Human urinary biomarkers of dioxin exposure: Analysis by metabolomics and biologically driven data dimensionality reduction

(1)

Article

Reference

Human urinary biomarkers of dioxin exposure: Analysis by metabolomics and biologically driven data dimensionality reduction

JEANNERET, Fabienne, et al.

Abstract

Untargeted metabolomic approaches offer new opportunities for a deeper understanding of the molecular events related to toxic exposure. This study proposes a metabolomic investigation of biochemical alterations occurring in urine as a result of dioxin toxicity. Urine samples were collected from Czech chemical workers submitted to severe dioxin occupational exposure in a herbicide production plant in the late 1960s. Experiments were carried out with ultra-high pressure liquid chromatography (UHPLC) coupled to high-resolution quadrupole time-of-flight (QTOF) mass spectrometry. A chemistry-driven feature selection was applied to focus on steroid-related metabolites. Supervised multivariate data analysis allowed biomarkers, mainly related to bile acids, to be highlighted. These results supported the hypothesis of liver damage and oxidative stress for long-term dioxin toxicity. As a second step of data analysis, the information gained from the urine analysis of Victor Yushchenko after his poisoning was examined. A subset of relevant urinary markers of acute dioxin toxicity from this extreme phenotype, including glucuro- and [...]

JEANNERET, Fabienne, et al . Human urinary biomarkers of dioxin exposure: Analysis by metabolomics and biologically driven data dimensionality reduction. Toxicology Letters , 2013, vol. 230, no. 2, p. 234-243

PMID : 24201000

DOI : 10.1016/j.toxlet.2013.10.031

Available at:

http://archive-ouverte.unige.ch/unige:33236

Disclaimer: layout of this document may differ from the published version.

(2)

ToxicologyLettersxxx (2013) xxx–xxx

ContentslistsavailableatScienceDirect

Toxicology Letters

j o ur na l h o me p a g e : w w w . e l s e v i e r . c o m / l o c a t e / t o x l e t

Human urinary biomarkers of dioxin exposure: Analysis by

metabolomics and biologically driven data dimensionality reduction

Fabienne Jeanneret

^a,b,c,1

, Julien Boccard

^d,1

, Flavia Badoud

^c

, Olivier Sorg

^a,c

, David Tonoli

^a

, Daniela Pelclova

^e

, Stepanka Vlckova

^e

, Douglas N. Rutledge

^d

, Caroline F. Samer

^c,f

,

Denis Hochstrasser

^a,g

, Jean-Hilaire Saurat

^c

, Serge Rudaz

^b,c,∗

aDepartmentofHumanProteinSciences,FacultyofMedicine,UniversityofGeneva,Geneva,Switzerland

bSchoolofPharmaceuticalSciences,UniversityofGeneva,UniversityofLausanne,Geneva,Switzerland

cSwissCentreforAppliedHumanToxicology,UniversityofGeneva,Geneva,Switzerland

dAgroParisTechUMR1145,LaboratoiredeChimieAnalytique,Paris,France

eCharlesUniversityPrague,DepartmentofOccupationalMedicineoftheFirstFacultyofMedicineandGeneralUniversityHospital,Prague,CzechRepublic

fClinicalPharmacologyandToxicology,GenevaUniversityHospitals,Geneva,Switzerland

gDepartmentofGeneticandLaboratoryMedicine,GenevaUniversityHospitals,Geneva,Switzerland

h i g h l i g h t s

•Highlight of biomarkers of dioxin exposurebymetabolomics.

•Datadimensionalityreductiondone by chemical and biological knowledge.

•Steroidandbileacidpatternaremod- iﬁedbydioxinexposure.

g r a p h i c a l a b s t r a c t

a r t i c l e i n f o

Articlehistory:

Available online xxx Keywords:

Steroidomics Extremephenotype Dimensionalityreduction Dioxin

Toxicity Biomarkers

a b s t r a c t

Untargetedmetabolomicapproachesoffernewopportunitiesforadeeperunderstandingofthemolec- ulareventsrelatedtotoxicexposure.Thisstudyproposesametabolomicinvestigationofbiochemical alterationsoccurringinurineasaresultofdioxintoxicity.UrinesampleswerecollectedfromCzech chemicalworkerssubmittedtoseveredioxinoccupationalexposureinaherbicideproductionplantin thelate1960s.Experimentswerecarriedoutwithultra-highpressureliquidchromatography(UHPLC) coupledtohigh-resolutionquadrupoletime-of-ﬂight(QTOF)massspectrometry.Achemistry-drivenfea- tureselectionwasappliedtofocusonsteroid-relatedmetabolites.Supervisedmultivariatedataanalysis allowedbiomarkers,mainlyrelatedtobileacids,tobehighlighted.Theseresultssupportedthehypoth- esisofliverdamageandoxidativestressforlong-termdioxintoxicity.Asasecondstepofdataanalysis, theinformationgainedfromtheurineanalysisofVictorYushchenkoafterhispoisoningwasexamined.

Asubsetofrelevanturinarymarkersofacutedioxintoxicityfromthisextremephenotype,including glucuro-andsulfo-conjugatedendogenoussteroidmetabolitesandbileacids,wasassessedforitsability todetectlong-termeffectsofexposure.Themetabolomicstrategypresentedinthisworkallowedthe determinationofmetabolicpatternsrelatedtodioxineffectsinhumanandthediscoveryofhighlypre- dictivesubsetsofbiologicallymeaningfulandclinicallyrelevantcompounds.Theseresultsareexpected

∗Correspondingauthorat:SchoolofPharmaceuticalSciences,UniversityofGeneva,20Bdd’Yvoy,1211Geneva4,Switzerland.

Tel.:+41223796572;fax:+41223796808.

E-mailaddress:serge.rudaz@unige.ch(S.Rudaz).

1 Bothauthorscontributedequallytothiswork.

http://dx.doi.org/10.1016/j.toxlet.2013.10.031

(3)

2 F.Jeanneretetal./ToxicologyLettersxxx (2013) xxx–xxx

toprovidevaluableinformationforadeeperunderstandingofthemoleculareventsrelatedtodioxin toxicity.Furthermore,itpresentsanoriginalmethodologyofdatadimensionalityreductionbyusing extremephenotypeasaguidetoselectrelevantfeaturespriortodatamodeling(biologicallydrivendata reduction).

