ContentslistsavailableatScienceDirect
NeuroImage
journalhomepage:www.elsevier.com/locate/neuroimage
Longitudinal
network
re-organization
across
learning
and
development
Ethan
M.
McCormick
a,∗,
Sabine
Peters
b,c,
Eveline
A.
Crone
b,c,d,
Eva
H.
Telzer
a a Department of Psychology and Neuroscience, University of North Carolina, 235 E. Cameron Avenue, Chapel Hill, NC 27599, United States b Department of Developmental and Educational Psychology, Leiden University, 2333AK Leiden, the Netherlandsc Leiden Institute for Brain and Cognition, 2333ZA Leiden, the Netherlands
d School of Social and Behavioural Sciences, Erasmus University Rotterdam, Rotterdam, the Netherlands
a
b
s
t
r
a
c
t
Whileitiswellunderstoodthatthebrainexperienceschangesacrossshort-termexperience/learningandlong-termdevelopment,itisunclearhowthesetwo mechanismsinteracttoproducedevelopmentaloutcomes.Herewetestaninteractivemodeloflearninganddevelopmentwherecertainlearning-relatedchanges areconstrainedbydevelopmentalchangesinthebrainagainstanalternativedevelopment-as-practicemodelwhereoutcomesaredeterminedprimarilybythe accumulationofexperienceregardlessofage.Participants(8–29years)participatedinathree-wave,acceleratedlongitudinalstudyduringwhichtheycompleteda feedbacklearningtaskduringanfMRIscan.Adoptinganovellongitudinalmodelingapproach,weprobedtheuniqueandmoderatedeffectsoflearning,experience, anddevelopmentsimultaneouslyonbehavioralperformanceandnetworkmodularityduringthetask.Wefoundnonlinearpatternsofdevelopmentforbothbehavior andbrain,andthatgreaterexperiencesupportedincreasedlearningandnetworkmodularityrelativetonaïvesubjects.Wealsofoundchangingbrain-behavior relationshipsacrossadolescentdevelopment,whereheightenednetworkmodularitypredictedimprovedlearning,butonlyfollowingthetransitionfromadolescence toyoungadulthood.Theseresultspresentcompellingsupportforaninteractiveviewofexperienceanddevelopment,wherechangesinthebrainimpactbehavior incontext-specificfashionbasedondevelopmentalgoals.
1. Introduction
Thebrainisadynamicsystemcapableofreshapingitselfacrosstime toadapttoitsexternalenvironment.Forsomedevelopmentalprocesses (e.g.,cognitivecontrolorrisk-taking;Casey,2015;orsocioemotional development;BlakemoreandMills,2014),thesechangesunfoldacross longtimehorizons(e.g.,monthsoryears).However,functional devel-opmentdoesnotrequireyears,orevenmonths,toshowmeasurable changes.Indeed,abroadliteraturehasdemonstratedthatbrain func-tionrapidlyadaptstotaskdemandsandfeedbacktosupportskill acqui-sitionorgoal-directedbehavior(e.g.,Dawetal.,2006;Bassettetal., 2011; McCormick and Telzer,2017a, 2017b, 2018; Telesford et al., 2017;Gerratyetal.,2018).However,itremainsunclear towhat ex-tenttheseshort-term,learning-relatedchangesinbrainactivation over-lapwiththelong-term,maturationalplasticity seenacross yearsand decadesofdevelopment(Galván,2010).Here,wetesttwopotential ex-planationsforhowexperienceanddevelopmentinteractacrosstimeto explainchangesinlearningperformanceandthefunctionalbrain sys-temsthatsupportthatperformanceacrosstime.Toprobethese interac-tions,weadoptanovelapplicationoflongitudinalmodelingthatallows ustoconsiderchangesacrossminutes,years,andthecourseof devel-opmentsimultaneously.Thisapproachoffersanintegratedperspective oflearninganddevelopmentasco-dependentprocessesofneuraland behaviorplasticitywhichinteractacrosstime.
∗Correspondingauthor.
E-mail address: emccormick@unc.edu(E.M.McCormick).
