ContentslistsavailableatSciVerseScienceDirect
Developmental
Cognitive
Neuroscience
jou rn a l h om epa g e:h t t p : / / w w w . e l s e v i e r . c o m / l o c a te / d c n
Brain
function
during
probabilistic
learning
in
relation
to
IQ
and
level
of
education
Wouter
van
den
Bos
a,b,c,1,
Eveline
A.
Crone
a,b,d,∗,1,
Berna
Güro˘glu
a,b aInstituteofPsychology,LeidenUniversity,TheNetherlandsbLeidenInstituteforBrainandCognition,TheNetherlands cDepartmentofPsychology,StanfordUniversity,USA
dDepartmentofPsychology,UniversityofAmsterdam,TheNetherlands
a
r
t
i
c
l
e
i
n
f
o
Articlehistory:Received24May2011 Receivedinrevisedform 29September2011 Accepted29September2011 Keywords: IQ Development Learning fMRI Brainmaturation
a
b
s
t
r
a
c
t
Knowinghowtoadaptyourbehaviorbasedonfeedbackliesatthecoreofsuccessful learn-ing.Weinvestigatedtherelationbetweenbrainfunction,greymattervolume,educational levelandIQinaDutchadolescentsample.Intotal45healthyvolunteersbetweenages13 and16wererecruitedfromschoolsforpre-vocationalandpre-universityeducation.For eachindividual,IQwasestimatedusingtwosubtestsfromtheWISC-III-R(similaritiesand blockdesign).Whileinthemagneticresonanceimaging(MRI)scanner,participants per-formedaprobabilisticlearningtask.Behavioralcomparisonsshowedthatparticipantswith higherIQusedamoreadaptivelearningstrategyafterreceivingpositivefeedback. Analy-sisofneuralactivationrevealedthathigherIQwasassociatedwithincreasedactivationin DLPFCanddACCwhenreceivingpositivefeedback,specificallyforruleswithlowreward probability(i.e.,unexpectedpositivefeedback).Furthermore,VBManalysesrevealedthat IQcorrelatedpositivelywithgreymattervolumewithintheseregions.Theseresults pro-videsupportforIQ-relatedindividualdifferencesinthedevelopmentaltimecoursesof neuralcircuitrysupportingfeedback-basedlearning.Currentfindingsareinterpretedin termsofaprolongedwindowofflexibilityandopportunityforadolescentswithhigher IQscores.
© 2011 Elsevier Ltd. All rights reserved.
1. Introduction
1.1. Thedevelopmentofperformancemonitoring
Performancemonitoring,whichinvolvesourabilityto adjust our behavior following changing task demands, is one of the core components for successful learning. Performancemonitoringinvolvesthedetectionoferrors or performance feedback in tasks where rules need to
∗ Correspondingauthorat:DepartmentofPsychology,Leiden Univer-sity,Wassenaarseweg52,2333AK,Leiden,TheNetherlands.
E-mailaddress:[email protected](E.A.Crone).
1 Bothauthorscontributedequally.
beinferredover trials, orwhen ruleschange unexpect-edly.Acrosschildhoodandadolescence,thereisasteady increaseintheabilitytomonitortheoutcomesofactions with the purpose of behavioral adjustment, which has beendemonstratedusingrule-switchtasks(Huizingaetal., 2006;Koolschijn etal., 2011)andprobabilistic learning tasks (vanden Bos et al., 2009; Eppinger et al., 2009). These findings have been interpreted in the context of slowlydevelopingexecutivecontrolfunctions,withsteady advancesinadolescence(Crone,2009).
Neuroimagingstudieshave demonstratedthata net-workofregionsinthelateralprefrontal(lat-PFC),superior parietal,andthedorsalanteriorcingulatecortex(dACC) is engagedwhen individualsprocess performance feed-back.Thesestudiestypicallymakeuseofsimplelearning
paradigms,inwhichastimulusrequiresaresponse,which isthenfollowedbypositiveornegativefeedback(Holroyd andColes,2002).Theactivationoftheseregionsistypically relatedtonegativefeedbackprocessinginadults(Holroyd etal.,2004;Zanolieetal.,2008).Thishasbeenassociated withamonitoringsystemwhichsignalsthatoutcomesare worsethanexpectedandthatadjustmentisnecessary.
Giventhedevelopmentalchangesinperformance mon-itoringuntillateadolescence,itwouldbeexpectedthatthis error-orfeedbackmonitoringsystemmaturesslowly.In supportofthishypothesis,neuroimagingstudiesreported anage-related increase in activationfollowing negative feedbackinthelat-PFC,parietalcortexanddACC(Crone et al., 2008), showing that these regions are differen-tiallyengagedacrossdevelopment.Thesefindingsarealso consistentwithprior neuroimagingstudies which have reportedthatthedevelopmentofotherexecutivecontrol functions,suchasworkingmemory, taskswitchingand responseinhibition, isalsoassociated witha protracted developmentofregionswithintheprefrontalcortex(for areview,seeBungeandWright,2007).
However,twopriorstudiesreportednotonlyless acti-vationinlat-PFCanddACCfollowingnegativefeedback,but alsomoreactivationinchildrenintheseareasfollowing positivefeedback(VanDuijvenvoordeetal.,2008;vanden Bosetal.,2009).Theseneuroimagingfindingswere inter-pretedaspointingtoadevelopmentalshiftfromattention thatisgiventopositivefeedback(inchildhood)towards attention that is given to negative feedback (in adult-hood),withadolescencebeingatransitionphase.Notably, thesefindingsargue against astrictmaturational view-pointwhichpredictsthatthelateralPFCanddACCcannot beengagedduetostructuralimmaturity(seethe matur-tionalviewpointbyJohnson,2011),instead,theseregions are engaged under different task demands, consistent withskilllearningorinteractivespecializationviewpoints (Johnson,2011;seealsoDumontheiletal.,2010). 1.2. IndividualdifferencesineducationandIQ
Recently, much attention is given to the individual differencesinlearningperformance,aschildrenfrom dif-ferent levels of education, and thus different levels of IQ,maybenefitdifferentlyfromlearningcues.These dif-ferences may beparticularly important for the field of educationalneuroscience,inwhichoneofthegoalsisto mapchanges in brainfunction toindividualdifferences inlearningtrajectories(Goswami,2006).Executive con-trolfunctions, and associated brain activity, have been tightlylinkedtoIQdifferencesinadults.Forexample, pre-viousneuroimagingstudieshavereportedthatadultswith higherfluidintelligenceshowstrongerrecruitmentofthe lat-PFCanddACCduringexecutive controlperformance (Duncan,2003;Grayetal.,2003).
