HAL Id: ineris-01863940
https://hal-ineris.archives-ouvertes.fr/ineris-01863940
Submitted on 29 Aug 2018
HAL is a multi-disciplinary open access
archive for the deposit and dissemination of
sci-entific research documents, whether they are
pub-lished or not. The documents may come from
teaching and research institutions in France or
abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est
destinée au dépôt et à la diffusion de documents
scientifiques de niveau recherche, publiés ou non,
émanant des établissements d’enseignement et de
recherche français ou étrangers, des laboratoires
publics ou privés.
Multiscale modelling approaches for assessing cosmetic
ingredients safety
Frédéric Y. Bois, Juan G. Diaz Ochoa, Monika Gajewska, Simona Kovarich,
Klaus Mauch, Alicia Paini, Alexandre R.R. Pery, Jose Vicente Sala Benito,
Sophie Teng, Andrew P. Worth
To cite this version:
Frédéric Y. Bois, Juan G. Diaz Ochoa, Monika Gajewska, Simona Kovarich, Klaus Mauch, et al..
Multiscale modelling approaches for assessing cosmetic ingredients safety. Toxicology, Elsevier, 2017,
392, pp.130-139. �10.1016/j.tox.2016.05.026�. �ineris-01863940�
Multiscale
modelling
approaches
for
assessing
cosmetic
ingredients
safety
Frédéric
Y.
Bois
a,*
,
Juan
G.
Diaz
Ochoa
c,
Monika
Gajewska
b,
Simona
Kovarich
d,
Klaus
Mauch
c,
Alicia
Paini
b,
Alexandre
Péry
a,
Jose
Vicente
Sala
Benito
b,
Sophie
Teng
a,
Andrew
Worth
ba
INERIS,DRC/VIVA/METO,ParcALATA,BP2,60550Verneuil-en-Halatte,France
b
EuropeanCommissionJointResearchCentre,InstituteforHealthandConsumerProtection,SystemsToxicologyUnit,ViaEnricoFermi2749,Ispra,VA,Italy
cInsilicoBiotechnologyAG,Meitnerstrasse8,70563Stuttgart,Germany dS-INSoluzioniInformatiche,viaG.Ferrari14,36100Vicenza,Italy
ARTICLE INFO Articlehistory: Received7October2015
Receivedinrevisedform30November2015 Accepted31May2016
Availableonline4June2016 Keywords:
PBPKmodel Multiscalemodel Insilicotoxicityprediction
ABSTRACT
TheEuropeanUnion’sbanonanimaltestingforcosmeticingredientsandproductshasgeneratedastrong momentumforthedevelopmentofinsilicoandinvitroalternativemethods.Oneofthefocusofthe COSMOSprojectwasabinitiopredictionofkineticsandtoxiceffectsthroughmultiscalepharmacokinetic modelingandinvitrodataintegration.Inourexperience,mathematicalorcomputermodelingandin vitroexperimentsarecomplementary.Wepresenthereasummaryofthemainmodelsandresults obtainedwithintheframeworkoftheprojectonthesetopics.Afirstsectionpresentsourworkatthe organelleandcellularlevel.Wethengotowardmodelingcelllevelseffects(monitoredcontinuously), multiscalephysiologicallybasedpharmacokineticandeffectmodels,androutetorouteextrapolation. WefollowwithashortpresentationoftheautomatedKNIMEworkflowsdevelopedfordissemination andeasyuseofthemodels.Weendwithadiscussionoftwochallengestothefield:ourlimitedabilityto dealwithmassivedataandcomplexcomputations.
ã2016TheAuthors.PublishedbyElsevierIrelandLtd.ThisisanopenaccessarticleundertheCC BY-NC-NDlicense(http://creativecommons.org/licenses/by-nc-nd/4.0/).
1.Introduction
The European decision to ban animal testing for cosmetic ingredientshasgeneratedastrongmomentumforthe develop-mentofinsilicoandinvitroalternativemethods.Oneoftheaimsof theCOSMOSproject(fundedbytheEuropeanCommissionandby CosmeticsEuropeunderthe7thFrameworkProgramme)wasto developapproachesfortheabinitiopredictionofkineticsandtoxic effectsthroughmultiscalepharmacokineticmodelingandinvitro data integration. We present here a summary of the relevant modelsand results obtainedby the COSMOS team. Our major activitieswerefocusedonmodelingtoxicokineticsand toxicody-namics(effects)invitroandinvivo,whicharerequiredtoperform quantitativeinvitrotoinvivoextrapolation(QIVIVE)(Adleretal., 2011;Bessemsetal.,2014;Coeckeetal.,2012;Quignotetal.,2014). Kineticmodelingisarelativelywelldevelopedfield,withwell established compartmental or physiologically based
pharmacokinetic(PBPK)models(Corley,2010;GibaldiandPerrier, 1982;Peters,2011).InPBPKmodels,thetransportandoverallfate of the substance administered is governed by anatomic and physiological considerations. The models can have a generic structure,which makesthemeasiertouse(noneedtodevelop newequationsforanewsubstance)(Beaudouinetal.,2010;Corley, 2010; Jamei et al.,2009; Willmann etal., 2007).Most of their parametersarephysiological,meaningthattheydonotdependon thechemicalconsidered,butsolelyonthesubjectexposed(atleast whenexposuretothechemicaldoesnotalterappreciablythebody functions,suchasbloodflows).Compilationsofaverageparameter valuesforseveralspeciesareavailable,andevenvaluesforspecific sub-groups(children,pregnantwomen,elderlypeople...)(Bois et al., 2010).The few remainingparameters, which depend on chemicalstructure,aresufficientlymechanistictobeobtainedby quantitative structure-property relationships (QSPR), or invitro experiments (Hamon et al., 2015). In vitro kinetic models are different from PBPK models, in that they are not particularly physiological,butrather representtheinvitrosystem modeled. Some generic models have been proposed for simple in vitro systems – including by us,the virtual cell based assay(VCBA)
*Correspondingauthor.
E-mailaddress:frederic.bois@ineris.fr(F.Y. Bois).
http://dx.doi.org/10.1016/j.tox.2016.05.026
0300-483X/ã2016TheAuthors.PublishedbyElsevierIrelandLtd.ThisisanopenaccessarticleundertheCCBY-NC-NDlicense( http://creativecommons.org/licenses/by-nc-nd/4.0/).
