virtualtoxlab – in silico prediction of the endocrine-disrupting

7
ENDOCRINE DISRUPTORS: BASIC RESEARCH AND METHODS 322 CHIMIA 2008, 62, No. 5 Chimia 62 (2008) 322–328 © Schweizerische Chemische Gesellschaft ISSN 0009–4293 VirtualToxLab – in silico Prediction of the Endocrine-Disrupting Potential of Drugs and Chemicals Angelo Vedani*, Morena Spreafico, Ourania Peristera, Max Dobler, and Martin Smiesko Abstract: In the last decade, we have developed and validated an in silico concept based on multidimensional QSAR (mQSAR) for the prediction of the toxic potential of drugs and environmental chemicals. Presently, the VirtualToxLab includes eleven so-called virtual test kits for estrogen (α/β), androgen, thyroid (α/β), glucocorticoid, aryl hydrocarbon, mineralocorticoid and peroxisome proliferator-activated receptor γ as well as for the enzymes cytochrome P450 3A4 and 2A13. The surrogates have been tested against a total of 824 compounds and are able to predict the binding affinity close to the experimental uncertainty with only six of the 194 test compounds giving calculated results more than a factor of 10 off the experimental binding affinity and the maximal individual devia- tion not exceeding a factor of 15. These results suggest that our approach is suited for the in silico identification of endocrine-disrupting effects triggered by drugs and environmental chemicals. Most recently, the technology has been made available through the Internet for academic laboratories, hospitals and environmental organizations. Keywords: in silico prediction of the endocrine-disrupting potential · Molecular modeling · Multidimensional QSAR · VirtualToxLab lymphocytes as well as neuronal degenera- tion as a response to stress, the peroxisome proliferator-activated receptor, which is associated with hepatocarcinogenesis in rodents, and the aryl hydrocarbon receptor which is involved in a whole range of tox- ic effects. [1] Harmful effects of drugs and chemicals can often be associated with their binding to other than their primary target macromolecules involved in biosynthesis, signal transduction, transport, storage, and metabolism. [28] Toxicity testing mandatory by inter- national regulations for drug development and chemical safety is still associated with stressful animal tests. While many in vitro approaches have been devised for tar- geting the various aspects of toxic effects (e.g. endocrine disruption), they require a chemical or drug molecule to be physically present (i.e. synthesized) before testing, are time consuming, and the results are often associated with large standard errors, par- ticularly in the low-to-mid activity range (mM–µM). In contrast to in vivo and in vitro assays, computational approaches can be applied to hypothetical substances as their three- dimensional (3D) structure can readily be generated in silico. Nowadays computer power permits larger batches of compounds (e.g. parts of corporate or public databases) to be scanned for their toxic potential in moderate time spans. Toxicity-modeling algorithms are typically based on quan- titative structureactivity relationships, neuronal networks, artificial intelligence or rule-based expert systems. The develop- ment of the VirtualToxLab technology has been previously been described. [9–10] In this account, we focus on the validation status and its implementation on the Internet. Methods The VirtualToxLab technology is based on a mixed-model approach: The binding mode of a drug or chemical of interest to- wards a target protein (enzyme, receptor) is identified using flexible docking to the 3D structure of the bioregulatory macromole- cule. Its binding affinity and the associat- ed toxic (or endocrine-disrupting) potential is then quantified using multidimensional QSAR (mQSAR). Most underlying tech- nologies were developed at our laboratory and are published in detail. [1115] Quasar — a receptor-modeling concept developed at the Biographics Laboratory 3R — is based on 6D-QSAR and explicitly allows for the simulation of induced fit. [12,15] Quasar generates a family of quasi-atomis- tic receptor surrogates that are optimized *Correspondence: Prof. Dr. A. Vedani Biographics Laboratory 3R Friedensgasse 35 CH-4056 Basel Tel.: +41 61 261 42 56 E-Mail: [email protected] Introduction Toxic agents, particularly those that exert their actions with a great deal of specificity, sometimes act via receptors to which they bind with high affinity. This phenomenon is referred to as receptor-mediated toxicity. Examples of soluble intracellular receptors, which are important in mediating toxic re- sponses, include the glucocorticoid recep- tor which is also involved in mediating tox- icity-associated effects such as apoptosis of doi:10.2533/chimia.2008.322

Upload: nguyenkhanh

Post on 10-Feb-2017

224 views

Category:

Documents


3 download

TRANSCRIPT

Page 1: VirtualToxLab – in silico Prediction of the Endocrine-Disrupting

ENDOCRINE DISRUPTORS BASIC RESEARCH AND METHODS 322CHIMIA 2008 62 No 5

Chimia 62 (2008) 322ndash328copy Schweizerische Chemische Gesellschaft

ISSN 0009ndash4293

VirtualToxLab ndash in silico Predictionof the Endocrine-Disrupting Potential ofDrugs and Chemicals

Angelo Vedani Morena Spreafico Ourania Peristera Max Dobler and Martin Smiesko

Abstract In the last decade we have developed and validated an in silico concept based on multidimensionalQSAR (mQSAR) for the prediction of the toxic potential of drugs and environmental chemicals Presently theVirtualToxLab includes eleven so-called virtual test kits for estrogen (αβ) androgen thyroid (αβ) glucocorticoidaryl hydrocarbon mineralocorticoid and peroxisome proliferator-activated receptor γ as well as for the enzymescytochrome P450 3A4 and 2A13 The surrogates have been tested against a total of 824 compounds and are ableto predict the binding affinity close to the experimental uncertainty with only six of the 194 test compounds givingcalculated results more than a factor of 10 off the experimental binding affinity and the maximal individual devia-tion not exceeding a factor of 15 These results suggest that our approach is suited for the in silico identification ofendocrine-disrupting effects triggered by drugs and environmental chemicals Most recently the technology hasbeen made available through the Internet for academic laboratories hospitals and environmental organizations

Keywords in silico prediction of the endocrine-disrupting potential middot Molecular modeling middot MultidimensionalQSAR middot VirtualToxLab

lymphocytes as well as neuronal degenera-tion as a response to stress the peroxisomeproliferator-activated receptor which isassociated with hepatocarcinogenesis inrodents and the aryl hydrocarbon receptorwhich is involved in a whole range of tox-ic effects[1] Harmful effects of drugs andchemicals can often be associated with theirbinding to other than their primary target minusmacromolecules involved in biosynthesissignal transduction transport storage andmetabolism[2minus8]

Toxicity testing minus mandatory by inter-national regulations for drug developmentand chemical safety minus is still associatedwith stressful animal tests While many invitro approaches have been devised for tar-geting the various aspects of toxic effects(eg endocrine disruption) they require achemical or drug molecule to be physicallypresent (ie synthesized) before testing aretime consuming and the results are oftenassociated with large standard errors par-ticularly in the low-to-mid activity range(mMndashmicroM)

In contrast to in vivo and in vitro assayscomputational approaches can be appliedto hypothetical substances as their three-dimensional (3D) structure can readily begenerated in silico Nowadays computerpower permits larger batches of compounds(eg parts of corporate or public databases)

to be scanned for their toxic potential inmoderate time spans Toxicity-modelingalgorithms are typically based on quan-titative structureminusactivity relationshipsneuronal networks artificial intelligenceor rule-based expert systems The develop-ment of the VirtualToxLab technology hasbeen previously been described[9ndash10] In thisaccount we focus on the validation statusand its implementation on the Internet

Methods

The VirtualToxLab technology is basedon a mixed-model approach The bindingmode of a drug or chemical of interest to-wards a target protein (enzyme receptor) isidentified using flexible docking to the 3Dstructure of the bioregulatory macromole-cule Its binding affinity minus and the associat-ed toxic (or endocrine-disrupting) potentialminus is then quantified using multidimensionalQSAR (mQSAR) Most underlying tech-nologies were developed at our laboratoryand are published in detail[11minus15]

Quasar mdash a receptor-modeling conceptdeveloped at the Biographics Laboratory3R mdash is based on 6D-QSAR and explicitlyallows for the simulation of induced fit[1215]

Quasar generates a family of quasi-atomis-tic receptor surrogates that are optimized

Correspondence Prof Dr A VedaniBiographics Laboratory 3RFriedensgasse 35CH-4056 BaselTel +41 61 261 42 56E-Mail angelobiografch

Introduction

Toxic agents particularly those that exerttheir actions with a great deal of specificitysometimes act via receptors to which theybind with high affinity This phenomenonis referred to as receptor-mediated toxicityExamples of soluble intracellular receptorswhich are important in mediating toxic re-sponses include the glucocorticoid recep-tor which is also involved in mediating tox-icity-associated effects such as apoptosis of

doi102533chimia2008322

ENDOCRINE DISRUPTORS BASIC RESEARCH AND METHODS 323CHIMIA 2008 62 No 5

by means of a genetic algorithm The hy-pothetical receptor site is characterized by athree-dimensional surface which surroundsthe ligand molecules at van der Waals dis-tance and which is populated with atomisticproperties mapped onto it The topology ofthis surface mimics the three-dimensionalshape of the binding site the mapped prop-erties representother informationof interestsuch as hydrophobicity electrostatic poten-tial and hydrogen-bonding propensity Thefourth dimension refers to the possibility ofrepresenting each ligand molecule as an en-semble of conformations orientations andprotonation states thereby reducing the bi-as in identifying the bioactive conformationand orientation (rarr 4D-QSAR) Within thisensemble the contribution of an individualentity to the total energy is determined bya normalized Boltzmann weight As mani-festation and magnitude of induced fit mayvary for different molecules binding to atarget protein the fifth dimension in Qua-sar allows for the simultaneous evaluationof up to six different induced-fit protocols(rarr 5D-QSAR) The most recent extensionof the Quasar concept to six dimensions(rarr 6D-QSAR) allows the simultaneousconsideration of different solvation mod-els This can either be achieved explicitlywhere parts of the surface area are mappedwith solvent properties whereby positionand size are optimized by the genetic al-gorithm or implicitly Here the solvationterms (ligand desolvation and solvent strip-ping) are independently scaled for each

different model within the surrogate fam-ily reflecting varying solvent accessibilityof the binding pocket Like for the fourthand fifth dimension a modest lsquoevolutionarypressurersquo is applied to achieve convergenceIn the Quasar concept the binding energyis calculated as followsEbinding = Eligandndashreceptor minus Eligand desolvation

minus TΔS ndash Eligand strain ndash Einduced fitwhereEligandndashreceptor = Eelectrostatic + Evan der Waals

+ Ehydrogen bonding+ E polarization

The contributions of the individual en-tities within a 4D ensemble (conformerorientiomer protomer andor tautomer)are normalized to unity using a BoltzmanncriterionEbinding total = sumEbinding individual middot

exp (ndashwi middotEbinding individualEbinding individual maximal)

