Download A multivariate predictive modeling approach reveals a novel CSF peptide signature for both PDF

A multivariate predictive modeling approach reveals a novel CSF peptide signature for both
Name: A multivariate predictive modeling approach reveals a novel CSF peptide signature for both
Pages: 18
Year: 2017
Language: English
File Size: 2.6 MB
Downloads: 0
Page 2

usedasresearchcriteriaforthediagnosisofAD.Anumberofstudieshavefoundsignifi cantlyreducedCSFA42levelsinADpatientscomparedtonormalcontrolswithfewexcep tions(Seeandformeta reviewandmeta analysesrespectively;seeforexceptions). IncomparisontoA42levels,studieshaveconsistentlyfoundincreasedCSFt tauandp tau levelsinADpatientscomparedtonormalcontrols(Seeandformeta reviewandmeta analysesrespectively).Furthermore,elevatedlevelsoft tauandp tauhavealsobeenobserved inMCIpatientsthatdevelopedADcomparedtostableMCIpatientsandnormalcontrols .Giventhepossibilitythatavarietyofotherpathologicalprocessesmaybesimultaneously ongoingintheADbrain(e.g.,oxidativestress,inflammationandsynapticdysfunction),apart fromthesethreecoreCSFbiomarkers,otherbiomarkerscouldreflectpathogenesisofADand revealnewbiomarkersforAD. Proteomicapproachespermitlarge scaleassessmentoftheinvolvementofhundredsof proteinsand/orpeptidesincomplexbiologicalprocesses,andmaygeneratehypothesesboth aboutdiseasemechanismsandpotentialtherapeutictargets.Thistypeofapproachhasbeen usedextensivelytodevelopbiomarkersandshapedevelopmentofscientifichypothesesinthe cancerliterature.Onechallengetoinvestigatorsutilizingproteomicapproachesisthe sheermassofdatathatareobtainedusingthesemethods,bothintermsofextractionofcoher enttrendsinthedataandintermsofthepotentialforspuriousassociationsidentifiedviamul tiplecomparisons.Weandothershaveaddressedthesepotentialproblemsbyusingmachine learningalgorithmstodeveloppeptide signatures correspondingtodiseasestate,andby employingstrictcriteriatoavoidthepotentialforfalsediscovery.Increasingly,proteo lyticfragments,ratherthanwholeproteins,arebeingusedfordiseaseclassificationbecauseof theexpansioninthecomplexityofthesignaturesavailable.Therefore,inthecurrent reportweexploretheuseofaproteomictechniqueappliedtoproteolyticfragmentsintheCSF fortheclassificationandpredictionofdiseaseprogressioninAD. ProteinprofilingoftheCSFusingadvancedproteomicstechniquessuchas2Dgelelec trophoresis,massspectrometry,andliquidchromatography massspectrometrycouldhelp identifynovelADbiomarkers.Whilestudiesusingproteomicstechniqueshaveidentifieda numberofadditionalADcandidates(e.g.,neuronalpentraxinreceptor(NPTXR)andheart typefattyacidbindingprotein(FABPH)),manyofthesestudieshavebeendoneon smallcohortsinvolvingsmallarraysofCSFmarkers,usinglesspowerfulcomputa tionalapproachesanddidnotvalidatethemarkersinanindependentcohort.Tocircumvent theseissues,weperformedcrosssectionalanalysisofCSFsamplesobtainedfromlargeand wellcharacterizedpopulationsofAD,MCI,andage matchednormalcontrol(NL)subjects fromtheAlzheimer'sDiseaseNeuroimagingInitiative(ADNI)study.Weanalyzedadiverse arrayofpeptidestodetermineifsingleormulti analyteCSFpeptidesignaturescouldbeused