ContentslistsavailableatScienceDirect
Toxicology
model, see below – but more complex systems (e.g., bi-compartmental systems, micro-chips) require specific develop-ments(Armitageetal.,2014;Creanetal.,2014;Ouattaraetal., 2011;Truisietal.,2015;Wilmesetal.,2013;Zaldívaretal.,2010). Thestateof theart isfarless advanced for“extrapolatable” toxicodynamic models. Traditional toxicodynamic models are similartocompartmentalmodels(andinfactextendthosewithad hoc“effect”compartments).Theyaredatafittedandthusspecific foragivenexperimentorclinicaltrial(CsajkaandVerotta,2006); theyarenotdesignedforextrapolations(exceptverybasictime and dose extrapolations). The equivalent of PBPK models in toxicodynamicsarebiologically-basedmodels.Amongtheearliest ofsuchmodelswerethebiologicallybasedcarcinogenesismodels (Armitage,1985;MoolgavkarandKnudson,1981).However,given the obscurity and complexity of the cancer process (still not elucidated),thosemodelswereatthesametime toosimple(to avoid criticisms) and too complex (to be used in a regulatory framework). They were never really used for risk assessment, exceptintheextremelysimplifiedformofthemultistagecancer dose-responsemodel(CrumpandHowe,1984).Anewgeneration of modelsis emerging with“systems biology” models(Geenen et al., 2012; Jusko, 2013) and “physiome” (or virtual human) models(Bassingthwaighte,2000;HunterandBorg,2003).Systems biologymodelsarebottom-upmodelsrootedinbiochemistryand benefitingfromourincreasingunderstandingofcellularsignaling andtranscriptionalcontrolpathways(facilitatedbytheexplosion of omics data). Physiome modelsare inherently top-down and multiscaleand thereforetheclosestequivalent toPBPKmodels (PBPK modelscan in fact be thoughtof as vascular body-level solutetransportmodels).Theystartedashigh-leveldescriptionsof organphysiologyand areincreasingtheirresolution tothecell level(arguablytherightleveltostartunderstandingtheoriginof most toxic injuries). Originating from two different research communities(biochemistsvs.physiologists)thesetwoapproaches areslowlymergingasthey meeteach otheratthetissuelevel. Furthermore,thetwoapproachesdidnotescapetheattentionof the members of the 21st century toxicology panel of the US NationalAcademyof Sciences,whoplacedthem attheheartof theirvisionstatement(NationalResearchCouncil(NRC), 2007), gaining much attention, given theauthority of itsauthors.The ensuingconsiderationoftoxicitypathwaysandmodesofaction (MOA)methappilytheadverseoutcomepathway(AOP)thinking invogueintheecotoxicologycommunity(Tollefsenetal.,2014), andisstillhesitatingaboutchangingname... Meanwhile,virtual organprogramshavebeenheavilyfunded(e.g.,bytheVirtualLiver project of theGerman Ministryof Research)(Holzhütter etal., 2012) given their potential impact for predictive drug safety assessment.Inshort,wearewitnessingaconvergenceofsystems biology and virtual organs modeling around the concept of quantitativeMOAs,fullyamenabletoQIVIVEandriskassessment. Thispaperfollowsabottom-upintegrationlogic:Afirstsection presents our work at the organelle/cellular level. We then go toward modeling cell levels effects (monitored continuously), multiscalePBPKandeffectmodels,androutetoroute extrapola-tion.WeendwithashortpresentationoftheautomatedKNIME workflows developed for dissemination and easy use of our models.
2.Fromorganellestocells
Before manifesting themselves at the cellular level, most toxicityeffects start at thescale of organelles.Mitochondriain particularareoftentargetsoftoxicity.Theyperformtwocritical functionsinthecell:theproductionofmorethan90%ofthecell’s energy, and the control of cell survival as an integral part of programmedcelldeath(apoptosis).Threegeneraladverseeffects
resultfrommitochondrialtoxicity:1.Disruptedenergy metabo-lism;2.Increasedfreeradicalgeneration;and3.Alteredapoptosis. Weaddressedthedisruptionofmitochondrialenergymetabolism bymeasuringandsimulatingmitochondrialmembranepotential (MMP). Themeasurementof MMPprovides informationonthe mitochondrion's ability to carry out oxidative phosphorylation (whichcoupleselectrontransfertoATPsynthesis),and transfer ions and substrates across its inner membranes (Nicholls and Ward,2000).Thus,oneofthemostcommonmethodstodetect mitochondrial toxicity is the monitoring of the cells' MMP. A variety of fluorescent dyes can be used to that effect in high throughputscreening.Forexample,cationicdyesdistributetothe mitochondrial matrix in accordance with Nernst's equation (MitchellandMoyle,1969),sothattheMMPisgivenby: MMP¼ð
a
VÞRFTlog CcytCmit
ð1Þ where R is the gasconstant, T thetemperature, F the Faraday constant,Ccyttheconcentrationofthechemicalinthecellcytosol,
Cmit its concentration in mitochondria,
a
a proportionalityconstant,andVthecellviability.Cmitiscomputedbyintegration
ofthefollowingdifferentialequation:
@
Cmit@
t ¼Kmit CaqCmitð2Þ whereCaqistheconcentrationintheaqueousphaseofthecell,and
Kmitisadiffusionrateconstantdependentonthechemicalandcell
lineused.
We report here results on the in vitro MMP disruption of HepaRG cells by caffeine, carbonyl cyanide-p-tri fluoromethoxy-phenylhydrazone(FCCP),amiodaroneandestragole.TheMMPwas measured and modeled using anextensionof the VCBA model (Zaldívaretal.,2011,2010).Thatmodel,likesomeothers(Armitage etal.,2014;Creanetal.,2014;Hamonetal.,2014;Pomponioetal., 2014;Truisietal.,2015;Wilmesetal.,2013)takesintoaccountthe fateofthetestcompoundinvitro:partitioningbetweenplasticvial walls,headspace,serumproteinsandcells.Itincludesamodelof cellgrowthanddeathandhasbeenlinkedtoathresholdmodelfor cellkilling.In thecourseoftheCOSMOSproject,we developed VCBAmodelsforamiodarone,caffeine,FCCP,coumarin,estragole, ethanol,andnicotine.TosimulateMMPdata,amitochondrialsub cellularcompartment and Nernst’s equation wereaddedtothe VCBAmodel.InvitroHepaRGMMPdatawereusedtooptimizetwo parameters of Nernst's equation (
a
and Kmit) by least-squareminimization.Fig.1showsthemeasuredandsimulatedMMPasa function of the exposure concentration of the four chemicals assayed. The model was able to correctly reproduce the amiodarone and estragole data.The caffeine datashowa peak at0.01M,whichcouldnotbereproducedbythemodel.TheFCCP inducedfastdecreaseofMMPatconcentrationslowerthan0.1mM wasnotwellcapturedbythemodeleither.Moreexperimentswith different chemicals will be needed to fully understand the determinantsofpredictionaccuracy.
3.Modelinginvitrokineticsandcontinuouslymeasuredcell effects
Twocomplementarymodelswereusedtoanalyzetoxiceffects in vitro, at the cell level. The first model is the VCBA model introducedabove(Zaldívaretal.,2011,2010).Thismodeliswell suitedtoanalyzefixedpointcytotoxicitydata(asinhigh-content imaging assays).The secondmodel canbeusedfor continuous cytotoxicity monitoring, using electrical impedance measure-ments(Xingetal.,2006).
ThelattermodeldescribesHepaRGcellviabilitylossfollowing exposure to hepatotoxic molecules. It was applied it to three
cosmetic related substances: coumarin, isoeugenol and benzo-phenone-2.Ourmodelcouplesdynamicdescriptionsofthemajor invitrokineticprocessesinvolvedwithasimplemodelofviability loss(seeTengetal.,2015fordetails).
ThedatawereobtainedusingHepaRGcells(Guillouzoetal., 2007), exposed to the selected chemicals either once (at concentrationsrangingfrom0.008to8mM witha 48h follow-up)orduringrepeatedexposures(atconcentrationsrangingfrom 0.128to8mM,withafollow-upof4weeks).Cellularviabilitywas monitoredbyimpedancemeasurementswithaninvitrolabel-free cell-basedmonitoringsystem(xCELLigenceTM,ACEABiosciences,
RocheDiagnostics).TheimpedanceofHepaRGcellswasconverted toanormalizedcellindex(NCI).AdecreaseinNCIisindicativeof celldetachmentordeath(Xingetal.,2006).
Wemodeledtheinvitrodynamicdecreaseinconcentrationfor thesubstanceofinterestbyanapparentlinearprocess(lumping unspecificphysico-chemical reactions, plastic binding, evapora-tionand linearmetabolism)or bya Michaelis-Mentenequation whensaturablecellularmetabolismwasobserved.
Forcoumarinandisoeugenol,thetime-courseofcellviability wasmodeledusingEq.(5),parameterizedbyakillingrate(b)anda noeffectconcentration,NEC.
dN dt ¼ bðCiNECÞN; ifCi>NEC; 0otherwise ð5Þ Nbeingthenumberofcellsinthewell.Forbenzophenone-2,we usedaslightlymorecomplexmodeltodescribecellspreadingat sub-toxicconcentrations(Tengetal.,2015).Themodelparameters were estimated by least square fitting of the model to the impedancedata.