Most developmental neuroimaging studies have not takenintoaccountindividualdifferencesinIQ.Twosets of studies, focusing on functional and structural brain development,providesomeinsightintothedevelopmental differences in relation to IQ. First, one functional neu-roimagingstudiescomparedmath-giftedadolescentswith a control group. The math-gifted group was tested for
quantitativereasoningabilityusingtheSchoolandCollege AbilityTest.ThesescoreswereconvertedtoIQscores,and theirscoreswereinorabovethe98thpercentile.The con-trolgrouphadquantitativereasoningabilityconvertedto IQscoresintherange45–55thpercentile.Themath-gifted adolescentsshowedhigheractivationandhighereffective connectivityindorsolateralPFC,dACCandsuperior pari-etalcortexwhenperformingmentalrotationofcomplex 3D figures compared toa control group(O’Boyle et al., 2005;Prescottetal.,2010).Preusseetal.(2011),however, observed a different pattern. They showedthat adoles-centswithhighfluid intelligence,asmeasuredwiththe RavenAdvancedProgressiveMatricestask, hadstronger activationinparietalcortexwhenperforminga geomet-ricanalogicalreasoningtaskcomparedtoadolescentswith averagefluidintelligence,whereasadolescentswith aver-agefluidintelligencehadstrongeractivationinthedACC duringthistaskcomparedtoadolescentswithhighfluid intelligence. Insum, thefindings of studiesfocusingon individual differencesin IQarestill inconclusive, which couldpartlybeduetothedifferencesinmeasuresofIQ (crystallizedvs.fluid)thathavebeenusedinpriorstudies. Second,structuralneuroimagingstudieshavemapped corticalbrainstructureswithrespecttoIQindeveloping childrenandadolescents.Thesestudieshavefoundusing longitudinalmeasuresthattherewasashiftfrom predomi-nantlynegativerelationsbetweenIQandcorticalthickness in children topositive correlations in adults, indicating thatadolescenceisanimportanttransitionphase. Specif-ically,IQwaspositivelyassociatedwithhigherplasticity withaprolongedphaseofcorticalincreaseinadolescence. Together, thesefindings wereinterpreted in terms of a dynamicneuroanatomicalexpressionofintelligence(Shaw et al.,2006,seealsoKaramaet al.,2011;Pangelinan et al.,2011).Here,wetestthehypothesisthattheprolonged windowofbraindevelopmentforadolescentswithhigher IQincomparisonwithadolescentswithlowerIQisalso observedinfunctionalbraindevelopment.
1.3. Thepresentstudy
Thegoalofthisstudywasthereforetoexamine indi-vidualdifferences inbrain functionduring performance monitoring in relationto IQ and level of education. In theNetherlands,adolescentsparticipateindifferentschool systemsaftertheageof12–13,basedonscoreson cog-nitivetasksthatarecompletedattheendofelementary school(age11or12).Theseschoolsystemscanbebroadly divided in pre-vocational and pre-university education. Adolescents wererecruited fromthese two school sys-temsinordertosampleawiderangeofIQsandtofurther exploredifferencesbetweenlevelsofeducation.Giventhat thesamplingfromdifferentschoolsystemswasatoolto recruitvaryingIQlevels,wedidnothavespecific hypothe-sesconcerningdifferencesineducationsystems.Yet,we founditimportanttospecificallytestforthisby compar-ingbothdifferencesrelatedtoIQanddifferencesrelatedto schoolsystems.
Fig.1.Atthebeginningofeachtrialacentrallylocatedcuewaspresented withajitteredintervalbetween500and6000ms,followedbya com-binedpresentationofastimuluspairandaresponsewindowofmax. 2500ms,afterwhichfeedbackwaspresentedfor1000ms.Afterthe feed-back,ashortfillerwaspresented,intheformofablankscreen,inorder tocompensatefordifferentreactiontimesbetweentrialsandbetween participants(fillerduration=2500ms−reactiontime).Participantschose onestimulusbypressingtheleftorrightbuttonandreceivedpositiveor negativefeedbackaccordingtoprobabilisticrules.Twopairsofstimuli werepresentedtotheparticipants:(1)theABpairwith80%positive feed-backforAand20%forBand(2)theCDpairwith70%positivefeedback forCand30%forD.
etal., 2009).Thistaskwaschosenbecauseitallows for adirectcomparisonwiththeliteraturewherethe devel-opmentalpatternhasalreadybeeninvestigated(vanden Bos et al., 2009).The taskrequired theparticipants to learn a stimulus–response rules fortwo sets of stimuli, forwhichonestimulus–responseset(containing stimuli AandB)wasassociatedwithan80–20%andthesecond stimulus–responseset(containingstimuliCandD)witha 70–30%rewardmapping(seeFig.1).Duringthetask, func-tionalneuroimagingdatawereacquiredandtheanalyses focusedontheprocessingofpositiveandnegativefeedback forhighprobability(80%and70%,stimuliAandC)andlow probability(20%or30%,stimuliBandD)feedback.
Basedonpriorstudies,weexpectedadolescentsfrom pre-vocationaleducationtohavelowerestimatedIQscores and tohave aslowerlearningcurvein theprobabilistic learningtask,relativetoadolescentsfrompre-university education.Thepriorimagingresultsofthedevelopmental studyusingthistaskshoweddevelopmentaldifferencesin feedbackprocessinginthelatPFCandtheACC,whichwill thereforebespecificregionsofinterestinthecurrentstudy. Usingaprobabilisticlearningtask,vandenBosetal.(2009) foundthatwhenparticipantsexplorearulewithlow prob-abilityforreward,theserewardsleadtomoreactivation inlat-PFCanddACCinchildrenandadolescentsrelative toadults.Incontrast,thisdifferencewasnotobservedfor rewardtrialswhenapplyingtherulewithhighprobability forreward.
Under the assumption that adolescents with higher IQ(frompre-universityeducation)relativetoadolescents with lower IQ (from pre-vocational education) have a prolongeddevelopmentaltrajectoryof prefrontalcortex maturation(Preusseetal.,2011;Shawetal.,2006;Karama etal.,2011),wepredictedthathighIQadolescentswould showanelevatedresponsetopositivefeedbacktothelow probabilityrewardrule.
Inaddition,weperformedavoxelbasedmorphometry (VBM)analysesinordertoinvestigatetherelationbetween IQand corticalgreymattervolumes.Based onprevious studies weexpected that IQwould be positively corre-latedwithcorticalgreymattervolume(Shawetal.,2006;
Karamaetal.,2011;Pangelinanetal.,2011),andwetested whetherthesedifferencescouldaccountforthedifferences infunctionalactivation.
2. Methods
2.1. Participants
Forty-fivehealthy right-handedadolescents between ages13and16years(Mage=14.39;22female,23male) participatedinthestudy.Participantswererecruitedfrom thecommunityandwithhelpofvariouslocalschools. Par-ticipantscamefrom twocategories basedontheDutch education system,where children are separated in dif-ferent school systems based on test scores at the end of elementary school (approximately age 11–12). Pre-vocational education educates children for 4 years and preparesthemforaworkingenvironment.Pre-university educationeducateschildren for5–6years andprepares themforfurtheruniversityeducation.Inthecurrent sam-ple,18 participantsattended pre-vocational schools (M age=14.5;9female,9male),and27participantsattended pre-universityschools(Mage=14.3;13female,14male). Partofthedatawaspreviouslyreportedinastudyonage comparisons(vandenBosetal.,2009).