Raptor an alternative technology de-veloped by our laboratory[13] explicitly andanisotropically allows for induced fit by adual-shell representation of the receptorsurrogate mapped with physicochemi-cal properties (hydrophobic character andhydrogen-bonding propensity) onto it InRaptor induced fit is not limited to stericaspects but includes the variation of thephysico-chemical fields along with it Theunderlying scoring function for evaluat-ing ligand-receptor interactions includesdirectional terms for hydrogen bonding

hydrophobicity and thereby treats solva-tion effects implicitly This makes the ap-proach independent from a partial-chargemodel and as a consequence allows tosmoothly model ligand molecules bindingto the receptor with different net chargesIn Raptor the binding energy is determinedas followsEbinding = Eligandndashreceptor ndash TΔS ndash Einduced fit

whereEligandndashreceptor = f (Ehydrogen bonding (shell 1)

+ E hydrophobic (shell 1))+ (10 ndash f) sdot(Ehydrogen bonding (shell 2)+ E hydrophobic (shell 2))

and f is the interpolation weight betweenthe two shells[13]

Depending upon whether or not the 3Dstructure of the target protein is availableat atomic resolution two fundamentallydifferent concepts are used to identify thisbioactive conformation flexible dockingfor systems with known 3D structure andpharmacophore hypothesis builder other-wise For ten of the proteins included inthe VirtualToxLab (estrogen αβ androgenthyroid αβ glucocorticoid mineralocorti-coid peroxisome proliferator-activated re-ceptor γ cytochrome P450 3A4 and 2A13)the crystal structures are available for thearyl hydrocarbon receptor it is not

Flexible docking aims at identifying allpotential binding modes (orientations con-formations) of a small molecule within thebinding pocket of a protein The underlyingprotocol should account for two aspects ofligandminusprotein binding i) the simulation ofinduced fit mdash allowing the protein to adaptits shape to the different orientations andconformations of the small molecule duringthe search procedure mdash and ii) the consid-eration of solvent effects (typically water)In our approach (software Yeti[1115]) thesampling is conducted based on a Monte-CarloMetropolis protocol which allows toinitially considering apparently less favor-able poses or conformations Such calcula-tions are quite computationally expensiveas for an lsquoexhaustive searchrsquo typically5000ndash10000 conformationsorientationsare generated and 500ndash1000 are fully mini-mized (Fig 1) The directional force field minuswhich is incorporated in all of our concepts(Yeti Quasar Raptor Symposar) minus wouldseem to be of utmost importance for quan-tifying hydrogen bonds and metal-ligandinteractions (Fig 2) appropriately

For the aryl hydrocarbon receptor(AhR) for which no experimental structureis available the alignment was performedbased on the molecular skeletons[1215]

Within the VirtualToxLab ligands bindingto the AhR are docked using the Symposartechnology[14] Symposar allows both 3D

Fig 1 Flexible docking minus flowchart of module AutoDock as implementedin Yeti

ENDOCRINE DISRUPTORS BASIC RESEARCH AND METHODS 324CHIMIA 2008 62 No 5

data sets (a single orientation and confor-mation per ligand molecule) and 4D datasets to be generated It stores the grid pro-duced and optimized by the ligands of thetraining set which can then be applied toany new test and prediction set After theflexible mapping the conformation of eachmolecule is minimized within the variousgrid fields

The VirtualToxLab

Details of model validation are pub-lished (cf [10] and references cited thereinor [16]) Except for the aryl hydrocarbonreceptor where no experimental structureof the protein is available all systems are

based on docking studies using the hu-man protein As we used biological datameasured in a single laboratory for mostsystems the structural diversity is not yetsufficient to consider all models validated(PPARγ TRαβ ERβ GR MR) For somesystems the activity range would not seemto be sufficiently large (PPARγ ERβ) Thecurrent validation status is summarizedin Table 1 and Fig 3 the appearance ofQuasar and Raptor models is depicted inFig 4

Internet Access Portal

Tofacilitateaccess toour technologywehave developed an InternetAccess Protocol

(IAP) immediately available for academiclaboratories (Fig 5) While the technologyitself is freely available we have to requesta modest fee for acquiring and maintainingthe underlying hardware and third-partysoftware as well as for customer supportWhile the VirtualToxLab runs fully auto-mated the 3D structure of the compound ofinterest must be generated and provided bythe end user Several freely available pro-grams exist to accomplish this task it isnonetheless vital to check the correctnessof the structure (constitution connectivitystereochemistry)

In a first step (cf Fig 6) the struc-ture is checked for correctness and thetautomeric and protonation state minus atphysiological pH minus is identified (softwareEpik[22]) Then the compound is subject-ed to a conformational-search protocol(to identify the global minimum confor-mation) in aqueous solution using theAMBER force field[23] as implementedin the MacroModel software[2425] NextMNDOESP[26] or CM1[27] atomic partialcharges are computed In the main stepthe compound is automatically docked tothe target protein(s) thereby allowing theinduced fit and sampling of all energeti-cally feasible binding modes (YetiAuto-Dock[1115]rarr 4D data set) Thereafterphysicochemical properties (solvationenergy internal strain in the bound modeentropic contribution to ligand binding)are determined Finally the compoundrsquostoxic potential is estimated by calculatingits binding affinity towards the macro-molecular target (Quasar[1215] and Rap-tor[13]) Based on the information avail-able on the data set used to evaluate themodel (training and test compounds) thepotential toxicological class (from 0 = be-nign to 4 = extremely toxic) is assignedThe results are then made available to theuser via the IAP (Fig 7) both as absolutenumbers (binding affinity towards the tar-get protein) and the estimated toxic poten-tial (color coded) At this point all userdata (3D structure of the compound ofinterest) and all files associated with theprocedures described above (compoundcoordinates models physicochemicalproperties protocol files) are irreversiblydiscarded (for security issues cf below)The user log file may be exported (to a textfile) at any time

Test Compounds

As an example we tested six com-pounds bisphenol-A a polymer additive3453rsquo4rsquo5rsquo-hexabromodiphenylether aflame-retarding chemical erythrosine afood dye methylparaben a preservativeand fungicide paracetamol an NSAIDand methyltrienolone a very potent ste-

Fig 2 Directional force field (as implemented in Yeti Quasar Raptor andSymposar)

Table 1 VirtualToxLab Summary of the currently available test kits (enzymereceptor models)

System compoundstraining+test=totalcompound classes

q2 rmstrain-ing

maxtrain-ing

p2 rmstest

maxtest

fnfpa

Ref

Aryl hydrocarbon 105+35=140 eight 0824 18 102 0769 23 135 02 [10]

Androgen 88+26=114 eight 0858 17 78 0792 16 139 11 [17]

Estrogen αEstrogen αb

Estrogen βb 80+26=106 six

80+23=103 five

089507870785

200911

862548

089206820827

291108

953824

000000

[15]ndash[10]

Glucocorticoid 82+28=110 four 0745 12 59 0650 22 55 00 [18]

Mineralocorticoid 40+12=52 three 0798 16 56 0701 25 95 00 [16]

PPARγ 75+20=95 two 0832 14 62 0723 14 39 00 [19]

Thyroid αThyroid β 64+18= 82 four

09190909

1820

4377

08140796

2527

10088

0110

[20][20]

CYP3A4 38+10=48 eighteen 0825 27 70 0659 38 71 00 [21]

CYP2A13 18+6=24 six 0900 06 15 0830 03 07 00 [16]

q2 = cross-validated r2 p2 = predictive r2 the rms and maximal deviation from the experimentalbinding affinity is given as factor in Ki or IC50afn = false-negative fp = false-positive compounds a factor 100 or more off the experimentalvaluebdifferent compounds (diphenolic azoles) than for the 80+26 model above

ENDOCRINE DISRUPTORS BASIC RESEARCH AND METHODS 325CHIMIA 2008 62 No 5

6

Fig 3 Comparison of experimental and calculated binding affinities of the models underlying the VirtualToxLab From left to right and top to bottomAhR AR ERα ERβ GR MR PPARγ TRα TRβ 3A4 (simulation without the eight threshold ligands cf ref [21]) 2A13 The ligands of the training setare shown as open circles those of the test set as filled circles The error bars correspond to the variation of the 200ndash500 models comprising the modelfamilies Dashed lines are drawn at plusmn 10 log unit from the experimental value

ENDOCRINE DISRUPTORS BASIC RESEARCH AND METHODS 326CHIMIA 2008 62 No 5

Fig 6 Appearance of the Internet Access Portalto the VirtualToxLab

Fig 4 Quasar model (6D-QSAR left) and Raptorsurrogate (dual-shell 5D-QSAR right) Thebound ligand is shown as stick representation(atom coloring gray = carbon white = hydrogenred = oxygen blue = nitrogen) The quasi-atomistic properties of the receptor are mappedonto the surface(s) blue = positively chargedsalt bridge red = negatively charged salt bridgebrownish colors = hydrophobic properties pink= hydrogen-bond flip flop The Quasar modelrepresents the Aryl hydrocarbon receptor[10] theRaptor model depicts the thyroid receptor β[17]

Fig 5 Flow-chart of the VirtualToxLab

ENDOCRINE DISRUPTORS BASIC RESEARCH AND METHODS 327CHIMIA 2008 62 No 5

roid The three-dimensional structures ofall compounds were generated using Bio[28]

and processed with the standard protocol ofthe VirtualToxLab The resulting toxic po-tentials are given in Table 2

Outlook

Up to date the VirtualToxLab concepthas only produced a few false-positive re-sults At the current level however false-negative predictions are still obtained asa compound of interest cannot be testedagainst all potential receptors it may bindto in vivo (some macromolecular targetswill remain unknown for others no ex-perimental structure exists or too fewaffinity data are available to establish aQSAR) We therefore plan to extend thecurrent concept by implementing a set ofvirtual filters which can recognize be-nign compounds Among others criteriainclude the molecular weight drug-likeproperties (rarr Lipinskirsquos rule of five) andthe presenceabsence of characteristicstructural motifs We are also consideringa combination of our VirtualToxLab withLigandScout a fast screening technologybased on protein-ligand structures[29] Anynew models and results are continuouslyupdated on our website[30] from wherethe VirtualToxLab documentation can bedownloaded

Effective April 1 2008 the technologyhas been made available through the Internetfor academic laboratories hospitals and en-vironmental organizations Interested partiesshould request the licensing agreement Af-ter signing the document they will receive apersonal accountpassword and the URL toaccess the VirtualToxLab on-line

In the VirtualToxLab the data are trans-ferred using the SSH (Secure Shell) proto-col After completion of the task all dataminus compound coordinates models physico-chemical parameters minus are irreversibly dis-carded The entries in the user log file (con-taining the calculated binding affinities)may be deleted by the user at any time

