How to characterize the function of a brain region

Review

How

to

Characterize

the

Function

of

a

Brain

Region

Sarah

Genon,

*

Andrew

Reid,

Robert

Langner,

Katrin

Amunts,

and

Simon

B.

Eickhoff

Many brainregions have been defined,but acomprehensiveformalizationof each

region’sfunctioninrelationtohumanbehaviorisstilllacking.Currentknowledge

comesfromvariousfields,which havediverseconceptionsof‘functions’.We

brieflyreviewthesefieldsandoutlinehowtheheterogeneityofassociationscould

beharnessedtodisclosethecomputationalfunctionofanyregion.Aggregating

activationdatafromneuroimagingstudiesallowsustocharacterizethe

func-tionalengagementofaregionacrossarangeofexperimentalconditions.

Fur-thermore, large-sample data can disclose covariation between brain region

features and ecological behavioral phenotyping. Combining these two

approachesopensanewperspectivetodeterminethebehavioralassociations

ofa brain region, and henceits functionand broader rolewithin large-scale

functionalnetworks.

WhatDoesAny PartoftheBrain Do?

Eversince humanshavescientificallyinvestigatedthemind,understandinghowitis orga-nizedatthelevelofitsbiologicalsubstrate(i.e.,thebrain)hasremainedchallenging.Forovera century, great progress has been made in mapping the human brain (based on various characteristics),leadingtoarapidlyexpandingnumberofparcellationschemesandatlases detailingtheorganization ofcortical areasandmodules [1].Severalstudieshave demon-stratedthatthestructuralsegregationofthecerebralcortexintodifferentareas (distinguish-ablebased ontheir biologicalproperties,such as molecular,cellular,or fiberarchitecture [2,3]) is closely related toits functional segregation [4] and, in turn, its organization into functionalnetworks[5].

CurrentconceptualizationsofbrainfunctionasaBayesianmachine,inwhichbrainareasare seenas connectedandrelativelyspecialized computationalunits, are incontrastwiththe actual availableknowledge about functional specialization. Studiesover the past century show that theunderstanding of brain–behavior relationships hasbeen an interdisciplinary endeavor,resultinginrichandheterogeneouspatternsofbehavioralfunctionsformanybrain regions. After reviewing the most common approaches that have contributed to this endeavor,weproposethatassessingtherelativefunctionalspecialization ofbrainregions requiresa criticalchange in viewpoint,whereinthe a priori definedconstruct isthe brain regionandtheunknownsarethebehavioralfunctionsassociatedwithit.Inthatperspective, recentadvancesindataaggregationoffernovelopportunitiesforasystematic characteriza-tionofbrainregionsacrossarangeofbehavioralconditionsandphenotypicalfeatures.Such anintegrativeapproachcouldbringustoapivotalstageinthehistoryofbrainmappingand cognitive neuroscience, in which we lift the conceptual fog that has clouded structure– functionrelationshipsinthebrain,andfocusonfutureformalconceptualizationsoffunctional segregationandintegration.

Highlights

While itislargelyaccepted that the

brainistopographicallyorganizedinto

distinctareasthatareintegratedinto

networks,theunique contributionof

each areato behavior is yet to be

elucidated.

Diverselinesofresearchusingdifferent

approaches have contributed to

numerousbehavioralassociationsfor

anybrainarea.

Emerging databases of task-based

activationdataofferthepossibilityof

characterizing the engagement of

brain regionsacross a broadrange

ofexperimentalbehavioralconditions.

Newlargesamplesofbothimaging

and phenotypical data provide an

opportunitytocomplementthe

activa-tionpatternbyexamining

cross-sub-ject associations between

imaging-derivedneurobiological markers and

ecologicalbehavioralcharacteristics.

1

InstituteofNeuroscienceand Medicine(INM-1,INM-7),Research CentreJülich,Jülich,Germany

2

InstituteofSystemsNeuroscience, MedicalFaculty,HeinrichHeine UniversityDüsseldorf,Düsseldorf, Germany

3

SchoolofPsychology,Universityof Nottingham,Nottingham,UK

4

C.andO.VogtInstituteforBrain Research,HeinrichHeineUniversity Düsseldorf,Düsseldorf,Germany

@

Twitter:@INM7_ISN

*Correspondence:

[email protected](S.Genon).

Glossary

Ecologicalvalidity:an

epistemologicalconceptreferringto

thequalityofthemethods,materials,

andsettingsofastudyusedto

reproducetheexaminedreal-life

phenomena.

Epistemological:referringtothe

studyofthenatureandgroundsof

knowledge,especiallyregarding

limitsandvalidity.

Internalvalidity:aqualitycriterion

forstudydesignsthatindicatesthe

degreetowhichobservedeffectsin

somedependentvariablecanbe

assumedtobecausedbythe

experimentalmanipulation;typically

highestinrandomizedexperiments

andlowest,orabsent,incorrelative

designs.

fMRI:canbeusedtomeasure

hemodynamicchangesrelatedto

neuralactivityduringaparticular

mentaltask.Oxyhemoglobinand

deoxyhemoglobinhavedifferent

magneticproperties,whichcanbe

capturedwithanfMRIscanner.

Duringmentalactivity,theratio

betweenoxyhemoglobinand

deoxyhemoglobinismodified,

allowinginferenceabouttheregion

ofthebraininwhichneuralactivity

changesduringaparticularmental

task.

MRI:atechniqueofimagingbody

tissues(suchasbraintissues)and

physiologicalprocesses.Ituses

magneticfields,radiowaves,and

fieldgradients.Asdifferenttissues

havedifferentmagneticproperties,

structuralMRIcanbeusedto

generateananatomicalimageofthe

brainthatdifferentiatesgraymatter,

whitematter,andcerebro-spinal

fluid.

Ontology:anexplicitspecificationof

theconceptualentitiesthatare

postulatedbyatheory.Aformal

ontologyspecifiesthestructureof

thetheoryintermsoftheelementary

entitiesandtheirconceptual

relationships.

Positronemissiontomography (PET):atechniquethatmeasures

physiologicalfunctionbydetecting

bloodflowandmetabolism.Itis

basedonthedetectionof

radioactivityemittedaftera

radioactivetracerhasbeeninjected

intothebody.Itcanthereforebe

usedtoexaminebloodflowand

oxygenconsumptionindifferent

BrainAreasasConnectedComputationalUnits

The first theories regarding functional specialization of brain areas (which later led to the conceptoffunctionalsegregation)hadalreadybeenproposedinthe early19th centuryby Gall(whoseviewwaslaterreferredtoas‘phrenology’)[6].However,many‘functions’thatwere associatedwithcertainpartsoftheouterskullwouldnotbeconsideredasfunctionsfroma modernpointofview(withtheexceptionoflanguage).Thefollowingdecadeswereenlivenedby debatesonlocalizationismversusconnectionism(foradetailedreview,see[7]).Thepioneering work performed byPaul Broca andCarl Wernicke inthe 19th century evidenced specific behavioralimpairmentfollowingfocalbrainlesions,but,atthesametime,itwasalsorealized thattheattributionofaspecificfunctiontoacorticalareaisrelatedtoitsanatomicalconnectivity withdistantbrainregions.ThiswasillustratedbyWernicke,whointroducedthefirstnetwork viewforlanguagecomprehensionandproduction[8].

Followingthisview,theconceptofdisconnectionsyndromesrefutedstrictlocalizationismasa complete or sufficient account of cortical organization [7]. Accordingly, the human brain mappingfieldcurrentlyreliesontheassumptionthatthebrainisgovernedbytwofundamental principlesoffunctionalorganization:segregationandintegration[7,9,10].Theformerrefersto the fact that thecerebral cortex isnot ahomogeneous entity butcan be subdividedinto regionally distinct modules (cortical areas or subcortical nuclei), based on functional and structuralproperties[11,12].Thelatteremphasizesthatnobrainregionisbyitselfsufficient toperformaparticularcognitive,sensory,ormotorfunction.Rather,allmentalcapacitiesrely onadynamicinterplayandexchangeofinformationbetweendifferentregions[13,14].