to(i)distinguishpatientswithADfromNL(diseasestateclassification)and(ii)predictfuture conversionfromMCItoADinaseparatepopulationofpatients(predictionoffuture progression).MethodsDatawereobtainedfromtheADNIdatabase(adni.loni.usc.edu).ADNIwaslaunchedin2003 asapublic privatepartnership,ledbyPrincipalInvestigatorMichaelW.Weiner,MD.The primarygoalofADNIhasbeentotestwhetherserialmagneticresonanceimaging(MRI),pos itronemissiontomography(PET),otherbiologicalmarkers,andclinicalandneuropsychologi calassessmentscanbecombinedtomeasuretheprogressionofMCIandearlyAD.Forup to dateinformation,seewww.adni info.org.Thisstudywasconductedacrossmultipleclinical sitesandwasapprovedbytheInstitutionalReviewBoardsofalloftheparticipating AnovelCSFproteomicbiomarkerforthediagnosisandpredictionofprogressionofAlzheimer'sDiseasePLOSONE|https://doi.org/10.1371/journal.pone.0182098August3,20172/18(DepartmentofDefenseawardnumberW81XWH 12 2 0012).ADNIisfundedbytheNational InstituteonAging,theNationalInstituteof BiomedicalImagingandBioengineering,and throughgenerouscontributionsfromthefollowing: AbbVie,Alzheimer'sAssociation;Alzheimer'sDrug DiscoveryFoundation;AraclonBiotech;BioClinica, Inc.;Biogen;Bristol MyersSquibbCompany; CereSpir,Inc.;EisaiInc.;ElanPharmaceuticals, Inc.;EliLillyandCompany;EuroImmun;F. Hoffmann LaRocheLtdanditsaffiliatedcompany Genentech,Inc.;Fujirebio;GEHealthcare;IXICO Ltd.;JanssenAlzheimerImmunotherapyResearch &Development,LLC.;Johnson&Johnson PharmaceuticalResearch&DevelopmentLLC.; Lumosity;Lundbeck;Merck&Co.,Inc.;Meso ScaleDiagnostics,LLC.;NeuroRxResearch; NeurotrackTechnologies;Novartis PharmaceuticalsCorporation;PfizerInc.;Piramal Imaging;Servier;TakedaPharmaceutical Company;andTransitionTherapeutics.The CanadianInstitutesofHealthResearchisproviding fundstosupportADNIclinicalsitesinCanada. Privatesectorcontributionsarefacilitatedbythe FoundationfortheNationalInstitutesofHealth ( www.fnih.org).Thegranteeorganizationisthe NorthernCaliforniaInstituteforResearchand Education,andthestudyiscoordinatedbythe Alzheimer'sDiseaseCooperativeStudyatthe UniversityofCalifornia,SanDiego.ADNIdataare disseminatedbytheLaboratoryforNeuroImaging attheUniversityofSouthernCalifornia.The fundershadnoroleinstudydesign,datacollection andanalysis,decisiontopublish,orpreparationof themanuscript. Competinginterests:Fundingforthisworkwas derivedinpartfromthefollowingcommercial sources:AraclonBiotech;BioClinica,Inc.;Biogen; Bristol MyersSquibbCompany;CereSpir,Inc.; EisaiInc.;ElanPharmaceuticals,Inc.;EliLillyand Company;EuroImmun;F.Hoffmann LaRocheLtd anditsaffiliatedcompanyGenentech,Inc.; Fujirebio;GEHealthcare;IXICOLtd.;Janssen AlzheimerImmunotherapyResearch& Development,LLC.;Johnson&Johnson PharmaceuticalResearch&DevelopmentLLC.; Lumosity;Lundbeck;Merck&Co.,Inc.;Meso ScaleDiagnostics,LLC.;NeuroRxResearch; NeurotrackTechnologies;Novartis PharmaceuticalsCorporation;PfizerInc.;Piramal Imaging;Servier;TakedaPharmaceutical Company;andTransitionTherapeutics.Funding fromthesesourcesdoesnotalterouradherenceto PLOSONEpoliciesonsharingdataand materials.