Disclaimer It is understood the Virtual-ToxLab may only generate reliable resultsfor compound classes used to train the mod-el of the respective target protein (which aregiven in the pertinent publications[9ndash21])The standard deviation of a given result isalso a safe quality indicator

AcknowledgementThe authors wish to express their gratitude

to Markus A Lill Assistant Professor at theDepartment of Medicinal Chemistry andMolecular Pharmacology Purdue University forhis contribution to the VirtualToxLab technology(Symposar and Raptor software) and the manyfruitfuldiscussionsonthegenerationandvalidationof receptor-surface models and mQSAR

Fig 7 Main dialog box of the IAP to the VirtualToxLab

Table 2 VirtualToxLab Toxic potential profile of six selected compounds

Bisphenol A Hexabromo-diphenylether

Erythrosine

Aryl hydrocarbonreceptor

(ndash) +++ ++++

Androgen receptor +++ +++ +++

Estrogen receptor α ++ ++ not determinable

Estrogen receptor β +++ +++ not determinable

Glucocorticoid receptor ++++ ++++ +++

Mineralocorticoidreceptor

(+) +++ not determinable

PPARγ ++ ++ (++++)

Thyroid receptor α ndash ndash (++)

Thydorid receptor β (+++) (+++) (+++++)

CYP450 3A4 ++ +++ (++)

CYP450 2A13 +++ +++ (+++)

Paracetamol Methylparaben Methyltrienolone

Aryl hydrocarbonreceptor

(+) ndash +

Androgen receptor +++ ++ ++++

Estrogen receptor α ++ ndash ++++

Estrogen receptor β +++ +++ +++

Glucocorticoid receptor +++ ++ +++

Mineralocorticoidreceptor

(+) ndash +++

PPARγ ++ + ++

Thyroid receptor α ndash ndash ndash

Thydorid receptor β (+++) (+) +++

CYP450 3A4 + ndash +++

CYP450 2A13 +++ +++ +++

Toxic potential in silico ndash none Ki or IC50 gt 100mM

+ very low 100mM gt Ki or IC50 gt 1mM

++ low 1 mM gt Ki or IC50 gt 10 microM

+++ medium 10 microM gt Ki or IC50 gt 100 nM

++++ high 100 nM gt Ki or IC50 gt 1nM

+++++ very high Ki or IC50 lt 1nM

Values in brackets indicate a high standard deviation and should be interpreted with caution ldquonotdeterminablerdquo indicates very high esdrsquos

Bisphenol A Hexabr Erythrosi

ENDOCRINE DISRUPTORS BASIC RESEARCH AND METHODS 328CHIMIA 2008 62 No 5

This research has been made possiblethrough grants by the Swiss National ScienceFoundation the Margaret and Francis-Fleitmann Foundation LucerneSwitzerlandthe Doerenkamp-Zbinden Foundation ZuumlrichSwitzerland and the Jacques en Dolly GazanFoundation ZugSwitzerland which are allgratefully acknowledged

Received March 22 2008

[1] J A Gustaffson Toxicol Lett 1995 135465

[2] B Rihova Adv Drug Deliv Rev 199829 273

[3] B Fischer Andrologia 2000 32 279[4] E V Hestermann J J Stegemann M E

Hahn Toxicol Appl Pharmacol 2000168 160

[5] K Lukasink A Pitkanen J Neurochem2000 74 2445

[6] D L Rymer T A Good J Biol Chem2001 276 2523

[7] A J Hampson M Grimaldi J Neurosci2002 22 257

[8] J D Oliver R A Roberts PharmacolToxicol 2002 91 1

[9] A Vedani M Dobler M A LillPharmacol Toxicol 2006 99 195

[10] A Vedani M Dobler M Spreafico OPeristera M Smiesko ALTEX 24 2007153

[11] A Vedani D W Huhta J Am ChemSoc 1990 112 4759

[12] A Vedani M Dobler J Med Chem2002 45 2139

[13] M A Lill A Vedani M Dobler J MedChem 2004 47 6174

[14] M A Lill A Vedani J Chem Inf Model2006 46 2135

[15] A Vedani M Dobler M A Lill J MedChem 2005 48 3700

[16] httpwwwbiografchindexphpid=projects

[17] M A Lill F Winiger A Vedani B ErnstJ Med Chem 2005 48 5666

[18] M Spreafico M Smiesko M A LillB Ernst A Vedani J Med Chemsubmitted

[19] A Vedani A-V Decloux M SpreaficoB Ernst Toxicol Lett 2007 173 17

[20] A Vedani M Zumstein M A Lill BErnst ChemMedChem 2007 2 78

[21] M A Lill M Dobler A VedaniChemMedChem 2006 1 73

[22] Epik version 15 Schroumldinger LLC NewYork NY 2007

[23] S J Weiner P A Kollmann D A CaseU C Singh C Ghio G Alagona SProfeta Jr P Weiner J Am Chem Soc1984 106 765

[24] MacroModel version 95 SchroumldingerLLC New York NY 2007

[25] F Mohamadi N G J Richards WC Guida R Liskamp M Lipton CCaufield G Chang T Hendrickson WC Still J Comput Chem 1990 11 440

[26] J J P Stewart J Comput-Aided MolecDesign 1990 4 1

[27] C J Cramer D G Truhlar J Comput-Aided Molec Design 1992 6 629

[28] Bio version 30 Biographics Laboratory3R BaselSwitzerland 2007

[29] G Wolber T Langer J Chem Inf Model2005 45 160

[30] httpwwwbiografch

Page 2: VirtualToxLab – in silico Prediction of the Endocrine-Disrupting

ENDOCRINE DISRUPTORS BASIC RESEARCH AND METHODS 323CHIMIA 2008 62 No 5

by means of a genetic algorithm The hy-pothetical receptor site is characterized by athree-dimensional surface which surroundsthe ligand molecules at van der Waals dis-tance and which is populated with atomisticproperties mapped onto it The topology ofthis surface mimics the three-dimensionalshape of the binding site the mapped prop-erties representother informationof interestsuch as hydrophobicity electrostatic poten-tial and hydrogen-bonding propensity Thefourth dimension refers to the possibility ofrepresenting each ligand molecule as an en-semble of conformations orientations andprotonation states thereby reducing the bi-as in identifying the bioactive conformationand orientation (rarr 4D-QSAR) Within thisensemble the contribution of an individualentity to the total energy is determined bya normalized Boltzmann weight As mani-festation and magnitude of induced fit mayvary for different molecules binding to atarget protein the fifth dimension in Qua-sar allows for the simultaneous evaluationof up to six different induced-fit protocols(rarr 5D-QSAR) The most recent extensionof the Quasar concept to six dimensions(rarr 6D-QSAR) allows the simultaneousconsideration of different solvation mod-els This can either be achieved explicitlywhere parts of the surface area are mappedwith solvent properties whereby positionand size are optimized by the genetic al-gorithm or implicitly Here the solvationterms (ligand desolvation and solvent strip-ping) are independently scaled for each

different model within the surrogate fam-ily reflecting varying solvent accessibilityof the binding pocket Like for the fourthand fifth dimension a modest lsquoevolutionarypressurersquo is applied to achieve convergenceIn the Quasar concept the binding energyis calculated as followsEbinding = Eligandndashreceptor minus Eligand desolvation

minus TΔS ndash Eligand strain ndash Einduced fitwhereEligandndashreceptor = Eelectrostatic + Evan der Waals

+ Ehydrogen bonding+ E polarization

The contributions of the individual en-tities within a 4D ensemble (conformerorientiomer protomer andor tautomer)are normalized to unity using a BoltzmanncriterionEbinding total = sumEbinding individual middot

exp (ndashwi middotEbinding individualEbinding individual maximal)

Raptor an alternative technology de-veloped by our laboratory[13] explicitly andanisotropically allows for induced fit by adual-shell representation of the receptorsurrogate mapped with physicochemi-cal properties (hydrophobic character andhydrogen-bonding propensity) onto it InRaptor induced fit is not limited to stericaspects but includes the variation of thephysico-chemical fields along with it Theunderlying scoring function for evaluat-ing ligand-receptor interactions includesdirectional terms for hydrogen bonding

hydrophobicity and thereby treats solva-tion effects implicitly This makes the ap-proach independent from a partial-chargemodel and as a consequence allows tosmoothly model ligand molecules bindingto the receptor with different net chargesIn Raptor the binding energy is determinedas followsEbinding = Eligandndashreceptor ndash TΔS ndash Einduced fit

whereEligandndashreceptor = f (Ehydrogen bonding (shell 1)

+ E hydrophobic (shell 1))+ (10 ndash f) sdot(Ehydrogen bonding (shell 2)+ E hydrophobic (shell 2))

and f is the interpolation weight betweenthe two shells[13]

Depending upon whether or not the 3Dstructure of the target protein is availableat atomic resolution two fundamentallydifferent concepts are used to identify thisbioactive conformation flexible dockingfor systems with known 3D structure andpharmacophore hypothesis builder other-wise For ten of the proteins included inthe VirtualToxLab (estrogen αβ androgenthyroid αβ glucocorticoid mineralocorti-coid peroxisome proliferator-activated re-ceptor γ cytochrome P450 3A4 and 2A13)the crystal structures are available for thearyl hydrocarbon receptor it is not

Flexible docking aims at identifying allpotential binding modes (orientations con-formations) of a small molecule within thebinding pocket of a protein The underlyingprotocol should account for two aspects ofligandminusprotein binding i) the simulation ofinduced fit mdash allowing the protein to adaptits shape to the different orientations andconformations of the small molecule duringthe search procedure mdash and ii) the consid-eration of solvent effects (typically water)In our approach (software Yeti[1115]) thesampling is conducted based on a Monte-CarloMetropolis protocol which allows toinitially considering apparently less favor-able poses or conformations Such calcula-tions are quite computationally expensiveas for an lsquoexhaustive searchrsquo typically5000ndash10000 conformationsorientationsare generated and 500ndash1000 are fully mini-mized (Fig 1) The directional force field minuswhich is incorporated in all of our concepts(Yeti Quasar Raptor Symposar) minus wouldseem to be of utmost importance for quan-tifying hydrogen bonds and metal-ligandinteractions (Fig 2) appropriately

For the aryl hydrocarbon receptor(AhR) for which no experimental structureis available the alignment was performedbased on the molecular skeletons[1215]

Within the VirtualToxLab ligands bindingto the AhR are docked using the Symposartechnology[14] Symposar allows both 3D