While traditionally thought of as a period of vulnerability (Steinberg et al., 2008; Casey et al., 2008; Shulman et al., 2016), adolescence isalso aperiodassociated withincreasesin flexible be-havior and thecapacity tolearn from feedback in the environment (Johnsonand Wilbrecht,2011; Crone andDahl, 2012; Casey,2015; Vigilantetal.,2015),withneuralchangesassociatedwithage support-ingincreasedlearning(VanDuijvenvoordeetal.,2008;Petersetal., 2016;McCormickandTelzer,2017a;PetersandCrone,2017).In gen-eral.theabilitytolearnandengageinothercomplexcognitivetasks (Caseyetal.,2005;Lunaetal.,2010),improveswithagethroughthe firstdecadesoflife.However,thisco-occurrencedoesnotbyitself im-plythatmaturationisnecessaryfortheage-relatedimprovements in learningseenduringdevelopment.Withincreasedagealsocomesmore experienceandpracticeatskillsneededtosupporttaskperformance. Underthisview,developmentinvolvestheaccumulationofpracticeor trainingofneuralsystems,andtheneuralmechanismsforthisprocess shouldcloselyresemblethoseinvolvedinshort-termlearning.In con-creteterms,thiswouldimplythatdevelopmentallyyoungerindividuals canbetrainedtoperformaswellasolderindividualsgivensufficient practice.
Incontrast,aninteractiveviewoflearninganddevelopmentwould suggestthatcertainkindsofneuralchangesinresponsetolearningare constrainedbydevelopmentalchangesinthebrain.Inotherwords,it shouldbepracticallyimpossibletotrainachildtoperformatadult lev-elsbecausetheyhavenotexperiencedthematurationalchangesinthe brainnecessarytosupportthatperformance.Thiswouldsuggestthat
https://doi.org/10.1016/j.neuroimage.2021.117784
Received23October2020;Receivedinrevisedform7January2021;Accepted11January2021 Availableonline24January2021
1053-8119/© 2021TheAuthor(s).PublishedbyElsevierInc.ThisisanopenaccessarticleundertheCCBY-NC-NDlicense (http://creativecommons.org/licenses/by-nc-nd/4.0/)
certainkindsofneuralchangesinresponsetolearningwillberelatively uniquetoolderindividuals.Thesetwoalternativeaccountsarenot mu-tuallyexclusive,sincetrainingstudiesin youngerindividualsclearly demonstratethatthere issomecapacitytoimprovecognitive perfor-manceandshiftbrainfunctioneveninthematuringbrain(Jollesand Crone,2012).However,thefirstexplanationofhowlearningand devel-opmentinteractwouldpredictthatthiscapacitytotrainshouldbequite extensive,whereasthesecondexplanationwouldpredictthat biologi-calmaturationimposesstricterlimitationsontheabilitytotrainyoung individualto“adult” levelsofperformance.Itisimportanttonote, how-ever,thatthese limitationsmaynotbemaladaptive,butratherserve someotherdevelopmentalfunctionwhere“immature” brainstatesor behavioralperformanceareimportantforflexiblelearningand adapta-tion(e.g.,JohnsonandWilbrecht,2011;JollesandCrone,2012;Crone andSteinbeis,2017;McCormickandTelzer,2017a).
Amajorchallengeinmodelingexperienceanddevelopment simulta-neouslyisthatinrealdata,theyareoftenconfounded(Bell,1953;Jolles andCrone,2012;Telzeretal.,2018)indevelopmentalmodels.In lon-gitudinalstudieswhichuseacohort-sequential(orpanel)design,where individualsarerepeatedlyassessedatthesameages,olderparticipants arealsomore-experiencedparticipants(bothinlifeandpracticeinthe specificmeasuresofinterest).Inneuroimagingcontexts,these experi-enceeffectscanconfounddevelopmentaleffectsinanumberofways, includingreducinganxietyaboutthescanner environment,changing baselineconditions(the“taskB” problem),orreduce errorsontasks throughfamiliarityratherthanchangeinunderlyingability(Jollesand Crone,2012;Telzeretal.,2018).Fortunately,wecanleveragean alter-native,theacceleratedlongitudinaldesign,toaddressthesechallenges. Inacceleratedlongitudinalstudies,individualsvaryintheageoffirst as-sessmentandarefollowedlongitudinallythereafter.Byadoptingthis de-sign,wecande-coupleexperiencefromage(orothermeasureof devel-opmentalstage)sufficientlytosuccessfullymodeltheaccumulationof experienceanddevelopmentalmaturationsimultaneously(McCormick,
preprint).