Importantly,theseprinciples(functionalsegregationandintegration)donotcontradicteach other,since integrationcanbeconceptualized asinteraction betweenrelativelyspecialized regions,eachsubservingadistinctprocess[9,15].Accordingly,eachareacanperformalimited rangeoffunctions,buttheconcretebehavioraloutputdependsonwhichinputshavebeen processed(fromafferentconnectivity)andwhichsignalissenttowhichotherareas(basedon efferentconnectivity).Inthis‘intrinsic’and‘connectivity’-basedfunctionalspecializationofbrain areas(developedin[16]),theselattercanbeconceptualizedasrelativelyspecialized compu-tationalunits, theobservedbehavioraleffects ofwhichdepend onthecoactivity(and thus informationsentandreceived)ofotherareas.

Consideringbrainareasascomputationalunitsraisesthequestionofthemechanismofthe computation,orbasicallythequestionof‘whatdoesthebraindoandhow?’Arelatively well-acknowledgedviewaddressingthisquestionisthatthebrainworksasaBayesianmachine [17],computingprobabilitiesthatminimizeuncertainty[18]andsupportdecisionmaking[19]. Thisviewhasbeensuccessful,forinstance,inexplainingperceptualprocessesasintegrative processingofprobabilisticdistributions[20].Ithasalsobeenadaptedtoexplainhigheraspects ofcognition,suchashumanoptimisticbias[21]and,relatedly,theroleofthelateralprefrontal cortexinupdatingbeliefs[22].

regions.Specifically,‘function’referstoacomputationaloperationperformedbyagivenregion, whichcontributestotheobservedbehavioraloutput.Toincorporatethisviewpoint,weusethe term‘operation-function’.

Inthefollowingsections,wefirstillustratetheheterogeneityofbehavioralfunctionsthathave beenassignedtobrainregions,accordingtopreviousapproaches;theseapproachesarethen reviewed.Wethenconsiderhowfunctionalspecializationfromsuchconceptualizationsmight beusedinconjunctionwithrecentadvancesindataaggregationmethodstosearchforthe coreoperation-functionofbrainregions.

FunctionalSpecializationsasPolyhedra

For any brain region, we can think of many different behavioral functions, based on the perspectivefrom whichweconsiderthisbrainregion.Inpractice,mostofthesebehavioral functionscansomehowberelatedtooneanotherandseemtocompriseacorecomputational function(i.e.,anoperation-function)thatgroundsallbehavioralassociationsbutremainslatent andisnotdirectlyobserved.Inotherwords,ourcurrentknowledgeofthefunctional speciali-zationofagivenbrainregioncanbeconceptualizedasapolyhedronwithitsmanysides(i.e., manybehavioralfunctions),thesumofwhichcanonlybeappreciatedbyinvestigationfrom manydifferentperspectives,butwhosecorecenterremainsintangible.

Thisconceptualpolyhedroncanbeillustratedbyoneofthemoststudiedpartsofthebrain,the hippocampus.Ithasbeenassociatedwithdifferentmemoryfunctions,suchasepisodic[23], autobiographical[24],explicit[25],contextual[26],orassociative[27]memory,andalsowith several‘processes’,includingdeclarative[28] orincrementallearning[29],recollection[30], encoding[31],retention[32],consolidation[33],noveltydetection[34],binding[35], compar-ator[36],mismatchdetection [37],pattern separation [38],and inferential processes[39]. Furthermore,thehippocampushasbeenassociatedwithparticular‘behavioraldomains’and ‘tasks’ such as spatial navigation [40], spatial discrimination [41], scene imagination [42], prospection[43],and allocentric representation[44]. Finally,it hasbeenassumed thatthe hippocampus supports more complex behavioral constructs, such as creative thinking or

flexiblecognition[45].

How hassucha‘functionalpolyhedron’beencreated? Actually,the hippocampus,likemanybrain regions,isacomplexstructurethatcanbeengaged inmany differentbehavioralfunctions, accordingtothecontextandtoitsinteractionwithotherbrainregions.Accordingly,differentfields haveuseddifferentapproachesandconceptualframeworkstoinferbrain–behaviorrelationships, capturingoneofitsmanybehavioralassociations.Thatis,ascribingbehavioralfunctionstoabrain regionhasbeenamultidisciplinaryendeavor,resultinginmultipleontologies(Box1)anddifferent levels of descriptionof mentalfunctions, rangingfrom high-level behavioral descriptions to individualtasksandisolatedprocesseshypothesizedbycognitivemodels[46].Generally,when theseconstructshavebeenrelatedtothebrain,thedisciplinehasshapedtheontologies,andthe inferentialapproachhasdriventhelevelofdescription,producingheterogeneousconceptual associationsforanybrainregion.Inthefollowingsections,wereviewtheinferentialapproaches usedinthosedifferentdisciplines,andtheirconcepts,advantages,anddrawbacks.

HowHaveFunctionsofBrainRegionsBeenInferred?

TheLesion-DeficitApproach

One of the first approaches linking brain and behavior was the lesion-deficit approach. FollowingthepioneeringworkofBrocaandWernickeinthe 19thcentury,oneofthemost

partsofthebrainduringamental

task.

Transcranialdirectcurrent stimulation:atechniqueof

neurostimulationinwhichelectrodes

areplacedontheheadofa

participanttoinducecurrents.It

changesneuronexcitabilityby

modifyingthemembranepolarization

potential.

Transcranialmagneticstimulation (TMS):atechniqueusedtostimulate

thebrainlocally.Amagneticcoilis

placedclosetothescalp(without

physicalcontactwiththehead)to

inducecurrents,whichchangethe

polarizationoftheneuronsinthe

areaunderthecoil.Theneuraleffect

ofTMSdependsonthefrequencyof

thestimulation.Itcanbeexcitatory

famousexamplesinthe20thcenturywas thestudyofpatientH.M.by BrendaMilnerand colleagues.Thispatienthadmedialtemporalloberesection(mainlythehippocampus),which resultedinsevereanterogradememorydeficits[47].Thisledtotheinferencethatthe hippo-campusplaysacrucialroleintheacquisitionofnewmemories.Asexemplifiedinthisfamous study,a‘behavioralfunction’isthusinferredfromanobserved‘dysfunction’(hereanterograde amnesia)followingrestrictedbraindamageorloss.Themainstrengthofthisapproachisthe causalnatureoftherelationshipbetweenthetwostudiedvariables(brainandbehavior).Thatis, observingadysfunctionfollowingdamagetoaspecificbrainregionallowsonetoinferacrucial roleofthisregionintherespectivebehavioralfunctionwithintheintactbrain.Thisstrengthgoes

withthe epistemological(seeGlossary)limitationofbeing onlyquasi-experimental, asan

experimentalapproachsupportingcausalityimpliestheabilitytodemonstratethatalternative explanationshavebeeneliminated.Specifically,inthelesion-dysfunctionapproach,theeffect ofprelesionfactors(e.g.,subclinicalstrokesand/orcognitiveimpairment)cannotberuledout. Inotherwords,thebehavioraldeficit(i.e.,theeffectonthedependentvariable)could(partly)be driven byfactors other than the lesion per se, thereby threatening internal validity [48]. Furthermore,atthebrainlevel,severalissuesarisingfromthespatiallystructureddistributionof lesions[49]andtheinfluenceofneuroplasticity(functionalandstructuraladaptationtodamage) canlimittheinferentialpoweroflesion-deficitmapping(Box2).Despitetheselimitations,this approachhasshapedmanyofthemostpre-eminentassumptionsaboutfunctional speciali-zation,andisstillconsideredabenchmark(duetoitscausalmechanism)againstwhichfindings obtainedfromotherapproachesarediscussed[49].

TheStimulationApproach

Amorerecent,experimentalapproach,mirroringthelesion-deficit approachbutappliedto healthyparticipants,isthestudyofbehavioralconsequencesofvirtuallesionscreatedwith brain stimulation techniques. In particular, brain activity can be locally impaired with

Box1.MatchingBrainOrganizationandCognitiveOntologies

Oneimportant issueinthe studyof brain–behavior relationships iswhether currentlyavailable ontologies and

taxonomiescanbemappedontothebrain.Thatis,ifwecouldcapturetheexacttopographicalorganizationof

thebrain,itseemsunlikelythatwewouldfindconceptsatthecurrentlevelofdescriptionofbehaviorthatcouldbe

mappedtotheidentifiedbiologicalunits.Itseemslikelythatthebehavioralstructureasitisimplemented(i.e.,‘coded’)in

thebraindoesnotfitwithpastandcontemporarycognitivetheoriesofbehavioralprocesses.