Page 3

institutions.Informedwrittenconsentwasobtainedfromallparticipantsateachsite.Thefol lowingindividualethicsboardsapprovedthestudy:AlbanyMedicalCollegeInstitutional ReviewBoard,BostonUniversityMedicalCampusInstitutionalReviewBoard(BUIRB),But lerHospitalInstitutionalReviewBoard,ClevelandClinicInstitutionalReviewBoard,Colum biaUniversityInstitutionalReviewBoard,Dartmouth HitchcockMedicalCenterCommittee fortheProtectionofHumanSubjects,DukeUniversityHealthSystemInstitutionalReview Board,EmoryUniversityInstitutionalReviewBoardGeorgetownUniversityInstitutional ReviewBoard,HumanInvestigationCommitteeYaleUniversitySchoolofMedicine,Human SubjectsCommittee,UniversityofKansasMedicalCenter,IndianaUniversityInstitutional ReviewBoard,ResearchComplianceAdministration,InstitutionalReviewBoardofBaylor CollegeofMedicine,InstitutionalReviewBoardoftheMountSinaiSchoolofMedicine,Johns HopkinsUniversitySchoolofMedicineInstitutionalReviewBoards,Lifespan RhodeIsland HospitalInstitutionalReviewBoard,MayoClinicInstitutionalReviewBoard,NathanKline InstituteRocklandPsychiatricCenterInstitutionalReviewBoard(NKIRPCIRB),NewYork UniversityLangoneMedicalCenterSchoolofMedicine,InstitutionalReviewBoardHuman ResearchProgram,NorthwesternUniversityInstitutionalReviewBoardOffice,Officeofthe WashingtonUniversitySchoolofMedicineIRB(OWUMCIRB),OregonHealthandScience UniversityInstitutionalReviewBoard,PartnersHumanResearchCommittee,ResearchEthics BoardJewishGeneralHospital,ResearchEthicsBoardSunnybrookHealthSciencesCentre, RoperSt.FrancisInstitutionalReviewBoard,RushUniversityMedicalCenterInstitutional ReviewBoard,StanfordUniversity,AdministrativePanelonHumanSubjectsinMedicalRe search,TheOhioStateUniversityInstitutionalReviewBoard,TheUniversityofTexasSouth westernMedicalCenterInstitutionalReviewBoard,UCLAOfficeoftheHumanResearch ProtectionProgramInstitutionalReviewBoard,UCSDHumanResearchProtectionsPro gram,UniversityHospitalsCaseMedicalCenterInstitutionalReviewBoard,UniversityofAla bamaatBirminghamInstitutionalReviewBoard,UniversityofBritishColumbia,Clinical ResearchEthicsBoard(CREB),UniversityofCaliforniaDavisOfficeofResearchIRBAdmin istration,UniversityofCaliforniaIrvineOfficeOfResearchInstitutionalReviewBoard(IRB), UniversityofCaliforniaSanFranciscoCommitteeonHumanResearch(CHR),University ofIowaInstitutionalReviewBoard,UniversityofKentuckyOfficeofResearchIntegrity,Uni versityofMichiganMedicalSchoolInstitutionalReviewBoard(IRBMED),Universityof PennsylvaniaInstitutionalReviewBoard,UniversityofPittsburghInstitutionalReviewBoard, UniversityofRochesterResearchSubjectsReviewBoard(RSRB),UniversityofSouthFlorida DivisionofResearchIntegrity&Compliance,UniversityofSouthernCaliforniaHealthSci enceCampusInstitutionalReviewBoard,UniversityofWesternOntarioResearchEthics BoardforHealthSciencesResearchInvolvingHumanSubjects(HSREB),UniversityofWis consinHealthSciencesInstitutionalReviewBoard,WakeForestUniversityInstitutional ReviewBoard,WeillCornellMedicalCollegeInstitutionalReviewBoard,WesternInstitu tionalReviewBoardandWesternUniversityHealthSciencesResearchEthicsBoard.PatientpopulationParticipantsincludedpatientswithAD(definedbyNINCDS ADRDA1)andMCI(using Petersencriteria),andNLfromtheADNIstudythatreceivedclinical,neuropsychologi cal,andbiomarkerassessmentswhichwererepeatedevery6monthsforupto36months.NL individualswerefreeofmemorycomplaintsordepressionandhadaMini MentalStateExam ination(MMSE)scoreabove25andaClinicalDementiaRating(CDR)scoreof0.MCIindi vidualscouldhaveMMSEscoresof23to30andrequiredaCDRof0.5andaninformant verifiedmemorycomplaintsubstantiatedbyabnormaleducation adjustedscoresonthe AnovelCSFproteomicbiomarkerforthediagnosisandpredictionofprogressionofAlzheimer'sDiseasePLOSONE|https://doi.org/10.1371/journal.pone.0182098August3,20173/18