Thecurrentstudyteststhetwocompetinghypothesesofhow expe-rienceanddevelopmentinteracttodriveneuralplasticityduring learn-ing.Participantsacrossawideagerange(8–29years)participatedina three-wave,acceleratedlongitudinalneuroimagingstudyduringwhich they completeda feedback learningtask.Byleveraging the acceler-atedlongitudinaldesignandanovelextensionofmixed-effectsmodels (McCormick,preprint),wedifferentiatebetweenthreetemporallevels ofneuralplasticity:1)short-termpractice-relatedchangeswithinascan session(within-individual)acrossblocksoffeedbacklearning;2) long-termchangeswithin individuals,acrossmeasurementoccasions (i.e., waves);and3)themixed(i.e.,within-andbetween-individual)effect ofchangesassociatedwithage.Byconsideringthesethreelevelsinthe samemodel,wecanpartitioneffectsateachlevel.(1)Within-session changesreflecthowbrainandbehavioradaptduringlearningthetask structure,(2)between-session changesreflect changesdue to experi-enceafterrepeatedexposuretothetaskandtestingenvironments,(3) whileagereflectsthedevelopmentaleffect.Importantly,including ef-fectsatthesecondlevelallowsustode-confoundageandexperience, givingamore reliableestimateof thedevelopmentaleffect. Because learningis anintegrative process,involvingtheinteractionsbetween manybrainregions (Bassettet al.,2011; Gerratyetal.,2014, 2015; McCormickandTelzer,2017a;Gerratyetal.,2018;McCormick,Gates, &Telzer, 2019),wetestthis developmentalmodelinthecontext of brainnetworks.Specifically,wemodeltheinteractionofpractice, expe-rience,anddevelopmenteffectsonnetworkmodularity.Modularityis ameasureofthedegreeofnetworksegregationintodistinctfunctional units(BullmoreandBassett,2011).Higherlevelsofmodularityinbrain networkspredictsincreasedlearning(Bassettetal.,2011;Ellefsenetal., 2015)andworkingmemory(Braunetal.,2015)performanceinadults. Ouranalyticapproachtoaddressingthesequestionsinvolvedseveral steps.First,wefitmixed-effectsmodelswithonlylinearandquadratic effectsofageonbehavioralperformanceandnetworkmodularity
dur-inglearningseparately(Petersetal., 2016;PetersandCrone,2017) forcomparisontomorecomplexmodels.Wethenincludedpredictors ofwithin-andbetween-sessionchangeasmaineffectstoconsiderthe uniqueeffectsofpractice,experience,anddevelopment,beforefitting amodelthatincludedinteractiontermsbetweenourpredictors.This thirdmodelallowedustoprobehowtheeffectsofthelower-level pre-dictors change across development in a continuous fashion. Finally, weestimatedabrain-as-predictormodelwhereweprobedhowbrain states(e.g.,highversuslowmodularity)differentiallypredicted learn-ingperformanceacrosspractice,experience,anddevelopment.This fi-nalmodeltestsacoredifferencebetweenthetwoexplanationsof devel-opmentalimprovementinlearningperformance.Inthe development-as-practiceview,networkmodularityshouldpredictlearningperformance consistentlyregardless ofwhenin thedevelopmentaltrajectory(i.e., there isnomoderationbyage).This iscontrastedbytheinteraction viewofdevelopmentandexperience,wherewewouldexpectthat mod-ularitywouldpredictperformancedifferentiallydependingonage. 2. Methods
2.1. Sample