1. Introduction

Dioxinsaretoxicandextremelypersistentandastheydegrade veryslowlyandaccumulatethroughthefoodchain,manypeople mayberepeatedlyexposedtolowamountsofdioxin-likecom- pounds,makingthisaquestionofpublichealthconcern(Giesetal., 2007;Steenland etal., 2001).Theirinﬂuenceon humanhealth atlow levelsremainspoorly understood.Theeffects of2,3,7,8- tetrachlorodibenzo-p-dioxin (TCDD) are mainlymediated byits bindingtothearylhydrocarbonreceptor(AhR)(BockandKohle, 2009).Hepaticdamages,neurologicaleffects,diabetes,atheroscle- rosisandhypertensionwereassociatedwithhumanacutedioxins exposure.TCDDhasbeenclassiﬁedasahumancarcinogenbythe WorldHealthOrganisationandtheUSNationalToxicologicalPro- gram(IARC,1997).Thisfactiscontroversial,becauseallevidences forarelationshipbetweenTCDDandhumancarcinogenesisare indirect.TCDDhasaverylongbiologicalhalf-lifeandisnotmuta- genic,andnostudycoulddemonstratethatTCDDisacarcinogen inhumans(Boffettaetal.,2011).

Themostseverehumanintoxicationsoccurredafterindustrial accidentsduetouncontrolledformationofTCDD,e.g.Seveso(Italy) in1976(Bertazzietal.,2001),Neratovice(CentralBohemiainthe formerCzechoslovakia)in1965–1968(Pelclovaetal.,2001)and Ludwigshafen(Germany)in1953(Zoberetal.,1997).Themassive spreadingofAgentOrange,aTCDD-contaminatedherbicide,during theVietnamWaralsoledtosevereintoxications(Gochfeld,2001).

Additional TCDD exposure cases were reported, including two womenexposedinVienna(Austria)inatextileresearchinstitute in1997–1998(Geusauetal.,2001)andthedeliberatepoisoning ofUkrainianpresidentVictorYushchenko(VY)inSeptember2004 (Sauratetal.,2012;Sorgetal.,2009).AllthesecasesofacuteTCDD intoxicationsresultedinchloracne.Thisskinlesionplaystherole ofasentinelsignfordioxinordioxin-likecompoundsexposure, asthereiscurrentlynootherreliablehumanbiomarkerofsuch intoxication.Theconﬁrmationisusuallydonebydirectdioxinmea- surementinblood,aprotocolwhichisnotroutinelyimplemented inmosthospitallaboratories.

Relevantandsensitivebiomarkersarethereforehighlydesir- abletodetecttheonsetofdioxintoxicitybeforetheappearanceof clinicalmanifestationsandmetabolomicsconstitutesapromising approachforthatpurpose.Metabolomicsisablossomingresearch ﬁeldwhichhasbeendemonstratedasavaluableapproachtomon- itorthe chemical diversity of complex biological systems from aholisticangle(Goodacreet al.,2004).It aimsatdepictingthe moleculareventsassociatedwithbiologicalprocessescharacter- izinglivingsystems,atthetissue,theorganorthewholeorganism level.Thisdescriptioncanberelatedtophysiologicalconditionsor tothereactiontopathological,environmentalordevelopmental perturbations(NicholsonandLindon,2008).Thankstoitsability tomonitoralterationsofendogenousmetaboliteslevels,butalso xenobioticsandtheirbiotransformationproducts,metabolomics wasquicklypromotedasarelevanttoolforassessingbiochemical phenotypesrelatedtotoxicity(Lindonetal.,2003).

Duetotheever-increasingsensitivityandresolutionofmod- ernanalyticalplatformscommonlyusedinmetabolomics,suchas massspectrometry(MS),akeychallengewhenachievinguntar- getedanalysesistreatingtheveryhighnumberofrecordedsignals (Boccardetal.,2010).Biologicallyrelevanthypothesesareharder toﬁndwhenhandlingthesemassivelymultivariatedata.Variable

selectionintendstoreducethesizeofthedatabyselectingsubsets ofrelevantvariables,i.e.biomarkersabletorepresentthesalient characteristicsofthephenomenonunderstudy.Knowledge-based dimensionalityreductionisveryusefultoreducethesizeofthe hypothesisspaceandprovidemoreparsimonioussolutionstobio- logicalproblems(Mitchell,1997).

TodateonlyfewmetabolomicstudiesassessingTCDDtoxicity basedoncelllinesorrodentmodelswerereported.Mostresults highlightedchangesrelatedtoaminoacidsandlipidmetabolism includingfattyacidsorbileacids(Forgacsetal.,2012;Linetal., 2011a,b;Ruiz-Aracamaetal.,2011).Ametabolomicinvestigation ofserumsamplesfromworkersexposedtoTCDDwasachieved recentlybutnomajoralterationcouldbedetected(SaberiHosnijeh etal.,2013).

Findingaclinicallyrelevantmetabolicpatternofdioxintoxicity inhumansconstitutesakeyissueintheperspectiveofdiagnos- ticandprognosticproceduresandlong-termexposureisprobably themostrelevantsituationinthecontextofhumanhealthcare.

Theobjectiveofthisworkwastoinvestigatelong-termeffectsof dioxintoxicity relatedtooccupational exposuretoTCDD.Urine samples from a cohort of Czech workers were analyzed on a platformhyphenatingultra-highpressureliquidchromatography (UHPLC) to quadrupole time-of-flight (QTOF) mass spectrometry. Since dioxins have been regularly described as perturbing thesteroidprofile,avalidatedmethoddedicatedtotheanalysis ofglucuro-and sulfo-conjugatedformsof someandrogens was applied(Badoudetal.,2011).AstheQTOFtechnologyallowsthe untargeted screening of metabolites, it could help findingnew biomarkers,asreportedrecentlyfortestosteroneintakeinthecon- textofanti-doping(Boccardetal.,2011).Anoriginaldatatreatment strategyusingbiologicalknowledgeaswellasstatisticaltoolswas implementedtoobtainasmallsubsetofbiomarkersabletochar- acterizecasesofTCDDintoxication.Forthispurpose,anextreme phenotypecasewasusedasareference.Phenotypicextremesare expectedtoprovidecharacteristicpatternsoftoxicityforthedetec- tionofreliablebiomarkers.Themostcontributingfeaturesdetected duringtheinvestigationofacaseofacuteintoxication(extreme phenotype) wereselected for furtheranalysisof thelong-term effects ofdioxin exposure. Therefore,thedata analysisstrategy integratedchemicalandphenotypicinformationforthedetection ofrelevanttoxicitybiomarkers.