Page 4

WechslerMemoryScaleRevised LogicalMemoryII.ADpatientscouldhaveMMSEscores of20to27andaCDRof0.5or1.0.Ofthe135MCIsubjectsfromwhomtheCSFproteomic datawereavailableatbaseline,122subjectsstayedinthestudyforatleast36months.CSFsamplesCSFsamples(0.5mL)wereobtainedinthemorningafteranovernightfastandprocessed usingtheCaprionProteomicsplatformthatusesmassspectrometrytoevaluatetheabilityofa panelofpeptidestodiscriminatediseasestatesanddiseaseprogression.ProceduresforCSF sampling,transport,andstoragehavebeendescribedpreviously.TheCSFmultiplexmul tiplereactionmonitoring(MRM)panelwasdevelopedbyCaprionProteomicsincollabora tionwiththeBiomarkerConsortiumProjectTeam.Atotalof320peptidesgeneratedfrom trypticdigestsof143proteinswereusedinthisstudy(seeS1Tableforlistofpeptidesandpro teins).Thesepeptidesincludeaseriesofpeptidesrepresentinginflammatorymarkersandpep tidesidentifiedinanearlierphaseoftheprogramthatusedmultiplexedimmunoassaybased platform(performedbyRulesBasedMedicine). Detailsregardingthetechnology,qualitycontrolandvalidationcanbefoundintheUseof TargetedMassSpectrometryProteomicStrategiestoIdentifyCSF BasedBiomarkersinAlz heimer'sDiseaseDataPrimer(http://adni.bitbucket.org/csfmrm.html).Inbrief,asdescribed inthedataprimerandinSpellmanetal.(2015),CSFsamplesweredepletedofplasma proteinsusingaMultipleAffinityRemovalSystem(MARS 14)column,trypsindigested(1:25 protease:proteinratio),lyophilized,desaltedandanalyzedbyLC/MRM MSanalysisona QTRAP5500LC MS/MSsystematCaprionProteomics.MRMisamassspectrometry based platformthathasbeenshowntobereproduciblewithinandacrosslaboratoriesandinstru mentplatforms.MRMexperimentswereperformedontriplequadrupole(Q)massspec trometers.Thefirst(Q1)andthird(Q3)massanalyzerwereusedtoisolateapeptideionanda correspondingfragmention.ThefragmentionsweregeneratedinQ2bycollisioninduced dissociation(CID).The320peptidesmetallthequalitycontrolcriteriasetbytheADNIwork inggroup.AnalysisFortheunivariateanalysistoidentifyindividualpeptidesthatareeitherdifferentially expressedbetweenADandNLsubjects,orbetweenMCI ADprogressorsversusnon progres sors,theanalysisofcovariancemodel(ANCOVA)wasusedwithageandgenderascovariates andthegroupstobecomparedasfixedeffect.Thismodelwasfitonthelog2transformed quantile normalizedintensitiesofthepeptideexpressionvalues.Outlierswereidentifiedand excludedbasedontheresidualsfromthisANCOVAmodelwhosevalueswereeitherlessthan Q1 1.5x(Q3 Q1)oraboveQ1+1.5x(Q3 Q1),whereQ1andQ3arethefirstandthirdquar tilesofthedistributionofresiduals.Thesignificanceofpeptideswasassessedandisreported intermsofthefalsediscoveryrateestimate(q value),andtherelevantsummarystatistics suchasthereceiveroperatorcharacteristicareaunderthecurve(ROCAUC),foldchange,and theeffectsize,alongwithp valuesarealsoreported. Multivariatepredictivemodelinganalysiswasthencarriedouttoderiveasignature(combi nationofpeptidesandanyadditionalcovariates)thatoptimallydifferentiatestheADversus NLsubjects.Thelistofcandidatepredictorsconsideredforselectioninthissignatureincluded thelistof320peptidesoftheCSFproteomicpanel,plusage,genderandapolipoproteinE (APO E)status(totally323predictors).Analgorithmbasedonthelogisticregressionmodel withlasso basedpenaltywasemployedforthisanalysis.Toensurethestabilityand robustnessoftheselectionofasubsetofpredictorsfortheoptimalsignatureviathisalgorithm, AnovelCSFproteomicbiomarkerforthediagnosisandpredictionofprogressionofAlzheimer'sDiseasePLOSONE|https://doi.org/10.1371/journal.pone.0182098August3,20174/18