Atotal of299participants(ages 8–29years;153female) partici-patedina3-wave,acceleratedlongitudinalMRIstudy.Participantswere scannedevery2years,spanninga5-yearperiod(Fig.1).Atwave1,28 participantswereexcludedforanumberoffactorsincludingnot com-pletingtheMRIsession(N=4),excessivemovementduringthescan session(>3mmrelativemotionin anydirection/rotation)(N=22), ADDdiagnosisdisclosure(N=1),andreportedmedicineuse(N=1), re-sultinginafinalsampleof271participantsattheinitialdatacollection (140female;Mage =14.17,SD=3.63,range=8.01–25.95years).Atwave2 (2yearslater),254participantswerescanned(33couldnotbescanned duetobraces;11declinedtoreturn).Ofthescannedparticipants,an additional21wereexcluded(12formotion;2forpreprocessingerrors; 5for T2artifacts;1formedicine use;1forADDdiagnosis),leaving afinalsampleof233participants(121female;Mage =16.15,SD=3.62,
range=10.02–26.61years).Duringthefinalwave(2yearslater),243 participantswerescanned(11couldnotbescannedduetobraces;45 declinedtoreturn).Ofthese,11wereexcluded(3didnotcompleteMRI session;4formotion;2forprocessingerrors;1formedicineuse;1for ADDdiagnosis),foratotalfinalsampleof232participants(121female;
Mage =18.15,SD=3.68,range=11.94–28.72years).Acrossthedataset, 183participantshaddataatallthreewaves,78participantshaddata attwowaves,31participantshaddataatonlyonewave,and7were excludedatallthreewaves.Atotalof736scanswereincludedforfinal dataanalyses.Whenconsideredatthetraillevel,thesescansyielded 4799observationsformodeling.
IQscoresweremeasuredatthefirsttwowavesofdatacollection, us-ingtheWISC-III(forparticipants<16years;NW1=195;NW2=119)or WAIS-III(forparticipants≥16years;NW1=76;NW2=114).All partic-ipantswerewithinthenormalrangeatwave1(M=109.8,SD=10.34,
range=80–142.5)andwave2(M=108.3,SD=10.27,range=80–147.5). Further details andthedistributions of thedescriptivevariables are availableinthesupplementalmaterial.
2.2. Feedbacklearningtask
ParticipantscompletedafeedbacklearningtaskduringanfMRI ses-sion(Petersetal.,2014,2016).Oneachtrial,participantssawascreen withthreeemptyboxesandone(outofapossiblesetofthree)stimulus underneath(Fig.2).Participantsweretoldthateachstimuluswithina givensethadacorrespondingcorrectlocationamongtheemptyboxes andthattheirgoalonthetaskwastoappropriatelysorteachstimulus intoitslocation.Foreachstimulus-locationchoice,participantseither receivedpositive(a“+” sign)ornegative(a“-” sign)feedbackbased ontheirchoice.Positivefeedbackindicatedcorrectstimulusplacement,
Fig.1. Structureofrepeatedmeasureswithin the accelerated longitudinal design. Partici-pantsareorderedinascendingorderbasedon theirageatwave1.Sexisdenotedbyseparate colors.
Fig.2. DuringtheFeedbackLearningtask,participantslearnedthecorrectplacementofeachstimuli(e.g.,theelephant)throughfeedback.Participantsreceived eitherpositiveornegativefeedbackoneachtrial.
whilenegativefeedbackindicatedincorrectplacement.Eachstimulus withinasetassociatedwithaunique,deterministicallycorrectlocation. Stimuliwithinasetwerepresentedinpseudorandomorder,constrained suchthatnostimuluswithinasetwaspresentmorethantwiceinarow. Afteramaximumof12trialsperblock,orafterallthreestimuliwithin asetwerecorrectlyplacedtwice(indicatingthatalllocationswere suc-cessfullylearned),stimulussetswereswappedoutforanewsetwith threenewstimuli.Participantssawatotalof15blocksof3-stimulisets (foramaximumpossibleof180trials)atwaves1and2,and10blocks (maximumpossible120trials)atwave3.PriortotheMRIsession, par-ticipantspracticedthreeexamplesetsofstimuli.Eachtrialconsistedof thefollowing:1)a500-msfixationcross,2)stimuluspresentationfor 2500mswhileparticipantsmadelocationdecisions,and3)feedback presentationfor1000ms. Trialswereseparatedbyintervalsjittered basedonOptSeq(Dale,1999),withdurationsthatvariedbetween0 and6s.