Asreviewedin[16],cognitivescientistshavetraditionallyformalizedthecomponentsofbehavioralfunctionusing

behavioralstudiesofnormalandneurologicallyimpairedindividuals.Asfurtherreviewedin[77,93],theresultsof

activationstudieshavechallengedtheclassicmodels,astheyhaveevidencedoverlapbetweenneuralsystems

activatedbytasksthatsharenoknowncognitivecomponents.Inagreementwithaprevioussuggestionof

system-aticallyassigninglabelsthatencompasstheoperationsthateachareaperforms[8],weproposetobuildasystematic

ontologywithabottom-upperspective(startingfromthebiologicalsubstrate:thebrain).

Onechallengeinthefuturewillindeedbetointegrateconceptsanddatafromdisparatebrainsciencedisciplineswithin

aunifiedframeworkofbrainbiology[94].Followingupthatperspective,buildingneurocognitivemodels(i.e.,models

combiningcognitiveandanatomicalmodels)ofnormalfunctioningandpathologyreliesontheorganizationofthe

cognitivecomponentsintoasingleframework[16].Suchatheoreticalpositionclearlyimpliesthatbiologicalevidence

shoulddrivetherevisionofcognitivetheories,providedthatthenewconceptualizationderivedfrombrain–behaviordata

hasbeenrobustlytested.

Inadditiontothesetheoreticalconsiderations,oneshouldalsoconsidertheclinicalutilityasa‘qualitymarker’for

psychologicalontologies.Thatis,abehavioral/cognitivetaxonomythatisinagreementwithbrainorganizationshould,in

principle,alsohelpresearcherstounderstand,diagnose,orclassifyneurocognitivepathologies.Thus,inthefuture,

assigningoperation-functiontobrainregionsshouldgohandinhandwiththeevolutionofformalcognitivetaxonomies,

transcranialmagneticstimulation(TMS),transcranial directcurrentstimulation,or transcranialalternating currentstimulation.Likewise, theopposite effect (i.e.,facilitationby increasingcorticalexcitabilitytoenhancebehavioralperformance)isalsopossible,depending ontheprotocol[50,51].Theseapproachestestcause–effectrelationshipsbyexperimentally manipulatinglocalbrainactivityandexaminingitseffectsonbehavior[52].Whilerepresentinga powerful experimental approach, internal validity can also be limited by the influence of individualcorticalgeometryandthe relativelackof focality,as wellas bythelimitedrange ofregionsthatcanbetargeted(Box2).Atthebehaviorallevel,incontrasttothelesion-deficit approach,stimulationapproachesdonottapintoeverydaybehaviorinnaturalsettings.Forthe sakeof experimentalcontrol andconstraintsof thelaboratory settingneededfor stimulus delivery,behavioralfunctionsareusuallyoperationalizedfromcognitivemodels(e.g.,‘memory recall’[53]),andinferenceismadeonisolatedbehavioralparameterssuchasreactiontime. Thus, compared with the lesion-deficit approach, experimental brain stimulation can offer higherinternalvalidity,buthaslimitedecologicalvalidity.

TheActivationApproach

Inrecentdecades,neuroimagingtechniquessuchaspositronemissiontomography(PET) andfMRIhaveproducedarapidgrowthinthestudyofbrain–behaviorrelationships[54],by revealinglocalizationsofbrainactivitychangesinduced bymentaloperations.fMRIquickly becamepreferredoverPETformappingtask-evokedactivitybecauseithasabetterspatialand temporalresolution[55];andcanhencelocalizeactivitychangesduringspecificmentalevents, suchas successfulmemory encoding[56].Withthistechnologicalprogress,cognitive psy-chologistsgaineda newtool to test andrefinecognitive models and theories[57].In this particularframework,theactivationapproachcanbeconsideredasexperimentalbecauseit allowstheresearchertofreelymanipulateanindependentvariable(behavioralcondition)and observeitseffectonthedependentvariable(brainactivation).Forexample,fMRIcanprovide supportfordual-processcognitivemodels(suchasrecollectionversusfamiliarity)by demon-stratingthatthetwoprocessesevokedistinctpatternsofactivity.However,addressingsuch questionswithfMRIrequireswell-controlleddesigns,usingtwoconditionswhichdifferonlyin

Box2.RealandVirtualLesions:StrongbutComplicatedEvidence

Localizedbrainlesionsandstimulationapproachesproducing‘virtual’lesionsmaybeconsideredasprovidingthe

strongestevidencethatthefunctionofabrainregioniscausallyrelatedtoaparticularbehavioralfunction,ifdamageto

theregionindeeddisruptstheperformanceoftherespectivebehavior.Thislevelofinterpretationisinaccessiblefor

eitheractivationstudiesorbrain–behaviorcorrelations,whichareobservationalinnatureandmaynotdifferentiate

epiphenomenaorspuriouseffects.

Inturn,lesion–symptomassociationsarecomplicatedbythehighplasticityofthehumanbrain(e.g.,[95,96]),which

resultsinsubstantialremodelingofcircuitryasearlyasdaysaftertheinsult.Inaddition,lesionlocationsareneither

uniformnorrandom,buteitherfollowaspatialstructuredeterminedbythevasculartree,incaseofischemiclesion,or

mainlyoccurinsurfacestructures,inthecaseoftraumaticbraininjury[49,97].Furthermore,otherfactorsmay

contributetotheobservedpattern:forexample,regionsthatappeartobestructurallyintactmighthavetheirfunction

impairedbydisconnectionfrom,ordamageto,animportant‘coworker’.Astheseregionsarealsounabletofunction

despitebeingstructurallyintact,theperturbationsascribedtothedamagedregionsmaybeoverestimated[97,98].

Noninvasivestimulationtechniques offeran experimentalapproach toperturbationandavoid thelimitationsof

neuropsychologicallesionmappingsuchasplasticity(thoughthereisevidenceforshort-termhomeostaticeffects),

buttheirapplicationislargelylimitedtosurfacestructures[99,100].Furthermore,thequestionablefocalityoftranscranial

magneticstimulation-induced currents[10]andinfluencesofcortical geometry substantiallyimpedethe spatial

specificityofbrainstimulation.Finally,stimulationofdenselyconnectedregionswillresultinuncontrolledpropagation

ofthepulse into spatiallydistantregions [5,6],inducing (uncontrolled)network effects.Thus,a well-controlled

experimentbasedonastimulationapproachshouldincludeanexaminationofthepropagationofthepulse,such

respect to the target processes (i.e., a ‘pure insertion’; cf. [58]). For example, isolating ‘recollection’inanfMRIscannercouldbeoperationalizedbycontrastingrecognitionofintact wordpairs(whichengagebothrecollectionandfamiliarity)withrecognitionofrecombinedword pairs(drivenonlybyfamiliarity)[59,60].Hence,mentalfunctionsarestudiedusingveryprecise andthereforerestrictedexperimentalimplementationsinferredfromcognitivetheories. Con-sequently, as in stimulation studies, the mental operations a participant engages in are conceptuallydistantfromeverydayfunctions(suchasthevividrecollectionofarecentmeeting), thereforelimiting ecologicalvalidity.Withregardto internalvalidity,the assumptionofpure insertion (the assumption that extra processes can be inserted purely, without changing existingprocesses oreliciting newprocesses), uponwhich the interpretation of activation relies,hasbeenquestioned(cf.[58]).Forexample,theinternalvalidityofanfMRIstudycanbe questionedbythefactthatepiphenomenaoftheexperimentalsetting(suchashigher atten-tionaldemandsinataskthaninacontrolcondition)cannoteasilybedissociatedfrom task-specificeffects.