Page 5

abootstrapprocedurewasusedtoestimatethelassopenaltyparameter.Theperformance oftheoptimalpeptidesignaturefromthisalgorithmthatdifferentiatestheADandNLsubjects wasevaluatedviaarigorousfive foldinternalstratifiedcross validationprocedure.Inthispro cedure,allstepsofthemodelbuildingandsignaturederivationprocesswerefullyembedded withinthecross validation.Thepredictionsofalltheleft outfoldsfromthiscross valida tionprocedurewerefirstaggregated,andtheperformancemeasuressuchastheoverall classificationaccuracy,sensitivity,specificity,andthepositiveandnegativepredictivevalues wereevaluatedontheseaggregatedpredictions.Thisinternalcross validationprocedurewas repeated20times,andthemeanandstandarddeviationoftheseperformancemeasuresare reported. TheaboveoptimalpeptidesignaturederivedtodifferentiatetheADandNLsubjectswas thentestedonaseparateindependentgroupofMCIsubjectsatbaselinetopredicttheirfuture progressiontoAD.AsthepeptidesignaturewouldreturnthepredictionresultsassimplyAD orNL,thepredictionofanMCIsubjectasNLwasconsideredas SignatureNegative atbase line,andthepredictionofanMCIsubjectasADwasconsideredas SignaturePositive at baseline.Theaccuracyofthispredictionwasthenassessedrelativetothetrueprogressionsta tusoftheMCIsubjectstoADoverthenext36months. Theperformanceofthispeptidesignaturewasfurtherevaluatedintermsofitsabilitytodif ferentiatethefuture timetoprogression fromMCItoADofthesebaselinesignaturepositive andsignaturenegativeMCIsubjectsviaKaplan Meieranalysis.Forthisevaluation,thepro gressionofMCIsubjectstoADovertheentirefuturetimecourseuntilthelastfollow upvisit wastakenintoconsideration.ThisevaluationoftheADversusNLpeptidesignatureonthe futureprogressionofaseparategroupofMCIsubjectstoADwouldnotonlyserveasaninde pendentverificationoftheutilityofourpeptidesignature,butalsoputittoagreatertesttosee whetheritisrobustenoughtoaddressadifferentandmoreimportantquestionrelatedtopre dictingthefuturediseaseprogressioninAD.Results Disease statedemographicsDatafrom287subjectswereanalyzed,withthelargestproportion(135/287or47.1%)coming fromMCIsubjects.Ofthe66ADsubjects,65werediagnosedas probable and1wasdiag nosedas possible AD.ThesubjectswerebalancedacrosstheNL,MCIandADgroupsin termsofage(rangeofmeans=74.79 75.80years,p>0.05)andeducation(rangeofmeans= 15.11 16.0years,p>0.05).Thereweremoremales(59.9%)thanfemales(40.1%)inthestudy, thoughsimilarnumbersofmaleandfemaleMCIsubjectsconvertedtoADoverathree year period(52.3vs.65.8%,p=0.166,Chi squaredtest).Asshownpreviously,thepresenceof theAPO E4alleletrackedwithdiseasestate(71.2%AD,52.6%MCIand31.8%NL,p< 0.0001,Chi squaredtest).Inaddition,thepresenceofthisallelealsotrackedwithMCItoAD progressionovera36 monthperiod(37.5%ofnon E4vs.56.3%ofE4progressedtoAD, p=0.028,Chi squaredtest),seeTables1and2.Disease stateclassification:UnivariateanalysisAlargenumberofpeptideswerefoundtobedifferentiallypresentinADvs.NLsubjects.As expected,oneAPO EpeptidesequencewaspresentinsubstantiallyhigheramountsinADvs. NLsubjects(APOE_LGADMEDVR:17.29folddifferenceinmedianvalue,q=9.45E 07,see Table3).ThisfindingwaspreviouslyknownsincethissequenceisfoundonlyinAPOE4+sub jects.Otherpeptides,someknowntobeinvolvedinneuronalfunction(e.g.,CA2D1, thevoltage dependentcalciumchannelsubunitalpha 2/delta 1),andothersnotclassically AnovelCSFproteomicbiomarkerforthediagnosisandpredictionofprogressionofAlzheimer'sDiseasePLOSONE|https://doi.org/10.1371/journal.pone.0182098August3,20175/18


Page 6

associatedwithneuronalfunction(e.g.,FABPH),differedbetweenADandNLsubjects.Using aq value<0.05criteria,39outof320peptidesreachedstatisticalsignificancewiththisfalse discoveryratecorrection,while11outof320hadq valueslessthan0.005(seeFig1forthetop 8peptides).Disease stateclassification:MultivariateanalysisCreationofanoptimizedmultivariatesignatureimproveddiseasestatedifferentiationcompared toindividualpeptides.Inclusionofall320peptidesequences,demographicdata(age,gender, education)andAPO E4statusproducedanoptimized16 peptidesignature.Thesizeofoursig naturesandcontentsweredeterminedviaatotallydata drivenmannerviathemathematicalopti mizationandalgorithmdescribedintheMethodssectionindetail.Thesignaturecomponentsare showninFig2,coefficientsareshowninS2Table.Thoughthismodel'sabilitytodifferentiateAD fromNLwasrelativelymodest,withtheareaunderitsreceiver operatingcharacteristiccurve (ROCAUC)of0.89+/ 0.01(basedon20iterationsof5foldcrossvalidation),thisvaluewas higherthanthatseenofanyindividualmarker(highestwasAPO Ewith0.73). Theperformanceofthe16 peptidemultivariatesignaturewascomparedtoallpermuta tionsofA42,t tauandp tau(181)intheCSF,includingtheirratios,andpublishedcut points .Acrossallmeasures,the16 peptidemultivariatesignatureoutperformedtheothermark erssignificantly(Table4).Inaddition,includingA42,t tauandp tau(181)withthe16 pep tidesignaturedidnotresultinasignificantimprovementinperformance.MCI ADprogression:UnivariateanalysisWecomparedCSFprofilesforMCIpatientsthatconvertedtoADbythe36monthvisitvs. MCIpatientsthatdidnotconvert.Threemarkershadmarginalq valuesof0.0508:hemoglo binsubunitalpha(HBA),neuronalpentraxin2(NPTX2)andpoliovirusreceptor related protein1(PVRL1,Table5).Interestingly,theAPO Epeptide(LGADMEDVR),whichdemon stratedexcellentdifferentiationbetweenADvs.NL,ranked199/320forpredictingconversionTable1.Disease statedemographics. AD(n=66)MCI(n=135)NL(n=86) Gender(n)M379144 F294442 Apo E(n)E4477121 Non E4196465 Age(years,mean+/ SD)75.097.5274.797.3675.805.55 Education(years,mean+/ SD)15.112.9616315.642.97 BaselineMMSE(mean+/ SD)23.521.8526.911.7429.051.02 https://doi.org/10.1371/journal.pone.0182098.t001Table2.Three yearMCIconvertervs.nonconverterdemographics. MCItoADconverters(n=64)MCInon converters(n=71) Gender(n)M4051 F2420 Apo E(n)E44031 Non E42440 Age(years,mean+/ SD)74.92+/ 7.5774.68+/ 7.21 Education(years,mean+/ SD)15.59+/ 3.0216.36+/ 2.89 BaselineMMSE(mean+/ SD)26.36+/ 1.6827.41+/ 1.64 https://doi.org/10.1371/journal.pone.0182098.t002 AnovelCSFproteomicbiomarkerforthediagnosisandpredictionofprogressionofAlzheimer'sDiseasePLOSONE|https://doi.org/10.1371/journal.pone.0182098August3,20176/18