2. Experimental

2.1. Chemicalsandreagents

5␣-Androstan-3␣-ol-17-one sulfate sodium salt (androsterone sulfate),4- androsten-17␤-ol-3-onesulfatesodiumsalt(testosteronesulfate),4-androsten- 17␣-ol-3-one sulfate sodium salt (epitestosterone sulfate), 5␤-androstan-3␣- ol-17-one sulfate sodium salt (etiocholanolone sulfate), 5-androsten-3␤-ol- 17-one sulfate sodium salt (dehydroepiandrosterone sulfate), 4-androsten- 17␤-ol-3-oneglucosiduronate(testosteroneglucuronide),4-androsten-17␣-ol-3- oneglucosiduronate(epitestosterone glucuronide),5␤-androstan-3␣-ol-17-one glucosiduronate (etiocholanolone glucuronide), 5-androsten-3␤-ol-17-one glucosiduronate(dehydroepiandrosteroneglucuronide)and5␣-androstan-17␤-ol-3- oneglucosiduronate(dihydrotestosteroneglucuronide)werepurchasedfromSter- aloids(Newport,RI,USA).4-Androsten-17␣-ol-3-onesulfate(epitestosteronesul- fate),5␣-androstan-3␣-ol-17-one-3␣-d-glucuronide(androsteroneglucuronide), [16,16,17␣-²H3]androst-4-en-17␣-ol-3-onesulfate triethylammonium salt (TS- d3), [16,16,17␤-²H3] androst-4-en-17␤-ol-3-one sulfate (ES-d3), [2,2,3␤,4,4-

2H5]5␤-androstan-3␣-ol-17-onesulfate(EtioS-d5),[2,2,4,4-²H4]5␣-androstan-3␣- ol-17-onesulfate(AS-d4),[16,16,17␣-²H3]androst-4-en-17␤-ol-3-oneglucuronide (TG-d3), [16,16,17␣-²H3]androst-4-en-17␣-ol-3-one glucuronide (EG-d3), and

(4)

F.Jeanneretetal./ToxicologyLettersxxx (2013) xxx–xxx 3

[2,2,4,4-²H4]5␣-androstan-3␣-ol-17-one glucuronide (AG-d4) were purchased from LGC Standard GmbH (Wesel, Germany). Water, acetonitrile (ACN) and methanol (MeOH) were ULC–MS quality and were obtained from Biosolve (Valkenswaald,TheNetherlands).Formicacid(FA)andammoniumhydroxidewere purchased,respectively,fromMerck(Darmstadt,Germany)andfromSigma–Aldrich (Buchs,Switzerland).

2.2. Urinesamples 2.2.1. Czechcohort

Anuncontrolleddecompositionreactionoccurredduringtheproductionofthe herbicidetrichlorophenolaceticacidinaCzechplantduringtheyearsbetween 1965and1968.ItleadedtotheformationofTCDD.TheTCDDbloodlevelsof80 workerswereestimatedtoabout5000pg/gbloodlipidatthetimeofexposure (Pelclovaetal.,2001).Amongthesepatients,11arestillmedicallyfollowedbythe CharlesUniversityofPragueinCzechRepublic(Pelclovaetal.,2002,2006,2007, 2009,2011;Urbanetal.,2007).24hurinesampleswerecollectedin2011andkept at−80^◦C.These11workerswerebornbetween1940and1946andwereexposed todioxinforperiodsfrom10daysto23months(Pelclovaetal.,2009).Samples fromhealthyvolunteersmatchedforage(65–70yearsold)werealsocollectedto constitutethecontrolgroup(CTRL65group,n=11).Thesesampleswereprovidedby theGenevaUniversityHospital(anonymousurineswithnon-pathologicalresults, protocolnumberCE11-111approvedbytheethicalcommission).

2.2.2. VictorYushchenko

VYwaspoisonedthe6thSeptember2004.HisTCDDbloodlevelwasabout 108,000pg/gfatfourmonthsafterintoxication(Sorgetal.,2009).Variousurine sampleswerecollectedinUkraineorinSwitzerlandthemonthsandyearsfollow- ingthepoisoning(Sauratetal.,2012).Thesesampleswerethenkeptat−80^◦C inGeneva.Elevensampleswereavailableforanalysis:theﬁrstsampleswerecol- lectedabout15monthsafterthepoisoningandthelastsampleabout3yearslater.

ThesampleswerelabeledwiththeabbreviationVYfollowedbythenumberof monthsafterintoxication.Thedayofintoxication(06.09.2004)wassettotime0.

Samples:VY-15.5(22.12.2005),VY-21.5(23.06.2006),VY-22.9(05.08.2006),VY-23 (08.08.2006),VY-28.5a(20.01.2007),VY-28.5b(21.01.2007),VY-28.5c(22.01.2007), VY-32.2(14.05.2007), VY-34.1(11.07.2007), VY-39.3(14.12.2007) andVY-46.1 (10.07.2008).SamplesVY-28.5atoVY-28.5ccorrespondedtothreecollectsof24h atthreesuccessivedays.Samplesfromhealthyvolunteersmatchedforage(50–55 yearsold)werealsocollectedtoconstitutethecontrolgroup(CTRL50group,n=12).

ThesesampleswereprovidedbytheGenevaUniversityHospital(anonymousurines withnon-pathologicalresults,protocolnumberCE11-111approvedbytheethical commission).

2.3. Analysisofurinesamples

Urinesampleswereanalyzedintriplicatebyusingasolidphaseextractionpro- tocolpriortoUHPLC–MS.AthoroughdescriptioncanbefoundinBadoudetal., 2011.Brieﬂy,1mLofurinewasdilutedwith1mLof2%FAspikedwith10␮Lofthe internalstandardsolution(ES-d3,EtioS-d5,AS-d4,TG-d3,EG-d3andAG-d4).Oasis HLBcartridgesof30mg(96-wellplate)wereconditionedwith500␮LofMeOH andequilibratedwith1mLof2%FA.Thedilutedurinesampleswereaddedtothe cartridges.Washingwasdoneby1mLof2%FA,followedby1mLofa5%ammo- niumhydroxide/MeOH(90/10,v/v)solution.Metaboliteswereelutedwith500␮L ofMeOH/water(40/60,v/v)andthenevaporatedtodrynessunderazotestream.

Thedryresidueswereresuspendedin100␮LofACN/water(30/70,v/v).

2.4. Instrumentalconditions

AnAcquityUHPLC-QTOF-MS(Xevo^TM,Waters)systemwasusedfortheanalysis ofsteroidandothermetabolitesinurine.CompoundswereseparatedonanAcquity UPLCcolumn(BEHC18150mm×2.1mm,1.7␮m),precededbyaVanGuardpre- column(BEHC18,5×2.1mm,1.7␮m).AgradientofmobilephaseA(0.1%FAin water)andmobilephaseB(0.1%FAinACN)wasusedlinearlyfrom5%to37%B over25min,followedbyawashingstepof95%Bfor3minandare-equilibrationof 8min.Thesamplevolumeinjectedwas10␮Linthefullloopmodeandthesamples weremaintainedat4^◦Cintheautosampler.