TheDegeneracyPrinciple

Historically,lesion-deficitapproaches(betheyneuropsychologicalorstimulationbased)were generallyconsideredtobeimportantwithrespecttothestudyoffunctionalspecializationof brainregions,becauseobservingarelationshipbetweenafocallesionandabehavioraldeficit suggeststhatthisregionisnecessaryforperformance.Hence,thelesionapproachwaslong consideredthegoldstandardforidentifyingthenecessityofabrainregionforagivenfunction. Bycontrast,theactivationapproachwasconsideredtheoptimalapproachforidentifyingwhich brainareasweresufficientforagivenbehavioralfunction.Accordingly,itwasinitiallyhopedthat acombinationoflesion-deficitmappingandactivationapproacheswouldidentifythe neces-saryandsufficientbrainregionsforagivenbehavioraltask[58].However,thisviewhadtobe revisedsubjecttothedegeneracyprinciple(i.e.,thefactthatoneuniquebehavioraloutputor outcomecanbeachievedbydifferentneurocognitivesystems[61]).Thisdegeneracytheory resultedinaconceptualmourninginthehumanbrainmappingfield,asitwasnowconsidered that‘theremaybenonecessaryandsufficientbrainarea’foranybehavioralfunction[58].

TheStructure-BehaviorCorrelationApproach

underminedbyalackofexperimentalcontrol,whichimpliesthatpossiblealternativeprocesses andstrategiesmayyieldthesameorsimilarbehavioraloutcomes[61].Theinternalvalidityis furtherreducedbytheconsiderationthatbehavioralmeasurementscouldberelativelynoisy proxiesofthelatentconstruct(s)theyaimtotarget[68].Thisconcernisespeciallynoteworthy forscoresbasedonsubjectivereports,thereliabilityofwhichisfrequentlyquestioned[68]. Moreover,neurobiologicalfeatureslikelocalvolumeorcorticalthicknessestimatesareinfl u-encedbynumerouspossiblefactors,whichmayshowcomplexrelationshipswiththecovariate ofinterest.This,inturn,mayleadtospuriousbrain–behaviorassociations(cf.discussion[69]).

SummaryandConclusions

Insummary,associationsbetweenbehavioralfunctionsandbrainregionshavebeenstudiedby differentresearchfields,whichhavedifferentconceptsofbehavioralfunctionsandusedifferent inferenceapproaches.Beyondtechnical strengths andweaknesses,the potentialof these differentapproachestoassociateparticularbehavioralfunctionswithparticularbrainregions hasbeendiscussedwithrespecttoecologicalvalidityandinternalvalidity(twoepistemological qualitiesconsideredinbehavioralsciences).Withalltheirdifferentstrengths,theseapproaches havetogethercontributedtoassignmultiplebehavioralfunctions,correspondingtodifferent levelsofdescription,tobrainregions.

Althoughthecomplementarityofthedifferentapproachescanbeseenasofferingrichnessin behavioralassociationsforbrainregions,theoriginalgoalofindividualstudieswastypicallyto focusonabehavioralfunctionandmapittoabrainregion,ratherthanelucidatingtheexact functionofagivenregion.Thatis,theaprioridefinedconstructwasamentaloperation,andthe objectofinferencewasthebrainregionthatwasrelated toit. Wewouldcontendthatthis modusoperandihasonlyaverylimitedcapacitytoanswertheinitialquestion:‘Whatdoesany partofthebraindo?’Inparticular,anyinferenceabouttheroleofanybrainregionthatisderived usingthismodusoperandiiscomplicatedbytheprincipleofdegeneracy[61].

Forexample,investigation into the associationbetween the hippocampusand associative memoryretrievalcanbeobscuredbythefactthatrecallinganassociationoftwoitemscanbe performedeitherbyretrievingaunitizeditemintegratingbothcomponents,orbyretrievingthe two associated items; with the second cognitive strategy being more likely to recruit the hippocampus[70,71]. Accordingly,regardless ofthe approach used,finding arole forthe hippocampus inthe behavioral outcome depends onthe neurocognitive aspectsthat the behavioralparadigm ormeasure mostlycaptures, and theneurocognitive system that the individual(s)recruit(differentparticipantscanrecruitdifferentneurocognitivesystems). Conse-quently,identifyinga roleof thehippocampus inassociativeretrievalcould requireseveral studies, covering a very heterogeneous range of behavioral paradigms or measuresand performedacrossdifferentpopulationsamples.

RecentToolsAidingProgress

ActivationDataAggregation

Anoverviewofthebehavioralfunctionsengagingagivenbrainregioncouldbeachievedby scanningagroupofindividualsforalargerangeofbehavioralconditionsthattargetdifferent mentalfunctions.Thisapproachhasrecentlybeenundertakenwith12participantsaspartof theIndividualBrainChartingproject.iWithitsongoingdevelopmentofspecificdecodingtools, thisandrelatedprojectscouldsignificantlycontributetoourunderstandingofthebehavioral engagementofbrainregions,providingrichandheterogeneouspatternsofregion–behavior associations at the individual level. Nevertheless, ensuring that patterns ofassociation go beyondtheidiosyncrasiesofthespecificexperimentaldesignsandparticipantswillrequiredata integrationacrossmanyindependentstudies[73].Thus,a‘subjectlevel’functionalpolyhedron thatcapitalizesonaggregationofexperimentswithin-subjectcouldcomplementfindingsfrom across-studiesdataaggregation.

Integrationacrossstudieshasnowbecomepossibleduetotwoinitiativescompilingtheresults ofpublishedactivationstudies:BrainMap[74,75]andNeurosynth[76].Althoughdifferingin theirapproachtodataextractionandlabeling(Box2),bothcontainthecoordinatesoflocal activation maxima as reported by many thousands of neuroimaging papers, along with descriptions ofthe behavioralconditions thatyielded the respectiveactivity patterns(e.g., ‘saccades’).Byapplyingstatisticaltestsaccountingforthebaserateofactivationforagiven regionandthe baserate ofeach behavioralconditioninthe database,the consistencyof particularbehavioralassociationsacrossthousandsofstudiescanbeanalyzedforanyregion ofinterest.Such approachescouldbeseenas ‘functionalbehavioralprofiling’(‘functional’ referringtotheuseofactivationdata).

Usingsuchanapproach,ithasrecentlybeendemonstratedthattheanteriorinsulaisengaged inaverywiderangeoffMRItasks[77],suggesting ageneric functionalrole, suchastask engagementmaintenance;thatcouldaccountforallthemorespecificmentalprocessesthat havepreviouslybeendiscussedforthisregion.Asillustratedinthisexample,thepatternsof associationsacrossawiderangeoftaskscanfosternewhypotheses,approximatingasmuch aspossiblethecoreroleoftheregion(andthusitsoperation-function),beyondthebehavioral

ontologyoftheoriginalstudies orthedatabase.Inarecentstudy, screeningtherangeof

studiesactivatingtheleftrostraldorsalpremotorcortex(PMd)revealedthatthissubregionwas activatedwheneverthetaskrequiredabstractionfromtheactualspatial(e.g.,scene imagina-tion),temporal(e.g.,explicitmemory),ormentalframe(e.g.,deception)([78],Figure1).This observationwasonlypossibleafterintegratingactivationsacrossdifferenttasksandbehavioral domains,and wecanspeculatethat this‘abstraction’ propertyactually reflects theuseof sequentialprocessing(spatialortemporal)inthePMdforvarioustypesofpredictionsbeyond thecurrentframework,inlinewiththeBayesianbrainhypothesis.

Neurosynthshowcomplementarylimitationsandadvantages,suggestingthattheir combina-tioncouldprovideamorecomprehensiveprofilingandbetteroverviewthanthepreviousfocus oneitheroftheminisolation.Finally,whileaquantitativesummaryofactivationdatamaythus disclosepatternsacrosstasksfromdifferentresearchfields,itdoesnotallowfordisentangling whethertheengagementofthebrainregionplaysacrucialroleintaskperformanceorwhether itis justan epiphenomenonrelatedto experimentalimplementation (suchas moreintense visualfixationorcognitiveengagement).Thishighlightsthepotentialbenefitofacorrelational approachusingmorenaturalistictaskstocomplementourviewonbehavioralassociationsfora givenbrainregion.Thepotentialoutcomesandlimitationsofsuchanapproacharediscussed inthenextsection.

Explicit

memory

Objects/scenes

imagina

Ɵ

on

Decep

Ɵ

on

Brain region

Behavioral pro

fi

le

Ac

Ɵ

va

Ɵ

on database

Behavioral condi

Ɵ

ons

25 + 32 = ?

Similar to 2-back?