On the basis of goodness of fit, linear loss was used for coumarin. Saturable loss gave better fits for isoeugenol and benzophenone-2. As shown for coumarin in Fig. 2 (for short-exposures) and in Fig. 3 (for repeated exposures), our model described rather well the kinetics of real-time impedance
measurementsinHepaRGcellsatvariousexposuredoses.Similar resultswereobtainedforisoeugenolandbenzophenone-2(Teng etal.,2015).
Howeverforisoeugenolandbenzophenone-2,theacutemodels failedtopredictlong-termexposuresandvice-versa.Thiscouldbe duetovariousphenomenanotaccountedforbythemodel,suchat thereversibilityof plasticbinding,or thecomplexdynamicsof cellular response and adaptation to stress. We have actually observedthatinvitrorepeatedexposuresexperimentstendtolead tocomplextime-dose-responserelationshipmuchmoredifficult tointerpretandanalysethan acutetoxicitydata(Hamonet al., 2014;Wilmesetal.,2013).Thisshouldnotbetakenasaproblemto avoid, but as a challenge to resolve, because such complex responsesareprobablymorerepresentativeoftheinvivocellular responses.Inanycase,whileawaitingbettermodelsanddeeper understanding,itisrecommendedtousetheproperexperimental datatopredictthecorrespondingeffect.
Fig.1.Observed(blackdots)andVCBAmodelsimulatedvalues(redcirclesand lines)ofmitochondrialmembranepotentialinHepaRGcellsexposedinvitrotofour testcompounds; carbonyl cyanide-p-trifluoromethoxyphenylhydrazone (FCCP), caffeine,amiodarone,andestragole.(Forinterpretationofthereferencestocolour inthisfigurelegend,thereaderisreferredtothewebversionofthisarticle.)
Fig.2. NormalizedCellIndex(NCI)ofHepaRGcellsexposedoncetocoumarinat variousconcentrations.Thedataandthemodelpredictionsarerepresentedby pointsandlines,respectively(adaptedfromTengetal.,2015).
Fig.3. NormalizedCellIndex(NCI)ofHepaRGcellsexposedevery2–3daysto coumarinatvariousconcentrationsfor4weeks.Thedataandthemodelpredictions arerepresentedbypointsandlines,respectively.
Fig.4showsacomparisonbetweentheTengetal.andVCBA modelestimatesofHepaRGcellsviabilityafterexposuretovarious concentrationsofcoumarin.TheVCBAmodelusedstaticCellomics assaydatatoassesscellviability.Cell-leveltoxicodynamicswere described by two parameters: a no-effectconcentration and a killingrate. ComparedtoCellomics,impedancemetrics provide slightlyhighervaluesofcellviability.However,thetwotestsgive comparable results when experimental variability is taken into account.
4.FrominvitrotoinvivowithmultiscalePBPKmodels 4.1.InsilicopredictionsofPBPKmodelparameters
Quantitativestructureactivity/propertyrelationships(QSARs/ QSPRs)aretheoretical(insilico)modelsthatrelatethestructureof chemicalstotheirbiologicactivitiesorproperties(e.g., physico-chemical,partitioningorfateproperty).QSARmodelscanbeused toparameterizePBPKmodelsbyprovidingpredictionsfor basic physico-chemicalproperties(e.g.,ionizationconstant,octanolover water partition coefficient, distribution coefficient or aqueous solubility) as well as key ADME properties (e.g., extent of gastrointestinal absorption, oral bioavailability, plasma protein binding,metabolismratebycytochromesorclearance).Anumber ofQSARmodelsandsoftwaretoolsareavailablefortheprediction ofADMErelatedproperties(Mostrag-SzlichtyngandWorth,2010; Worthetal.,2011).InthecontextoftheCOSMOSprojectweused theACD/LabsPerceptasoftware(ACD/LabsPercepta,release2014. AdvancedChemistry Development Inc., Toronto, Canada. www. acdlabs.com)topredictseveralphysico-chemicalparameters(e.g., LogKow, pKa, Fup) and ADME properties (e.g., passive oral
absorption,oral bioavailability, total body clearance) that were usedforthecalibrationofPBPKmodels.
4.2.Fromcellstotissuestowholebody
Asshownintheprevioussections,notonlytheassessmentof concentrations in organelles, but also the assessment of cell concentrationsinatissueisrelevanttoperforminvitrotoinvivo extrapolations(Martinetal.,2015).Sincetheliverasawholeisa rather homogeneous organ, it is possible torepresent it as an averagesinglecellcontainingrelativelycomplexmicro-organelles
envelopingtransportandmolecularpathways(Kraussetal.,2012). ThisapproachbalancesthecomplexityofPBPKmodelsandthatof molecular interactions, while reducing to a minimum the complexityoftheorganstructure.
However, when the liver structure is analyzed in detail it becomes clearthatliversubstructures areheterogeneous(Pang andRowland,1977;Pangetal.,2007).Inthelivertissue,bloodis distributedbyportalveinsintofunctionalsubunits,calledlobules, whichcarryoutdiversefunctionsincludingthedetoxificationof xenobioticsatthecellularlevel.Thelobulescapillariesdistribute bloodanddissolvedsubstancestovariouscells,andinparticular hepatocytes.Inalobule,bloodtransportproducesanaturaloxygen gradient, introducing a zonation of the lobule, which in turn inducesaheterogeneousdistributionofCYPexpressionfactorsin thehepatocytesradiallyalignedinthelobule.Themetabolismof substances and thetoxicresponseof theorgandepend onthis heterogeneity.
Thatissue hasattractedtheattentionof several groups and projects (Holzhütter et al., 2012), and has motivated several experimental initiatives, such as novel forms of cell cultures (Soldatowetal.,2013).Mathematicalmodelsforthemulti-scale descriptionoftheliverhavealsobeendeveloped(DiazOchoaetal., 2013;HuntandRopella,2008;Pangetal.,2007;Wambaughand Shah,2010).
Several publications have shown how such models can be implementedtoimprovepredictionsoftheinvivotoxicresponse to small molecules. In our case (Diaz Ochoa et al., 2013), we describedthemetabolismanddetoxificationofacetaminophenin the lobule using six sinusoids, which are capillary structures, orderedtoformahexagonalshapedlobule.Ineachhepatocyte,a metabolicnetworksub-modelallowstheestimationoftherateof productionofrelevantmetabolitesandtheproductionofoxidative species. On this basis, we can estimate the clearance by the simulatedliverandapplyittothelivercompartmentinaPBPK model.Theexplicitcalculationofclearanceusingaparallelliver model requiresaconcurrentparallelization. Thatopensa novel computationalwaytointegrate livercellswithanevergrowing moleculardescriptioncomplexity,dependingontheinformation availableforthereconstructionofmolecularpathways.Thiskind ofapproachisessentiallya“divideandconquer”strategywhere different sub-models are defined in separate programs run in parallel.Currentdevelopmentsarenotrestrictedtothegeneration ofmodelsoralgorithmsimplementedinconventionalmachines. For instance the use of cloud computing has also opened the possibility to simulate very complex liver models with a high degreeofbiologicaldetail(RopellaandHunt,2010).
Thatapproachallowedustoestimatelocalconcentrationsof acetaminophen that potentiallytriggercell necrosis,helping to visualize liverdamage afterhumanexposuretoa toxicdoseof acetaminophen(Fig.5).Duetothecouplingofthetoxicresponse and detoxification it was possible to qualitatively estimatethe effectofthetoxicresponseontheconcentrationofthesubstance inthewholeorganism.Implementingcelldynamicsand mechan-icalcellinteractionsmadeitalsofeasibletovisualizeindetailliver necrosis,closingthegapbetweeninsilicopredictionsandinvivo outcomes(Drasdoetal.,2014).
However, for quantitative estimation, it is sometimes more appropriate to directlycouple the sub-cellular unitsto a PBPK model. Simpler approaches combining PBPK modeling and systemsbiologyhaveinvestigatedthecouplingofasinglecomplex livercellwithinternalmetabolism(Bois,2009;Péryetal.,2013). Slightly more complex modelsdescribe the liver as a 1D tube representingaliversinusoidwherethetransportofsubstancesis coupledtothemetabolismin thehepatocytes(Andersenet al., 1997;Jonesetal.,2012).Fig.6illustratesourimplementationof suchamodel.