TheQTOFXevowasequippedwithanelectrosprayionization(ESI)source operatedinthenegativemode.Followingoperatingconditionswereused:thedes- olvationgasflowwas800L/hwithatemperatureof360^◦C,thecapillaryvoltagewas definedat2.4kV,theconevoltagewaskeptconstantat50V,thesourcetempera- turewas120^◦C,theconegasflowandthecollisiongasflowweresetto10L/hand 0.25mL/min,respectively.Twofunctionswereusedfortheacquisitions:awide- passquadrupolemodewithlowcollisionenergy(5eV)forthefirstfunction(range m/z95–1000)andacollisionenergyrampof5–70eVforthesecondfunction(MSÊ, rangem/z50–1000).Datawerecollectedincentroidmodewithascantimeof0.2s andaninterscandelayof0.02s.Dynamicrangeenhancement(DRE)wasusedand recalibrationofthedataweremadewiththeinfusionofasolutionof2ng/mLof Leucine-enkephalin(Sigma–Aldrich,Buchs,Switzerland)throughtheLockSpray probeat20␮L/min.

2.5. Dataanalysis

Rawdatapre-treatment,includingbaselinecorrection,peakdetection,chro- matogramalignment,peakintegrationandnormalizationwereachievedwith MarkerLynxXS^TM(SCN854,Waters).Theparametersweresetasfollows:peak detectionfrom1to29.5mininthem/z95–1000massrange,themasstoleranceset to0.05amu,chromatogramalignmentperformedusingEG-d3internalstandard, thepeakwidthat5%heightsetto25s,thepeak-to-peakbaselinesetto15,themass windowsetto0.05amu,retentiontimewindowsetto0.8min,noiseelimination levelsetto15,themarkerintensitythresholdsetto500countsandthenormal- izationperformedaccordingtothesumofthetotalreconstitutedspectra.Afirst filteringprocedurewasachievedwiththeMarkerLynxXS^TMReplicatefunctionto selectonlyreliablepeaks.Individualsampleswerethenconsideredasbiological replicatesandpeakswerekeptforfurtheranalysiswhentheywerepresentatleast in50%ofadefinedgroup(VYsamples,Czechworkers,etc.).Thisstrategyallowed theremovalofnoisyandirrelevantvariablesfromthedatamatrixandareduced datatableof3682ionswasobtained.

Chemistry-drivendimensionalityreductionwasperformedwithin-houserout- inesundertheMATLAB^®7environment(TheMathWorks,Natick,USA).Orthogonal partialleastsquaresdiscriminantanalysis(OPLS-DA)modelswerecomputedwith theSIMCA-Psoftware(version12,Umetrics,Umeå,Sweden).Aleave-one-outcross- validationwasachievedtoassessthepredictiveabilityofthemodels.CV-ANOVA (Erikssonetal.,2008)andpermutationtestswereperformedtoensurethevalidity ofthemodels.

3. Resultsanddiscussion

3.1. ExploratoryanalysisoftheCzechcohort:long-termeffectof dioxintoxicityfollowinganacuteoccupationalexposure

Between1965and1968,a groupof Czechchemicalworkers (approximately 80 people) were subjected tointoxication with dioxinsinaherbicideproductionplantinCentralBohemia,aregion of formerCzechoslovakia. Earlyexaminationsweredonein the 1970sandallmensufferedfromchloracne.Thissymptomwasstill reportedfortwopatients,30yearsafterexposurewithextremely highplasmalevelsofTCDD.Plasmalipidalterationswereobserved duringthefollow-upperiodandseveralpatientsmanifestedneu- ropsychological,and neurologicalabnormalities (Pelclova etal., 2001).TCDDlevelsbetween25andabout400timesthenormal populationandchroniceffectsincludinghyperlipidaemia,hyper- tensionanddiabeteswerefurtherdescribed40yearsafterexposure (Pelclovaetal.,2009).Thesepreviousstudiesunderlinedthere- foretheelevatedpersistenceofTCDDinthebodyassociatedwith long-termeffectsonhumanhealth.Ithastobenotedthatthese effectscanbeobservedseveraldecadesaftersevereintoxication andmetabolicdisordersareexpectedtosubsist.Ofthe80intoxi- catedworkers,urinesamplesof11workerscollectedin2011were availableforthepresentstudy.Thesesampleswereanalyzedto evaluatethelong-termeffectofdioxintoxicity(about40yearsafter exposure).

Severaladvantagesarederivedfromtheuseofurineasaref- erencebioﬂuidtomonitortoxicity,suchaseasyandnon-invasive samplingandreducedsamplepreparationcomparedtobloodas noproteinprecipitationisrequired(Ryanetal.,2011).However, thecomplexityofurinesamplesremainshighwhenperforming untargetedexperiments,evenafterensuringtheproperselectionof consistentdatafromtherawsignal.Additionally,theircontentcan begreatlymodiﬁedbyseveralphysiologicalandenvironmentalfac- tors,suchasage,diet,diurnalvariationandmedication.Measuring metabolites in urine in an untargeted manner remains there- foreachallengingtaskduetothenumerousconfoundingfactors.

Inclinicalpractice,aknowledge-baseddimensionalityreduction maythereforebehighlydesirabletoavoididiosyncraticresponses relatedtoatherapeuticinterventionduringuntargetedanalyses.

Chemicaldatamayconstituteavaluablesourceofinformationto deﬁnecriteriaforselectingvariablesandsimplifyingmultivariate models.

Achemistry-drivendimensionalityreductionwasrecentlypro- posed to ﬁlter untargeted metabolomic data and speciﬁcally

(5)

4 F.Jeanneretetal./ToxicologyLettersxxx (2013) xxx–xxx assessurinarysteroidmetabolitesinthecontextofdopingcon-

trol(Boccard et al., 2011). Asdioxin is known to alter steroid metabolism,asimilarprocedurewasappliedinthisstudytofocus onthesteroidsubsetofurinarycompounds(Egelandetal.,1994;

Grochowalskietal.,2001;Kleemanetal.,1990;Manhetal.,2013;

Sechmanetal.,2011).Inpracticalterms,rawdatawereprocessed withMarkerLynxXS^TMtogeneratealistof3682peakspersam- ple(seeSection2formoredetails).Eachofthe3682unidentiﬁed peakswasthencompared witha seriesofreferencem/zvalues extracted from theLipidMaps database,related to endogenous glucuronidated and sulfated steroid conjugates from the sterol lipidsclass(Fahyetal.,2009).Peaksmatchinga referenceentry withinatoleranceintervalof±0.025amuwereconsideredasputa- tivesteroid-relatedmetabolitesandretainedforfurtheranalysis, whileretentiontimeswerenotconsideredintheselectionpro- cess.Aseriesof284steroid-relatedcompoundswasobtainedafter thissecond selectionstep correspondingtoapproximately7.7%

ofthedetectedfeatures.Thesteroidnatureoftheselectedcom- poundscouldbecheckedbytheanalysisoffragmentationpatterns obtainedintheMS^Emode.