2.3. Behavioralanalyses 2.3.1. Taskmetricsofbehavior
Ourprimarymetricoftaskperformancewasthelearningrate par-ticipantsdisplayedinformingcorrectstimulus-locationassociations.To calculate learningrate,we distinguishedbetween twophasesoftask performance:thelearningandtheapplicationphase(Petersetal.,2014, 2016).Thelearningphasewasdefinedastrialswherethecorrect loca-tionforagivenstimuluswasstillunknown,andparticipantsneeded torely ontrial-and-erroror hypothesistesting tocorrectlyplacethe stimulus.Trialsinthelearningphasecouldresultineitherpositive (in-dicatingafuture staystrategy)ornegative(promptingafutureshift strategy)feedback. Incontrast,theapplicationphasewas definedas trialswherethecorrectlocationforthepresentedstimulusisalready known(asestablishedbyanearlierlearningtrial)andparticipants cor-rectly placethatstimulusagain.Learning ratewascalculatedas the
proportionoftrialsinthelearningphasewherefeedbackwascorrectly appliedinthefollowingtrialinvolvingthesamestimulus(eitheras re-peatedplacementfollowingpositivefeedbackorasalteredplacement followingnegativefeedback)outofall thetrialsduringthelearning phase.
2.3.2. Linearmixed-effectsmodel
Totestourdevelopmental/experienceinteractionmodel,wefita lin-earmixed-effectsmodeltoparticipants’learningratedata.Wefollowed amodel-buildingproceduresimilartotheoneusedinpreviouswork inthissample(Petersetal.,2016;PetersandCrone,2017).This pro-cedureinvolvedabuild-upapproachwherewetestedmaineffectsand theninteractionsoftimetoestablishtheoptimaldevelopmentalform beforebringinginadditionalpredictors.WefitarandomeffectsANOVA modelwitharandominterceptwhichservedasacomparisonfor subse-quentmodels.Fordescriptivepurposesattherandomeffectslevel,we fitathreelevelmodelwhereblockswerenestedwithinwaveandthen withinperson,howeverforcomparisonwithfuturemodels,wealsofita twolevelmodelwherewaveandagewereincludedatlevel1.To com-pareacrosslevelsofchange,weconstructedamodelusingthelme4 softwarepackagethroughR(version1.1–21;Batesetal.,2015),where stimulusblocks(N=1-max15)andwave(N=1–3)werenestedwithin individual,andagewasincludedasatime-varyingcovariate.Because wave(i.e.,repeatedexposuretothetask)wasapredictorofinterest, wedidnotnestwithrespecttowavesincethatwouldresultina vari-ablethatactsasbothanestingfactorandlineareffectofinterest.We includedinteractionsbetweenwave,age,andblocks.Tocapturemore complexchangesin behaviorbetweenblocks ofthetask,weutilized piece-wiseregressionatlevel1(Flora,2008;Lietal.,2001),including predictorswhichmodelthelineareffectsacrossthefirstandsecondhalf ofthetaskseparately.Thismodelresultedinthefollowingequation:
ReducedForm: 𝐿𝑒𝑎𝑟𝑛𝑖𝑛𝑔𝑅𝑎𝑡𝑒𝑖𝑗 =𝛾00+𝛾10𝐹𝑖𝑟𝑠𝑡𝐻𝑎𝑙𝑓𝑖𝑗 +𝛾20𝑆𝑒𝑐𝑜𝑛𝑑𝐻𝑎𝑙𝑓𝑖𝑗 +𝛾30𝑊𝑎𝑣𝑒𝑖𝑗 +𝛾40𝐴𝑔𝑒𝑖𝑗 +𝛾50𝐴𝑔𝑒2𝑗 +𝛾60𝐹𝑖𝑟𝑠𝑡𝐻𝑎𝑙𝑓𝑖𝑗 ∗𝑊𝑎𝑣𝑒𝑖𝑗 +𝛾70𝑆𝑒𝑐𝑜𝑛𝑑𝐻𝑎𝑙𝑓𝑖𝑗 ∗𝑊𝑎𝑣𝑒𝑖𝑗 +𝛾80𝐹𝑖𝑟𝑠𝑡𝐻𝑎𝑙𝑓𝑖𝑗 ∗𝐴𝑔𝑒𝑖𝑗 +𝛾90𝑆𝑒𝑐𝑜𝑛𝑑𝐻𝑎𝑙𝑓𝑖𝑗 ∗𝐴𝑔𝑒𝑖𝑗 +𝛾100𝐹𝑖𝑟𝑠𝑡𝐻𝑎𝑙𝑓𝑖𝑗 ∗𝐴𝑔𝑒2𝑖𝑗 +𝛾110𝑆𝑒𝑐𝑜𝑛𝑑𝐻𝑎𝑙𝑓𝑖𝑗 ∗𝐴𝑔𝑒2𝑖𝑗 +𝛾120𝑊𝑎𝑣𝑒𝑖𝑗 ∗𝐴𝑔𝑒𝑖𝑗 +𝛾130𝑊𝑎𝑣𝑒𝑖𝑗 ∗𝐴𝑔𝑒2𝑖𝑗 +𝑢0𝑗 +𝑟𝑖𝑗
While previous work has discouraged using wave as a predic-torin longitudinalmodels(insteadusing preciseage;see Mehtaand West,2000),herewedrawameaningfuldistinctionbetweenwaveand age.Wewouldexpectchangesinbehavioraftereachsubsequent expo-surethetaskenvironment(Telzeretal.,2018),whichduetothe sam-plingmethodinacceleratedlongitudinaldesignsisdisassociatedwith agetosome degreebecausealarge agerangeis representedateach wave.
2.4. fMRIdataacquisitionandprocessing 2.4.1. MRIdataacquisition
Scans across all three waves were acquired using the same Philips 3T MRI scanner, utilizing identical scan settings. The Feed-back Learning Task includedT2∗-weighted echoplanar images (EPI; slicethickness=2.75mm;38slices; sequentialacquisition; TR=2.2s; TE=30 ms; FOV=220 × 220 × 114.68 mm). Additionally, struc-tural images were acquired, including a high-resolution 3D T1-FFE anatomical scan (TR=9.76 ms; TE=4.59 ms; 140 slices; voxel size=0.875×0.875×1.2mm;FOV=224×177×168mm;flipangle=8). Priortoundergoingthescanprocedure,participantswereintroducedto thescannerenvironment(e.g.,spaceandnoises)throughamockscan session.
2.4.2. fMRIdatapreprocessingandanalysis
PreprocessingandanalysesutilizedasuiteoftoolsfromFSLFMRIBs SoftwareLibrary(FSLv6.0;https://fsl.fmrib.ox.ac.uk/fsl/),Stepstaken duringpreprocessingincludedskullstrippingofallimagesusingBET; andslice-to-slicemotioncorrectionofEPIimagesusingMCFLIRT; co-registerationinatwo-stepsequencetothehigh-resolutionT2-weighted andT1-FFEanatomicalimagesusingFLIRTinordertowarptheminto thestandardstereotacticspacedefinedbytheMontrealNeurological Institute(MNI) andtheInternationalConsortium forBrainMapping; andtheapplicationofa128shigh-passtemporalfiltertoremovelow frequencydriftwithinthetime-series.