Fig 1 Flexible docking minus flowchart of module AutoDock as implementedin Yeti

ENDOCRINE DISRUPTORS BASIC RESEARCH AND METHODS 324CHIMIA 2008 62 No 5

data sets (a single orientation and confor-mation per ligand molecule) and 4D datasets to be generated It stores the grid pro-duced and optimized by the ligands of thetraining set which can then be applied toany new test and prediction set After theflexible mapping the conformation of eachmolecule is minimized within the variousgrid fields

The VirtualToxLab

Details of model validation are pub-lished (cf [10] and references cited thereinor [16]) Except for the aryl hydrocarbonreceptor where no experimental structureof the protein is available all systems are

based on docking studies using the hu-man protein As we used biological datameasured in a single laboratory for mostsystems the structural diversity is not yetsufficient to consider all models validated(PPARγ TRαβ ERβ GR MR) For somesystems the activity range would not seemto be sufficiently large (PPARγ ERβ) Thecurrent validation status is summarizedin Table 1 and Fig 3 the appearance ofQuasar and Raptor models is depicted inFig 4

Internet Access Portal

Tofacilitateaccess toour technologywehave developed an InternetAccess Protocol

(IAP) immediately available for academiclaboratories (Fig 5) While the technologyitself is freely available we have to requesta modest fee for acquiring and maintainingthe underlying hardware and third-partysoftware as well as for customer supportWhile the VirtualToxLab runs fully auto-mated the 3D structure of the compound ofinterest must be generated and provided bythe end user Several freely available pro-grams exist to accomplish this task it isnonetheless vital to check the correctnessof the structure (constitution connectivitystereochemistry)

In a first step (cf Fig 6) the struc-ture is checked for correctness and thetautomeric and protonation state minus atphysiological pH minus is identified (softwareEpik[22]) Then the compound is subject-ed to a conformational-search protocol(to identify the global minimum confor-mation) in aqueous solution using theAMBER force field[23] as implementedin the MacroModel software[2425] NextMNDOESP[26] or CM1[27] atomic partialcharges are computed In the main stepthe compound is automatically docked tothe target protein(s) thereby allowing theinduced fit and sampling of all energeti-cally feasible binding modes (YetiAuto-Dock[1115]rarr 4D data set) Thereafterphysicochemical properties (solvationenergy internal strain in the bound modeentropic contribution to ligand binding)are determined Finally the compoundrsquostoxic potential is estimated by calculatingits binding affinity towards the macro-molecular target (Quasar[1215] and Rap-tor[13]) Based on the information avail-able on the data set used to evaluate themodel (training and test compounds) thepotential toxicological class (from 0 = be-nign to 4 = extremely toxic) is assignedThe results are then made available to theuser via the IAP (Fig 7) both as absolutenumbers (binding affinity towards the tar-get protein) and the estimated toxic poten-tial (color coded) At this point all userdata (3D structure of the compound ofinterest) and all files associated with theprocedures described above (compoundcoordinates models physicochemicalproperties protocol files) are irreversiblydiscarded (for security issues cf below)The user log file may be exported (to a textfile) at any time

Test Compounds

As an example we tested six com-pounds bisphenol-A a polymer additive3453rsquo4rsquo5rsquo-hexabromodiphenylether aflame-retarding chemical erythrosine afood dye methylparaben a preservativeand fungicide paracetamol an NSAIDand methyltrienolone a very potent ste-

Fig 2 Directional force field (as implemented in Yeti Quasar Raptor andSymposar)

Table 1 VirtualToxLab Summary of the currently available test kits (enzymereceptor models)

System compoundstraining+test=totalcompound classes

q2 rmstrain-ing

maxtrain-ing

p2 rmstest

maxtest

fnfpa

Ref

Aryl hydrocarbon 105+35=140 eight 0824 18 102 0769 23 135 02 [10]

Androgen 88+26=114 eight 0858 17 78 0792 16 139 11 [17]

Estrogen αEstrogen αb

Estrogen βb 80+26=106 six

80+23=103 five

089507870785

200911

862548

089206820827

291108

953824

000000

[15]ndash[10]

Glucocorticoid 82+28=110 four 0745 12 59 0650 22 55 00 [18]

Mineralocorticoid 40+12=52 three 0798 16 56 0701 25 95 00 [16]

PPARγ 75+20=95 two 0832 14 62 0723 14 39 00 [19]

Thyroid αThyroid β 64+18= 82 four

09190909

1820

4377

08140796

2527

10088

0110

[20][20]

CYP3A4 38+10=48 eighteen 0825 27 70 0659 38 71 00 [21]

CYP2A13 18+6=24 six 0900 06 15 0830 03 07 00 [16]

q2 = cross-validated r2 p2 = predictive r2 the rms and maximal deviation from the experimentalbinding affinity is given as factor in Ki or IC50afn = false-negative fp = false-positive compounds a factor 100 or more off the experimentalvaluebdifferent compounds (diphenolic azoles) than for the 80+26 model above

ENDOCRINE DISRUPTORS BASIC RESEARCH AND METHODS 325CHIMIA 2008 62 No 5

6

Fig 3 Comparison of experimental and calculated binding affinities of the models underlying the VirtualToxLab From left to right and top to bottomAhR AR ERα ERβ GR MR PPARγ TRα TRβ 3A4 (simulation without the eight threshold ligands cf ref [21]) 2A13 The ligands of the training setare shown as open circles those of the test set as filled circles The error bars correspond to the variation of the 200ndash500 models comprising the modelfamilies Dashed lines are drawn at plusmn 10 log unit from the experimental value

ENDOCRINE DISRUPTORS BASIC RESEARCH AND METHODS 326CHIMIA 2008 62 No 5

Fig 6 Appearance of the Internet Access Portalto the VirtualToxLab

Fig 4 Quasar model (6D-QSAR left) and Raptorsurrogate (dual-shell 5D-QSAR right) Thebound ligand is shown as stick representation(atom coloring gray = carbon white = hydrogenred = oxygen blue = nitrogen) The quasi-atomistic properties of the receptor are mappedonto the surface(s) blue = positively chargedsalt bridge red = negatively charged salt bridgebrownish colors = hydrophobic properties pink= hydrogen-bond flip flop The Quasar modelrepresents the Aryl hydrocarbon receptor[10] theRaptor model depicts the thyroid receptor β[17]

Fig 5 Flow-chart of the VirtualToxLab

ENDOCRINE DISRUPTORS BASIC RESEARCH AND METHODS 327CHIMIA 2008 62 No 5

roid The three-dimensional structures ofall compounds were generated using Bio[28]

and processed with the standard protocol ofthe VirtualToxLab The resulting toxic po-tentials are given in Table 2

Outlook

Up to date the VirtualToxLab concepthas only produced a few false-positive re-sults At the current level however false-negative predictions are still obtained asa compound of interest cannot be testedagainst all potential receptors it may bindto in vivo (some macromolecular targetswill remain unknown for others no ex-perimental structure exists or too fewaffinity data are available to establish aQSAR) We therefore plan to extend thecurrent concept by implementing a set ofvirtual filters which can recognize be-nign compounds Among others criteriainclude the molecular weight drug-likeproperties (rarr Lipinskirsquos rule of five) andthe presenceabsence of characteristicstructural motifs We are also consideringa combination of our VirtualToxLab withLigandScout a fast screening technologybased on protein-ligand structures[29] Anynew models and results are continuouslyupdated on our website[30] from wherethe VirtualToxLab documentation can bedownloaded

Effective April 1 2008 the technologyhas been made available through the Internetfor academic laboratories hospitals and en-vironmental organizations Interested partiesshould request the licensing agreement Af-ter signing the document they will receive apersonal accountpassword and the URL toaccess the VirtualToxLab on-line

In the VirtualToxLab the data are trans-ferred using the SSH (Secure Shell) proto-col After completion of the task all dataminus compound coordinates models physico-chemical parameters minus are irreversibly dis-carded The entries in the user log file (con-taining the calculated binding affinities)may be deleted by the user at any time

Disclaimer It is understood the Virtual-ToxLab may only generate reliable resultsfor compound classes used to train the mod-el of the respective target protein (which aregiven in the pertinent publications[9ndash21])The standard deviation of a given result isalso a safe quality indicator

AcknowledgementThe authors wish to express their gratitude

to Markus A Lill Assistant Professor at theDepartment of Medicinal Chemistry andMolecular Pharmacology Purdue University forhis contribution to the VirtualToxLab technology(Symposar and Raptor software) and the manyfruitfuldiscussionsonthegenerationandvalidationof receptor-surface models and mQSAR

Fig 7 Main dialog box of the IAP to the VirtualToxLab

Table 2 VirtualToxLab Toxic potential profile of six selected compounds

Bisphenol A Hexabromo-diphenylether

Erythrosine

Aryl hydrocarbonreceptor

(ndash) +++ ++++

Androgen receptor +++ +++ +++

Estrogen receptor α ++ ++ not determinable

Estrogen receptor β +++ +++ not determinable

Glucocorticoid receptor ++++ ++++ +++

Mineralocorticoidreceptor

(+) +++ not determinable

PPARγ ++ ++ (++++)

Thyroid receptor α ndash ndash (++)

Thydorid receptor β (+++) (+++) (+++++)

CYP450 3A4 ++ +++ (++)

CYP450 2A13 +++ +++ (+++)

Paracetamol Methylparaben Methyltrienolone

Aryl hydrocarbonreceptor

(+) ndash +

Androgen receptor +++ ++ ++++

Estrogen receptor α ++ ndash ++++

Estrogen receptor β +++ +++ +++

Glucocorticoid receptor +++ ++ +++

Mineralocorticoidreceptor

(+) ndash +++

PPARγ ++ + ++

Thyroid receptor α ndash ndash ndash

Thydorid receptor β (+++) (+) +++

CYP450 3A4 + ndash +++

CYP450 2A13 +++ +++ +++

Toxic potential in silico ndash none Ki or IC50 gt 100mM

+ very low 100mM gt Ki or IC50 gt 1mM

++ low 1 mM gt Ki or IC50 gt 10 microM

+++ medium 10 microM gt Ki or IC50 gt 100 nM

++++ high 100 nM gt Ki or IC50 gt 1nM

+++++ very high Ki or IC50 lt 1nM

Values in brackets indicate a high standard deviation and should be interpreted with caution ldquonotdeterminablerdquo indicates very high esdrsquos

Bisphenol A Hexabr Erythrosi

ENDOCRINE DISRUPTORS BASIC RESEARCH AND METHODS 328CHIMIA 2008 62 No 5

This research has been made possiblethrough grants by the Swiss National ScienceFoundation the Margaret and Francis-Fleitmann Foundation LucerneSwitzerlandthe Doerenkamp-Zbinden Foundation ZuumlrichSwitzerland and the Jacques en Dolly GazanFoundation ZugSwitzerland which are allgratefully acknowledged