DatawerePareto-scaledandanOPLS-DAmodelwascomputed todistinguishexposedpatients(n=11)fromagroupofhealthy volunteers(n=11)onthebasisof the284variables (model#1).

Thismodelingapproach allowsa straightforwardinterpretation oftheresultsbyseparatingthepredictivevariationfromortho- gonalsourcesofsystematicvariability(TryggandWold,2002).The predictivescoresand loadingvectorsarelesspronetoundesir- ablevariationsandrelevantpatternscanbemoreeasilyrelated toa meaningfulbiologicalcontext.Thecontrolgroupwascom- posedof healthy menmatched for age (65–70years old)with non-pathologicalresultsforurine dipstick chemical analyses.A modelwith2latentvariables(onepredictiveandoneorthogonal) wasobtainedandaclearseparationofthetwoclasseswasobserved onthescoreplot(Fig.1A).Themodelvaliditywasestimatedby LOOCVandsatisfactorystatisticalindiceswerereported,i.e.global accuracyof95.5%,speciﬁcityof 90.9%,sensitivityof100.0% and CV-ANOVAp-value p<0.05. Due tothe relativelylimited num- berof samplesavailable for buildingthemodel (n=22) and as cross-validationmaybeoverlyoptimisticwhenassessingpredic- tiveability,theriskofoverﬁttingmustbeconsidered.Despitethis limitation,thecontributionofthe284variableswasassessedby theinvestigationoftheS-plotrelatedtothepredictivecomponent (Fig.1B).Thisrepresentationallowsboththeamplitudeofvariation andthereliability ofeachvariabletobesummarizedin aprac- ticalway bycombiningcovariance andcorrelation information, respectively(Wiklundetal.,2008).Bythesemeans,themostdis- criminantvariablescouldbehighlighted(blackcircles,

䊉

).Someof them,suchasetiocholanoloneglucuronideandandrosteroneglu- curonide,wereunambiguouslyidentiﬁedbycomparingtheirmass spectraandretentiontimeswithauthenticstandards.

Additionally,othersteroid metaboliteswerepartiallyidenti- ﬁed on the basis of accurate mass measurements and several unknownfeaturesweretentativelyassociatedtobileacids.Dioxin- inducedincreasedlevelsofglycocholicacidanditsglucuronide metabolite(m/z464.299at13minand m/z640.333at6.5min), glycodeoxycholic acid (m/z 448.306 at 20.3min), chenodeoxycholic acidsulfate or ursodeoxycholic acid sulfate (m/z 471.24 at23.2min)andbothglucuro-andsulfoconjugatesofglycochen- odeoxycholic acid (m/z 624.338 at 22min and m/z 528.262 at 20.3min).Decreasedconcentrationsduetodioxinexposurewere associatedwithglucuroconjugatedformsofthefollowingsteroids:

hydroxyandrosteroneorhydroxyetiocholanolone(m/z481.241at 10.1min),hydroxyDHEAorhydroxytestosterone(m/z479.228at 9.6min),etiocholanolone(m/z465.246at20.6min) andestrone (m/z 445.19at 6.3min). Glycoursodeoxycholic acidglucuronide (m/z 624.336 at 14min) and sulfate (m/z 528.261 at 11.8min)

metabolites were also decreased. Steroids and bile acids have numerousstructuralisomersordiastereoisomersthatcannotbe resolvedbymassspectrometryfragmentation.Thereforethemen- tionedidentifiedmetabolitescouldalsocorrespondtoanisomer oradiastereoisomer.Theidentificationofthesemarkerscanonly bemadewithauthenticstandards,butmodifiedlevelsofthisclass ofcompoundsweredetected.

These results were in accordance with previous studies, as alteredcholesterolmetabolismandbileacidbiosynthesiswasasso- ciated toTCDD toxicity in severalin vitro and invivo models.

TranscriptomicshighlightedalteredmRNAlevelsofgenesrelated tocholesterolandfattyacidsbiosynthesisandglucosemetabolism intheliverofTCDDdaily-exposedmice(Satoetal.,2008).Expres- sion changes in genes involved in cholesterol metabolism and bileacidssynthesisweresimilarlyreportedinratsafterexposure tohigh-doseofTCDD(Fletcheretal.,2005;Moffatetal.,2010).

TCDDexposurewasalsoassociated withthedisruptionoflipid metabolisminguineapigsbyalteringgenesinvolvedinadipogene- sisandlipogenesis(Nishiumietal.,2008).Asawhole,theseresults supportthehypothesisofcholesterolhomeostasisdysregulation followingTCDDexposuremediatedbyAhR.

Asalreadyobservedbyseveralauthorsinvariousmetabolomic studies,thenumberofvariablesinthecurrentmodel(284)istoo highforanexhaustiveidentificationofallputativebiomarkersand for a directinterpretation.Anotherfiltering stepwasthus per- formedbytakingadvantageofthedataoriginatingfromanidenti- fieddioxinacuteintoxication,i.e.urinesofVictorYushchenko.

3.2. ExploratoryanalysisofVictorYushchenkourinesamples:

extremephenotypeduetoanacutepoisoningwithTCDD

VictorYushchenko(VY),candidateforthepresidencyinUkraine waspoisoned withasinglehighdoseofdioxinduringadinner inSeptember2004.Hebecameseverelyill,sufferingofgastritis, colitis,hepatitis,neuropathyandpancreatitis.Chloracneappeared afewweekslater.Thesymptomswereclearlydescribedandprior analysessuggestedanacuteintoxicationwithpureTCDD(Saurat etal.,2012;Sorgetal.,2009).

Asignificantapproachinepidemiologyconsistsintheidenti- ficationandthespecificstudyofpatientswithclinicallyrelevant phenotypes,ratherthanunselectedpatients,toprogressmoreeffi- cientlyintheunderstandingofadisease(Perez-Graciaetal.,2010).

Asmultifactorial effects can produceintricate phenomena hin- deringproperinterpretationinhumanstudies,thesephenotypic extremesareexpectedtofacilitatedetectionofthecharacteristic symptomsofadisease.Inthecontextoftoxicology,acollection ofvariousphenotypescanrangefromchroniclowdosetoacute toxicexposure.Moreparticularly,thecharacterizationofextreme phenotypessuchasacuteintoxicationepisodesmayprovidevalu- ableinformationfor adeeperunderstandingofthecellularand moleculareventsoccurringaftertoxicexposure.Inotherwords, thediscoveryofbiomarkersofanacuteintoxicationcouldprovide informationforabetterunderstandingofTCDDexposure.