Chisquare analysisconfirmedthatthegender distri-butionwas similar acrossthe two groups (2(1)=.056,
p=.82),andgroupsdidnotdifferinmeanage(t(44)=.541, p=.59).Allparticipantsreportednormalor corrected-to-normalvisionandparticipantsortheircaregiversreported theabsenceofspecificlearning disorderand neurologi-calorpsychiatricimpairments.ParentsfilledouttheChild BehaviorCheckList(CBCL,Achenbach,1991)inorderto screenforpsychiatricconditions. Allparticipantsscored belowclinicallevelsonallsubscalesoftheCBCL,andhad scoreswithin1SDofthemeanofanormativestandardized sample.Participants and theircaregivers gave informed consentforthestudyandallprocedureswereapproved bythemedicalethicalcommitteeoftheLeidenUniversity MedicalCenter.InaccordancewithLeidenUniversity Med-icalCenterpolicy,allanatomicalscanswerereviewedand clearedbytheradiologydepartmentfollowingeachscan. Noanomalousfindingswerereported.
2.2. BehavioralassessmentofestimatedIQ
Participantscompleteda verbalandnon-verbal mea-sure (similarities and block design subscales) of the WechslerIntelligenceScaleforChildren(WISC)inorderto obtainanestimateoftheirintelligencequotient(Wechsler, 1991,1997).Inthecurrentsample,therewasahigh cor-relationbetweentheestimatedIQonthetwo subscales (r=.87),therefore theestimatedIQ scorein subsequent analysesistheaveragedIQscoreofthetwosubscales.2The minimumIQscorewas70andthemaximumIQscorewas 130.Aspredicted,thepre-universitygrouphadhigherIQ
2 ThebehavioralandfMRIanalysesusingeachIQmeasureseparately
Table1
GroupstatisticsonIQ,reactiontimes(RTs),ageandheadmotion.Groupsarebasedoneducationallevel,andsubgroupsaredefinedbyeducationallevel andIQ.Standarderrorsarepresentedinparentheses.
IQ Reactiontimesinms Headmotion
averageinmm
Headmotionmax inmm Age Pre-vocationaleducation 91.3(2.4) 805(47) .09(.02) 2.01 14.5(.2) Pre-universityeducation 107 (2.0) 773(39) .08(.01) 2.21 14.3(.2) Subgroups I(IQ70–90) 80(2.3) 818(74) .11(.02) 2.01 15(.4) IIa(IQ90–110) 98.6(1.3) 799(36) .07(.01) 2.0 14.1(.3) IIb(IQ90–110) 99.3(1.8) 779(41) .08(.01) 2.21 14.6(.4) III(IQ110–130) 115.6(2.6) 764(38) .07(.02) .81 14.6(.4)
scoresthanthepre-vocationalgroup,t(44)=4.89,p<.001
(seeTable1).
Thescoreswereanalyzedin moredetail toexamine howtheestimatedIQscoresrelatedtoschooltypes.As canbeseen inFig.2A,theestimatedIQscoresare nor-mallydistributed.Fig.2Bshowsthatalladolescentswho attendpre-vocationalschoolshaveIQscoresthatfallinbins 70–110,andalladolescentsfromthepre-universityschools haveIQscoresthatfallinbins90–130.Thus,asexpected,
Fig.2.(AandB)IQdistributionforthepre-vocationalandpre-university participants.Thesegroupswerefurtherbrokendownto:alowIQgroup (I;IQ70–90),twoaverageIQgroups(IIaandIIb;IQ90–110)andahighIQ group(III;IQ110–130).
ourIQestimationisstronglyrelatedtoeducationallevel. Interestingly,thedistributionoftheIQscoresallowsfora secondtypeofcategorization.Thatis,withinthe90–110 binsthereareparticipantswhodifferinschool-levelbut have overlapping IQestimations.Thispattern allowsus todisentangletheeffectsofschooldifferencesvs.IQ dif-ferences,bysubdividingtheparticipantsinfourgroups; LowIQ/pre-vocational(n=7;GroupI;IQ[70–90]),Medium IQ/pre-vocational(n=12;GroupIIa;IQ[90–110]),Medium IQ/pre-university(n=12;GroupIIb;IQ[90–110])andHigh IQ/pre-university(n=14;GroupIII;IQ[110–130])(seeFig.2 andTable1).
2.3. Taskprocedure:probabilisticlearningtask
Theprocedurefortheprobabilisticlearningtask(Fig.1; basedonFranketal.,2004;vandenBosetal.,2009)wasas follows:Thetaskconsistedoftwostimuluspairs(calledAB andCD).Thestimuluspairsconsistedofpicturesof every-dayobjects(e.g.,achairandaclock).Eachtrialstartedwith thedisplayofone ofthetwostimuluspairsand subse-quentlytheparticipanthadtochooseoneofthetwostimuli (e.g.,AorB),whichwerepresentedontheleftortheright sideofthescreen.Thestimuluspairswerepresentedin ran-domorder.Participantswereinstructedtochooseeither theleftortherightstimulusbypressingabuttonwiththe indexormiddlefingeroftherighthandwithina2500ms window,whichwasfollowedbya1000msfeedback dis-play.Thefeedbackdisplayconsistedofagreen√-signal forpositivefeedbackandaredcross(×)fornegative feed-back. Ifnoresponsewasgivenwithin2500ms,thetext “tooslow”waspresentedonthescreen.Thisoccurredon lessthan4%ofthetrials.Importantly,theschooltypesdid notdifferinnumberof‘tooslow’responses(t(43)=−.25, p=.79).
The feedback displayed was probabilistic. Choosing stimulusA ledtopositive feedbackon80% ofABtrials, whereaschoosingstimulusBledtopositivefeedbackon 20%ofthesetrials.TheCDpairprocedurewassimilar,but probabilityforrewardwaslower;choosingstimulusCled topositivefeedbackon70%ofCDtrials,whereaschoosing stimulusDledtopositivefeedbackon30%inthesetrials. Thus,themostoptimalchoiceinordertoobtainrewards wasAorC(correctrule),whereastheleastoptimalchoice wasBorD(alternativerule).
toreceivepositivefeedbackoneverytrial.Further, par-ticipantswereinformedthatalthoughstimulisometimes appearedontherightsideandsometimesontheleftside, thiswasanirrelevantdimension.Aftertheinstructionsand rightbeforethescanningsession,theparticipantsplayed 40practiceroundsonacomputerinaquietroomtoensure thattheyunderstoodthetask.