Received March 22 2008

[1] J A Gustaffson Toxicol Lett 1995 135465

[2] B Rihova Adv Drug Deliv Rev 199829 273

[3] B Fischer Andrologia 2000 32 279[4] E V Hestermann J J Stegemann M E

Hahn Toxicol Appl Pharmacol 2000168 160

[5] K Lukasink A Pitkanen J Neurochem2000 74 2445

[6] D L Rymer T A Good J Biol Chem2001 276 2523

[7] A J Hampson M Grimaldi J Neurosci2002 22 257

[8] J D Oliver R A Roberts PharmacolToxicol 2002 91 1

[9] A Vedani M Dobler M A LillPharmacol Toxicol 2006 99 195

[10] A Vedani M Dobler M Spreafico OPeristera M Smiesko ALTEX 24 2007153

[11] A Vedani D W Huhta J Am ChemSoc 1990 112 4759

[12] A Vedani M Dobler J Med Chem2002 45 2139

[13] M A Lill A Vedani M Dobler J MedChem 2004 47 6174

[14] M A Lill A Vedani J Chem Inf Model2006 46 2135

[15] A Vedani M Dobler M A Lill J MedChem 2005 48 3700

[16] httpwwwbiografchindexphpid=projects

[17] M A Lill F Winiger A Vedani B ErnstJ Med Chem 2005 48 5666

[18] M Spreafico M Smiesko M A LillB Ernst A Vedani J Med Chemsubmitted

[19] A Vedani A-V Decloux M SpreaficoB Ernst Toxicol Lett 2007 173 17

[20] A Vedani M Zumstein M A Lill BErnst ChemMedChem 2007 2 78

[21] M A Lill M Dobler A VedaniChemMedChem 2006 1 73

[22] Epik version 15 Schroumldinger LLC NewYork NY 2007

[23] S J Weiner P A Kollmann D A CaseU C Singh C Ghio G Alagona SProfeta Jr P Weiner J Am Chem Soc1984 106 765

[24] MacroModel version 95 SchroumldingerLLC New York NY 2007

[25] F Mohamadi N G J Richards WC Guida R Liskamp M Lipton CCaufield G Chang T Hendrickson WC Still J Comput Chem 1990 11 440

[26] J J P Stewart J Comput-Aided MolecDesign 1990 4 1

[27] C J Cramer D G Truhlar J Comput-Aided Molec Design 1992 6 629

[28] Bio version 30 Biographics Laboratory3R BaselSwitzerland 2007

[29] G Wolber T Langer J Chem Inf Model2005 45 160

[30] httpwwwbiografch

Page 3: VirtualToxLab – in silico Prediction of the Endocrine-Disrupting

ENDOCRINE DISRUPTORS BASIC RESEARCH AND METHODS 324CHIMIA 2008 62 No 5

data sets (a single orientation and confor-mation per ligand molecule) and 4D datasets to be generated It stores the grid pro-duced and optimized by the ligands of thetraining set which can then be applied toany new test and prediction set After theflexible mapping the conformation of eachmolecule is minimized within the variousgrid fields

The VirtualToxLab

Details of model validation are pub-lished (cf [10] and references cited thereinor [16]) Except for the aryl hydrocarbonreceptor where no experimental structureof the protein is available all systems are

based on docking studies using the hu-man protein As we used biological datameasured in a single laboratory for mostsystems the structural diversity is not yetsufficient to consider all models validated(PPARγ TRαβ ERβ GR MR) For somesystems the activity range would not seemto be sufficiently large (PPARγ ERβ) Thecurrent validation status is summarizedin Table 1 and Fig 3 the appearance ofQuasar and Raptor models is depicted inFig 4

Internet Access Portal

Tofacilitateaccess toour technologywehave developed an InternetAccess Protocol

(IAP) immediately available for academiclaboratories (Fig 5) While the technologyitself is freely available we have to requesta modest fee for acquiring and maintainingthe underlying hardware and third-partysoftware as well as for customer supportWhile the VirtualToxLab runs fully auto-mated the 3D structure of the compound ofinterest must be generated and provided bythe end user Several freely available pro-grams exist to accomplish this task it isnonetheless vital to check the correctnessof the structure (constitution connectivitystereochemistry)

In a first step (cf Fig 6) the struc-ture is checked for correctness and thetautomeric and protonation state minus atphysiological pH minus is identified (softwareEpik[22]) Then the compound is subject-ed to a conformational-search protocol(to identify the global minimum confor-mation) in aqueous solution using theAMBER force field[23] as implementedin the MacroModel software[2425] NextMNDOESP[26] or CM1[27] atomic partialcharges are computed In the main stepthe compound is automatically docked tothe target protein(s) thereby allowing theinduced fit and sampling of all energeti-cally feasible binding modes (YetiAuto-Dock[1115]rarr 4D data set) Thereafterphysicochemical properties (solvationenergy internal strain in the bound modeentropic contribution to ligand binding)are determined Finally the compoundrsquostoxic potential is estimated by calculatingits binding affinity towards the macro-molecular target (Quasar[1215] and Rap-tor[13]) Based on the information avail-able on the data set used to evaluate themodel (training and test compounds) thepotential toxicological class (from 0 = be-nign to 4 = extremely toxic) is assignedThe results are then made available to theuser via the IAP (Fig 7) both as absolutenumbers (binding affinity towards the tar-get protein) and the estimated toxic poten-tial (color coded) At this point all userdata (3D structure of the compound ofinterest) and all files associated with theprocedures described above (compoundcoordinates models physicochemicalproperties protocol files) are irreversiblydiscarded (for security issues cf below)The user log file may be exported (to a textfile) at any time

Test Compounds

As an example we tested six com-pounds bisphenol-A a polymer additive3453rsquo4rsquo5rsquo-hexabromodiphenylether aflame-retarding chemical erythrosine afood dye methylparaben a preservativeand fungicide paracetamol an NSAIDand methyltrienolone a very potent ste-

Fig 2 Directional force field (as implemented in Yeti Quasar Raptor andSymposar)

Table 1 VirtualToxLab Summary of the currently available test kits (enzymereceptor models)

System compoundstraining+test=totalcompound classes

q2 rmstrain-ing

maxtrain-ing

p2 rmstest

maxtest

fnfpa

Ref

Aryl hydrocarbon 105+35=140 eight 0824 18 102 0769 23 135 02 [10]

Androgen 88+26=114 eight 0858 17 78 0792 16 139 11 [17]

Estrogen αEstrogen αb

Estrogen βb 80+26=106 six

80+23=103 five

089507870785

200911

862548

089206820827

291108

953824

000000

[15]ndash[10]

Glucocorticoid 82+28=110 four 0745 12 59 0650 22 55 00 [18]

Mineralocorticoid 40+12=52 three 0798 16 56 0701 25 95 00 [16]

PPARγ 75+20=95 two 0832 14 62 0723 14 39 00 [19]

Thyroid αThyroid β 64+18= 82 four

09190909

1820

4377

08140796

2527

10088

0110

[20][20]

CYP3A4 38+10=48 eighteen 0825 27 70 0659 38 71 00 [21]

CYP2A13 18+6=24 six 0900 06 15 0830 03 07 00 [16]

q2 = cross-validated r2 p2 = predictive r2 the rms and maximal deviation from the experimentalbinding affinity is given as factor in Ki or IC50afn = false-negative fp = false-positive compounds a factor 100 or more off the experimentalvaluebdifferent compounds (diphenolic azoles) than for the 80+26 model above

ENDOCRINE DISRUPTORS BASIC RESEARCH AND METHODS 325CHIMIA 2008 62 No 5

6

Fig 3 Comparison of experimental and calculated binding affinities of the models underlying the VirtualToxLab From left to right and top to bottomAhR AR ERα ERβ GR MR PPARγ TRα TRβ 3A4 (simulation without the eight threshold ligands cf ref [21]) 2A13 The ligands of the training setare shown as open circles those of the test set as filled circles The error bars correspond to the variation of the 200ndash500 models comprising the modelfamilies Dashed lines are drawn at plusmn 10 log unit from the experimental value

ENDOCRINE DISRUPTORS BASIC RESEARCH AND METHODS 326CHIMIA 2008 62 No 5

Fig 6 Appearance of the Internet Access Portalto the VirtualToxLab

Fig 4 Quasar model (6D-QSAR left) and Raptorsurrogate (dual-shell 5D-QSAR right) Thebound ligand is shown as stick representation(atom coloring gray = carbon white = hydrogenred = oxygen blue = nitrogen) The quasi-atomistic properties of the receptor are mappedonto the surface(s) blue = positively chargedsalt bridge red = negatively charged salt bridgebrownish colors = hydrophobic properties pink= hydrogen-bond flip flop The Quasar modelrepresents the Aryl hydrocarbon receptor[10] theRaptor model depicts the thyroid receptor β[17]

Fig 5 Flow-chart of the VirtualToxLab

ENDOCRINE DISRUPTORS BASIC RESEARCH AND METHODS 327CHIMIA 2008 62 No 5

roid The three-dimensional structures ofall compounds were generated using Bio[28]

and processed with the standard protocol ofthe VirtualToxLab The resulting toxic po-tentials are given in Table 2

Outlook

Up to date the VirtualToxLab concepthas only produced a few false-positive re-sults At the current level however false-negative predictions are still obtained asa compound of interest cannot be testedagainst all potential receptors it may bindto in vivo (some macromolecular targetswill remain unknown for others no ex-perimental structure exists or too fewaffinity data are available to establish aQSAR) We therefore plan to extend thecurrent concept by implementing a set ofvirtual filters which can recognize be-nign compounds Among others criteriainclude the molecular weight drug-likeproperties (rarr Lipinskirsquos rule of five) andthe presenceabsence of characteristicstructural motifs We are also consideringa combination of our VirtualToxLab withLigandScout a fast screening technologybased on protein-ligand structures[29] Anynew models and results are continuouslyupdated on our website[30] from wherethe VirtualToxLab documentation can bedownloaded

Effective April 1 2008 the technologyhas been made available through the Internetfor academic laboratories hospitals and en-vironmental organizations Interested partiesshould request the licensing agreement Af-ter signing the document they will receive apersonal accountpassword and the URL toaccess the VirtualToxLab on-line

In the VirtualToxLab the data are trans-ferred using the SSH (Secure Shell) proto-col After completion of the task all dataminus compound coordinates models physico-chemical parameters minus are irreversibly dis-carded The entries in the user log file (con-taining the calculated binding affinities)may be deleted by the user at any time

Disclaimer It is understood the Virtual-ToxLab may only generate reliable resultsfor compound classes used to train the mod-el of the respective target protein (which aregiven in the pertinent publications[9ndash21])The standard deviation of a given result isalso a safe quality indicator