DuetothepathologiesVYsufferedandhispoliticalschedule, nostructuredexperimentalsetupcouldbedeﬁnedforsamplecol- lection:samplingandcollectiontimesoftheurinarysampleswere notedbutirregular.UrinessamplesofVYwerecollectedatdiffer- enttimepointsaftertheintoxication(about1yearto4yearsafter thepoisoning,seeSection2).

Preliminaryinvestigationsweredonetoevaluatetheurinesof VYasahomogeneousgroupandtoselectthemostrelevantfea- turesabletodistinguishVYfromacontrolgroupofthesameage.

Furthermore,thispreliminarystudywasmandatorytoexcludethe presenceofundesirabletrendortemporalbiasinthisparticular urineset.Forthelatter,notemporaldriftafterthepoisoningcould beclearlyestablishedusingvariousstatisticaltools.Todecreasethe

(6)

Fig.1. OPLS-DAmodel#1basedon284steroid-relatedvariables.(A)Scoreplot:urinesamplesfromCzechchemicalworkersexposedtodioxinaresymbolizedwithblack diamonds()andsamplestakenfromhealthyvolunteersbywhitesquares().(B)S-plot:discriminatingvariablesaresymbolizedbyblackcircles(䊉).

numberofexogenousmarkersoriginatingfromtherapyatapartic- ularsamplingtime,featuresdetectedinatleast50%ofVYsamples wereretained(seeSection2).Thisprocedureallowedmetabolites relatedtospeciﬁcdrugdeliverytoberemovedfromthedataand todecreaseintra-individualvariability.

Multivariatedataanalysiswasperformedtodifferentiateurine samplesfromacontrolgroupofvolunteersmatchedforageandthe samplescollectedduringYushchenko’smedicalcare.Paretoscaling wasappliedtothedatatableandanOPLS-DAmodelwascomputed.

Asatisfactorymodelwithonepredictiveandoneorthogonallatent variablewasobtained(model#2)andaclearseparationofthetwo classesofobservationswasobservedonthescoreplot(datanot shown).Cross-validationwascarriedoutindicatinga significant CV-ANOVAp-valuep<0.001.Nofalsenegativeorfalsepositivepre- dictionwasreported.AS-plotwasthenbuilttohighlightthemost relevant markers.Identifications wereinvestigated in multiples opensourcesdatabasessuchastheHumanMetabolomeDatabase, theMETLIN Metabolite Database or Chemspider for exogenous compounds.DespitetheremovalofunreliableMSsignalsduring data pre-treatment, first observations indicated putativestruc- turesmainlymatchingexogenouscompounds.Theseresultswere closelyrelatedtothevariousdrugtherapiesVYreceivedduringthe yearsfollowingpoisoning.

Thetwo firsttop-ranked elevatedmarkerswere assignedto propofol glucuronide (m/z 353.150 at 16min) and 4-hydroxy- propofolhydrogensulfate(m/z273.072at7.2min).Propofolisa widespreadsedative-hypnotic agentusedfor theinduction and maintenanceof generalanesthesia.The metabolismofpropofol is wellstudiedandvarious urinary metabolitesweredescribed (Favettaetal.,2000).Thevariablecorrespondingtothepropofol glucuronidewasfoundtobeveryintenseinsomeurinesamples (e.g.VY-28.5a)anditsMSÊspectracorrespondedtotheexpected fragmentation pattern. As a complementary indication for this identification,the medicationreport confirmed that VYwas in surgeryatthecorrespondingsamplingtime points.Otheranes- thetic agents, such as sevoflurane (m/z 166.992 at 6.4min) or its metabolitehexafluoroisopropanol (HFIP, same chemical for- mula:C₃H₂F₆O)werealsohighlighted.Accordingtotheliterature, sevoflurane,HFIPandHFIPglucuronidecouldberetrievedinurine (Accorsietal.,2005).Somenon-steroidalanti-inflammatorydrugs werealsodetected,suchasibuprofenmetabolite(m/z381.151at 21.6min).Overall,severalofthemostcharacteristicallyelevated biomarkersofVYsampleswereduetomedication.Nevertheless, poisoningwithaveryhighdoseofTCDDcouldleadtoexpressed biomarkersthatshouldbedetectedwhateverthedrugtherapy.

It hasto benoted that previousresults of blood and urine VYsamples highlightedsomeabnormallevels ofsteroids(Sorg andSaurat,unpublishedresults).Therefore,thesteroid-basedfil- teringstrategy usedfortheanalysisofsamplesfromtheCzech Cohortwasimplemented.Paretoscalingwasappliedtothedata subset of 284 steroid-related features and an OPLS-DA model wasassessed.Asatisfactorymodelwithonepredictiveandone orthogonallatentvariablewasobtained(model#3)andaclearsep- arationof thetwoclasses ofobservations wasobservedonthe scoreplot(Fig.2).Cross-validationwascarriedoutindicatingahigh accuracyofprediction(90.9%)andahighlysignificantCV-ANOVA p-valuep<0.001.Nofalsenegative(sensitivityof100%)andonlya fewfalsepositivecases(specificityof81.8%)werereported.

AnS-plotwasinvestigatedtohighlightrelevantbiomarkersof acuteTCDDtoxicity.Theurinarylevelsofaseriesof12compounds wereincreasedin VYsamples,while12 others weredecreased when compared tocontrols. Thesevariables and theirputative identificationbasedonreferenceentriesofthesteroidfilteraredis- playedinTable1.Amongthem,somesteroidssuchasDHEAsulfate, androsteronesulfateandandrosteroneglucuronidewereidentified withauthenticstandards,whileotherswereevaluatedbyMSÊfrag- mentationandsearchintheliterature(seeSupplementaryMaterial forexamplesofputativeidentification).

Supplementarymaterial related tothis articlecan befound, in the online version, at http://dx.doi.org/10.1016/j.toxlet.

2013.10.031.

These results were consistent with the scarce clinical data reportingtheeffectsofacutedioxintoxicitywithalteredsteroid metabolism and liver damage. The strong induction of the cytochrome P4501A1(CYP1A1) coding genewasreportedin a study combining transcriptomics analysesand immunostaining of Yushchenko’s skin lesions (Sauratet al., 2012). Cytochrome P450enzymesformamultigene familymetabolizingthousands ofendogenouscompoundssuchassteroidsaswellasexogenous chemicalssuchasdrugsortoxicchemicals.Theobservedactiva- tionofCYP1A1wasassociatedwithdioxinmetabolizationandthe reductionofthesystemictoxicity.Alteredcholesterolmetabolism andbileacidbiosynthesiswasalsoassociatedtoTCDDtoxicityand cytochromeP450modulationinseveralinvitroandinvivomod- els(Fletcheretal.,2005;Moffatetal.,2010;Rifkind,2006).These resultssupportthehypothesisofthedioxin-induceddevelopment ofoxidativestressbyalterationsofcytochromeP450sexpression and bile acidshomeostasis dysregulation(Pelclovaet al.,2011;

Reichardetal.,2005).Furthermore,othertranscriptsimplicatedin theregulationofbileacidsynthesissuchasCYP7A1,CYP46A1and

(7)

Fig.2.VYOPLS-DAbasedon284steroid-relatedvariables(model#3).(A)Scoreplot:urinesamplesfromVYaresymbolizedwithblacktriangles()andsamplestakenfrom healthyvolunteersbywhitesquares().(B)S-plot:discriminatingvariablesaresymbolizedbyblackcircles(䊉).