M

Plant or animal?

Horse

Font color?

Green

What was her name? 1. Linda 2. Susan 3. Helen

Are you?

Self-con

fi

dent

Ac

Ɵ

va

Ɵ

on peaks

Same or different?

ζ

ζ

?

Figure1.IllustrationofBehavioralFunctionalPro_filingfortheLeftRostralDorsalPremotorCortex(PMd).Activationdatabases(suchasBrainMapand

Neurosynth)containacollectionofactivationpeaksthathavebeenreportedinstereotacticspaceinscientificpapers,aswellasinformationonbehavioralconditions

associatedwiththesepeaks(basedonthebehavioraltaskthattheparticipantshadtoperformintheMRIscanner).Foragivenbrainregion-of-interest(heretherostral

leftPMd),wesearchedamongallthepeaksofactivationreportedintheBrainMapdatabaseforthosethatwerelocatedinthisregion.Inthisdatabase,thebehavioral

conditionrelatedtoeachpeakisspecifiedintermsofbehavioralparadigmsandbehaviordomains.Examiningthebehavioralparadigmsandbehavioraldomainsin

whichthepeaksofactivationwereconsistentlyreportedintheregion-of-interestallowedustoestablishabehavioralprofileofthisregion.Asillustratedintheleftinferior

panel,theleftrostralPMdwasfoundtobeactivatedinexperimentaltasksprobingexplicitmemory,objectorsceneimagination,anddeception[78].Thefaceusedto

CorrelationinLarge-ScalePopulationSamples

Anemergingethosofdatasharinghaspromotedopenaccesstoagrowingnumberoflarge datasetsof neuroimagingand phenotypicaldata[82,83].TheHuman ConnectomeProject (HCP,[84]),the1000Brainsstudy[85],andtheUKBiobank[86]areinstancesofsuchinitiatives. Theyprovidemultimodalbrainimaging,informationonhistoryandcurrentlife-style, question-nairescores,andasubstantialrangeofstandardneuropsychologicalmeasuresthataddress several cognitive dimensions, such as working memory, executive functions, and verbal learning.BraincharacteristicsmeasuredwithMRIinlarge-scalepopulationdatashowanatural covariancewithcognitivephenotypes(Figure2A, [86]),whichallowsastandardcorrelation approachforidentificationofspecificbrainregionsthatarerelatedtobehavioraldimensionsof aprioriinterest(suchasconscientiousness[65],Figure2B).Itwouldalsoallowevaluationofa specificassociationbetweenaregionofinterestandaprioriselectedbehavioralvariables(such ashippocampusvolumeandmemoryperformance[87],Figure2C)withinahypothesis-driven framework.

Supportingthevalidityofsuchanapproachtocapturebrain–behaviorrelationships,measures tappingintosimilaraspectsofbehaviortendtoshowcorrelationinthesamebrainregion(such asimmediaterecallanddelayedrecallinalist-learningtask[87],orextraversionand consci-entiousnessintheassessmentofpersonality[65]).Suchrelationshipsopentheperspectiveof anexploratoryapproachsearchingforsignificantassociationsbetweenbrainmeasurementsin anaprioriselectedregionofinterestand awiderangeofpsychometricvariables. Thatis, capitalizingonthehypothesisthatneurobiologicalfeaturessuchasgray-mattervolumeand corticalthicknesscovarylocallywithbehavioralcharacteristics[62,88],thebehavioralfunctions inwhichagivenbrainregionpotentiallyplayarelativerolecouldbeinferredfromitsstructural brain–behaviorcorrelationacrosstherangeofphenotypicalvariables.Asthisapproachisbuilt onaverydistinctconceptualizationofmentalfunctionsthroughtheassessmentofcomplex,

Box3.BrainMapandNeurosynth

BrainMapandNeurosyntharebothcollectionsofresultsofactivationstudies,consistingofreportedcoordinatesand

informationabouttheexperimentalcontext.BrainMap[75,101]isbasedonmanualencodingofspatialcoordinatesand

behavioralconditions,accordingtoanexpert-definedontology.Thatis,eachreportedsetofcoordinatesisencoded

withrespecttotheemployedparadigm,theassessedcontrast,andotheraspects suchasstimuliorrequired

responses.Thislabelingiscross-validatedbyasecondinvestigator,yieldingarigorousstandardoflabeling,resulting

inratherslowgrowthandconfinementtoanaprioritaxonomy.

Neurosynth[76],bycontrast,isbasedonautomatedtext-mining.First,coordinatesareextractedfrompublished

neuroimagingarticlesusinganautomatedparser,withoutdistinctionbetweendifferentcontrastsorexperiments.Thus,

onemajordifferencebetweenbothdatabasesisthatBrainMapworksontheexperimentlevel(singlecontrast)and

Neurosynthonthestudylevel(completepaper).Inthelatterdatabase,eacharticleisthen‘tagged'withthosetermsthat

occurwithhighfrequencyintheabstractofthepaper.Thisautomatedparsinghastheadvantageofnotbeinglimitedby

apredefinedsetoflabels.Bycontrast,itcomeswiththedrawbackthatthedescriptionsaresolelybasedonthewords

usedtodescribethestudybytheauthors.Consequently,thelabelswillsummarizetheconceptualtermsthatthe

researchers(orthereviewers)wantedtoseeaddressed,notnecessarilythebehavioralfunctionorprocessthatwastruly

isolated.

Thus,bothdatabasesarebasedoncurrentpsychologicalontologiesandterminologybutmayprovideslightlydifferent

behavioralinformationforagivenbrainregion.Forexample,memory-relatedtermsassociatedwiththehippocampusin

BrainMapwouldbe‘explicitmemory’,whileNeurosynthwouldprovidetermssuchas‘remember’,and‘remembering’.

Whiletheformerlackssemanticprecision,thelattercouldbe(spuriously)drivenbythefactthatmanystudies

addressing‘remember’havefocusedonthemedialtemporallobe.Asregion-of-intereststudiesarenotexcluded

inNeurosynth,thereisapotentialforself-fulfillingprophecies.Hence,eachapproachhasitsownstrengthsand

3 4 5 6 7 8 9 13.5

11.5 9.5 7.5 5.5 3.5 1.5 -0.5 -2.5

40

30

20

10

0

-log

10

(P)

1400

1200

1000

800

600

400

10 20 30 40 50

Hippocampal/intracranial volume raƟo

Female Male

GMV in cuneus

ConscienƟousness

UK biobank sample (n = 5285)

HCP sample (n = 364)

ADNI MCI sample (n = 694)

(A)

(B) (C)

Dela

ye

d r

ec

all

Female Male

Ph

ysica

l

bo

ne

density and size

Lifestyle general

Ea

rly life

fa

cto

rs

Lifestyle

exercise and

work

Life

style

fo

od

and

dr

in

k

Lifestyle alcohol _Lifestyle

to

ba

cco

Ph

ysica

l

general

Ph

ysica

l ca

rd

ia

c

Blo

od

a

ssa

ys

Co

gni

Ɵ

ve

phenoty

pes

Bonf FDR

Figure2.StructuralBrain–BehaviorCorrelationApproach.(A)Manhattanplotrelatingasetofcerebralmeasuresderivedfromanatomicalimagestonon-brain

phenotypicalvariables(1100variablesclusteredintomajorgroupsalongthex-axis)intheUKBioBankcohort.Foreachvariable,thesignificanceofthecross-subject

correlationwitheachbrainmeasuresetisplottedverticallyinunitsof log10[86].(B)Positivecorrelationbetweenconscientiousnessscoresandgray-mattervolume

(GMV)inthecuneusinmalesonly,withinasamplefromtheHumanConnectomeProject(HCP)[65].(C)Significantrelationshipbetweenhippocampalvolumeand

immediaterecallperformanceattheReyAuditoryVerbalLearningTestinteractingwithgenderinasampleofparticipantswithmildcognitiveimpairment(MCI)fromthe

ecologicallymorevalidtasksthanthespecificcontrastsofferedbytheactivationapproach,it shouldrevealcomplementarypatternsofbehavioralassociationforany givenregion.Thus, ultimately,for a given brainregion,the pattern of behavioralassociations revealed bythis structuralcorrelationapproachshouldbeintegratedwiththepatternofbehavioralassociations derivedfromactivationdata,toofferamulticonceptualandmultimodalpatternofbehavioral associationsforanybrainregion.Thishybridapproachwould,inturn,helptodevelopnew hypothesesontheoperation-functionofanyregion.