Intotal,thetaskinthescannerconsistedoftwoblocks of100trialseach:50ABtrialsand50CDtrialsperblock. Thefirstandthesecondblockconsistedofdifferentsets ofpicturesandtherefore,participantshadtolearnanew mappinginbothtaskblocks.Thedurationofeachblock wasapproximately 8.5min.The stimuliwere presented in pseudo-random order with a jittered interstimu-lus interval (min=1000ms, max=6000ms) optimized withOptSeq2(surfer.nmr.mgh.harvard.edu/optseq/;Dale, 1999).Inordertoensurethatallblockswerecompletely independent, the stimuli were different in the practice blockandthetwoexperimentalblocks.Duringinter-trial intervals, a central fixation cross was shown. Because previousstudiesrevealedthatbehavioralstrategieswere dependentonruletype(correctoralternativerule),we dif-ferentiatedbetweenover-learnedhighprobabilities(Aand Ctrialscollapsed,correctrule)andalternativelow proba-bilities(BandDtrialscollapsed,alternativerule,cf.vanden Bosetal.,2009)inthebehavioralandimaginganalyses. 2.4. Dataacquisition
Participantswerefamiliarizedwiththescanner envi-ronmentonthedayofthefMRIsessionthroughtheuse ofamockscanner,whichsimulatedthesoundsand envi-ronmentofarealMRIscanner.Datawereacquiredusinga 3.0TPhilipsAchievascannerattheLeidenUniversity Med-icalCenter.Stimuliwereprojectedontoascreenlocatedat theheadofthescannerboreandviewedbyparticipantsby meansofamirrormountedtotheheadcoilassembly.First, alocalizerscanwasobtainedforeachparticipant. Subse-quently,T2*-weightedEcho-PlanarImages(EPI)(TR=2.2s, TE=30ms,80×80matrix, FOV=220, 352.75mm trans-verse slices with0.28mm gap) wereobtainedduring 2 functionalrunsof232volumeseach.Thefirsttwoscans werediscardedtoallowforequilibrationofT1saturation effects. A high-resolution T1-weighted anatomical scan and a highresolution T2-weighted matched-bandwidth high-resolutionanatomicalscan,withthesameslice pre-scriptionastheEPIs,wereobtainedfromeachparticipant afterthefunctionalruns.Stimuluspresentationandthe timingofallstimuliand responseeventswereacquired using E-Primesoftware. Headmotionwasrestricted by usingapillowandfoaminsertsthatsurroundedthehead. 2.5. fMRIdataanalysis
DatawerepreprocessedusingSPM5(Wellcome Depart-ment of Cognitive Neurology, London). The functional timeserieswererealignedtocompensateforsmallhead movements. Translational movement parameters never exceeded1voxel(<3mm)inanydirectionforanysubject or scan. Therewerenosignificant differencesin move-mentparametersbetweenthegroupsF(2,45)=.89,p=.32
(seeTable1).Functionalvolumeswerespatiallysmoothed usingan8mmfull-widthhalf-maximumGaussiankernel. FunctionalvolumeswerespatiallynormalizedtoEPI tem-plates.Thenormalizationalgorithmuseda12-parameter affinetransformationtogetherwithanonlinear transfor-mationinvolvingcosinebasisfunctionsandresampledthe volumesto3mmcubicvoxels.TheMNI305templatewas usedforvisualizationandallresultsarereportedinthe MNI305stereotaxicspace(Cosocoetal.,1997).
Statisticalanalyseswereperformedonindividual par-ticipants’datausingthegenerallinearmodelinSPM5.The fMRItimeseriesdataweremodeledbyaseriesofevents convolvedwithacanonicalhaemodynamicresponse func-tion(HRF).Thepresentationofthefeedbackscreenwas modeledas0-durationevents.Thestimuliandresponses werenotmodeledseparatelyastheseoccurredinoneprior oroverlappingEPIimagesasfeedbackpresentation.
Inthemodel,feedbackwasfurthersubdividedinto cor-rect(AandC)vs.alternative(BandD)ruleandpositive vs.negativefeedback.Thesetrialfunctionswereusedas covariatesinagenerallinearmodel,alongwithabasicset ofcosinefunctionsthathigh-passfilteredthedata,anda covariateforruneffects.Theleast-squaresparameter esti-matesofheightofthebest-fittingcanonicalHRFforeach conditionwereusedinpair-wisecontrasts.Theresulting contrastimages,computedonaparticipant-by-participant basis,weresubmittedtogroupanalyses.Atthegrouplevel, contrastsbetweenconditionswerecomputedby perform-ingone-tailedt-testsontheseimages,treatingparticipants asarandomeffect.
WeperformedregressionanalyseswithIQasa covari-ateofinteresttoidentifyregionsthatshowedIQrelated differencesinfeedbackprocessing.Subsequentlywe per-formedROI analysestofurtherinvestigatetheeffectsof schooldifferencesvs.IQdifferences,usingthefour sub-groupsdescribedabove.
2.6. Region-of-interestanalyses
We used the Marsbar toolbox with SPM5 (http: //marsbar.sourceforge.net;Brett et al.,2002)toperform RegionofInterest(ROI)analysestofurthercharacterizeIQ andschooltypedifferencesinpatternsofactivationand greymattervolume.WecreatedROIsoftheregionsthat wereidentifiedin thewholebrainanalysesfor illustra-tionpurposesorposthoctests.Themasksusedtogenerate functionalROIswasbasedontheregressionanalyseswith IQascovariateacrossallparticipants(p<.001,>10voxels). ForROI analyseswefocusedontwo aprioridefined regions(rightDLPFCanddACC),asaresult,effectswere consideredsignificantatan˛of .025,based on Bonfer-ronicorrectionformultiplecomparisons,p=.05/2,unless reportedotherwise. These regionswere chosena priori basedonourfindingsinourpriordevelopmentalstudies (vandenBosetal.,2009).
2.7. Voxelbasedmorphometry
Fig.3.(A)Learningcurveshowingaverageaccuracyper10trialblockforeachofthefourgroups(I,IIa,IIbandIII).(B)CorrelationbetweenIQandaccuracy onthelast60trialsofthetask.(C)CorrelationbetweenIQandpercentageofwin-staychoicesperruletype.(D)CorrelationbetweenIQandpercentageof lose-shiftchoicesperruletype.
participants were excluded because of the poor qual-ityoftheirstructuralT1images.Aftereachparticipant’s structuralT1imagewasnormalizedtotheMNItemplate, imagesweresegmentedintoseparatemapsforcerebral spinalfluid,greyandwhitematter.Althoughthe partici-pantswerenotofadultage,segmentationwasperformed withuseofpriorprobabilitymapsbecausethisincreased the quality of the segmentation process. Additionally, thehand-selectedanteriorcommissureofallbrainswas usedastheoriginfor theregistrationandsegmentation analyses.Furthermore, modulationfor non-linear warp-ingonlywasperformedusingtheJacobiandeterminants inordertoadjustvolumeestimationsforoverallheadsize (O’Brienetal.,2006).Finally,imageswereresampledinto 3×3×3voxelsandsmoothedusingan8mmfull-width half-maximumGaussiankernel,makingthevolumemasks comparabletothefunctionalactivationmaps.
3. Results
3.1. Behavior
3.1.1. Taskperformanceandeducationallevel
To investigate the differences in task performance for the different levels of education we calculated the percentageofcorrectchoices(thehighprobability stim-ulus)perblockof20trialsforeachparticipant,resultingin
fiveblocksintotal(Fig.3A).Becausethetworunsinthe scannerconsistedofnewstimuluspairs,theblocksofthe tworunswerecollapsed.
To investigate whether the participants learned over time, we performed a repeated-measure ANOVA with task block as within-subjects variable and school level as between-subjects variable. The task block (5 blocks)×school type (2 types: pre-vocational vs. pre-university) ANOVAshowed that participantslearned to make more correct choices over time, as indicated by a main effect of task block (F(4,176)=27.12, p<.001). However,therewerenoperformancedifferencesbetween school types (F(1, 44)=3.34, p=.08), and there wasno schooltype× taskblockinteraction(F(4,176)=.78,p=.98). The analyses of performance further focused onthe last60trails.Intheselast60trialstheparticipantshave acquiredsubstantialknowledgeoftheprobabilities asso-ciatedwiththedifferentstimuli.Indeed,participantsno longer showed differences in learning (main effect of task block, F(2, 88)=2.75, p=.10). The task block (last 3 blocks)×schooltype (2 types:pre-vocational vs. pre-university)ANOVAshowedthatpre-universityadolescents had higher accuracies than pre-vocational adolescents (main effectschool type:F(1, 44)=3.76, p<.05),but no interactionbetweentaskblockandschooltype(p>.30).
wecomparedthetwogroups(IIaandIIb)fromdifferent educationallevelsbutwiththesamelevelofIQ.Thetask block(3)×schooltype(IIaandIIb)revealednosignificant effects(allp’s>.2),suggestingthattheperformance differ-encesbetweeneducationallevelsaremainlydrivenbyIQ differences.Asexpected,correlationanalyseswithgeneral performancerevealedapositiverelationbetweenIQand thenumberofcorrectchoicesinthelast60trials(r=.48, p<.001,seeFig.3B).
