Genet. Metab. 2002, 77, 326 – 331.[57] S. Stockler, P. W. Schutz, G. S. Salomons, Subcell. Biochem. 2007, 46,

149 – 166.[58] C. I. Nabuurs, C. U. Choe, A. Veltien, H. E. Kan, L. J. C. van Loon, R. J. T.

Rodenburg, J. Matschke, B. Wieringa, G. J. Kemp, D. Isbrandt, A. Heer-schap, J. Physiol. 2013, 591, 571 – 592.

[59] M. K. Bjursell, H. J. Blom, J. A. Cayuela, M. L. Engvall, N. Lesko, S. Balasu-bramaniam, G. Brandberg, M. Halldin, M. Falkenberg, C. Jakobs, D.Smith, E. Struys, U. Von Dçbeln, C. M. Gustafsson, J. Lundeberg, A.Wedell, Am. J. Hum. Genet. 2011, 89, 507 – 515.

[60] T. Okyuama, Y. Akazawa, J. Uchida, K. Nishino, T. Kumagai, F. Imamura,Oncol. Res. 2011, 19, 399 – 402.

[61] T. Ando, K. Kojima, P. Chahota, A. Kozaki, N. D. Milind, Y. Kitade, Bioorg.Med. Chem. Lett. 2008, 18, 2615 – 2618.

[62] T. Ando, M. Iwata, F. Zulfiqar, T. Miyamoto, M. Nakanishi, Y. Kitade,Bioorg. Med. Chem. 2008, 16, 3809 – 3815.

[63] Y. Kitade, T. Ando, T. Yamaguchi, A. Hori, M. Nakanishi, Y. Ueno, Bioorg.Med. Chem. 2006, 14, 5578 – 5583.

[64] H. Kojima, A. Kozaki, M. Iwata, T. Ando, Y. Kitade, Bioorg. Med. Chem.2008, 16, 6575 – 6579.

[65] Z. W. Chen, K. Matsushita, T. Yamashita, T. A. Fujii, H. Toyama, O. Adachi,H. D. Bellamy, F. S. Mathews, Structure 2002, 10, 837 – 849.

[66] J. L. Whittingham, I. Leal, C. Nguyen, G. Kasinathan, E. Bell, A. F. Jones,C. Berry, A. Benito, J. P. Turkenburg, E. J. Dodson, L. M. Ruiz Perez, A. J.Wilkinson, N. G. Johansson, R. Brun, I. H. Gilbert, D. G. Pacanowska, K. S.Wilson, Structure 2005, 13, 329 – 338.

[67] A. Newby, Biochem. J. 1985, 226, 343 – 344.[68] D. Boison, Trends Pharmacol. Sci. 2006, 27, 652 – 658.[69] D. Boison, Prog. Neurobiol. 2008, 84, 249 – 262.[70] D. Boison, Glia 2012, 60, 1234 – 1243.[71] T. Li, G. Ren, T. Lusardi, A. Wilz, J. Q. Lan, T. Iwasato, S. Itohara, R. P.

Simon, D. Boison, J. Clin. Invest. 2008, 118, 571 – 582.[72] S. McGaraughty, M. Cowart, M. F. Jarvis, R. F. Berman, Curr. Top. Med.

Chem. 2005, 5, 43 – 58.[73] H. H. Lee, J. Moon, S. W. Suh, Proteins Struct. Funct. Genet. 2007, 69,

691 – 694.[74] H. Munier-Lehmann, P. O. Vidalain, F. Tangy, Y. L. Janin, J. Med. Chem.

2013, 56, 3148 – 3167.[75] N. Parveen, K. A. Cornell, Mol. Microbiol. 2011, 79, 7 – 20.[76] A. Lûpez, T. Tarragû, E. Giralt, Expert Opin. Ther. Pat. 2011, 21, 1023 –

1044.[77] L. E. Olson, D. Bedja, S. J. Alvey, A. J. Cardounel, K. L. Gabrielson, R. H.