CYP27A1wereexpressedintheskinbiopsyofVY(SorgandSaurat, unpublishedresults).The24variableshighlightedtobethemost importantfeaturestodiscriminateVYsampleswereproposedasa metabolicpatternforthedetectionofacutedioxintoxicityaswell asabasisfortheselectionoffeaturesintheCzechdataset.

3.3. Applicationofthephenotype-baseddimensionalityreduction totheCzechcohort

BasedonthebiomarkersobtainedfromtheanalysisofVYsam- ples,anOPLS-DAmodelwascomputedfortheCzechdataseton thebasisofthe24selectedvariables(model#4)asaphenotypic selectionﬁlter.Aschemeoftheknowledge-basedvariableselection procedureispresentedinFig.3.

Amodelwith2latentvariables(model#4,onepredictiveand oneorthogonal)wasobtainedwithasatisfactorypredictionaccu- racyof 90.9%estimated byLOOCVand asigniﬁcantCV-ANOVA p-valuep<0.05.Ithastobenotedthatthetwoclasseswerenot

perfectlyseparated onthescore plot(Fig.4A),but similarlyto model#1,nofalsenegatives(sensitivityof100%)andasmallnum- beroffalsepositivecases(speciﬁcityof81.8%)werereportedduring cross-validation.Theseresultsareparticularlyrelevantinthecon- textofchronictoxicitydetection,asnorealcaseofintoxication waswronglypredicted.Thisfurtherconﬁrmsthereliabilityofthe selectedbiomarkersandtheirabilitytocharacterizebothacuteand long-termdioxinintoxicationphenotypes.

Interestingly,severalcompoundshighlightedinmodel#1were also reported as relevant biomarkers in the restricted subset.

AccordingtotheS-plot(Fig.4B),a smallnumberofcompounds couldberelated totheclassseparation.The upperrightregion correspondingtodecreasedconcentrationsafterdioxinexposure includedglycoursodeoxycholicsulfate(m/z528.261at11.8min), hydroxyDHEAorhydroxytestosteroneglucuronide(m/z479.228 at9.6min),andhydroxyandrosteroneorhydroxyetiocholanolone glucuronide (m/z 481.241 at 10.1min). The lower left area related toincreased levels includedglucuronide metabolitesof

Table1

24Acutedioxintoxicitybiomarkers.

Markers m/z RT(min) Putativeidentiﬁcations(orisomericstructure) Proposedformula

1 367.156 15.4 DHEAS C19H28O5S

2 369.172 10.8 Dihydrotestosterone-Sorisomer C19H30O5S

3 369.172 17.3 Isomericstructure C19H30O5S

4 369.172 18.6 Androsterone-S C19H30O5S

5 383.151 7.8 Dihydroxyandrostenone-S C19H28O6S

6 445.190 8.8 Estrone-Gorisomer C24H30O8

7 463.232 17.2 Isomerofepitestosterone-G C25H36O8

8 464.299 13.0 Glycocholicacid C26H43NO6

9 465.245 21.2 Androsterone-G C25H38O8

10 465.248 14.3 Isomericstructure(likeisomerofdihydrotestosterone-G) C25H38O8

11 471.246 26.3 Chenodeoxycholicacid-Sorursodeoxycholicacid-S C24H40O7S

12 479.225 12.0 HydroxyDHEA-GorhydroxyTestosterone-Goroxo-androsterone-G C25H36O9

13 479.228 7.7 HydroxyDHEA-GorhydroxyTestosterone-Goroxo-androsterone-G C25H36O9

14 481.241 10.1 Hydroxyandrosterone-Gorhydroxyetiocholanolone-G C25H38O9

15 483.259 9.3 Hydroxyandrostane-G C25H40O9

16 495.294 25.0 Pregnanediol-3-G C27H44O8

17 528.261 11.8 Glycoursodeoxycholicacid-S C26H43NO8S

18 567.316 20.1 Deoxycholicacid-Gorchenodeoxycholicacid-G C30H48O10

19 583.310 17.2 Cholicacid-Gorisomer C30H48O11

20 583.312 9.7 Cholicacid-Gorisomer C30H48O11

21 624.338 21.1 Glycochenodeoxycholicacid-Gorisomer C32H51NO11

22 624.338 22.0 Glycochenodeoxycholicacid-Gorisomer C32H51NO11

23 640.333 6.5 Glycocholicacid3-Gorisomer C32H51NO12

24 640.333 13.4 Glycocholicacid3-Gorisomer C32H51NO12

-Sisforsulfateand-Gisforglucuronidecompound.Steroidconﬁrmedwithauthenticstandardsarewritteninbolditalic.

MarkerselevatedbyVYincomparisontothecontrolsarehighlightedingray.

(8)

Fig.3. Knowledge-basedvariableselectionprocedure.

chenodeoxycholic acid (m/z 567.316 at 20.1min), glycochenodeoxycholicacid(m/z624.338at22min)andestrone(m/z445.19 at 8.8min). Furthermore, glycocholic acid and its glucuronide metabolite(m/z464.299at 13minand m/z640.333at6.5min) anddihydrotestosterone sulfate(m/z369.172at10.8min)were alsotentativelyidentiﬁedasincreasedbiomarkers.

However,ithastobepointedoutthattheroleofeachvariable isnotnecessarilysimilarinbothcasesandcanevenbeopposite.

Actually,thecombinationofthebiomarkersismorerelevantthan theevaluationofasingleparametertakenindividually.VYandthe Czechcohortarecasesofacuteexposure.Urinesamplingwascar- riedoutabout40yearsaftertheexposureoftheCzechworkersand about1–5yearsafterthepoisoningofVY.Thisdifferenceinsamp- lingcouldleadtodifferentbiomarkersofacuteexposure, some related tolong-term toxicity and somefor short/medium term toxicity.Therespectiveroleofthebiomarkerswereinvestigated byexaminingtheirnormalizedloadings(p(corr))tothepredictive

latentvariableofbothmodel#3andmodel#4(Fig.5).Interestingly, somemarkershadasimilarcontributiontobothdioxinexposure cases.Ontheotherhand,biomarkerscharacterizedbyincreased levelsinVYsamplesweremainlyrelatedtosteroids,whilesamples fromtheCzechcohortfollowedadifferenttrend.