Fig.4. Comparisonof HepaRGcell viability exposedto coumarinat 24h, as measuredbyCellomicsandestimatedbytheVCBAmodel(blackdots)orimpedance measurementsandTeng’smodel(reddots).(Forinterpretationofthereferencesto colourinthisfigurelegend,thereaderisreferredtothewebversionofthisarticle.)
Such 1D-liver modelshave disadvantages aswell as advan-tages: they are less realistic because they assume that the dispersion rate is homogeneous, implying that the substance concentrationisatequilibriumatanytimeineachcompartment. In general the lack of hydrodynamics, as well as the incorrect representationofboundaryconditionsmayhidesomeeffects(like differencesin bloodvelocities withina sinusoid)related tothe transport of substances from the sinusoidal space to the hepatocyte,aswellashindertherepresentationoftheinherent fractal structure of the organ (Dokoumetzidis and Macheras, 2003).However,theeliminationoftheexplicitspatialdependence allowsasimple mathematicaldescriptionusingordinary differ-entialequations.Thiskindoflessgranular1D-liverstillcaptures theessentialsoftheliverzonationandsimultaneouslyincreases thecalculationperformancewhencellularnetworks(metabolism orcellregulation)arecoupledtothehepatocyte.Insummary,the selectionofthedimensionalityofthelivermodeldependsonthe level of complexity required for the quality of the in silico predictions.
InFig.7wepresentanexampleoftheuseofazonated1-Dliver foracetaminophen.Inthisexample,anoraldoseof5mg/kgwas
assumedtobegiventoamanweighing73kg.Thevaluesofthe PBPKmodelparametersarebasedonthosereportedbyPéryetal. (2013).Weassumedthatprimarymetabolismineachhepatocyte wasadequatelydescribedbyaMichaelis-Mententerm,multiplied byascalingfactorrepresentingthedifferencesinmetabolicratein eachzoneoftheliver.Withoutzonationthisfactorisequal1forall hepatocytes. We first compared the predictions of our model withoutzonationtothoseobtainedwithapreviouslypublished unstructuredmodel(Péryetal.,2013).
We assumed a heterogeneous metabolism along the liver sinusoid. Different signals like gradients of oxygen, nutrients, metabolites, hormones and cytokines have an influence on zonation since they modulate the activity of various enzymes involved in the metabolic pathways (Gebhardt and Matz-Soja, 2014).However,theestimationofzonationfactorsisnottrivial, sincetheyvarycontinuouslyinresponsetofluctuatingsinusoidal patterns (Oinonen and Lindros, 1998). Glucuronidation is the dominantpathwayof conjugationathighAPAP concentrations, andisfasterinthepericentralthanintheperiportalregion,ashas been shown in experimental studies (Anundi et al., 1993). Correspondingly, weselected a factor of 1.0 for thepericentral zone,0.5forthemidlobularzone,and0.5fortheperiportalzone (OinonenandLindros,1998).Asaconsequence,anoverallincrease inacetaminophenconcentrationsandriskoftoxicresponsewas observed(Fig.7).
That model has been implemented as a KNIME work-flow (Berthold et al.,2007), increasingthe portability of themodel. Based onthis methodology,itis possibletowritegeneralPBPK modelsthatcaneasilybeappliedtodifferentsubstances.
Asignificantlimitationisthedifficultytoobservezonationin vivo,limitingthepossibilitytovalidatethecorrespondingmodels. Severaladvancesinon-chipbioreactortechnologymimickinga1D livercouldhelpvalidateinsilicomodels(AllenandBhatia,2003). For example, advances in microfluidics have facilitated the developmentoforganson-chipcontainingcellscoupledtomicro vesselsmimickingthecapillariesofrealorgans(Dashetal.,2009; Kimura etal.,2015), openingthepossibilitytodirectlyobserve liver zonation and changes in metabolic rates. These advances representapotentialmethodtodelivercomplementaryresultsor even couple in vitro and in silico methods to increase the confidenceinbothmethods.
Fig.6.1D-livermodelofthetransportofasubstancefromportalveinstocentralveinsinthesinusoidaswellasitstransportintothehepatocytes(parametershavebeen definedaccordingtoTable1).Ineachhepatocytethemetabolismisdescribedbyamolecularnetwork.Thisenablesthedescriptionofaheterogeneousmetabolismand phenotypealongthesinusoid(zonation).
Fig.5.Visualizationofliverdamageatdifferenttimes(inminutes)aftertheoral applicationofacetaminophenfortwovirtualpatientswithdifferentCYPexpression factors.Theorangeandblackcenterarearepresentshealthyanddamagedliver tissue,respectively.Thesixredellipsoidsattheperipheryaretheperiportalveins (reprintedfromDiazOchoaetal.,2013;withpermission).
TheintroductionofadditionalstructuresinaPBPKmodelhasa doubleeffect:theycanimprovethepredictionsmadewithsucha model, but they can also limit its use if new and unknown parameters cannot be easily estimated (Poggesi et al., 2014). Despitethedifficultyofexperimentallyvalidatingstructuredliver models, theycan be useful to study qualitatively the effect of populationvariabilityincytochromesexpressiononliverzonation, toestimate risks involved in changes in zonation (induced for exampleby othersubstances or liverdamage), or tostudy the impactofzonationonlivertoxicresponse.
4.3.Route-to-routeextrapolation
Topredictsafelevelsofhumanexposuretogeneralchemicals wehavesofarusuallyreliedonanimalstudiescarriedoutmainly bytheoralroute.However,forcosmeticingredientsthemainroute of consumer exposure is expected to be dermal. For existing chemicals,ratherthancarryingoutadditionalanimalstudiesusing anewexposureroute(whichgoesagainst3Rprinciplesandthe European Unionban onanimal testingforcosmetics), route-to-routeextrapolationcanbeperformedbymodelingifreliabledata
Table1
Parametersrequiredforthemodelingofsinusoidsinour1Dlivermodel(Fig.6).Alltheparametersarescaledproportionallytothetotallivervolume. Parameter Valueorformula Source
Permeability(neutralsubstances) P=(logP6.7)cm/s.Validforneutralsubstances Trappetal.(2008)
Transportthroughsinusoid 0.05mm/s VollmarandMenger(2009)
Meansinusoidlength 368mm Grisham(2009)
Volumesinusoidalspace 2.121013L estimatedbyourselvesusinghepatocytevolume VolumeHepatocyte 5.091014L Lodish(2000)
Sinusoiddiameter 10mm Grisham(2009)
Thicknessofbarrierbetweensinusoidandhepatocyte 2mm Ostrovidovetal.(2004)
Approximatednumberofhepatocytesintheliver 109
Grisham(2009)
Fig.7.Modelpredictionofacetaminophenconcentration-timeprofilesinthreezones(1:periportal;2:midlobular;3:pericentral)oftheliversinusoidvesselsorhepatocytes, andinvenousblood,withzonationofmetabolism(redcurves)orwithoutzonation(blackcurves),aftera5mg/kgoraladministrationofacetaminophenofanadultman.The transportratetohepatocyteswas1.40101L/min,and1.08102L/minalongthesinusoids.(Forinterpretationofthereferencestocolourinthisfigurelegend,thereaderis referredtothewebversionofthisarticle.)
ontheADMEpropertiesofthesubstanceareavailable(Gerrityand Henry,1990).Obviouslyhere,technicalconsiderationsastothe typeofinvitroassaystoperformareessential.Themodeof route-to-routeextrapolationshouldbedecidedonacase-by-casebasis (Nielsenetal.,2008).Forexample,thethresholdoftoxicological concern(TTC)approachcaneventuallybeapplied(Worthetal., 2012), and Kroes et al. (2007) give examples of how the TTC approachcanbeappliedtocosmeticingredientsandimpurities. Theyproposedthatdefaultfactorsshouldbeusedtoadjustfroman externaltopicaldosetoaninternaldose,followedbyapplicationof theoralTTCvaluesderivedbyMunroetal.(1996).PBPKmodeling providesamoremechanisticapproachtoroute-to-route extrapo-lation(DoursonandFelter,1997;Reitzetal.,1988).Inourrecent workoncosmeticingredients,aseriesofPBPKmodelscoupledto VCBA models was developed toextrapolate between routes of exposurefor caffeine,coumarin,ethanol, hydroquinone, isopro-panolandnicotine(Gajewskaetal.,2014).