3.1.2. BehavioralstrategyandIQ
TofurtherexamineIQ-relateddifferencesintask perfor-manceweexploredtherelationbetweenIQandbehavioral strategiesinthelast60trials.Forthisanalysis,we exam-inedhowoftenparticipantschoseeitherthesamestimulus afterpositivefeedback(win-stay)ortheshiftedtotheother stimulusafternegativefeedback(lose-shift).
TherewereIQrelateddifferencesinbehavioral strate-giesfollowingpositivefeedback;theseanalysesrevealed apositivecorrelationbetweenwin-staychoicesandIQfor thecorrectrule(r=.54,p<.001),andanegativecorrelation betweenwin-staychoicesandIQforthealternativerule (r=−.43,p<.001,seeFig.3C).Thus,higherIQwasrelated toamoreoptimalshiftingbehaviorafterreceiving posi-tivefeedback(itisoptimaltostaywiththecorrectrule andshiftwiththealternativerule).Incontrast,therewas nosignificantrelationbetweenIQandlose-shiftbehavior (seeFig.3D).
3.2. fMRIresults
InthefMRIanalyseswefocusedonpatternsof activa-tioninthelast60trialsofthetaskinordertobeableto investigatetheneuralcorrelatesoffeedbackprocessingfor thedifferentruletypesandtomakeadirectcomparison withthedevelopmentalliterature(seealsovandenBos etal.,2009).
3.2.1. Effectsoffeedbackandruletype
First, we examined the brain regions involved in feedback processing across all participants. The com-parison [positive feedback>negative feedback] resulted in activation in several regions, including the ven-tral striatum, VMPFC, bilateral DLPFC and the parietal cortex (see Table 2). The opposite contrast [neg-ative feedback>positive feedback] resulted in acti-vation in the bilateral insula and the dorsal ACC (seeTable2).
Next, we investigated how feedback was modulated by rule type. Two rule types were distinguished: the correctrule(cor)andthealternativerule(alt).The com-parison[positivefeedback(alt)>positivefeedback(cor)] resultedinactivationintheventralstriatum(seeTable2), whereas the opposite contrast did not yield any sig-nificant effects. Analyses of the effect of rule type on negativefeedbackprocessingdidnotyieldanysignificant results.
Thus,inthecurrentstudy,theDLPFCwasinvolvedin processingpositivefeedbackandthedACCinprocessing negativefeedback.Thestriatumwasparticularlysensitive
topositivefeedbackwhenitwasleastexpected(i.e.,a pos-itivepredictionerror).
3.2.2. EffectsofIQonfeedbackprocessing
Toexaminehowfeedbackprocessingwasmodulated byIQdifferences,wefirstidentifiedindividualdifferences inneural activationoffeedback processingby perform-ing a regression analysis on the [positive feedback vs. negativefeedback]contrastwithestimated IQasa pre-dictorvariable.Thisanalysisdidnotyieldanysignificant results.
Second,we performeda regressionanalysisfor each feedbacktypeinthecontextofthecorrectandalternative rules.Examiningfeedbackinthecontextofcorrectvs. alter-nativeruleswaspreviouslyfoundtobemostpowerfulin dissociatingeffectsforpositiveandnegativefeedbacktrials separately(vandenBosetal.,2009).Theregression analy-sisonthe[positivefeedback(alt)>positivefeedback(cor)] contrastrevealedapositivecorrelationbetweenestimated IQandBOLDactivityinregionspreviouslyimplicatedin performance monitoring,includingtheright DLPFCand thedACC(seeFig.4A,Table2).AscanbeseeninFig.4B, higherIQwasassociatedwithmoreactivityinrightDLPFC anddACCfollowingpositivefeedbackafterselectingthe alternativerulecomparedtothecorrectrule.Aregression analysiswithnegativefeedbacktrials(alternativevs. cor-rectrule)didnotrevealregionsthatcorrelatedwithIQand brainactivation.
To furtherexplore the relationbetween IQ, level of educationandactivityinrightDLPFCanddACC,we per-formedseparatecorrelationsbetweenIQandbrainactivity foreach school group(seeFig.4B;coloreddots),based onROIs whichwerederived fromthewhole-brain con-trast[positivefeedback(alt)>positivefeedback(cor)]with IQasregressor.TheseanalysesshowedthatDPLFC activ-itywasrelated toIQinboth thepre-vocational andthe pre-university group (r=.54, p<.02 and r=.59, p<.01, respectively).However,fordACC,thecorrelationwithIQ wasonlysignificantforthepre-universitygroup(r=.44, p<.02), and not for the pre-vocational group (r=−.02, p=.93).ThesefindingssuggestthatactivityindACCis pos-siblyrelatedtobothIQandeducationallevel.
3.2.3. Educationallevelrelateddifferencesinneural activity
Table2
Brainregionsrevealedbywholebraincontrasts.MNIcoordinatorsformaineffects,peakvoxelsreportedatp<.001,uncorrected,atleast10contiguous voxels.
Anatomicalregion L/R Z MNIcoordinates
x y z
Positive>negative
Striatum L/R 7.49 12 15 −3
Dorsolateralprefrontalcortex R 4.61 46 23 31
Superiorparietalcortex R 4.54 41 −58 53
VentralmedialPFC L/R 5.07 0 51 −12
Visualcortex L/R 4.60 31 −93 −7
Negative>positive
Dorsalanteriorcingulatecortex L/R 4.43 6 21 31
Anteriorinsula L 4.78 40 23 2
R 4.60 −41 17 −1
Positivealternative>positivecorrect
Ventralstriatum L/R 4.43 −14 8 −4
RegressionIQ[positivealternative>positivecorrect]
Dorsalanteriorcingulatecortex L/R 5.03 −3 24 35
Dorsolateralprefrontalcortex R 5.71 45 21 26
Superiorparietalcortex R 4.23 37 −59 50
Superiorfrontalgyrus R 4.11 19 58 21
Middleocciptialgyrus R 4.16 24 −78 15
Fig.5.(A)Corticalmapshowing:(1)abluecoloredmaskthatidentifiesall corticalareasthatshowedasignificantcorrelationbetweengreymatter volumeandIQ,(2)aredcoloredmaskshowingrDPLFC,dACCand pari-etalcortexfromthefunctionalregressionanalyses,and(3)apurplemask showingtheoverlapbetweenmasks1and2.(B)Representationofthe correlationbetweenIQandaveragegreymattervolumesforthethree regionsofinterest.(Forinterpretationofthereferencestocolorinthis figurelegend,thereaderisreferredtothewebversionofthearticle.)