Page 7

fromMCItoAD.Thesedatasuggestthatindividualpeptidemarkersdoapoorjobofpredict ingMCItoADprogressionontheirown;hencethemotivationtocombinemarkersinamul tivariateanalysistoincreasetheirutility(below).MCI ADprogression:MultivariateanalysisThesame16 peptidemultivariatesignaturethatwasdevelopedfordiseasestateclassification wasemployedontheMCIsubjects,whichrepresentacompletelyindependentpopulation,atTable3.Normalvs.AlzheimerDisease,univariateanalysis.Shownaretheanalyteswithaq value0.05. SymbolSequenceFoldChangeROCAUCEffectsizep valueq value APOELGADMEDVR17.290.730.742.95E 099.45E 07 FABPHSLGVGFATR1.300.720.811.57E 082.51E 06 FABPHSIVTLDGGK1.350.730.862.49E 072.66E 05 PTPRNAEAPALFSR0.870.66 0.541.18E 050.0009 CA2D1FVVTDGGITR0.820.65 0.553.14E 050.0019 VGFNSEPQDEGELFQGVDPR0.800.67 0.623.52E 050.0019 VGFAYQGVAAPFPK0.830.64 0.534.72E 050.0022 NPTXRLVEAFGGATK0.780.69 0.736.15E 050.0025 CCKNAHLGALLAR0.820.65 0.478.84E 050.0031 PTPRNSELEAQTGLQILQTGVGQR0.880.63 0.509.83E 050.0031 NPTXRELDVLQGR0.840.69 0.701.34E 040.0039 PIMTVQLVVGDGR0.860.66 0.580.00030.0070 SCG1NYLNYGEEGAPGK0.820.67 0.600.00030.0070 SCG2VLEYLNQEK0.910.63 0.440.00040.0097 CH3L1ILGQQVPYATK1.090.620.480.00050.0100 VGFTHLGEALAPLSK0.880.63 0.500.00050.0104 FAM3CGINVALANGK0.860.63 0.440.00060.0108 AMDIVQFSPSGK0.850.66 0.570.00060.0108 AMDIPVDEEAFVIDFKPR0.900.62 0.400.00070.0121 CA2D1TASGVNQLVDIYEK0.880.65 0.460.00080.0121 CA2D1IKPVFIEDANFGR0.850.63 0.440.00080.0121 CMGASEALAVDGAGKPGAEEAQDPEGK0.870.64 0.430.00090.0137 CMGAYPGPQAEGDSEGLSQGLVDR0.820.62 0.380.00100.0145 NEGR1SSIIFAGGDK0.910.63 0.510.00120.0155 CH3L1SFTLASSETGVGAPISGPGIPGR1.080.600.430.00130.0155 SCG1HLEEPGETQNAFLNER0.810.620.020.00130.0155 CMGAEDSLEAGLPLQVR0.770.63 0.250.00130.0155 NPTX2LESLEHQLR0.810.64 0.520.00140.0156 NRCAMVFNTPEGVPSAPSSLK0.890.64 0.490.00160.0173 FAM3CSPFEQHIK0.950.61 0.440.00210.0225 PCSK1GEAAGAVQELAR0.870.63 0.490.00240.0252 NPTX1LENLEQYSR0.890.63 0.530.00260.0255 PCSK1ALAHLLEAER0.850.63 0.510.00320.0308 SCG3FQDDPDGLHQLDGTPLTAEDIVHK0.840.63 0.420.00350.0331 NPTX2TESTLNALLQR0.860.64 0.550.00370.0341 TTHYTSESGELHGLTTEEEFVEGIYK1.080.620.390.00510.0445 PDYNLSGSFLK0.870.61 0.350.00520.0445 PCSK1NSDPALGLDDDPDAPAAQLAR0.860.63 0.440.00550.0464 NRCAMYIVSGTPTFVPYLIK0.890.60 0.320.00570.0468 https://doi.org/10.1371/journal.pone.0182098.t003 AnovelCSFproteomicbiomarkerforthediagnosisandpredictionofprogressionofAlzheimer'sDiseasePLOSONE|https://doi.org/10.1371/journal.pone.0182098August3,20177/18