TowardTestingofInteractionModelsandFinerScales

Thecorrelationapproachand,moregenerally,anydata-drivenapproachusingbigdatasetsin whichresearchersjust‘letthedataspeak’havetheirownlimitations,astheneuroimagingand psychometricdatamaycontainsubstantialnoise[89],withconfoundingfactorspartlydriving theobservedeffect[68].Ultimately,thefunctionofanybrainregionshouldbeconsideredwithin an integrativeapproach, including not only patternsrevealed bylocal properties,but also interactionswithother brainregions.In otherwords,data-drivenapproachesthatadopta functionalsegregationviewandareappliedtoaggregatedobservations,offeragreat oppor-tunity for exploratory work and discovery science, but any resulting ‘operation-function’ hypothesis should be integrated into a functional model, tested with a hypothesis-based approach. As discussed previously, each approach has its own technical and scientific strengthsandlimitations,suggestingthatacomprehensiveevaluationofagivenhypothesis shouldcapitalizeonacombinationofdifferentapproaches,ratherthanfocusexclusivelyonany oneofthem(seeOutstandingQuestions). Technicaladvancescannowofferbetter experi-mentalcontrol,suchasbycombiningelectroencephalography(EEG)withfocalbrain stimula-tion[90].Furthermore,Bayesian-basedmethodologicalframeworks,suchasdynamiccausal modeling,have been successful inrecent years in the statistical testingof neurocognitive models,andcannowevenbeextendedtocombinedEEG–fMRIparadigms[91].Altogether, thesetechnicalandmethodologicaldevelopmentsholdgreatpromisefortestingmodelsof operation-functionscomputedbydifferentbrainareasininteraction.

ConcludingRemarks

Despitecenturiesofstudyofbrain–behaviorrelationships,aclearformalizationofthefunction ofmanybrainregions,accountingfor theengagementofthe regionindifferentbehavioral functions,islacking.Recentprogressindataaggregationmethodshasopenedawideavenue for a systematic, multiconceptual characterization of behavioral associations for any brain region.Onthe onehand, previousdecadesof fMRIandPET activationexperimentshave providedawealthofresultsthatcanbeusedtoshifttheperspectivetowardsearchingforthe rangeofbehavioralassociationsofbrainregionsusingaquantitativeapproach.Ontheother hand,large-scalepopulationdatasetsopennewpossibilitiesforacomplementaryapproach basedoncovariancebetweenneurobiologicalandbehavioralfeatures.Thishybridbehavioral profilingcouldleveragethemyriadofbehavioralaspectsofanybrainregion,hence progres-sivelyunveilingthenatureofthefunctionofbrainregionsandnetworks.

Acknowledgments

OurworkissupportedbytheDeutscheForschungsgemeinschaft(DFG,GE2835/1-1,EI816/4-1),theNationalInstituteof

MentalHealth(R01-MH074457),theHelmholtzPortfolioTheme‘SupercomputingandModellingfortheHumanBrain’,and

theEuropeanUnion’sHorizon2020ResearchandInnovationProgrammeunderGrantAgreementNo.7202070(HBP

SGA1).TheauthorsthankErinSundermannandAlessandraNostroforillustrativeplots.

Resources

i

https://project.inria.fr/IBC/data/

References

1. Amunts,K.et al. (2014)Interoperableatlasesofthehuman brain.Neuroimage 99,525–532

2. Lorenz,S.et al. (2017)Twonewcytoarchitectonicareasonthe humanmid-fusiformgyrus.Cereb.Cortex27,373–385

3. Toga,A.W.etal.(2006)Towardsmultimodalatlasesofthe humanbrain.Nat.Rev.Neurosci.7,952–966

4. Eickhoff,S.B.et al. (2007)Assignmentoffunctionalactivations toprobabilisticcytoarchitectonicareasrevisited.Neuroimage

36,511–521

5. Yeo,B.T.etal.(2011)Theorganizationofthehumancerebral cortexestimatedbyintrinsicfunctionalconnectivity.J. Neuro-physiol.106,1125–1165

Box4.BrainAreasinDifferentIndividuals

Thebrain’sspatialorganizationcanbeseenasamultiscaletopographiclayout.Mappingthisorganizationinto

specializedregions thatare, inturn,integrated into largernetworks isaddressedbyvariousbrain parcellation

approaches.Severalatlasesarenowavailable.Thoseatlasesareaveragedacrosspopulationsamples,representing

criticalmapsforstudyingbrainfunctionanddysfunction.Nevertheless,thisworkiscomplicatedbyinterindividual

variability,particularlywithrespecttogyralfoldingpatterns[102].Furthermore,theorganizationofthecerebralcortex,

ratherthanbeingseenasasetofclear-cutpatches,couldbeseenasemergingfromthecrossingofdifferentgradients

offeaturesorganizedalongdifferentaxes[105].

Severalmethodologicalapproacheshavebeenproposed[inparticularmorerecentlyintheframeworkoftheHuman

ConnectomeProject(HCP)[84]andtheUKBiobank[86]]toaccountforthesefactorsthatcomplicatethematchingof

areasindifferentindividuals.OnefirstlineofimprovementliesintheacquisitionofMRIdataathigherresolution,thatcan,

forexample,distinguishareasonoppositesidesofsulcalbanks.Additionally,registrationbasedonsurfacefeaturesand

examinationofseveralgradientsoffeatures,suchasmyelinandresting-statesignal,canfurtherimprovealignmentof

structuralandfunctionalunits,respectively[102].Alsoincontrastwithtraditionalalignment,whichisbasedon

macro-anatomicalfeatures,hyperalignment(basedonfunctionalfeatures,andhenceinafunctionalspace)[103] could

significantlycontributetothestudyoffunctionalunits.

Asaforementioned,boththeHCPandtheUKBiobankofferarangeofpsychometricdata,buttheHCPandthe

IndividualBrainChartingProjectalsoallowacharacterizationofactivationacrossarangeofbehavioralparadigms.

Thus,whileBrainMapandNeurosynthcanprovidesystematicactivation-basedfunctionalprofiling,averagedacross

individuals,thespatialmappingofthestructure-functionpatterncanbefurtherexaminedwithindatasetsthatallow

high-evel alignment. These improvementshave, forexample, been used toidentifya functional gradientlying

horizontallyfromtheposteriormiddlefrontalgyrustothecrownoftheprecentralgyrus(area55b),whichisengaged

whenparticipantslistentostories[102].

OutstandingQuestions

Towhatextentdomental functions

reflectregionalspecializationof

individ-ualbrainregions,andtheintegration

thereofintolarge-scalenetworks?

How can we integrate the different

conceptsand hypotheses regarding

the behavioral functions of brain

regionsfromdifferentfieldsofresearch

anddifferentmethodologies?

Howcouldwecharacterizethe

func-tionofabrainregionbyaccountingfor

bothsegregationimplementedinthe

local structure and integration

sup-portedbytheconnectivitypattern?

Cancurrentlyavailablecognitive

ontol-ogiesbemappedontothebrain?That

is,willwefindthatspecificcognitive

conceptscanbemappedonto

spe-cificbiologicalunits?

Howcanwefostertheintegrationand

joint/complementary use ofdifferent

methods,whichallhavetheirindividual

strengthsanddrawbacks?