2.4.3. Nuisanceregressors
Prior tomodelingthefMRIdatafurther,wetook severalstepsto reducetheinfluenceofmotion.Motion,asmeasuredbyframewise dis-placement(Poweretal.,2012),wasminimalacrossthesample(mean acrossparticipants=0.12mmFD;max=0.77mm;average percent-ageof volumeswith>0.3mm FD=4.95%).Wealsocontrolledfor 8nuisanceregressorsin theGLM andtime-seriesanalyses:6motion parametersgeneratedduringrealignmentandtheaveragesignalfrom boththewhitematterandcerebrospinalfluidmasks.Previouswork(see Ciricetal.,2017)hasshownthatthesestrategiesreducetheinfluence ofmotiononfunctionalconnectivityanalyses.
2.4.4. Graphconstruction
Wethenutilizedagraphtheoreticalapproach toinvestigate how networks in the brainchanged across levels of practice, experience, anddevelopment.UsingasubsetoftheBigBrainparcellationscheme (Sietzmanetal.,2020),anatlascomprisedof300,5-mmsphereparcels fromcorticalandsubcorticalregions,weextractedfunctionaltimeseries dataforeachblock(15intotal)inordertomodelchangesinnetwork structureacrosstimeduringthetask.Wechosetoexaminenetwork fea-turesbetweenregions withtheoreticalrelevancetotask performance during learning. This resulted in 147regions including those in the cingulo-opercular(14),defaultmode(55),fronto-parietal(27),salience (14),ventral(9)anddorsal(14)attention,hippocampal(6),andreward (8)sub-networks (fortherelevantROI coordinatesona whole-brain projection,seeFigureS1).Selectionofthesenetworkswereguidedby thoseregionsengagedinthefeedbacklearningtaskinpreviousresearch (Petersetal.,2016;Petersetal.,2016)orclassicallyengagedduring learninganddecision-making (DawandShohamy, 2008; Sadaghiani andD’Esposito,2014;McCormickandTelzer,2018a;McCormicketal., 2019).Thissubsetwaschosentobalanceincludingenoughregionsof interestwiththechallengesofcomputingwhole-brainnetworkson rel-ativelyshorttimeseries.Regionswereincludedorexcludedasagroup basedontheirnetworklabel(e.g.,allfronto-paretialregionswere in-cludedwhileallvisualregionswereexcluded).Asafollow-up sensitiv-ityanalysis,were-ranallanalyseswithmodularitycalculatedonthe whole-brainnetwork(whole-brainmodularityvalues areavailablein theposteddatafile;https://osf.io/62gwz/).Resultsremained substan-tivelyunchangedandweretainedthesubsettomaintainfidelitytothe originalanalysisplan.
Toextract, weconstructed a task regressormadefrom theonset anddurationofeachblockofstimuliconvolvedwithanHRFfunction. These regressorsweremultipliedwiththeentiretimeseriesextracted fromROIsinordertogiveasetoftime-seriesfilesforeachindividual ateachwave.Blockdurations(M=45.32s;SD=3.26;range=37.27– 60.23 s)werecomparabletoapproachesused indynamic functional connectivityanalyses(approximately30s;Shireretal.,2012; Gonzalez-Castilloetal.,2015).Correlationmatriceswereconstructedby comput-ingthezero-lagcross-correlationbetweeneachROI.Graphmetricswere calculatedacrossarangeofcosts(5–20%in5%increments;Cohenand D’Esposito,2016).Weutilizedthestandardcommunityassignmentfor distinguishingwithin-versusbetween-networkedges(Sietzmanetal., 2020).
Fig.3.ProbingInteractionswithAge.Fourdistinctageswerechosentoprobeinteractionswithage,includingduringearlyadolescence(12,[~ −1SD];red),middle adolescence(16,[~ meanage],green),lateadolescence(20,[+1SD];blue),andyoungadulthood(25,[~ +2SD],purple).Barheightrepresentstheproportionof observationsatthatlevel.Rugplothashes(belowx-axis)representindividualobservations.