Wereporthere,asanexample,theresultsforcaffeine,basedon the use of a full PBPK model. Concentration-time profiles of caffeineinblood,followingeitheroralordermalabsorptionwere simulatedbyourPBPKmodelattheoralNOAELdoseof2.1mgper kgof body mass and per day (Fig. 8).These simulations were performedforamalesubject,weighing75kg.Theconcentration versustimeprofilesforthetwoadministrationroutesareclearly different,whichindicatesthattheNOAELdoseshouldprobablynot bethesameafteroralordermalexposures,asdiscussedindetailin (Gajewskaetal.,2014).
PBPKmodelscanpredictconcentration-timeprofilesofagiven compoundattheorganlevelorcelllevel(seeabove).Inturn,the predictedintracellularconcentrationscanbelinkedtocellularor tissue level effect through the use of a concentration-effect relationship (i.e., a model), such as the VCBA one or similar (Gajewskaetal.,2015;Painietal.,2012),ormorecomplexpathway model(Hamonetal.,2014,2015).Suchamodelingapproachhas thepotentialtopredicteffectivetargetorgandose.Doseresponse curvesobtainedbasedoninsilicosimulationandinvitrotoxicity datacanbeusedtoderivebenchmarkdoselevels,usableaspoints ofdepartureforriskassessment.Acceptablehumandailyintake, tolerable intakes, health based recommendation and exposure limitscanbeset(Rietjensetal.,2011).In thespecificcase ofa cosmetic,theamountofaningredientusedinthefinalformulation canberestrictedinordertoobtainafinalproductthatdoesnot poseahealthconcerntotheconsumer.
5.Frominsilicomodelstoautomatedworkflows
Futurehumansafetyassessmentsarelikelytorelystronglyon theuseofmulti-scalemodels,implementedthrougha combina-tion ofcomputational tools,in order toperformextrapolations suchasQIVIVE.Thedevelopmentofautomatedsoftwaretoolsis nowseenasanimportant stepforharmonizingand expediting chemical safety assessments. The KNIME workbench (Berthold etal.,2007)providesauser-friendlygraphicalworkflowinterface fordataprocessingandanalysis.Mostofthemathematicalmodels developedwithintheframeworkofCOSMOS–mostlywritteninR (RDevelopmentCoreTeam,2013)–havebeenimplementedinthe KNIMEplatform.Theyalreadyareorwillsoonbepubliclyavailable throughtheCOSMOSKNIMEWebPortal(http://knimewebportal. cosmostox.eu),withdocumentationanduserguidanceavailableat
Fig.8. Concentration-timeprofilesofcaffeineinbloodaftereitherbolusoralor singleapplicationdermalexposuretoanoralNOAELdose(2.1mgperkgofbody mass)(adaptedfromGajewskaetal.,2014).
Table2
ListofthebiokineticsworkflowsdevelopedinCOSMOSandwhichareorwillbeavailableintheCOSMOSKNIMEWebPortalathttp://knimewebportal.cosmostox.eu. Type Chemical Exposure Species Output
PBPK Caffeine Oral,dermal Human,rat Timekinetics PBPK Coumarin Oral,dermal Human,rat Timekinetics PBPK Ethanol Oral,inhalation Human,rat Timekinetics PBPK Estragole Oral,dermal Human Timekinetics PBPK Hydroquinone Oral,dermal Human,rat Timekinetics PBPK Isopropanol Oral,dermal Human,rat Timekinetics PBPK Methyliodide Oral,dermal,inhalation Human,rat Timekinetics PBPK Nicotine Intra-venous(IV) Human Timekinetics PBPK Styrene Inhalation Human Timekinetics PBPK Methotrexate Oral,dermal Human Timekinetics PBPK Valproicacid Oral,dermal Human Timekinetics PBPK Piperonylbutoxide Dermal Human Timekinetics PBPK General Oral,dermal,inhalation Human Time-dosekinetics PBPKandbioaccumulation 90chemicals
General Oral Oral Human Human Bio-accumulationfactor Bio-accumulationfactor QIVIVE Caffeine Oral,dermal Human Time-dosekinetics QIVIVE Coumarin Oral,dermal Human Time-dosekinetics QIVIVE Estragole Oral,dermal Human Time-dosekinetics QIVIVE General Oral,dermal Human Time-dosekinetics
VCBA 30chemicals Invitro HepaRG,HepG2,3t3balb/c,A549,cardiomyocytes Time-dosekineticsandeffects 1Dliver Coumarin Oral,inhalation,IV Human Timekinetics
http://cosmosspace.cosmostox.eu(Table2).Fig.9showsascreen capture of the KNIME caffeine work-flow graphical interface. Workflowsofferagraphicalinterfaceforautomatizingdatainput, modelparameters’setting,simulationcallsandoutputtodifferent datastreams.
6.Conclusion
Theworkpresentedherewasperformedtodevelop asafety assessmentapproachforcosmeticingredientsthatdoesnotrelyon theuseofanimalexperimentsandwhichisthereforecompliant withtheEuropeanUnionbanonanimaltesting.Twoapproaches are available to maintain an acceptable safety level for those products:Thefirstwouldbetousehumanbasedinvivobiomarker assay. The second, and more realistic, is the combination of predictive chemistry (in silico) assessments, based on legacy knowledge frompast animal experiments, invitro testing, and QIVIVE.Wefocusedhereonthelatter.Mathematicalorcomputer modelingandinvitroexperimentsarecomplementary.Apurelyab initio chemistry-based approach is not currently feasible for toxicitypredictions,sincetheknowledgegapistoowideandhasto be filled with in vitro data. Such data by themselves are also insufficient. In vitro systems arestill farfrom able to mimic a humanbodyandthedatacollected,essentiallyfinite,needtobe interpolatedandextrapolatedinvariousways.Modelingcanhelp there. We have made a modest contribution to the large internationalendeavorinthis field,addressingvariousscalesof complexity: from the sub-cellular level (mitochondrial stress), throughthecellularone(cytotoxicitymodels),thetissueandorgan scale(liver),uptothewholebody(withPBPKmodels).
Themodelswepresentareallrelevanttothesafetyassessment ofcosmeticingredients,butatvariousdegrees,dependingofthe depthoftheassessmentsought.Dermalexposuresimulationsare available in all the PBPK models we developed. The KNIME workflowsforinvitroandinvivokinetics’simulations(Table2)are clearlythemostreadilyapplicable,inparticularforQIVIVEinthe contextof ab initioassessments (e.g.,topredictinvivo cellular levels which can be used in input to dose-response models calibratedwithinvitrodata)(Hamonetal.,2015;Quignotetal., 2014).Weareactuallyusingthemincasestudiesdemonstrating theapplicationsofthevarioustoolsdevelopedin COSMOSand
otherprojectsoftheSEURATcluster(theresultswillbereported laterinseparatepublications).Theothermodelspresentedhere aremorespecificandhavenarrowerscope.Thelivermodel,for example, shouldfor now be reserved for in depthmechanistic studies of particular chemicals and is too complex for high-throughput screening. On the other hand, the models we developed can be transposed to drugs or general chemicals withoutproblems.