Page 8

baselinetopredicttheirprogressiontoADover36months.AsshowninTable6,acrossall measures,the16 peptidesignatureoutperformedallpermutationsofA42,t tauandp tau (181)andpublishedcut points.Receiver operatorcurveswereconstructedusingallcom binationsofmarkersforA42anddifferentformsoftau,forthe16 peptidesignatureshown inFig2andforacombinationofthetwoforpredictingthe36 monthMCI ADprogression. Thelargestareaunderthecurvewasobservedforthe16 peptidesignature(0.74),withasimi larvalueseenforthecombined16 peptide+A42/taumarkers(0.73)andthelowestseenfor combinationsofA42/taumarkerswithoutthemultivariatesignature(0.64,p<0.05,Fig3). The16 peptideADvsNLmultivariatesignaturewasthentestedontheMCIsubjectsat baselinetopredicttheirprogressiontoADovertheentirefuturetimecourseuptothelastfol low upvisit.Theclassifierbuiltbasedonthe16 peptideADvsNLsignaturewasusedtoplace theMCIpatientsatbaselineintotwocategories;thosepredictedasNLwereconsideredas SignatureNegative andthosepredictedasADwereconsideredas SignaturePositive .As evidentfromFig4A,MCIsubjectsinthesignaturepositivegroupatbaselinehadamuchfaster mediantimetoprogression(MTP)toADthanthoseinthesignaturenegativegroup(21.32 monthsversus71.56months,p=3.3x10 7,hazardratio=3.38).Whilesimilaranalysisusing combinationsofA42,t tauandp tau(181)toplacetheMCIsubjectsintosignaturepositive andnegativegroupsatbaselinerevealfasterprogressionofthesignaturepositiveMCIsubjects toAD(MTPof25.69versus48.89months,p=0.0065,hazardratio=1.92,Fig4B),the16 pep tidesignatureprovidedamorerobustpredictorofMTP(Table7).The16 peptidesignature Fig1.Univariateanalysis.Valuesofthe8peptidemarkerswiththelowestq valuesinADvs.NLdiseasestateclassification.Individualsubjectsare shownasopencircles.Boxesrepresentthefirstandthirdquartiles.Thelinesthatextendoutfromthetopandbottomendsofboxindicatetherangeofthe range,minustheoutliers.Thepointsoutsidethelinesarethelowandhighoutliers. https://doi.org/10.1371/journal.pone.0182098.g001 AnovelCSFproteomicbiomarkerforthediagnosisandpredictionofprogressionofAlzheimer'sDiseasePLOSONE|https://doi.org/10.1371/journal.pone.0182098August3,20178/18


Page 9

alsooutperformedthepublishedcut pointsonA42,t tauandp tau(181),whichhada hazardratioof1.8.Thesedatasuggestthat16 peptidesignatureisastrongpredictoroffuture progressionfromMCItoADoverthesubsequentyearsandoutperformsthetraditionalCSF biomarkers.DiscussionInthisstudy,thediagnosticandpredictiveaccuracyofanarrayof300+peptidesintheCSF forthediagnosisofMCIandADandforthepredictionofprogressionfromMCItoADwas examined.Itwasfoundthatseveralindividualpeptides,includingmanynotclassicallyassoci atedwithneuronalfunction,showedhighstatisticalsignificanceindistinguishingbetweenAD Fig2.16 peptidesignature.Relativeimportanceofthecontributionofeachpeptideinthe16 peptidemultivariatesignaturefordifferentiatingADvs.NL thatissubsequentlyusedforpredictingprogressionofMCIsubjectstoAD.Peptidesareplottedintheorderoftheirimportance/contributiontothis multivariatesignatureinthelogisticregressionmodel.Asthe16thpeptiderelatedtoCATDappearstoprovideverylittleincrementalvalue(notedinred),the data drivenprocessthatledtoitsinclusioninthesignaturesuggestedanoverallbenefitofretainingitinthesignature.Thecoefficientsforeachofthese markersisgiveninS2Table. https://doi.org/10.1371/journal.pone.0182098.g002Table4.Performanceofmultivariatemodeltodifferentiatediseasestate.ThetoprowcorrespondstoallpermutationsofAb,tTauandpTau,andthe bottomrowreferstothe16 peptidesignatureshowninFig2. AccuracySensitivitySpecificityPPVNPV A 1 42,tTau,pTausignature0.78+/ 0.010.80+/ 0.020.75+/ 0.020.71+/ 0.010.84+/ 0.01 16 peptidesignature0.85+/ 0.020.86+/ 0.020.84+/ 0.030.80+/ 0.030.89+/ 0.01 https://doi.org/10.1371/journal.pone.0182098.t004 AnovelCSFproteomicbiomarkerforthediagnosisandpredictionofprogressionofAlzheimer'sDiseasePLOSONE|https://doi.org/10.1371/journal.pone.0182098August3,20179/18