AcknowledgementThe authors wish to express their gratitude

to Markus A Lill Assistant Professor at theDepartment of Medicinal Chemistry andMolecular Pharmacology Purdue University forhis contribution to the VirtualToxLab technology(Symposar and Raptor software) and the manyfruitfuldiscussionsonthegenerationandvalidationof receptor-surface models and mQSAR

Fig 7 Main dialog box of the IAP to the VirtualToxLab

Table 2 VirtualToxLab Toxic potential profile of six selected compounds

Bisphenol A Hexabromo-diphenylether

Erythrosine

Aryl hydrocarbonreceptor

(ndash) +++ ++++

Androgen receptor +++ +++ +++

Estrogen receptor α ++ ++ not determinable

Estrogen receptor β +++ +++ not determinable

Glucocorticoid receptor ++++ ++++ +++

Mineralocorticoidreceptor

(+) +++ not determinable

PPARγ ++ ++ (++++)

Thyroid receptor α ndash ndash (++)

Thydorid receptor β (+++) (+++) (+++++)

CYP450 3A4 ++ +++ (++)

CYP450 2A13 +++ +++ (+++)

Paracetamol Methylparaben Methyltrienolone

Aryl hydrocarbonreceptor

(+) ndash +

Androgen receptor +++ ++ ++++

Estrogen receptor α ++ ndash ++++

Estrogen receptor β +++ +++ +++

Glucocorticoid receptor +++ ++ +++

Mineralocorticoidreceptor

(+) ndash +++

PPARγ ++ + ++

Thyroid receptor α ndash ndash ndash

Thydorid receptor β (+++) (+) +++

CYP450 3A4 + ndash +++

CYP450 2A13 +++ +++ +++

Toxic potential in silico ndash none Ki or IC50 gt 100mM

+ very low 100mM gt Ki or IC50 gt 1mM

++ low 1 mM gt Ki or IC50 gt 10 microM

+++ medium 10 microM gt Ki or IC50 gt 100 nM

++++ high 100 nM gt Ki or IC50 gt 1nM

+++++ very high Ki or IC50 lt 1nM

Values in brackets indicate a high standard deviation and should be interpreted with caution ldquonotdeterminablerdquo indicates very high esdrsquos

Bisphenol A Hexabr Erythrosi

ENDOCRINE DISRUPTORS BASIC RESEARCH AND METHODS 328CHIMIA 2008 62 No 5

This research has been made possiblethrough grants by the Swiss National ScienceFoundation the Margaret and Francis-Fleitmann Foundation LucerneSwitzerlandthe Doerenkamp-Zbinden Foundation ZuumlrichSwitzerland and the Jacques en Dolly GazanFoundation ZugSwitzerland which are allgratefully acknowledged

Received March 22 2008

[1] J A Gustaffson Toxicol Lett 1995 135465

[2] B Rihova Adv Drug Deliv Rev 199829 273

[3] B Fischer Andrologia 2000 32 279[4] E V Hestermann J J Stegemann M E

Hahn Toxicol Appl Pharmacol 2000168 160

[5] K Lukasink A Pitkanen J Neurochem2000 74 2445

[6] D L Rymer T A Good J Biol Chem2001 276 2523

[7] A J Hampson M Grimaldi J Neurosci2002 22 257

[8] J D Oliver R A Roberts PharmacolToxicol 2002 91 1

[9] A Vedani M Dobler M A LillPharmacol Toxicol 2006 99 195

[10] A Vedani M Dobler M Spreafico OPeristera M Smiesko ALTEX 24 2007153

[11] A Vedani D W Huhta J Am ChemSoc 1990 112 4759

[12] A Vedani M Dobler J Med Chem2002 45 2139

[13] M A Lill A Vedani M Dobler J MedChem 2004 47 6174

[14] M A Lill A Vedani J Chem Inf Model2006 46 2135

[15] A Vedani M Dobler M A Lill J MedChem 2005 48 3700

[16] httpwwwbiografchindexphpid=projects

[17] M A Lill F Winiger A Vedani B ErnstJ Med Chem 2005 48 5666

[18] M Spreafico M Smiesko M A LillB Ernst A Vedani J Med Chemsubmitted

[19] A Vedani A-V Decloux M SpreaficoB Ernst Toxicol Lett 2007 173 17

[20] A Vedani M Zumstein M A Lill BErnst ChemMedChem 2007 2 78

[21] M A Lill M Dobler A VedaniChemMedChem 2006 1 73

[22] Epik version 15 Schroumldinger LLC NewYork NY 2007

[23] S J Weiner P A Kollmann D A CaseU C Singh C Ghio G Alagona SProfeta Jr P Weiner J Am Chem Soc1984 106 765

[24] MacroModel version 95 SchroumldingerLLC New York NY 2007

[25] F Mohamadi N G J Richards WC Guida R Liskamp M Lipton CCaufield G Chang T Hendrickson WC Still J Comput Chem 1990 11 440

[26] J J P Stewart J Comput-Aided MolecDesign 1990 4 1

[27] C J Cramer D G Truhlar J Comput-Aided Molec Design 1992 6 629

[28] Bio version 30 Biographics Laboratory3R BaselSwitzerland 2007

[29] G Wolber T Langer J Chem Inf Model2005 45 160

[30] httpwwwbiografch

Page 4: VirtualToxLab – in silico Prediction of the Endocrine-Disrupting

ENDOCRINE DISRUPTORS BASIC RESEARCH AND METHODS 325CHIMIA 2008 62 No 5

6

Fig 3 Comparison of experimental and calculated binding affinities of the models underlying the VirtualToxLab From left to right and top to bottomAhR AR ERα ERβ GR MR PPARγ TRα TRβ 3A4 (simulation without the eight threshold ligands cf ref [21]) 2A13 The ligands of the training setare shown as open circles those of the test set as filled circles The error bars correspond to the variation of the 200ndash500 models comprising the modelfamilies Dashed lines are drawn at plusmn 10 log unit from the experimental value

ENDOCRINE DISRUPTORS BASIC RESEARCH AND METHODS 326CHIMIA 2008 62 No 5

Fig 6 Appearance of the Internet Access Portalto the VirtualToxLab

Fig 4 Quasar model (6D-QSAR left) and Raptorsurrogate (dual-shell 5D-QSAR right) Thebound ligand is shown as stick representation(atom coloring gray = carbon white = hydrogenred = oxygen blue = nitrogen) The quasi-atomistic properties of the receptor are mappedonto the surface(s) blue = positively chargedsalt bridge red = negatively charged salt bridgebrownish colors = hydrophobic properties pink= hydrogen-bond flip flop The Quasar modelrepresents the Aryl hydrocarbon receptor[10] theRaptor model depicts the thyroid receptor β[17]

Fig 5 Flow-chart of the VirtualToxLab

ENDOCRINE DISRUPTORS BASIC RESEARCH AND METHODS 327CHIMIA 2008 62 No 5

roid The three-dimensional structures ofall compounds were generated using Bio[28]

and processed with the standard protocol ofthe VirtualToxLab The resulting toxic po-tentials are given in Table 2

Outlook

Up to date the VirtualToxLab concepthas only produced a few false-positive re-sults At the current level however false-negative predictions are still obtained asa compound of interest cannot be testedagainst all potential receptors it may bindto in vivo (some macromolecular targetswill remain unknown for others no ex-perimental structure exists or too fewaffinity data are available to establish aQSAR) We therefore plan to extend thecurrent concept by implementing a set ofvirtual filters which can recognize be-nign compounds Among others criteriainclude the molecular weight drug-likeproperties (rarr Lipinskirsquos rule of five) andthe presenceabsence of characteristicstructural motifs We are also consideringa combination of our VirtualToxLab withLigandScout a fast screening technologybased on protein-ligand structures[29] Anynew models and results are continuouslyupdated on our website[30] from wherethe VirtualToxLab documentation can bedownloaded

Effective April 1 2008 the technologyhas been made available through the Internetfor academic laboratories hospitals and en-vironmental organizations Interested partiesshould request the licensing agreement Af-ter signing the document they will receive apersonal accountpassword and the URL toaccess the VirtualToxLab on-line

In the VirtualToxLab the data are trans-ferred using the SSH (Secure Shell) proto-col After completion of the task all dataminus compound coordinates models physico-chemical parameters minus are irreversibly dis-carded The entries in the user log file (con-taining the calculated binding affinities)may be deleted by the user at any time

Disclaimer It is understood the Virtual-ToxLab may only generate reliable resultsfor compound classes used to train the mod-el of the respective target protein (which aregiven in the pertinent publications[9ndash21])The standard deviation of a given result isalso a safe quality indicator

AcknowledgementThe authors wish to express their gratitude

to Markus A Lill Assistant Professor at theDepartment of Medicinal Chemistry andMolecular Pharmacology Purdue University forhis contribution to the VirtualToxLab technology(Symposar and Raptor software) and the manyfruitfuldiscussionsonthegenerationandvalidationof receptor-surface models and mQSAR

Fig 7 Main dialog box of the IAP to the VirtualToxLab

Table 2 VirtualToxLab Toxic potential profile of six selected compounds

Bisphenol A Hexabromo-diphenylether

Erythrosine

Aryl hydrocarbonreceptor

(ndash) +++ ++++

Androgen receptor +++ +++ +++

Estrogen receptor α ++ ++ not determinable

Estrogen receptor β +++ +++ not determinable

Glucocorticoid receptor ++++ ++++ +++

Mineralocorticoidreceptor

(+) +++ not determinable

PPARγ ++ ++ (++++)

Thyroid receptor α ndash ndash (++)

Thydorid receptor β (+++) (+++) (+++++)

CYP450 3A4 ++ +++ (++)

CYP450 2A13 +++ +++ (+++)

Paracetamol Methylparaben Methyltrienolone

Aryl hydrocarbonreceptor

(+) ndash +

Androgen receptor +++ ++ ++++

Estrogen receptor α ++ ndash ++++

Estrogen receptor β +++ +++ +++

Glucocorticoid receptor +++ ++ +++

Mineralocorticoidreceptor

(+) ndash +++

PPARγ ++ + ++

Thyroid receptor α ndash ndash ndash

Thydorid receptor β (+++) (+) +++

CYP450 3A4 + ndash +++

CYP450 2A13 +++ +++ +++

Toxic potential in silico ndash none Ki or IC50 gt 100mM

+ very low 100mM gt Ki or IC50 gt 1mM

++ low 1 mM gt Ki or IC50 gt 10 microM

+++ medium 10 microM gt Ki or IC50 gt 100 nM

++++ high 100 nM gt Ki or IC50 gt 1nM

+++++ very high Ki or IC50 lt 1nM

Values in brackets indicate a high standard deviation and should be interpreted with caution ldquonotdeterminablerdquo indicates very high esdrsquos