Whenthesystematiccharacterization

ofabrainregion,basedondata

aggre-gation,hasgeneratedanew

hypothe-sis, how can we combine different

approachestotestthishypothesisin

6. Gall,F.J.(1818)AnatomieetPhysiologieduSystêmeNerveux en Général, et du Cerveau en Particulier: Avec des Observations sur la Possibilité de Reconnoitre Plusieurs Dispositions Intellec-tuelles et Morales de l'Homme et des Animaux par la Confi gu-ration de Leurs Têtes,F.Schoell(inFrench)

7. Friston,K.J.(2011)Functionalandeffectiveconnectivity:a review.Brain Connect. 1,13–36

8. Lichtheim,L.(1885)Onaphasia.Brain 7,433–484

9. Tononi,G.et al. (1994)Ameasureforbraincomplexity:relating functionalsegregationandintegrationinthenervoussystem.

Proc. Natl. Acad. Sci. U. S. A. 91,5033–5037

10. Fox,P.T.andFriston,K.J.(2012)Distributedprocessing; dis-tributedfunctions?Neuroimage 61,407–426

11. Finger,S.(2001)Origins of Neuroscience: A History of Explora-tions into Brain Function, OxfordUniversityPress

12. Amunts,K.andZilles,K.(2015)Architectonicmappingofthe humanbrainbeyondBrodmann.Neuron 88,1086–1107

13. Bressler,S.L.(1995)Large-scalecorticalnetworksand cogni-tion.BrainRes.Rev.20,288–304

14. Tononi,G.etal.(1998)Complexityandcoherency:integrating informationinthebrain.Trends Cogn. Sci. 2,474–484

15. Friston,K.(2002)Beyondphrenology:whatcanneuroimaging tellusaboutdistributedcircuitry?Annu. Rev. Neurosci. 25, 221–250

16. Price,C.J.andFriston,K.J.(2005)Functionalontologiesfor cognition:thesystematicdefinitionofstructureandfunction.

Cogn. Neuropsychol. 22,262–275

17. Friston,K.(2012)ThehistoryofthefutureoftheBayesianbrain.

Neuroimage 62,1230–1233

18. Friston,K.(2010) Thefree-energyprinciple:aunifiedbrain theory?Nat. Rev. Neurosci. 11,127–138

19. Beck,J.M.et al. (2008)Probabilisticpopulationcodesfor Bayesiandecisionmaking.Neuron 60,1142–1152

20. Knill,D.C.andPouget,A.(2004)TheBayesianbrain:theroleof uncertaintyinneuralcodingandcomputation.Trends Neurosci.

27,712–719

21. Sharot,T.et al. (2011)Howunrealisticoptimismismaintainedin thefaceofreality.Nat.Neurosci.14,1475–1479

22. Sharot,T.etal.(2012)Selectivelyalteringbeliefformationinthe humanbrain.Proc.Natl.Acad.Sci.U.S.A.109,17058–17062

23. Horner,A.J.andDoeller,C.F.(2017)Plasticityofhippocampal memoriesinhumans.Curr. Opin. Neurobiol. 43,102–109

24. Bonnici,H.M.et al. (2013)Representationsofrecentandremote autobiographicalmemoriesinhippocampalsubfields. Hippo-campus 23,849–854

25. Eichenbaum,H.(2004)Hippocampus:cognitiveprocessesand neuralrepresentationsthatunderliedeclarativememory. Neu-ron 44,109–120

26. Maren,S.andHolt,W.(2000)Thehippocampusandcontextual memoryretrievalinPavlovianconditioning.Behav. Brain Res.

110,97–108

27. Stella,F.andTreves,A.(2011)Associativememorystorageand retrieval:involvementofthetaoscillationsinhippocampal infor-mationprocessing.Neural Plast. 2011,683961

28. Viskontas,I.V.(2008)Advancesinmemoryresearch: single-neuronrecordingsfromthehumanmedialtemporallobeaidour understandingofdeclarativememory.Curr. Opin. Neurol. 21, 662–668

29. Meeter,M.etal.(2005)Integratingincrementallearningand episodicmemorymodelsofthehippocampalregion.Psychol. Rev.112,560–585

30. Montaldi,D.andMayes,A.R.(2010)Theroleofrecollectionand familiarityinthefunctionaldifferentiationofthemedialtemporal lobes.Hippocampus 20,1291–1314

31. Rebola,N.et al. (2017)Operationandplasticityofhippocampal CA3circuits:implicationsformemoryencoding.Nat. Rev. Neu-rosci. 18,208–220

32. Moscovitch,M.etal.(2006)Thecognitiveneuroscienceof remoteepisodic,semanticandspatialmemory.Curr. Opin. Neurobiol. 16,179–190

33. Kitamura,T.et al. (2017)Engramsandcircuitscrucialfor sys-temsconsolidationofamemory.Science 356,73–78

34. Kumaran,D.andMaguire,E.A.(2007)Whichcomputational mechanismsoperateinthehippocampusduringnovelty detec-tion?Hippocampus 17,735–748

35. Yonelinas,A.P.(2013)Thehippocampussupports high-resolu-tionbindingintheserviceofperception,workingmemoryand long-termmemory.Behav. Brain Res. 254,34–44

36. Vinogradova,O.S.(2001)Hippocampusascomparator:roleof thetwoinputandtwooutputsystemsofthehippocampusin selectionandregistrationofinformation.Hippocampus 11, 578–598

37. Mizumori,S.J.andTryon,V.L.(2015)Integrativehippocampal anddecision-makingneurocircuitryduringgoal-relevant predic-tionsandencoding.Prog.BrainRes.219,217–242

38. Chavlis,S.andPoirazi,P.(2017)Patternseparationinthe hippocampusthroughtheeyesofcomputationalmodeling.

Synapse71

39. Pezzulo,G.et al. (2017)Internallygeneratedhippocampal sequencesasavantagepointtoprobefuture-oriented cogni-tion.Ann. N. Y. Acad. Sci. 1396,144–165

40. Chersi,F.andBurgess,N.(2015)Thecognitivearchitectureof spatialnavigation:hippocampalandstriatalcontributions. Neu-ron 88,64–77

41. Kim,J.andLee,I.(2011)Hippocampusisnecessaryforspatial discriminationusingdistalcue-configuration.Hippocampus 21, 609–621

42. Hassabis,D.et al. (2007)Patientswithhippocampalamnesia cannotimaginenewexperiences.Proc. Natl. Acad. Sci. U. S. A.

104,1726–1731

43. Maguire,E.A.et al. (2016)Scenes,spaces,andmemorytraces: whatdoesthehippocampusdo?Neuroscientist 22,432–439

44. Oess,T.et al. (2017)Acomputationalmodelforspatial naviga-tionbasedonreferenceframesinthehippocampus, retrosple-nialcortex,andposteriorparietalcortex.Front. Neurorobot. 11, 4

45. Duff,M.C.etal.(2013)Hippocampalamnesiadisruptscreative thinking.Hippocampus23,1143–1149

46. Driver,J.etal.(2007)Introduction.Mentalprocessesinthe humanbrain.Philos. Trans. R. Soc. Lond. B Biol. Sci. 362, 757–760

47. Scoville,W.B.andMilner,B.(1957)Lossofrecentmemoryafter bilateralhippocampallesions.J. Neurol. Neurosurg. Psychiatry

20,11–21

48. Stangor, C. (2014) Research Methods for the Behavioral Sciences, NelsonEducation

49. Mah,Y.H.et al. (2014)Humanbrainlesion-deficitinference remapped.Brain 137,2522–2531

50. Luber,B.andLisanby,S.H.(2014)Enhancementofhuman cognitiveperformanceusingtranscranialmagneticstimulation (TMS).Neuroimage 85,961–970

51. Filmer,H.L.et al. (2014) Applicationsof transcranialdirect currentstimulationforunderstandingbrainfunction.Trends Neurosci. 37,742–753

52. Leary, M.R. (2001) Introduction to Behavioral Research Methods,AllynandBacon

53. Dedoncker,J.etal.(2016)Asystematicreviewand meta-analysisoftheeffectsoftranscranialdirectcurrentstimulation (tDCS)overthedorsolateralprefrontalcortexinhealthyand neuropsychiatricsamples:influenceofstimulationparameters.