(F(1,11)=4.84, p<.02)but didnot differentiatebetween
rule types(F(1,11)<1.0, p=.58).In contrast, inthe
pre-universitygroup,thedACCwassensitivetoruletypesuch
thatactivitywashigherfollowingfeedbackfromthe
alter-nativerule compared tothe correct rule(F(1,11)=7.23,
p<.01)butnottovalence(F(1,11)<1.0,p=.41,seeFig.4C).3
Taken together, activity following positive feedback in rDLPFC correlated with IQ, independent of the school fromwhichtheparticipantswererecruited.Activityinthe dACCalsocorrelatedwithIQ,butseemedtohaveaslightly differentactivationpatternforthetwoschooltypes. 3.3. Voxelbasedmorphometry
VBManalysesyieldedsignificantcorrelationsbetween IQandtheestimatedgreymattervolumeinseveralregions acrossthewholebrain(seeSupplementaryFig.S1).Most cortical regions showed a positive correlation between IQand grey matter volume.Thisrelation indicatesthat the children with a higher IQhave largercortical grey matter volume than those with a lower IQ. A negative correlationbetweenIQandgreymatterwasfoundin bilat-eralhippocampus/amygdala,leftventralstriatum,bilateral cerebellum,andlefttemporalpole.Thus,thesestructures haveahighergreymattervolumeforparticipantswitha lowerIQcomparedtothosewithahigherIQ.
Ascan beseen in Fig.5, thecorticalbrain areas for whichthegreymattervolumepositivelycorrelatedwith IQoverlapwiththetwofunctionallydefinedROIs(dACC andrDLPFC).Tofurtherinvestigatetherelationbetween IQandgreymattervolumeintheseROIs,weextractedthe
3WehaveadditionallyperformedthesamesetofanalysesontheDLPFC
anddACCROIsthatwereextractedfromthe[POS-NEG]and[NEG-POS] contrasts(reportedinTable2).Theseanalysesrevealedthesameresults (allp’s<.02)forboththeeffectsofeducationandIQ.
regionalestimatedgreymattervolumeforeachparticipant andcorrelatedthiswithIQlevel(seeFig.5).
Given the IQ-related grey matter differences, we testedwhetherfunctionalactivationdifferencescouldbe explainedbylocalchangesingreymattervolumeusing BPM. These analyses show that although there are IQ related differencesin corticalgraymatter,thesedo not explaintheIQrelatedchangesinfunctionalactivity(see Fig.S2).
4. Discussion
Thegoalofthisstudywastoexamineindividual differ-encesinperformancemonitoringinadolescentsinrelation tovaryingIQscoresanddifferenteducationalbackgrounds. ThecorrelationsbetweenneuralactivityandIQ,andthe comparisonsofadolescentsfrompre-vocationaland pre-universityeducationresultedin a setofbehavioral and neural differences. Behavioral analyses resulted in two importantpatterns:(1)IQwaspositivelycorrelatedwith overallaccuracyonthelast60trialsofthetask,and(2) theseIQrelated changesinlearning weredue to differ-encesinshiftingstrategy inrelationtopositive,butnot negativefeedback.Specifically,thedifferencesinshifting strategiesshowedthatparticipantswithahigherIQwere morelikelytostaywiththesameruleafterpositive feed-backwhenselectingthecorrectrule,butweremorelikely toswitchawayafterpositivefeedbackwhenselectingthe alternativerule.Bothareoptimalshiftingstrategiesand thustheycontributetotheincreasedaccuracylevels.The IQdifferenceinwin-staydecisionsincontextofthecorrect rulearecomparabletotheagerelatedincreaseinwin-stay decisionsthatwasreportedinapriordevelopmentalstudy withthesametask(vandenBosetal.,2009).
4.1. BrainfunctioninrelationtoIQandeducation
Basedontheresultsfrompreviousfunctionaland struc-turalimagingstudiesweexaminedwhetheradolescents withhigher IQhave a pattern of neural activationthat allowsforgreaterflexibilityforlearning(Shawetal.,2006). Basedonpreviousstudies(vandenBosetal.,2009;Van Duijvenvoordeetal.,2008)differenceswereexpectedto occurinthekeyareasoftheperformancemonitoring net-work:therightDLPFCandthedACC.
4.1.1. RightDLPFC
studies. This comparison with prior studies suggests that the current results support a prolonged, rather thanadvancedmaturationhypothesisofIQrelatedbrain changes(seealsoShawetal.,2006).Inadolescence, indi-vidualswithhigherIQmaybemorefocusedonexploring alternativeactions.
Priorresearchsuggeststhatdevelopmentaldifferences inpositiveandnegativefeedbackadjustmentaretheresult ofdifferencesinattentionregulation(Somsen,2007). Fol-lowingthishypothesis,thecurrentresultssuggestthatthe individualdifferencesinIQarerelatedtoafocuson differ-entinformationvalueofpositivefeedbacksignals.Whereas adolescentswitha higherIQfocusonpositivefeedback when exploring thealternative rule (thus thefeedback containinginformativevalueforexploration),the adoles-centswithalowerIQseemtofocusonpositivefeedback thatconfirmsthealreadylearnedcorrectrule.Possibly,the adolescentswithhigherIQwerebetteratdifferentiating betweenpositive feedbackthatisexpectedandpositive feedbackthatisunexpected.Anincreasedattention sys-temforunexpectedpositivefeedbackcanbeadvantageous whenexploringalternativeactions.Theadolescentswith higherIQwerealsobetterabletoswitchfollowingpositive feedbackwhenchoosingthealternativerule.Asaresult, theadolescentswithalowerIQmaybelessabletoupdate therelevantfeedbackinformationfromalternativerules, andthereforetheyarepossiblyless flexibleinselecting alternativeactions.
4.1.2. dACC
Similarto the rDLPFC, the dACC also showed an IQ relatedincreaseinactivityrelatedtopositivefeedbackafter selectingthealternativerule.Again,this patternof acti-vationis similartothe developmentalpattern reported previously(vandenBosetal.,2009;VanDuijvenvoorde etal.,2008),supportingthedelayed,orexplorative matu-rationhypothesisofIQrelatedbrainchanges.Interestingly, adolescents withhigher IQ also showed an increase in activationfollowingnegativefeedbackwhensamplingthe alternativerule.Inpreviousstudies,dACCactivityhasbeen relatedtoamoregeneralroleindetectingconflict(Brown andBraver,2005), andsignalingtheneedforbehavioral adjustment(HolroydandColes,2008;Kernsetal.,2004; Rushworth,2008).Thissuggeststhattheadolescentswith alowerIQarepossiblylessabletoidentifytherelevant feedbackinformationandtheyshowanactivationpattern inthedACCwhichissensitivetonegativefeedbackin gen-eral(alsowhenitshouldbeignored),whereasthehigher IQadolescentsonlyshowdACCactivitywhenthereisneed forbehavioraladjustment(afterselectingthealternative rule).