Bisphenol A Hexabr Erythrosi

ENDOCRINE DISRUPTORS BASIC RESEARCH AND METHODS 328CHIMIA 2008 62 No 5

This research has been made possiblethrough grants by the Swiss National ScienceFoundation the Margaret and Francis-Fleitmann Foundation LucerneSwitzerlandthe Doerenkamp-Zbinden Foundation ZuumlrichSwitzerland and the Jacques en Dolly GazanFoundation ZugSwitzerland which are allgratefully acknowledged

Received March 22 2008

[1] J A Gustaffson Toxicol Lett 1995 135465

[2] B Rihova Adv Drug Deliv Rev 199829 273

[3] B Fischer Andrologia 2000 32 279[4] E V Hestermann J J Stegemann M E

Hahn Toxicol Appl Pharmacol 2000168 160

[5] K Lukasink A Pitkanen J Neurochem2000 74 2445

[6] D L Rymer T A Good J Biol Chem2001 276 2523

[7] A J Hampson M Grimaldi J Neurosci2002 22 257

[8] J D Oliver R A Roberts PharmacolToxicol 2002 91 1

[9] A Vedani M Dobler M A LillPharmacol Toxicol 2006 99 195

[10] A Vedani M Dobler M Spreafico OPeristera M Smiesko ALTEX 24 2007153

[11] A Vedani D W Huhta J Am ChemSoc 1990 112 4759

[12] A Vedani M Dobler J Med Chem2002 45 2139

[13] M A Lill A Vedani M Dobler J MedChem 2004 47 6174

[14] M A Lill A Vedani J Chem Inf Model2006 46 2135

[15] A Vedani M Dobler M A Lill J MedChem 2005 48 3700

[16] httpwwwbiografchindexphpid=projects

[17] M A Lill F Winiger A Vedani B ErnstJ Med Chem 2005 48 5666

[18] M Spreafico M Smiesko M A LillB Ernst A Vedani J Med Chemsubmitted

[19] A Vedani A-V Decloux M SpreaficoB Ernst Toxicol Lett 2007 173 17

[20] A Vedani M Zumstein M A Lill BErnst ChemMedChem 2007 2 78

[21] M A Lill M Dobler A VedaniChemMedChem 2006 1 73

[22] Epik version 15 Schroumldinger LLC NewYork NY 2007

[23] S J Weiner P A Kollmann D A CaseU C Singh C Ghio G Alagona SProfeta Jr P Weiner J Am Chem Soc1984 106 765

[24] MacroModel version 95 SchroumldingerLLC New York NY 2007

[25] F Mohamadi N G J Richards WC Guida R Liskamp M Lipton CCaufield G Chang T Hendrickson WC Still J Comput Chem 1990 11 440

[26] J J P Stewart J Comput-Aided MolecDesign 1990 4 1

[27] C J Cramer D G Truhlar J Comput-Aided Molec Design 1992 6 629

[28] Bio version 30 Biographics Laboratory3R BaselSwitzerland 2007

[29] G Wolber T Langer J Chem Inf Model2005 45 160

[30] httpwwwbiografch

Page 5: VirtualToxLab – in silico Prediction of the Endocrine-Disrupting

ENDOCRINE DISRUPTORS BASIC RESEARCH AND METHODS 326CHIMIA 2008 62 No 5

Fig 6 Appearance of the Internet Access Portalto the VirtualToxLab

Fig 4 Quasar model (6D-QSAR left) and Raptorsurrogate (dual-shell 5D-QSAR right) Thebound ligand is shown as stick representation(atom coloring gray = carbon white = hydrogenred = oxygen blue = nitrogen) The quasi-atomistic properties of the receptor are mappedonto the surface(s) blue = positively chargedsalt bridge red = negatively charged salt bridgebrownish colors = hydrophobic properties pink= hydrogen-bond flip flop The Quasar modelrepresents the Aryl hydrocarbon receptor[10] theRaptor model depicts the thyroid receptor β[17]

Fig 5 Flow-chart of the VirtualToxLab

ENDOCRINE DISRUPTORS BASIC RESEARCH AND METHODS 327CHIMIA 2008 62 No 5

roid The three-dimensional structures ofall compounds were generated using Bio[28]

and processed with the standard protocol ofthe VirtualToxLab The resulting toxic po-tentials are given in Table 2

Outlook

Up to date the VirtualToxLab concepthas only produced a few false-positive re-sults At the current level however false-negative predictions are still obtained asa compound of interest cannot be testedagainst all potential receptors it may bindto in vivo (some macromolecular targetswill remain unknown for others no ex-perimental structure exists or too fewaffinity data are available to establish aQSAR) We therefore plan to extend thecurrent concept by implementing a set ofvirtual filters which can recognize be-nign compounds Among others criteriainclude the molecular weight drug-likeproperties (rarr Lipinskirsquos rule of five) andthe presenceabsence of characteristicstructural motifs We are also consideringa combination of our VirtualToxLab withLigandScout a fast screening technologybased on protein-ligand structures[29] Anynew models and results are continuouslyupdated on our website[30] from wherethe VirtualToxLab documentation can bedownloaded

Effective April 1 2008 the technologyhas been made available through the Internetfor academic laboratories hospitals and en-vironmental organizations Interested partiesshould request the licensing agreement Af-ter signing the document they will receive apersonal accountpassword and the URL toaccess the VirtualToxLab on-line

In the VirtualToxLab the data are trans-ferred using the SSH (Secure Shell) proto-col After completion of the task all dataminus compound coordinates models physico-chemical parameters minus are irreversibly dis-carded The entries in the user log file (con-taining the calculated binding affinities)may be deleted by the user at any time

Disclaimer It is understood the Virtual-ToxLab may only generate reliable resultsfor compound classes used to train the mod-el of the respective target protein (which aregiven in the pertinent publications[9ndash21])The standard deviation of a given result isalso a safe quality indicator

AcknowledgementThe authors wish to express their gratitude

to Markus A Lill Assistant Professor at theDepartment of Medicinal Chemistry andMolecular Pharmacology Purdue University forhis contribution to the VirtualToxLab technology(Symposar and Raptor software) and the manyfruitfuldiscussionsonthegenerationandvalidationof receptor-surface models and mQSAR

Fig 7 Main dialog box of the IAP to the VirtualToxLab

Table 2 VirtualToxLab Toxic potential profile of six selected compounds

Bisphenol A Hexabromo-diphenylether

Erythrosine

Aryl hydrocarbonreceptor

(ndash) +++ ++++

Androgen receptor +++ +++ +++

Estrogen receptor α ++ ++ not determinable

Estrogen receptor β +++ +++ not determinable

Glucocorticoid receptor ++++ ++++ +++

Mineralocorticoidreceptor

(+) +++ not determinable

PPARγ ++ ++ (++++)

Thyroid receptor α ndash ndash (++)

Thydorid receptor β (+++) (+++) (+++++)

CYP450 3A4 ++ +++ (++)

CYP450 2A13 +++ +++ (+++)

Paracetamol Methylparaben Methyltrienolone

Aryl hydrocarbonreceptor

(+) ndash +

Androgen receptor +++ ++ ++++

Estrogen receptor α ++ ndash ++++

Estrogen receptor β +++ +++ +++

Glucocorticoid receptor +++ ++ +++

Mineralocorticoidreceptor

(+) ndash +++

PPARγ ++ + ++

Thyroid receptor α ndash ndash ndash

Thydorid receptor β (+++) (+) +++

CYP450 3A4 + ndash +++

CYP450 2A13 +++ +++ +++

Toxic potential in silico ndash none Ki or IC50 gt 100mM

+ very low 100mM gt Ki or IC50 gt 1mM

++ low 1 mM gt Ki or IC50 gt 10 microM

+++ medium 10 microM gt Ki or IC50 gt 100 nM

++++ high 100 nM gt Ki or IC50 gt 1nM

+++++ very high Ki or IC50 lt 1nM

Values in brackets indicate a high standard deviation and should be interpreted with caution ldquonotdeterminablerdquo indicates very high esdrsquos

Bisphenol A Hexabr Erythrosi

ENDOCRINE DISRUPTORS BASIC RESEARCH AND METHODS 328CHIMIA 2008 62 No 5

This research has been made possiblethrough grants by the Swiss National ScienceFoundation the Margaret and Francis-Fleitmann Foundation LucerneSwitzerlandthe Doerenkamp-Zbinden Foundation ZuumlrichSwitzerland and the Jacques en Dolly GazanFoundation ZugSwitzerland which are allgratefully acknowledged

Received March 22 2008

[1] J A Gustaffson Toxicol Lett 1995 135465

[2] B Rihova Adv Drug Deliv Rev 199829 273

[3] B Fischer Andrologia 2000 32 279[4] E V Hestermann J J Stegemann M E

Hahn Toxicol Appl Pharmacol 2000168 160

[5] K Lukasink A Pitkanen J Neurochem2000 74 2445

[6] D L Rymer T A Good J Biol Chem2001 276 2523

[7] A J Hampson M Grimaldi J Neurosci2002 22 257

[8] J D Oliver R A Roberts PharmacolToxicol 2002 91 1

[9] A Vedani M Dobler M A LillPharmacol Toxicol 2006 99 195

[10] A Vedani M Dobler M Spreafico OPeristera M Smiesko ALTEX 24 2007153

[11] A Vedani D W Huhta J Am ChemSoc 1990 112 4759

[12] A Vedani M Dobler J Med Chem2002 45 2139

[13] M A Lill A Vedani M Dobler J MedChem 2004 47 6174

[14] M A Lill A Vedani J Chem Inf Model2006 46 2135

[15] A Vedani M Dobler M A Lill J MedChem 2005 48 3700

[16] httpwwwbiografchindexphpid=projects

[17] M A Lill F Winiger A Vedani B ErnstJ Med Chem 2005 48 5666

[18] M Spreafico M Smiesko M A LillB Ernst A Vedani J Med Chemsubmitted

[19] A Vedani A-V Decloux M SpreaficoB Ernst Toxicol Lett 2007 173 17

[20] A Vedani M Zumstein M A Lill BErnst ChemMedChem 2007 2 78

[21] M A Lill M Dobler A VedaniChemMedChem 2006 1 73

[22] Epik version 15 Schroumldinger LLC NewYork NY 2007

[23] S J Weiner P A Kollmann D A CaseU C Singh C Ghio G Alagona SProfeta Jr P Weiner J Am Chem Soc1984 106 765

[24] MacroModel version 95 SchroumldingerLLC New York NY 2007

[25] F Mohamadi N G J Richards WC Guida R Liskamp M Lipton CCaufield G Chang T Hendrickson WC Still J Comput Chem 1990 11 440