Reeves, Cancer Res. 2003, 63, 6602 – 6606.[78] M. Carlquist, T. Frejd, M. F. Gorwa-Grauslund, Chem.-Biol. Interact. 2008,

174, 98 – 108.[79] L. Eltis, V. Snieckus (University of British Columbia, Queen’s University

Kingston), US Pat. No. US 20100041631 A1, 2007.[80] C. C. Dobler, K. K. Wong, G. B. Marks, BMC Pulm. Med. 2009, 9, 32.

ChemMedChem 2015, 10, 1209 – 1217 www.chemmedchem.org Ó 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim1216

Full Papers

[81] C. M. M. Lawes, S. Thornley, R. Young, R. Hopkins, R. Marshall, W. C.Chan, G. Jackson, Prim. Care Respir. J. 2012, 21, 35 – 40.

[82] R. P. Young, R. Hopkins, T. E. Eaton, Eur. Respir. Rev. 2009, 18, 222 – 232.[83] P. N. Belloni, L. Garvin, C. P. Mao, I. Bailey-Healy, D. Leaffer in Chest,

2000, p. 235S – 241S.[84] S. V. Stinchcombe, M. Maden, Am. J. Respir. Cell Mol. Biol. 2008, 38, 185 –

191.[85] M. Hind, S. Stinchcombe, Curr. Opin. Investig. Drugs 2009, 10, 1243 –

1250.[86] D. R. Michael, T. G. Ashlin, M. L. Buckley, D. P. Ramji, Curr. Atheroscler.

Rep. 2012, 14, 284 – 293.[87] A. Kratzer, M. Buchebner, T. Pfeifer, T. M. Becker, G. Uray, M. Miyazaki, S.

Miyazaki-Anzai, B. Ebner, P. G. Chandak, R. S. Kadam, E. Calayir, N.Rathke, H. Ahammer, B. Radovic, M. Trauner, G. Hoefler, U. B. Kompella,G. Fauler, M. Levi, S. Levak-Frank, G. M. Kostner, D. Kratky, J. Lipid Res.2009, 50, 312 – 326.

[88] S. B. Joseph, E. McKilligin, L. Pei, M. A. Watson, A. R. Collins, B. A. Laffitte,M. Chen, G. Noh, J. Goodman, G. N. Hagger, J. Tran, T. K. Tippin, X.Wang, A. J. Lusis, W. A. Hsueh, R. E. Law, J. L. Collins, T. M. Willson, P. Ton-tonoz, Proc. Natl. Acad. Sci. USA 2002, 99, 7604 – 7609.

[89] L. Chen, W. H. Haught, B. Yang, T. G. P. Saldeen, S. Parathasarathy, J. L.Mehta, J. Am. Coll. Cardiol. 1997, 30, 569 – 575.

[90] T. J. Bench, A. Jeremias, D. L. Brown, Pharmacol. Res. 2011, 64, 561 – 566.[91] E. Dejonckheere, R. E. Vandenbroucke, C. Libert, Cytokine Growth Factor

Rev. 2011, 22, 73 – 81.[92] S. Lenglet, F. Mach, F. Montecucco, Mediators Inflammation 2013, 2013,

659282.[93] S. Bereswill, M. MuÇoz, A. Fischer, R. Plickert, L.-M. Haag, B. Otto, A. A.

Kìhl, C. Loddenkemper, U. B. Gçbel, M. M. Heimesaat, PLoS One 2010,5, e15099.

[94] U. Svajger, M. Jeras, Int. Rev. Immunol. 2012, 31, 202 – 222.[95] H. Li, N. Xia, U. Fçrstermann, Nitric Oxide 2012, 26, 102 – 110.[96] J. Chen, L. Zhang, Y. Zhang, H. Zhang, J. Du, J. Ding, Y. Guo, H. Jiang, X.

Shen, BMC Microbiol. 2009, 9, 91.[97] L. Zhang, W. Liu, T. Hu, L. Du, C. Luo, K. Chen, X. Shen, H. Jiang, J. Biol.

Chem. 2008, 283, 5370 – 5379.[98] J. C. Brown, G. Huang, V. Haley-Zitlin, X. Jiang, Appl. Environ. Microbiol.

2009, 75, 848 – 852.

Received: March 16, 2015

Revised: April 29, 2015

Published online on June 1, 2015

ChemMedChem 2015, 10, 1209 – 1217 www.chemmedchem.org Ó 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim1217

Full Papers