Exploratory analyses revealed that when comparing groupsof similarIQ,but fromdifferentschoolsystems, feedbackrelatedactivityinthedACCwasalsodifferentially sensitivetothetype/levelofeducation.Theexactreasonfor thisdifferenceisnotyetunderstoodbutpointinthe direc-tionofaninfluenceofenvironmentalfactors,althoughit shouldbenotedthat theparticipantswereallrecruited fromdifferentschools.Thissuggeststhattheroleofthe schoolsystemonbrainfunctionanditsdevelopmentisan interestingavenueforfuturestudies.
Finally,learningtheorieshavesuggestedthattwo sep-arate learning strategies contribute to feedback based learning (Dawet al.,2005; Maia,2009): a model-based strategythatoperatesonexplicittaskrepresentations,such as rules describing the reward contingencies given the current state(associated with DLPFC and dACC), and a model-freestrategythatusesfeedbackdirectlytocompute action values without any explicit model of the envi-ronment(associatedwithstriatumandVMPFC). Forthe currentstudywedecidedtofocusontherule-based net-work given that the majority of the current literature on development of IQis related to executive function-ing(cf.vandenBosetal.,2009).However,amodel-free reinforcement learning analysis of the same task sug-geststherearealsoimportantdevelopmentalchangesin striatum-VMPFCconnectivitythatunderliedifferencesin learning the reward contingencies (van den Bos et al., 2011).Although suchanalysesarebeyondthe scopeof thecurrentpaper,thechallengeforfuturestudieswillbe tounderstandhowdifferencesinreinforcementlearning mechanismsmaycontributetoIQrelatedchangesin feed-backlearning.
4.2. BrainstructureinrelationtoIQandeducation
In the current study VBM analyses wereperformed, which revealed thatgrey mattervolume waspositively correlatedwithIQinmanypartsofthecortex,including thefunctionallydefineddACCandrDLPFCROIs.Although severalstudieswithadolescentshaveconfirmedthe pos-itiverelationbetweengreymatterandintelligence(Shaw et al., 2006; Pangelinanet al., 2011), theexact relation betweenthequantityof greymatter andthequality of cognitive functions is still not well understood (Deary etal.,2010).Possibly,individualswithahigherIQshowa prolongedperiodofsynapticoverproductionandpruning duringadolescence,whichisreflectedbyincreasedgrey mattervolumes(Shawetal.,2006).Inturn,thisprolonged periodofplasticityisthoughttocontributetoincreased capacityforflexibilityandlearning(Johnstonetal.,2009). Thecombinedresultsofthefunctionalandstructural anal-ysesinthisstudyareinlinewiththesehypotheses.First, boththerDLPFCandthedACCshowedactivationpatterns that suggestthat a highIQisrelated tofaster learning, increased flexibility and adaptive behavior. Second, the grey mattervolumeof thesesame areascorrelated sig-nificantlywithIQ.However,notethatouranalysesalso revealedthatgreymattervolumebyitselfcouldnotfully explainthechangesinbrainfunction.Infuturestudiesit willthereforebeimportanttofurtherinvestigatethe rela-tionbetweenbrainstructure,functionandIQ,forinstance byincorporatingstructuralconnectivityanalyses(Jungand Haier,2007;Schmithorstetal.,2005).
4.3. Conclusionsandimplicationsforeducation
capacitydifference,intelligenceisrelatedtothecapacityof prefrontalareastodynamicallyadapttotaskdemandsand environmentalcontingencies.Therefore,ourdatasupport thehypothesisthatthedynamicpropertiesofthese corti-calareasarerelatedtoaprolongeddevelopmentalperiod ofcorticalplasticityassociatedwithahigherIQ.
Theseresultshavepotentiallyimportantimplications for educational practice (Goswami, 2006; Blakemore, 2010).It hasbeenshownthatexecutivefunctions,even morethanIQ(BlairandRazza,2007),arestronglyrelatedto capacitiesdeemedtobeimportantinschoollearningsuch asnumericalprocessing(Qinetal.,2004)andreading com-prehension(SchlaggarandChurch,2009).Furthermore,a seriesofneuroimagingstudiesinadolescentsandchildren haveshownthatthesecapacitiesrelyonthesamebrain areasthatareassociatedwithexecutivefunctions(fora meta-analysis, see Houdéet al., 2010).A better under-standingoftheindividualdifferencesinfunctioningofthe cognitivecontrolnetworkinrelationtoIQandageperiod maycontributetoabetterunderstandingofhowtodevelop specifictrainingprogramstoimproveschoolperformance (Diamondetal.,2007).
AppendixA. Supplementarydata
Supplementary data associated with this arti-cle can be found, in the online version, at doi:10.1016/j.dcn.2011.09.007.
References
Achenbach,T.M.,1991.ManualfortheChildBehaviorChecklist4–18/and 1991 Profile. University of Vermont, Department of Psychiatry, Burlington,VT.
Blair,C.,Razza,R.P.,2007.Relatingeffortfulcontrol,executivefunction, andfalsebeliefunderstandingtoemergingmathandliteracyability inkindergarten.ChildDev.78(2),647–663.
Blakemore,S.J.,2010.Thedevelopingsocialbrain:implicationsfor edu-cation.Neuron65,744–747.
Brett,M.C.,Anton,J.L.,Valabregue,R.,Poline,J.B.,2002.Regionofinterest analysisusinganspmtoolbox.Neuroimage16,497.
Brown,J.W.,Braver,T.S.,2005.Learnedpredictionsoferrorlikelihoodin theanteriorcingulatecortex.Science307,1118–1121.
Bunge,S.A.,Wright,S.B.,2007.Neurodevelopmentalchangesin work-ing memory and cognitive control. Curr. Opin. Neurobiol. 17, 243–250.
Cosoco,C.A.,Kollokian,V.,Kwan,R.K.S.,Evans,A.C.,1997.Brainweb:online interfaceofa3-DMRIsimulatedbraindatabase.NeuroImage5,425. Crone,E.A.,2009.Executivefunctionsinadolescence:inferencesfrom
brainandbehavior.Dev.Sci.12,825–830.
Crone,E.A.,Zanolie,K.,VanLeijenhorst,L.,Westenberg,P.M.,Rombouts, S.A.R.B.,2008.Neuralmechanismssupportingflexibleperformance adjustmentduringdevelopment.Cogn.Affect.Behav.Neurosci.2, 165–177.
Dale,A.M.,1999.Optimalexperimentaldesignforevent-relatedfMRI. Hum.BrainMapp..
Daw, N.D., Niv, Y., Dayan, P., 2005. Uncertainty-based competition betweenprefrontalanddorsolateralstriatalsystemsforbehavioral control.Nat.Neurosci.8,1704–1711.
Deary,I.J.,Penke,L.,Johnson,W.,2010.Theneuroscienceofhuman intel-ligencedifferences.Nat.Rev.Neurosci.11,201–211.
Diamond,A.,Barnett,W.S.,etal.,2007.Preschoolprogramimproves cog-nitivecontrol.Science318,1387–1388.
Dumontheil,I.,Hassan,B.,Gilbert,S.,Blakemore,S.J.,2010.Development oftheselectionandmanipulationofself-generatedthoughtsin ado-lescence.J.Neurosci.30,7664–7671.
Duncan,J.,2003.Intelligencepredictsbrainresponsestodemandingtasks. Nat.Neurosci.6,207–208.
Eppinger,B.,Mock,B.,Kray,J.,2009.Developmentaldifferencesin learn-inganderrorprocessing:evidencefromERPs.Psychophysiology46, 1043–1053.