[26] J J P Stewart J Comput-Aided MolecDesign 1990 4 1

[27] C J Cramer D G Truhlar J Comput-Aided Molec Design 1992 6 629

[28] Bio version 30 Biographics Laboratory3R BaselSwitzerland 2007

[29] G Wolber T Langer J Chem Inf Model2005 45 160

[30] httpwwwbiografch

Page 6: VirtualToxLab – in silico Prediction of the Endocrine-Disrupting

ENDOCRINE DISRUPTORS BASIC RESEARCH AND METHODS 327CHIMIA 2008 62 No 5

roid The three-dimensional structures ofall compounds were generated using Bio[28]

and processed with the standard protocol ofthe VirtualToxLab The resulting toxic po-tentials are given in Table 2

Outlook

Up to date the VirtualToxLab concepthas only produced a few false-positive re-sults At the current level however false-negative predictions are still obtained asa compound of interest cannot be testedagainst all potential receptors it may bindto in vivo (some macromolecular targetswill remain unknown for others no ex-perimental structure exists or too fewaffinity data are available to establish aQSAR) We therefore plan to extend thecurrent concept by implementing a set ofvirtual filters which can recognize be-nign compounds Among others criteriainclude the molecular weight drug-likeproperties (rarr Lipinskirsquos rule of five) andthe presenceabsence of characteristicstructural motifs We are also consideringa combination of our VirtualToxLab withLigandScout a fast screening technologybased on protein-ligand structures[29] Anynew models and results are continuouslyupdated on our website[30] from wherethe VirtualToxLab documentation can bedownloaded

Effective April 1 2008 the technologyhas been made available through the Internetfor academic laboratories hospitals and en-vironmental organizations Interested partiesshould request the licensing agreement Af-ter signing the document they will receive apersonal accountpassword and the URL toaccess the VirtualToxLab on-line

In the VirtualToxLab the data are trans-ferred using the SSH (Secure Shell) proto-col After completion of the task all dataminus compound coordinates models physico-chemical parameters minus are irreversibly dis-carded The entries in the user log file (con-taining the calculated binding affinities)may be deleted by the user at any time

Disclaimer It is understood the Virtual-ToxLab may only generate reliable resultsfor compound classes used to train the mod-el of the respective target protein (which aregiven in the pertinent publications[9ndash21])The standard deviation of a given result isalso a safe quality indicator

AcknowledgementThe authors wish to express their gratitude

to Markus A Lill Assistant Professor at theDepartment of Medicinal Chemistry andMolecular Pharmacology Purdue University forhis contribution to the VirtualToxLab technology(Symposar and Raptor software) and the manyfruitfuldiscussionsonthegenerationandvalidationof receptor-surface models and mQSAR

Fig 7 Main dialog box of the IAP to the VirtualToxLab

Table 2 VirtualToxLab Toxic potential profile of six selected compounds

Bisphenol A Hexabromo-diphenylether

Erythrosine

Aryl hydrocarbonreceptor

(ndash) +++ ++++

Androgen receptor +++ +++ +++

Estrogen receptor α ++ ++ not determinable

Estrogen receptor β +++ +++ not determinable

Glucocorticoid receptor ++++ ++++ +++

Mineralocorticoidreceptor

(+) +++ not determinable

PPARγ ++ ++ (++++)

Thyroid receptor α ndash ndash (++)

Thydorid receptor β (+++) (+++) (+++++)

CYP450 3A4 ++ +++ (++)

CYP450 2A13 +++ +++ (+++)

Paracetamol Methylparaben Methyltrienolone

Aryl hydrocarbonreceptor

(+) ndash +

Androgen receptor +++ ++ ++++

Estrogen receptor α ++ ndash ++++

Estrogen receptor β +++ +++ +++

Glucocorticoid receptor +++ ++ +++

Mineralocorticoidreceptor

(+) ndash +++

PPARγ ++ + ++

Thyroid receptor α ndash ndash ndash

Thydorid receptor β (+++) (+) +++

CYP450 3A4 + ndash +++

CYP450 2A13 +++ +++ +++

Toxic potential in silico ndash none Ki or IC50 gt 100mM

+ very low 100mM gt Ki or IC50 gt 1mM

++ low 1 mM gt Ki or IC50 gt 10 microM

+++ medium 10 microM gt Ki or IC50 gt 100 nM

++++ high 100 nM gt Ki or IC50 gt 1nM

+++++ very high Ki or IC50 lt 1nM

Values in brackets indicate a high standard deviation and should be interpreted with caution ldquonotdeterminablerdquo indicates very high esdrsquos

Bisphenol A Hexabr Erythrosi

ENDOCRINE DISRUPTORS BASIC RESEARCH AND METHODS 328CHIMIA 2008 62 No 5

This research has been made possiblethrough grants by the Swiss National ScienceFoundation the Margaret and Francis-Fleitmann Foundation LucerneSwitzerlandthe Doerenkamp-Zbinden Foundation ZuumlrichSwitzerland and the Jacques en Dolly GazanFoundation ZugSwitzerland which are allgratefully acknowledged

Received March 22 2008

[1] J A Gustaffson Toxicol Lett 1995 135465

[2] B Rihova Adv Drug Deliv Rev 199829 273

[3] B Fischer Andrologia 2000 32 279[4] E V Hestermann J J Stegemann M E

Hahn Toxicol Appl Pharmacol 2000168 160

[5] K Lukasink A Pitkanen J Neurochem2000 74 2445

[6] D L Rymer T A Good J Biol Chem2001 276 2523

[7] A J Hampson M Grimaldi J Neurosci2002 22 257

[8] J D Oliver R A Roberts PharmacolToxicol 2002 91 1

[9] A Vedani M Dobler M A LillPharmacol Toxicol 2006 99 195

[10] A Vedani M Dobler M Spreafico OPeristera M Smiesko ALTEX 24 2007153

[11] A Vedani D W Huhta J Am ChemSoc 1990 112 4759

[12] A Vedani M Dobler J Med Chem2002 45 2139

[13] M A Lill A Vedani M Dobler J MedChem 2004 47 6174

[14] M A Lill A Vedani J Chem Inf Model2006 46 2135

[15] A Vedani M Dobler M A Lill J MedChem 2005 48 3700

[16] httpwwwbiografchindexphpid=projects

[17] M A Lill F Winiger A Vedani B ErnstJ Med Chem 2005 48 5666

[18] M Spreafico M Smiesko M A LillB Ernst A Vedani J Med Chemsubmitted

[19] A Vedani A-V Decloux M SpreaficoB Ernst Toxicol Lett 2007 173 17

[20] A Vedani M Zumstein M A Lill BErnst ChemMedChem 2007 2 78

[21] M A Lill M Dobler A VedaniChemMedChem 2006 1 73

[22] Epik version 15 Schroumldinger LLC NewYork NY 2007

[23] S J Weiner P A Kollmann D A CaseU C Singh C Ghio G Alagona SProfeta Jr P Weiner J Am Chem Soc1984 106 765

[24] MacroModel version 95 SchroumldingerLLC New York NY 2007

[25] F Mohamadi N G J Richards WC Guida R Liskamp M Lipton CCaufield G Chang T Hendrickson WC Still J Comput Chem 1990 11 440

[26] J J P Stewart J Comput-Aided MolecDesign 1990 4 1

[27] C J Cramer D G Truhlar J Comput-Aided Molec Design 1992 6 629

[28] Bio version 30 Biographics Laboratory3R BaselSwitzerland 2007

[29] G Wolber T Langer J Chem Inf Model2005 45 160

[30] httpwwwbiografch

Page 7: VirtualToxLab – in silico Prediction of the Endocrine-Disrupting

ENDOCRINE DISRUPTORS BASIC RESEARCH AND METHODS 328CHIMIA 2008 62 No 5

This research has been made possiblethrough grants by the Swiss National ScienceFoundation the Margaret and Francis-Fleitmann Foundation LucerneSwitzerlandthe Doerenkamp-Zbinden Foundation ZuumlrichSwitzerland and the Jacques en Dolly GazanFoundation ZugSwitzerland which are allgratefully acknowledged

Received March 22 2008

[1] J A Gustaffson Toxicol Lett 1995 135465

[2] B Rihova Adv Drug Deliv Rev 199829 273

[3] B Fischer Andrologia 2000 32 279[4] E V Hestermann J J Stegemann M E

Hahn Toxicol Appl Pharmacol 2000168 160

[5] K Lukasink A Pitkanen J Neurochem2000 74 2445

[6] D L Rymer T A Good J Biol Chem2001 276 2523

[7] A J Hampson M Grimaldi J Neurosci2002 22 257

[8] J D Oliver R A Roberts PharmacolToxicol 2002 91 1

[9] A Vedani M Dobler M A LillPharmacol Toxicol 2006 99 195

[10] A Vedani M Dobler M Spreafico OPeristera M Smiesko ALTEX 24 2007153

[11] A Vedani D W Huhta J Am ChemSoc 1990 112 4759

[12] A Vedani M Dobler J Med Chem2002 45 2139

[13] M A Lill A Vedani M Dobler J MedChem 2004 47 6174

[14] M A Lill A Vedani J Chem Inf Model2006 46 2135

[15] A Vedani M Dobler M A Lill J MedChem 2005 48 3700

[16] httpwwwbiografchindexphpid=projects

[17] M A Lill F Winiger A Vedani B ErnstJ Med Chem 2005 48 5666

[18] M Spreafico M Smiesko M A LillB Ernst A Vedani J Med Chemsubmitted

[19] A Vedani A-V Decloux M SpreaficoB Ernst Toxicol Lett 2007 173 17

[20] A Vedani M Zumstein M A Lill BErnst ChemMedChem 2007 2 78

[21] M A Lill M Dobler A VedaniChemMedChem 2006 1 73

[22] Epik version 15 Schroumldinger LLC NewYork NY 2007

[23] S J Weiner P A Kollmann D A CaseU C Singh C Ghio G Alagona SProfeta Jr P Weiner J Am Chem Soc1984 106 765

[24] MacroModel version 95 SchroumldingerLLC New York NY 2007

[25] F Mohamadi N G J Richards WC Guida R Liskamp M Lipton CCaufield G Chang T Hendrickson WC Still J Comput Chem 1990 11 440

[26] J J P Stewart J Comput-Aided MolecDesign 1990 4 1

[27] C J Cramer D G Truhlar J Comput-Aided Molec Design 1992 6 629

[28] Bio version 30 Biographics Laboratory3R BaselSwitzerland 2007

[29] G Wolber T Langer J Chem Inf Model2005 45 160

[30] httpwwwbiografch