One apparent issue in the overview we presented is the definitionoftheboundariesbetweenmodels.Forexample,where should a PBPK model stop and where does systems biology modelingstart?Thequestionmayseempurelydefinitional,aswe knowthatacontinuumexistsfromtransportandmetabolismto metabolismcontrolandsignalingpathways,andthatcontinuum extends to cell–cell communications, to the tissue level etc. However, it is also an operational question: What should toxicologists,risk assessmentpractitioners anddecision makers expectfromPBPKmodeling?Certainlynotanall-encompassing tool.OuropinionisthatPBPKmodelsshouldfocusontransport and leave the details of metabolism and transport to systems biology, so that the norm should be coupled PBPK – systems biology –physiomemodels (Geenen etal., 2013;Hamonet al., 2015). Or maybe PBPK models should rather be viewed as particular (vascular system) physiome models? Not everyone mayagreewiththisview,aswetendtoendorsethelabelsimposed by customs and norms. Yet we should be open to changes, particularlywhentheybroadenourviewsandcapabilities.
Onelimitationinthisfieldisourabilitytodealwithcomplex computations based on massive data sets. This includes, for example, the integration of omics data. Beyond multivariate statistics,clusteringandadhocapproximatepathwayanalysis,the integrationofsuchdatainamechanisticframeworkisascientific andcomputationalchallenge(USEPA,2014).Virtualtissuesand virtual body models are obviouslyextremely complexand few researchorganizationshavethecapabilitiestorunthemtoafull extent not to mention risk assessment practitioners, so that simplified versions are still the norm. Computational and mathematical methods still need improvement to meet this challenge.
Themodelingofadverseeffectsinhumansisbasedonthefact that toxicity pathwaysand mechanismsare in facthigh-jacked
Fig.9.SchematicrepresentationofaKNIMEwork-flowdevelopedforPBPKmodelingforcaffeineeffectsafteroralanddermalexposure.Theworkflowautomatizesdata input,modelparameters’setting,simulationcallsandoutputtodifferentdatastreams.Thecorecalculationaredoneinthe“Rsnippet”nodesinthemiddle.
biological pathways, and toxic effects are a particular class of physiologicaleffects.Whiletheuseofmodelingapproachesandin vitromethodstoreplaceanimalexperimentsremainsa consider-ablechallenge, advancesinthefieldarebeingmadeduetothe effortsofaverylargecommunityofscientists.
Acknowledgements
This work was supported by the European Community’s SeventhFrameworkProgram(FP7/2007-2013)andbyCosmetics Europethroughthe COSMOSproject(grant agreementnumber 266835).
References
Adler,S.,Basketter,D.,Creton,S.,Pelkonen,O.,vanBenthem,J.,Zuang,V.,Andersen, K.,Angers-Loustau,A.,Aptula,A.,Bal-Price,A.,etal.,2011.Alternative (non-animal)methodsforcosmeticstesting:currentstatusandfutureprospects— 2010.Arch.Toxicol.85,367–485.
Allen,J.W.,Bhatia,S.N.,2003.Formationofsteady-stateoxygengradientsinvitro: applicationtoliverzonation.Biotechnol.Bioeng.82,253–262.
Andersen,M.E.,Birnbaum,L.S.,Barton,H.A.,Eklund,C.R.,1997.Regionalhepatic CYP1A1andCYP1A2inductionwith2,3,7,8-tetrachlorodibenzo-p-dioxin evaluatedwithamulticompartmentgeometricmodelofhepaticzonation. Toxicol.Appl.Pharmacol.144,145–155.
Anundi,I.,Lähteenmäki,T.,Rundgren,M.,Moldeus,P.,Lindros,K.O.,1993.Zonation ofacetaminophenmetabolismandcytochromeP4502E1-mediatedtoxicity studiedinisolatedperiportalandperivenoushepatocytes.Biochem.Pharmacol. 45,1251–1259.
Armitage,P.,1985.Multistagemodelsofcarcinogenesis.Environ.HealthPerspect. 63,195–201.
Armitage,J.M.,Wania,F.,Arnot,J.A.,2014.Applicationofmassbalancemodelsand thechemicalactivityconcepttofacilitatetheuseofinvitrotoxicitydataforrisk assessment.Environ.Sci.Technol.48,9770–9779.
Bassingthwaighte,J.B.,2000.Strategiesforthephysiomeproject.Ann.Biomed.Eng. 28,1043–1058.
Beaudouin,R.,Micallef,S.,Brochot,C.,2010.Astochasticwhole-body physiologicallybasedpharmacokineticmodeltoassesstheimpactof inter-individualvariabilityontissuedosimetryoverthehumanlifespan.Regul. Toxicol.Pharm.57,103–116.
Berthold,M.R.,Cebron,N.,Dill,F.,Gabriel,T.R.,Kötter,T.,Meinl,T.,Ohl,P.,Sieb,C., Thiel,K.,Wiswedel,B.,2007.KNIME:TheKonstanzInformationMiner.In StudiesinClassification,DataAnalysis,andKnowledgeOrganization(GfKL 2007).Springer.
Bessems,J.G.,Loizou,G.,Krishnan,K.,ClewellIII,H.J.,Bernasconi,C.,Bois,F.Y., Coecke,S.,Collnot,E.M.,Diembeck,W.,Farcal,L.R.,etal.,2014.PBTKmodelling platformsandparameterestimationtoolstoenableanimal-freerisk assessment:recommendationsfromajointEPAA–EURLECVAMADME workshop.Regul.Toxicol.Pharm.68,119–139.
Bois,F.Y.,2009.Physiologically-basedmodellingandpredictionofdrug interactions.BasicClin.Pharmacol.Toxicol.106,154–161.
Bois,F.Y.,Jamei,M.,Clewell,H.J.,2010.PBPKmodellingofinter-individualvariability inthepharmacokineticsofenvironmentalchemicals.Toxicology278,256–267.
Coecke,S.,Pelkonen,O.,BatistaLeite,S.,Bernauer,U.,Bessems,J.,Bois,F.Y., Gundert-Remy,U.,Loizou,G.,Testai,E.,Zaldívar,J.M.,2012.Toxicokineticsasakeytothe integratedtoxicityriskassessmentbasedprimarilyonnon-animalapproaches. Toxicol.InVitro27,1570–1577.
Corley,R.A.,2010.PharmacokineticsandPBPKmodels.In:McQueen,C.A.(Ed.), ComprehensiveToxicology,vol.12.Elsevier,Oxford,pp.27–58.
Crean,D.,Bellwon,P.,Aschauer,L.,Limonciel,A.,Moenks,K.,Hewitt,P.,Schmidt,T., Herrgen,K.,Dekant,W.,Lukas,A.,etal.,2014.Developmentofaninvitrorenal epithelialdiseasestatemodelforxenobiotictoxicitytesting.Toxicol.InVitro30, 128–137.
Crump,K.S.,Howe,R.B.,1984.Themultistagemodelwithatime-dependentdose pattern:applicationstocarcinogenicriskassessment.RiskAnal.4,163–176.
Csajka,C.,Verotta,D.,2006.Pharmacokinetic-pharmacodynamicmodelling: historyandperspectives.J.Pharmacokinet.Pharmacodyn.33,227–279.
Dash,A.,Inman,W.,Hoffmaster,K.,Sevidal,S.,Kelly,J.,Obach,R.S.,Griffith,L.G., Tannenbaum,S.R.,2009.Livertissueengineeringintheevaluationofdrug safety.ExpertOpin.DrugMetab.Toxicol.5,1159–1174.
DiazOchoa,J.G.,Bucher,J.,Péry,A.R.R.,ZaldivarComenges,J.M.,Niklas,J.,Mauch,K., 2013.Amulti-scalemodelingframeworkforindividualized,spatiotemporal predictionofdrugeffectsandtoxicologicalrisk.Front.Pharmacol.3,204.
Dokoumetzidis,A.,Macheras,P.,2003.Amodelfortransportanddispersioninthe circulatorysystembasedonthevascularfractaltree.Ann.Biomed.Eng.31,284– 293.
Dourson,M.L.,Felter,S.P.,1997.Route-to-routeextrapolationofthetoxicpotencyof MTBE.RiskAnal.17,717–725.