Frank,M.J.,Seeberger,L.C.,O’Reilly,R.C.,2004.Bycarrotorbystick: cognitive reinforcement learning in parkinsonism. Science 306, 1940–1943.
Goswami,U.,2006.Neuroscienceandeducation:fromresearchto prac-tice?Nat.Neurosci.Rev.7,406–413.
Gray,J.R.,Chabris,C.F.,Braver,T.S.,2003.Neuralmechanismsofgeneral fluidintelligence.Nat.Neurosci.6,316–322.
Holroyd,C.B.,Coles,M.G.H.,2002.Theneuralbasisofhumanerror pro-cessing:reinforcementlearning, dopamine, andtheerror-related negativity.Psychol.Rev.109,679–709.
Holroyd,C.B.,Coles,M.G.,2008.Dorsalanteriorcingulatecortex inte-gratesreinforcementhistorytoguidevoluntarybehavior.Cortex.44, 548–559.
Holroyd,C.B.,Nieuwenhuis,S.,Yeung,N.,Nystrom,L.,Mars,R.B.,Coles, M.G.H.,etal.,2004.Dorsalanteriorcingulate cortexshowsfMRI responsetointernal andexternalerrorsignals.Nat.Neurosci.7, 497–498.
Houdé,O.,Rossi,S.,Lubin,A.,Joliot,M.,2010.Mappingnumerical pro-cessing,reading,andexecutivefunctionsinthedevelopingbrain:an fMRImeta-analysisof52studiesincluding842children.Dev.Sci.13, 876–885.
Huizinga,M.,Dolan,C.V.,vanderMolen,M.W.,2006.Age-relatedchange inexecutivefunction:developmentaltrendsandalatentvariable analysis.Neuropsychologia44,2017–2036.
Johnson,M.H.,2011.Interactivespecialization:adomain-general frame-workforhumanfunctionalbraindevelopment.Dev.Cogn.Neurosci. 1,7–21.
Johnston,M.V.,Ishida,A.,Ishida,W.N.,etal.,2009.Plasticityandinjuryin thedevelopingbrain.BrainDev.31,1–10.
Jung,R.E.,Haier,R.J.,2007.Theparieto-frontalintegrationtheory(P-FIT) ofintelligence:convergingneuroimagingevidence.Behav.BrainSci. 30,135–187.
Karama,S.,Colom,R.,etal.,2011.Corticalthicknesscorrelatesof spe-cificcognitive performance accounted for by the general factor ofintelligenceinhealthychildrenaged6to18.NeuroImage 55, 1443–1453.
Kerns,J.G.,Cohen,J.D.,MacDonaldIII,A.W.,Cho,R.Y.,Stenger,V.A.,Carter, C.S.,2004.Anteriorcingulateconflictmonitoringandadjustmentsin control.Science303,1023–1026.
Koolschijn,P.C.,Schel,M.A.,DeRooij,M.,Rombouts,S.A.R.B.,Crone,E.A., 2011.A3-yearlongitudinalfMRIstudyonperformancemonitoring andtest–retestreliabilityfromchildhoodtoearlyadulthood.J. Neu-rosci.31,4204–4212.
Maia, T.V., 2009. Reinforcement learning, conditioning, and the brain:successesandchallenges.Cogn. Affect.Behav.Neurosci.9, 343–364.
O’Boyle,M.W.,Cunningtond,R.,Silk,T.J.,Vaughan,D.,Jackson,G., Syn-geniotis,A.,Egan,G.F.,2005.Mathematicallygiftedmaleadolescents activateauniquebrainnetworkduringmentalrotation.Cogn.Brain Res.25,583–587.
O’Brien,L.M.,Ziegler,D.A.,Deutsch,C.K.,Kennedy,D.N.,Goldstein,J.M., Seidman,L.J.,Hodge,S.,Makris,N.,Caviness,V.,Frazier,J.A.,Herbert, M.R.,2006.Adjustmentforwholebrainandcranialsizeinvolumetric brainstudies:areviewofcommonadjustmentfactorsandstatistical methods.Harv.Rev.Psych.14,141–151.
Pangelinan,M.M.,Zhang,G.,vanMeter,J.W.,Clark,J.E.,Hatfield,B.D., Hau-fler,A.J.,2011.Beyongageandgender:relationshipsbetweencortical andsubcorticalbrainvolumeandcognitive-motorabilitiesin school-agedchildren.NeuroImage54,3093–3100.
Prescott,J.,Gavrilescu,M.,Cunningtond,R.,O’Boyle,M.W.,Egan,G.F., 2010.Enhancedbrainconnectivityinmath-giftedadolescents:an fMRIstudyusingmentalrotation.Cogn.Neurosci.iFirst,1–20. Preusse, F., van der Meer, E., Deshpande, G., Krueger, F.,
Warten-burger,I.,2011.Fluidintelligenceallowsflexiblerecruitmentofthe parieto-frontalnetworkinanalogicalreasoning.Front.Neurosci.5, 1–14.
Qin,Y.,Carter,C.S.,etal.,2004.Thechangeofthebrainactivationpatterns aschildrenlearnalgebraequationsolving.Proc.Natl.Acad.Sci.U.S.A. 101,5686–5691.
Rushworth,M.F.,2008.Intention,choiceandthemedialfrontalcortex. Ann.N.Y.Acad.Sci.1124,181–207.
Schlaggar, B.L., Church, J.A., 2009. Functional neuroimaging insights intothedevelopmentofskilledreading.Curr.Dir.Psychol.Sci.18, 21–26.
pediatricpopulation:adiffusiontensorMRIstudy.Hum.BrainMapp. 26,139–147.
Shaw,P.,Greenstein,D.,Lerch,J.,Clasen,L.,Lenroot,R.,Gogtay,N.,Evans, A.,Rapoport,J.,Giedd,J.,2006.Intellectualabilityandcortical devel-opmentinchildrenandadolescents.Nature440,676–679. Somsen,R.J.,2007.Thedevelopmentofattentionregulationinthe
wis-consincardsortingtask.Dev.Sci.10,664–680.
vandenBos,W.,Güro˘glu,B.,vanderBulk,B.G.,Rombouts,S.A.R.B.,Crone, E.A.,2009.Betterthanexpectedorasbadasyouthought?The neu-rocognitivedevelopmentofprobabilisticfeedbackprocessing.Front. Neurosci.3,52.
vandenBos,W.,Cohen,M.X.,Kahnt,T.,Crone,E.A.,2011.Striatum-medial prefrontalcortexconnectivitypredicts developmentalchangesin reinforcementlearning.CerebralCortex(E-pubaheadofprint).
VanDuijvenvoorde,A.,Zanolie,K.,Rombouts,S.A.R.B.,Raijmakers,M., Crone,E.A.,2008.Valuingthepositiveoradjustingthenegative?A neurocognitiveanalysisoffeedback-basedlearning.J.Neurosci.28, 9495–9503.
Wechsler,D.,1991.WechslerIntelligenceScaleforChildren—Third Edi-tion.Manual.ThePsychologicalCorporation,SanAntonio. Wechsler, D.,1997.Wechsler AdultIntelligenceScale—ThirdEdition.
AdministrationandScoringManual.ThePsychologicalCorporation, SanAntonio.