Page 10

andNL.A16 peptidemultivariatesignaturebasedonthesepeptideswasidentifiedwithan overallclassificationaccuracyof85%,withimprovedaccuracy,sensitivity,specificityandposi tiveandnegativepredictivevaluescomparedtomoretraditionalCSFmarkers.Morenotably, whenthissame16 peptidesignaturewastestedonanindependentgroupof135MCIsubjects, itoutperformedthetraditionalA/taumarkersforpredictingthefutureprogressionfrom MCItoAD;apositiveresultonthis16 peptidemultivariatesignatureatbaselineresultedina 3.38 foldfasterprogressiontoAD.Thoughsomeofthesepeptideshavebeendescribedprevi ouslyasindividualbiomarkers(seebelow),thecurrentdatasuggesttheircombinationoutper formspreviousCSFmarkersandpointtothepossibilitythatothernovelmarkersmayhavea previouslyunrecognizedroleindiagnostictestingaswellasinunderstandingthepathophysi ologyofAD.ReviewofspecificanalytesidentifiedOverthepastseveralyears,proteomicapproacheshaveidentifiedanalphabetsoupofpotential markersthatmaybeabletopermitearlydiagnosisofADorpredictconversionfromMCIto AD.Manyofthepotentialmarkersidentifiedbythesestudieshaveknownor suspectedrolesineitherADorinpathologicalprocessesthoughttobedisruptedinAD.For example,asexpected,oneoftheAPOEpeptidesexamined(LGADMEDVR),whichisTable5.MCItoADconvertersvs.non converters,univariateanalysis,lowest20q values. SymbolSequenceFoldChangeROCAUCEffectsizep valueq value HBAFLASVSTVLTSK1.660.630.470.00060.19 NPTX2LESLEHQLR0.800.65 0.520.00130.19 HBAVGAHAGEYGAEALER2.680.640.490.00170.19 HBBSAVTALWGK2.230.630.440.00460.28 HBBVNVDEVGGEALGR2.110.630.460.00490.28 PRDX1DISLSDYK1.120.610.360.00610.28 NPTX2TESTLNALLQR0.710.63 0.470.00700.28 NRCAMSLPSEASEQYLTK0.900.59 0.340.00710.28 HBATYFPHFDLSHGSAQVK1.460.610.440.01280.37 CO3IHWESASLLR0.550.64 0.410.01330.37 CFABVSEADSSNADWVTK0.880.63 0.450.01370.37 HBBEFTPPVQAAYQK2.180.610.520.01380.37 PVRL1ITQVTWQK0.920.63 0.450.01640.40 CFABYGLVTYATYPK0.840.60 0.360.02220.42 CO2HAIILLTDGK0.920.60 0.370.02270.42 NPTXRELDVLQGR0.850.61 0.390.02450.42 CAH1YSSLAEAASK1.350.590.310.02750.42 C1QBVPGLYYFTYHASSR0.920.59 0.260.02840.42 TTHYVEIDTK1.100.590.240.02870.42 PRDX6LSILYPATTGR1.290.590.230.02870.42 https://doi.org/10.1371/journal.pone.0182098.t005Table6.PerformanceofmultivariatemodeltodifferentiateMCItoADconvertersvs.non converters.Thetoprowcorrespondstoallpermutationsof Ab,tTauandpTau,andthebottomrowreferstothe16 peptidesignatureshowninFig2. AccuracySensitivitySpecificityPPVNPV A 1 42,tTau,pTausignature0.620.780.490.580.71 16 peptidesignature0.700.780.630.650.76 https://doi.org/10.1371/journal.pone.0182098.t006 AnovelCSFproteomicbiomarkerforthediagnosisandpredictionofprogressionofAlzheimer'sDiseasePLOSONE|https://doi.org/10.1371/journal.pone.0182098August3,201710/18


Tags: Download A multivariate predictive modeling approach reveals a novel CSF peptide signature for both PDF, A multivariate predictive modeling approach reveals a novel CSF peptide signature for both free pdf download, A multivariate predictive modeling approach reveals a novel CSF peptide signature for both Pdf online download, A multivariate predictive modeling approach reveals a novel CSF peptide signature for both 2017 download, A multivariate predictive modeling approach reveals a novel CSF peptide signature for both.pdf, A multivariate predictive modeling approach reveals a novel CSF peptide signature for both read online.
About | Contact | DMCA | Terms | Privacy | Mobile Specifications
Copyright 2021 FilePdf