Common biomarkers may be due to similar toxic pathway and dissimilaritiesto differentmechanisms. Thearyl hydrocar- bonreceptordoesnotonlyactasatranscriptionfactor binding thedioxinresponsiveelement(DRE)andactivatingmetabolizing enzymessuchasCYP1A1andCYP1B1.AhRisalsoinvolvedinthe regulationofotherpathwaysinaDRE-independentmanner.As anexample,AhRregulatescholesterolbiosynthesisbyinteracting withthesterolelement-bindingprotein2(SREPB2)transcription factor(Tanosetal.,2012).Furthermore,cross-talksbetweenAhR andsomenuclearreceptorssuchastheestrogenreceptor(ER)and theandrogenreceptor(AR)arewelldescribed(Ohtakeetal.,2011;

Pateletal.,2007).ARitselfinﬂuencesthecholesterolhomeostasis

Fig.4.OPLSmodel#4basedon24acutedioxintoxicitybiomarkers.(A)Scoreplot:urinesamplesfromCzechchemicalworkersexposedtodioxinaresymbolizedwithblack diamonds()andsamplestakenfromhealthyvolunteersbywhitesquares().(B)S-plot:discriminatingvariablesaresymbolizedbyblackcircles(䊉).

(9)

Fig.5.Biomarkers’normalizedloadingsfrommodel#3andmodel#4.Commoncontributionsarehighlightedwithdottedlines.Forputativebiomarkerscandidates,see Table1.

byinteractingwiththeliver Xreceptor (LXR),which canregu- latesulfotransferasesandtherebyandrogensactivity(Krycerand Brown,2011).Inmice,itwasreportedthatTCDDderegulatesgenes ofphaseIandIImetabolismsuchasgenesimplicatedincholesterol metabolism(CYP7A1),aswellastwonuclearreceptors,i.e.theshort heterodimerpartner(SHP)andfarnesoidXreceptor(FXR)(Fletcher etal.,2005).FXR,similarlytoLXR,interactswithsulfotransferases andreducesandrogensactivity(Runge-Morriset al.,2013).The numerouscross-talksandinteractionsbetweennuclearreceptors involvedinsteroidsandbileacidsregulations(FXR,LXR)indicate two-waysregulationsbetweenbileacidsmetabolismandsteroid signaling.Thistwo-wayregulationandthedescriptionofdioxins effectsonbothpathwaysconfirmedtheperturbationsofsteroids andbileacidsdescribed inanimalssuchasrodentsorchickens (Fletcheretal.,2005;Forgacsetal.,2012;Sechmanetal.,2011).Dis- turbancesofsteroidsandbileacidsreportedinthisstudycouldbe supportedbyinferringsimilarmechanismsbetweenhumansand animalsexposedtodioxins,takingintoaccountthegreatvariation ofdosesbetweenthespeciestoobserveatoxiceffect(Mukerjee, 1998).Somenuclearreceptorscouldbeimplicatedinthisprocess, suchasARandLXRorFXR,butthemechanismsstillneedtobe elucidated.Theseresultsprovidedpromisingopportunitiesinthe perspectiveofabetterunderstandingoftheeffectsofdioxinexpo- sureonhumanhealthasareducedlistofpotentialbiomarkersof dioxinintoxicationwasobtained.Atthisstage,theclassicalwayof biomarkersvalidationinvolvestheirproperidentificationthanks tocomparisonswithauthenticstandard.Thebiologicalrelevance andtheclinicalvalidityoftheidentifiedmarkersarethenassessed byinvivostudies(animalincase ofdioxin)and/orinvitrosys- tems.Analternativevalidationistheanalysisofthesemarkersinan independentcohort.Inthefuture,thebiomarkerssubsetshouldbe evaluatedandvalidatedinotherhumancohortsexposedtodioxin.

4. Conclusion

Ametabolomicapproachwasproposedtoassesstheeffectsof acutedioxinexposure.Appropriatedatapre-treatmentprocedures and a chemistry-driven ﬁlterwere implemented toreduce the numberofvariablestoconsiderandfocusonthesteroidmetabolite contentofurinesamples.Aseriesof284steroid-relatedvariables wasinvestigatedanda predictiveOPLS-DAmodelwasobtained

fromtheCzechcohort sufferingfromoccupationaldioxinexpo- sure.Thevariables’contributionshighlightedrelevantbiomarkers mainlyrelatedtobileacids.Asasecondstepofdatamining,prior knowledgecollectedduringthetreatmentofVYafterTCDDacute toxicexposureallowedbiologicallymeaningfulcompoundstobe selected.The24mostrelevantbiomarkerswerethenusedtoinves- tigatethelong-termeffectsofexposuretodioxin. Theseresults showedalteredlevelsofendogenoussteroidmetabolitesandmod- ifiedurinarybileacidsprofiles.Takentogether,thesefindingsare compatible withan increasedexpression ofcytochrome P450s, persistenthepatotoxicity,bileacidhomeostasisdysregulationand oxidativestress.Areliablemodelcouldbeevaluated;however,due tothesmallnumberofobservations,additionaltargetedinvesti- gationsofthebiomarkersneedtobeperformedforconfirmation purpose.Theselectionofcharacteristicphenotypeswasdemon- stratedasaveryeffectiveapproachtodetecttoxicitybiomarkers moreeasily.Theintegrationofothertypes ofknowledge-based filterssuchasmetabolicpathwaysisexpectedtooffervaluable alternativestodevelop dedicatedstrategies for theselection of meaningfulvariablesaccordingtothebiologicalcontext.Thelong- termfollow-upofpatientssufferingfromseveredioxinintoxication overdecadesisexpectedtoprovidenewavenuesofresearchto haveabetterunderstandingoftheeffectsofthesechemicalson humanhealth.

Conﬂictofinterest

Theauthorsdeclarenoconﬂictofinterest.

Acknowledgements

TheauthorsthankJean-DanielGrafandhisteamforhavingpro- videdtheurinecontrolsamples.TheSwissFoundationforGrants in Biologyand Medicine,and Novartis are thankedfor suppor- tingthis work(grant PASMP3-140064 toJulien Boccard).Fund PrvoukP25/1LF/2forcollectingandshippingsamplesfromPrague toGeneva.

References

Accorsi,A.,Morrone,B.,Domenichini,I.,Valenti,S.,Rafﬁ,G.B.,Violante,F.S.,2005.

Urinarysevoﬂuraneandhexaﬂuoro-isopropanolasbiomarkersoflow-level