Drasdo,D.,Hoehme,S.,Hengstler,J.G.,2014.Howpredictivequantitativemodelling oftissueorganisationcaninformliverdiseasepathogenesis.J.Hepatol.61,951– 956.
Gajewska,M.,Worth,A.,Urani,C.,Briesen,H.,Schramm,K.-W.,2014.Applicationof physiologically-basedtoxicokineticmodellinginoral-to-dermalextrapolation ofthresholddosesofcosmeticingredients.Toxicol.Lett.227,189–202.
Gajewska,M.,Paini,A.,SalaBenito,J.V.,Burton,J.,Worth,A.,Urani,C.,Briesen,H., Schramm,K.-W.,2015.Invitro-to-invivocorrelationoftheskinpenetration, liverclearanceandhepatotoxicityofcaffeine.FoodChem.Toxicol.75,39–49.
Gebhardt,R.,Matz-Soja,M.,2014.Liverzonation:novelaspectsofitsregulationand itsimpactonhomeostasis.WorldJ.Gastroenterol.20,8491.
Geenen,S.,Taylor,P.N.,Snoep,J.L.,Wilson,I.D.,Kenna,J.G.,Westerhoff,H.V.,2012. Systemsbiologytoolsfortoxicology.Arch.Toxicol.86,1251–1271.
Geenen,S.,Yates,J.W.,Kenna,J.G.,Bois,F.Y.,Wilson,I.D.,Westerhoff,H.V.,2013. Multiscalemodellingapproachcombiningakineticmodelofglutathione metabolismwithPBPKmodelsofparacetamolandthepotential glutathione-depletionbiomarkersophthalmicacidand5-oxoprolineinhumansandrats. Integr.Biol.5,877–888.
Gerrity,T.R.,Henry,C.J.,1990.PrinciplesofRoute-to-RouteExtrapolationforRisk Assessment.Elsevier,NewYork.
Gibaldi,M.,Perrier,D.,1982.Pharmacokinetics.MarcelDekker,NewYork.
Grisham,J.W.,2009.Organizationalprinciplesoftheliver.In:Arias,I.M.(Ed.),The Liver.JohnWiley&Sons,Ltd.,Chichester,UK,pp.1–15.
Guillouzo,A.,Corlu,A.,Aninat,C.,Glaise,D.,Morel,F.,Guguen-Guillouzo,C.,2007. ThehumanhepatomaHepaRGcells:ahighlydifferentiatedmodelforstudiesof livermetabolismandtoxicityofxenobiotics.Chem.Biol.Interact.168,66–73.
Hamon,J.,Jennings,P.,Bois,F.Y.,2014.Systemsbiologymodelingofomicsdata: effectofcyclosporineaontheNrf2pathwayinhumanrenalcells.BMCSyst. Biol.8,76.
Hamon,J.,Renner,M.,Jamei,M.,Lukas,A.,Kopp-Schneider,A.,Bois,F.Y.,2015. Quantitativeinvitrotoinvivoextrapolationoftissuestoxicity.Toxicol.InVitro 30,203–216.
Holzhütter,H.-G.,Drasdo,D.,Preusser,T.,Lippert,J.,Henney,A.M.,2012.Thevirtual liver:amultidisciplinary,multilevelchallengeforsystemsbiology.Wiley Interdiscip.Rev.Syst.Biol.Med.4,221–235.
Hunt,C.A.,Ropella,G.E.P.,2008.UsinganInSilicoLivertoEvaluateaHepatic EnzymeInductionMechanism.EngineeringinMedicineandBiologySociety (IEEE),pp.2415–2418.
Hunter,P.J.,Borg,T.K.,2003.Integrationfromproteinstoorgans:thePhysiome Project.Nat.Rev.Mol.CellBiol.4,237–243.
Jamei,M.,Marciniak,S.,Feng,K.R.,Barnett,A.,Tucker,G.,Rostami-Hodjegan,A., 2009.TheSimcyppopulation-basedADMEsimulator.ExpertOpin.DrugMetab. Toxicol.5,211–223.
Jones,H.M.,Barton,H.A.,Lai,Y.,Bi,Y.A.,Kimoto,E.,Kempshall,S.,Tate,S.C., El-Kattan,A.,Houston,J.B.,Galetin,A.,etal.,2012.Mechanisticpharmacokinetic modelingforthepredictionoftransporter-mediateddispositioninhumans fromsandwichculturehumanhepatocytedata.DrugMetab.Dispos.40,1007– 1017.
Jusko,W.J.,2013.Movingfrombasictowardsystemspharmacodynamicmodels.J. Pharm.Sci.102,2930–2940.
Kimura,H.,Ikeda,T.,Nakayama,H.,Sakai,Y.,Fujii,T.,2015.Anon-chipsmall intestine-livermodelforpharmacokineticstudies.J.Lab.Autom.20,265–273.
Krauss,M.,Schaller,S.,Borchers,S.,Findeisen,R.,Lippert,J.,Kuepfer,L.,2012. Integratingcellularmetabolismintoamultiscalewhole-bodymodel.PLoS Comput.Biol.8,e1002750.
Kroes,R.,Renwick,A.G.,Feron,V.,Galli,C.L.,Gibney,M.,Greim,H.,Guy,R.H., Lhuguenot,J.C.,deSandt,J.J.M.V.,2007.Applicationofthethresholdof toxicologicalconcern(TTC)tothesafetyevaluationofcosmeticingredients. FoodChem.Toxicol.45,2533–2562.
Lodish,H.,2000.MolecularCellBiology.Freeman,NewYork,NY.
Martin,S.A.,McLanahan,E.D.,Bushnell,P.J.,Hunter,E.S.,El-Masri,H.,2015.Species extrapolationoflife-stagephysiologically-basedpharmacokinetic(PBPK) modelstoinvestigatethedevelopmentaltoxicologyofethanolusinginvitroto invivo(IVIVE)methods.Toxicol.Sci.143,512–535.
Mitchell,P.,Moyle,J.,1969.EstimationofmembranepotentialandpHdifference acrossthecristaemembraneofratlivermitochondria.Eur.J.Biochem.7,471– 484.
Moolgavkar,S.H.,KnudsonJr.,A.G.,1981.Mutationandcancer:amodelforhuman carcinogenesis.J.Natl.CancerInst.66,1037–1052.
Mostrag-Szlichtyng,A.,Worth,A.,2010.ReviewofQSARModelsandSoftwareTools forPredictingBiokineticProperties.JRCInstituteforHealthandConsumer Protection,Luxembourg.
Munro,I.C.,Ford,R.A.,Kennepohl,E.,Sprenger,J.G.,1996.Correlationofstructural classwithno-observed-effectlevels:aproposalforestablishingathresholdof concern.FoodChem.Toxicol.34,829–867.
NationalResearchCouncil(NRC),2007.ToxicityTestinginthe21stCentury:AVision andAStrategy.NationalAcademyPress,Washington,DC.
Nicholls,D.G.,Ward,M.W.,2000.Mitochondrialmembranepotentialandneuronal glutamateexcitotoxicity:mortalityandmillivolts.TrendsNeurosci.23, 166–174.
Nielsen,E.,Østergaard,G.,Larsen,J.C.,2008.ToxicologicalRiskAssessmentof Chemicals.InformaHealthcare,NewYork.
Oinonen,T.,Lindros,K.O.,1998.ZonationofhepaticcytochromeP-450expression andregulation.Biochem.J.329,17–35.
Ostrovidov,S.,Jiang,J.,Sakai,Y.,Fujii,T.,2004.Membrane-basedPDMS microbioreactorforperfused3Dprimaryrathepatocytecultures.Biomed. Microdevices6,279–287.
Ouattara,D.A.,Choi,S.H.,Sakai,Y.,Péry,A.R.,Brochot,C.,2011.Kineticmodellingof invitrocell-basedassaystocharacterizenon-specificbindingsandADME processesinastaticandaperfusedfluidicsystem.Toxicol.Lett.205,310–319.