Brain Stimul. 9,501–517

54. Rosen,B.R.andSavoy,R.L.(2012)fMRIat20:hasitchanged theworld?Neuroimage 62,1316–1324

55. Raichle,M.E.(2009)Abriefhistoryofhumanbrainmapping.

56. Brewer,J.B.etal.(1998)Makingmemories:brainactivitythat predicts how wellvisual experience will be remembered.

Science 281,1185–1187

57. Henson,R.(2005)Whatcanfunctionalneuroimagingtellthe experimentalpsychologist?Q. J. Exp. Psychol. 58A,193–233

58. Friston,K.J.andPrice,C.J.(2011)Modulesandbrainmapping.

Cogn. Neuropsychol. 28,241–250

59. Yonelinas,A.P.(2002)Thenatureofrecollectionandfamiliarity: areviewof30yearsofresearch.J. Mem. Lang. 46,441–517

60. Genon,S.et al. (2013)Itemfamiliarityandcontrolledassociative retrievalinAlzheimer’s disease:anfMRIstudy.Cortex 49, 1566–1584

61. Price,C.J.andFriston,K.J.(2002)Degeneracyandcognitive anatomy.Trends Cogn. Sci. 6,416–421

62. Kanai,R.andRees,G.(2011)Thestructuralbasisof inter-individualdifferencesinhumanbehaviourandcognition.Nat. Rev. Neurosci. 12,231–242

63. Boekel,W.et al. (2015)Apurelyconfirmatoryreplicationstudyof structuralbrain-behaviorcorrelations.Cortex66,115–133

64. Yuan,P.andRaz,N.(2014)Prefrontalcortexandexecutive functionsinhealthyadults:ameta-analysisofstructural neuro-imagingstudies.Neurosci. Biobehav. Rev. 42,180–192

65. Nostro,A.D.etal.(2017)Correlationsbetweenpersonalityandbrain structure:acrucialroleofgender. Cereb. Cortex 27,3698–3712

66. Delis,D.C.et al. (1987)California Verbal Learning Test (CVLT) Manual, PsychologicalCorporation

67. Genon,S.et al. (2013)VerballearninginAlzheimer’sdisease andmildcognitiveimpairment:fine-grainedacquisitionand short-delayconsolidationperformanceandneuralcorrelates.

Neurobiol. Aging 34,361–373

68. Westfall,J.andYarkoni,T.(2016)Statisticallycontrollingfor confoundingconstructsisharderthanyouthink.PLoS One 11, e0152719

69. Genon,S.et al. (2017)Searchingforbehaviorrelatingtogrey mattervolumeina-prioridefinedrightdorsalpremotorregions: lessonslearned.Neuroimage 157,144–156

70. Staresina,B.P.andDavachi,L.(2008)Selectiveandshared contributionsofthehippocampusandperirhinalcortexto epi-sodicitemandassociativeencoding.J.Cogn.Neurosci.20, 1478–1489

71. Staresina,B.P.andDavachi,L.(2010)Objectunitizationand associativememoryformationaresupportedbydistinctbrain regions.J. Neurosci. 30,9890–9897

72. Poldrack,R.A.(2011)Inferringmentalstatesfromneuroimaging data:fromreverseinferencetolarge-scaledecoding.Neuron

72,692–697

73. Schwartz,Y.(2013)Mappingparadigmontologiestoandfrom thebrain.InAdvances in Neural Information Processing Sys-tems 26 (NIPS 2013),pp.1673–1681,NeuralInformation Proc-essingSystemsFoundation

74. Fox, P.T. andLancaster, J.L. (2002) Opinion: mappingcontext and content:theBrainMapmodel.Nat. Rev. Neurosci. 3,319–321

75. Laird,A.R.et al. (2005)BrainMap:thesocialevolutionofa humanbrainmappingdatabase.Neuroinformatics 3,65–78

76. Yarkoni,T.et al. (2011)Large-scaleautomatedsynthesisof humanfunctionalneuroimagingdata.Nat. Methods 8,665–670

77. Poldrack,R.A.(2010)Mappingmentalfunctiontobrain struc-ture: how cancognitive neuroimagingsucceed? Perspect. Psychol. Sci. 5,753–761

78. Genon,S.etal.(2017)Theheterogeneityoftheleftdorsal premotorcortexevidencedbymultimodalconnectivity-based parcellationandfunctionalcharacterization.Neuroimage Pub-lishedonlineFebruary14,2017.http://dx.doi.org/10.1016/j. neuroimage.2017.02.034

79. Lieberman,M.D.andEisenberger,N.I.(2015)Thedorsal ante-riorcingulatecortexisselectiveforpain:resultsfromlarge-scale reverse inference. Proc. Natl. Acad. Sci. U. S. A. 112, 15250–15255

80. Wager,T.D.etal.(2016)PainintheACC?Proc.Natl.Acad.Sci. U. S. A. 113,E2474–E2475

81. Lieberman,M.D.et al. (2016)ReplytoWageret al.:Painandthe dACC:theimportanceofhitrate-adjustedeffectsandposterior probabilitieswithfairpriors.Proc. Natl. Acad. Sci. U. S. A. 113, E2476–E2479

82. Poldrack, R.A. and Gorgolewski, K.J. (2014) Making big data open: datasharinginneuroimaging.Nat. Neurosci. 17,1510–1517

83. Milham,M.P.(2012)Openneurosciencesolutionsforthe con-nectome-wideassociationera.Neuron 73,214–218

84. VanEssen,D.C.et al. (2013)TheWU-Minnhumanconnectome project:anoverview.Neuroimage 80,62–79

85. Caspers,S.et al. (2014)Studyingvariabilityinhumanbrain aging in a population-basedGerman cohort-rationale and designof1000BRAINS.Front. Aging Neurosci. 6,149

86. Miller,K.L.et al. (2016)Multimodalpopulationbrainimagingin the UK Biobank prospective epidemiological study. Nat. Neurosci.19,1523–1536

87. Sundermann,E.E.etal.(2016)Betterverbalmemoryinwomen thanmeninMCIdespitesimilarlevelsofhippocampalatrophy.

Neurology86,1368–1376

88. Evans,A.C.(2013)Networksofanatomicalcovariance. Neuro-image 80,489–504

89. Genon,S.et al. (2017)Searchingforbehaviorrelatingtogrey mattervolumeina-prioridefinedrightdorsalpremotorregions: lessonslearned.Neuroimage 157,144–156

90. Hill,A.T.et al. (2016)TMS-EEG:awindowintothe neurophysi-ologicaleffectsoftranscranialelectricalstimulationinnon-motor brainregions.Neurosci. Biobehav. Rev. 64,175–184

91. Friston,K.et al. (2017)Dynamiccausalmodellingrevisited.

Neuroimage PublishedonlineFebruary17,2017. http://dx. doi.org/10.1016/j.neuroimage.2017.02.045

92. Stachenfeld,K.L.et al. (2017)Thehippocampusasapredictive map.Nat. Neurosci. 20,1643–1653

93. Poldrack,R.A.andYarkoni,T.(2016)Frombrainmapsto cognitiveontologies:informaticsandthesearchfor mental structure.Annu. Rev. Psychol. 67,587–612

94. Changeux,J.-P.(2017)Climbingbrainlevelsoforganisation fromgenestoconsciousness.TrendsCogn.Sci.21,168–181

95. May,A.(2011)Experience-dependentstructuralplasticityinthe adulthumanbrain.TrendsCogn.Sci.15,475–482

96. Voss,M.W.etal.(2013)Bridginganimalandhumanmodelsof exercise-inducedbrainplasticity.Trends Cogn. Sci. 17,525–544

97. Rorden,C.andKarnath,H.O.(2004)Usinghumanbrainlesions toinferfunction:arelicfromapasterainthefMRIage? Nat. Rev. Neurosci. 5,813–819

98. Herbet,G.et al. (2015) Rethinkingvoxel-wiselesion-deficit analysis:anewchallengeforcomputationalneuropsychology.

Cortex 64,413

99. Siebner,H.R.et al. (2009)Howdoestranscranialmagnetic stimulationmodifyneuronalactivityinthebrain?Implications forstudiesofcognition.Cortex 45,1035–1042

100.Sandrini,M.et al. (2011)Theuseoftranscranialmagnetic stimulationincognitiveneuroscience:anewsynthesisof meth-odologicalissues.Neurosci. Biobehav. Rev. 35,516–536

101.Laird,A.R.et al. (2011)TheBrainMapstrategyfor standardiza-tion,sharing,andmeta-analysisofneuroimagingdata.BMC Res. Notes 4,349

102.Glasser,M.etal.(2016)Amulti-modalparcellationofhuman cerebralcortex.Nature536,171–178

103.Conroy,B.R.etal.(2013)Inter-subjectalignmentofhuman corticalanatomy usingfunctional connectivity.Neuroimage

81,400–411

104.Burton,A.M.et al. (2010)TheGlasgowfacematchingtest.

Behav. Res. Methods 42,286–291