Péry,A.,Brochot,C.,Zeman,F.A.,Mombelli,E.,Desmots,S.,Pavan,M.,Fioravanzo,E., Zaldívar,J.-M.,2013.Predictionofdose-hepatotoxicresponseinhumansbased ontoxicokinetic/toxicodynamicmodelingwithorwithoutinvivodata:acase studywithacetaminophen.Toxicol.Lett.220,26–34.
Paini,A.,Punt,A.,Scholz,G.,Gremaud,E.,Spenkelink,B.,Alink,G.,Schilter,B.,van Bladeren,P.J.,Rietjens,I.M.C.M.,2012.InvivovalidationofDNAadduct formationbyestragoleinratspredictedbyphysiologicallybasedbiodynamic modelling.Mutagenesis27,653–663.
Pang,K.S.,Rowland,M.,1977.Hepaticclearanceofdrugs.I.Theoretical considerationsofawell-stirredmodelandaparalleltubemodel.Influenceof hepaticbloodflow,plasmaandbloodcellbinding,andthehepatocellular enzymaticactivityonhepaticdrugclearance.J.Pharmacokinet.Biopharm.5, 625–653.
Pang,K.S.,Weiss,M.,Macheras,P.,2007.Advancedpharmacokineticmodelsbased onorganclearance,circulatory,andfractalconcepts.AAPSJ.9,E268–E283.
Peters,S.A.,2011.PhysiologicallyBasedPharmacokinetic(pbpk)Modelingand Simulations:Principles,Methods,andApplicationsinthePharmaceutical Industry.Wiley,Hoboken,NJ.
Poggesi,I.,Snoeys,J.,VanPeer,A.,2014.Thesuccessesandfailuresofphysiologically basedpharmacokineticmodeling:thereisroomforimprovement.ExpertOpin. DrugMetab.Toxicol.10,631–635.
Pomponio,G.,Savary,C.C.,Parmentier,C.,Bois,F.,Guillouzo,A.,Romanelli,L., Richert,L.,DiConsiglio,E.,Testai,E.,2014.Invitrokineticsofamiodaroneandits majormetaboliteintwohumanlivercellmodelsafteracuteandrepeated treatments.Toxicol.InVitro30,36–51.
Quignot,N.,Hamon,J.,Bois,F.Y.,2014.Extrapolatinginvitroresultstopredict humantoxicity.In:Bal-Price,A.,Jennings,P.(Eds.),InVitroToxicologySystems. SpringerScience,NewYork,pp.531–550.
RDevelopmentCoreTeam,2013.R:ALanguageandEnvironmentforStatistical Computing.RFoundationforStatisticalComputing,Vienna,Austria.
Reitz,R.H.,MacDougal,J.N.,Himmelstein,M.W.,Nolan,R.J.,Schumann,A.M.,1988. Physiologicallybasedpharmacokineticmodelingwithmethylchloroform: implicationsforinterspecies,highdose/lowdose,anddoseroute extrapolations.Toxicol.Appl.Pharmacol.95,185–199.
Rietjens,I.M.C.M.,Louisse,J.,Punt,A.,2011.Tutorialonphysiologicallybasedkinetic modelinginmolecularnutritionandfoodresearch.Mol.Nutr.FoodRes.55, 941–956.
Ropella,G.E.,Hunt,C.A.,2010.Cloudcomputingandvalidationofexpandablein silicolivers.BMCSyst.Biol.4,168.
Soldatow,V.Y.,LeCluyse,E.L.,Griffith,L.G.,Rusyn,I.,2013.Invitromodelsforliver toxicitytesting.Toxicol.Res.2,23–39.
Teng,S.,Barcellini-Couget,S.,Beaudouin,R.,Brochot,C.,Desousa,G.,Rahmani,R., Pery,A.R.R.,2015.BK/TDmodelsforanalyzinginvitroimpedancedataon cytotoxicity.Toxicol.Lett.235,96–106.
Tollefsen,K.E.,Scholz,S.,Cronin,M.T.,Edwards,S.W.,deKnecht,J.,Crofton,K., Garcia-Reyero,N.,Hartung,T.,Worth,A.,Patlewicz,G.,2014.Applyingadverse outcomepathways(AOPs)tosupportintegratedapproachestotestingand assessment(IATA).Regul.Toxicol.Pharm.70,629–640.
Trapp,S.,Rosania,G.R.,Horobin,R.W.,Kornhuber,J.,2008.Quantitativemodelingof selectivelysosomaltargetingfordrugdesign.Eur.Biophys.J.37,1317–1328.
Truisi,G.L.,Consiglio,E.D.,Parmentier,C.,Savary,C.C.,Pomponio,G.,Bois,F.,Lauer, B.,Jossé,R.,Hewitt,P.G.,Mueller,S.O.,etal.,2015.Understandingthebiokinetics ofibuprofenaftersingleandrepeatedtreatmentsinratandhumaninvitroliver cellsystems.Toxicol.Lett.233,172–186.
UnitedStatesEnvironmentalProtectionAgency(USEPA),2014.NextGeneration RiskAssessment:IncorporationofRecentAdvancesinMolecular,
Computational,andSystemsBiology.UnitedStatesEnvironmentalProtection Agency,WashingtonDC.
Vollmar,B.,Menger,M.D.,2009.Thehepaticmicrocirculation:mechanistic contributionsandtherapeutictargetsinliverinjuryandrepair.Physiol.Rev.89, 1269–1339.
Wambaugh,J.,Shah,I.,2010.Simulatingmicrodosimetryinavirtualhepaticlobule. PLoSComput.Biol.6,e1000756.
Willmann,S.,Höhn,K.,Edginton,A.,Sevestre,M.,Solodenko,J.,Weiss,W.,Lippert,J., Schmitt,W.,2007.Developmentofaphysiology-basedwhole-bodypopulation modelforassessingtheinfluenceofindividualvariabilityonthe
pharmacokineticsofdrugs.J.Pharmacokinet.Pharmacodyn.34,401–431.
Wilmes,A.,Limonciel,A.,Aschauer,L.,Moenks,K.,Bielow,C.,Leonard,M.O.,Hamon, J.,Carpi,D.,Ruzek,S.,Handler,A.,etal.,2013.Integratedomicprofilingreveals novelinsightsofcyclosporineAinducedcellularstress.J.Proteom.79,180–194.
Worth,A.,Fuart-Gatnik,M.,Lapenna,S.,LoPiparo,E.,Mostrag-Szlichtyng,A., Serafimova,R.,2011.TheUseofComputationalMethodsintheToxicological AssessmentofChemicalsinFood:CurrentStatusandFutureProspects.JRC InstituteforHealthandConsumerProtection,Luxembourg.
Worth,A.,Cronin,M.,Enoch,S.,Fioravanzo,E.,Fuart-Gatnik,M.,Pavan,M.,Yang,C., 2012.ApplicabilityoftheThresholdofToxicologicalConcern(TTC)Approachto Cosmetics—PreliminaryAnalysis.JRCInstituteforHealthandConsumer Protection,Luxembourg.
Xing,J.Z.,Zhu,L.,Gabos,S.,Xie,L.,2006.Microelectroniccellsensorassayfor detectionofcytotoxicityandpredictionofacutetoxicity.Toxicol.InVitro20, 995–1004.
Zaldívar,J.M.,Mennecozzi,M.,Rodrigues,R.,Bouhifd,M.,2010.ABiology-based DynamicApproachfortheModellingofToxicityinCell-basedAssays.PartI: FateModelling.JRCInstituteforHealthandConsumerProtection,Luxembourg.
Zaldívar,J.,Mennecozzi,M.,Macko,P.,Rodrigues,R.,Bouhifd,M.,Baraibar,J.,2011.A Biology-basedDynamicApproachfortheModellingofToxicityinCellAssays. PartII:ModelsforCellPopulationGrowthandToxicity.JRCInstituteforHealth andConsumerProtection,Luxembourg.