2.4.5. Graphmetric
Forourmeasureofbraindevelopment,wecalculatednetwork mod-ularity,orthedegreetowhichcommunitieswithinthebrainnetwork aresegregated.Modularityiscomputedastherelativenumberofedges betweennodesofthesamecommunitycomparedtothenumberof to-taledgeswithinthebraingraph.Wecalculatedmodularity(Q∗)using positiveandnegativeweightededgesas:
𝑄∗= 1 2𝑊+ ∑ 𝑖𝑗 ( 𝑤+ 𝑖𝑗 −𝑒+𝑖𝑗 ) 𝛿(𝑚𝑖 ,𝑚𝑗 ) − 1 2𝑊++2𝑊− ∑ 𝑖𝑗 ( 𝑤− 𝑖𝑗 −𝑒−𝑖𝑗 ) 𝛿(𝑚𝑖 ,𝑚𝑗 )
wherew+isthenumberofpositivelyweightededgesandw−isthe num-berofnegativelyweightededges.The𝑒𝑖𝑗 termrepresentstheexpected numberofedgesbetweentwonodesiandj,andthe𝛿(𝑚𝑖 ,𝑚𝑗 )termis1 ifthenodesiandjareinthesamemoduleand0iftheyarenotinthe samemodule.Noticethatnegativelyweightededgesaregivenless influ-encethanpositivelyweightededgesincomputingmodularity(Rubinov &Sporns,2011).
2.5. Developmentalmodel
Wethenutilizedthesamemulti-levelmodelingapproachusedfor thebehaviortomodelchangeinbrainnetworksacrossblocks,waves, andage: 𝑀𝑜𝑑𝑢𝑙𝑎𝑟𝑖𝑡𝑦𝑖𝑗 =𝛾00+𝛾10𝐹𝑖𝑟𝑠𝑡𝐻𝑎𝑙𝑓𝑖𝑗 +𝛾20𝑆𝑒𝑐𝑜𝑛𝑑𝐻𝑎𝑙𝑓𝑖𝑗 +𝛾30𝑊𝑎𝑣𝑒𝑖𝑗 +𝛾40𝐴𝑔𝑒𝑖𝑗 +𝛾50𝐴𝑔𝑒2𝑗 +𝛾60𝐹𝑖𝑟𝑠𝑡𝐻𝑎𝑙𝑓𝑖𝑗 ∗𝑊𝑎𝑣𝑒𝑖𝑗 +𝛾70𝑆𝑒𝑐𝑜𝑛𝑑𝐻𝑎𝑙𝑓𝑖𝑗 ∗𝑊𝑎𝑣𝑒𝑖𝑗 +𝛾80𝐹𝑖𝑟𝑠𝑡𝐻𝑎𝑙𝑓𝑖𝑗 ∗𝐴𝑔𝑒𝑖𝑗 +𝛾90𝑆𝑒𝑐𝑜𝑛𝑑𝐻𝑎𝑙𝑓𝑖𝑗 ∗𝐴𝑔𝑒𝑖𝑗 +𝛾100𝐹𝑖𝑟𝑠𝑡𝐻𝑎𝑙𝑓𝑖𝑗 ∗𝐴𝑔𝑒2𝑖𝑗 +𝛾110𝑆𝑒𝑐𝑜𝑛𝑑𝐻𝑎𝑙𝑓𝑖𝑗 ∗𝐴𝑔𝑒2𝑖𝑗 +𝛾120𝑊𝑎𝑣𝑒𝑖𝑗 ∗𝐴𝑔𝑒𝑖𝑗 +𝛾130𝑊𝑎𝑣𝑒𝑖𝑗 ∗𝐴𝑔𝑒2𝑖𝑗 +𝑢0𝑗 +𝑟𝑖𝑗 2.6. Probinginteractions
Tobetterunderstandpotentialinteractioneffectsinvolvingagein themodelsofbehaviorandbrain,weprobedtheeffectsat four dis-tinct ages (Fig. 3). Ageswere chosen tobe evenlyspaced (approxi-mately standarddeviationdistances within thesample)androughly correspondtodifferentdevelopmentalperiodsincludingearly,middle, andlateadolescences,aswellasyoungadulthood(e.g.,Shulmanetal., 2016).Interactioneffectsin themodelarecontinuousacross theage rangeandthereforeleverageinformationacrossthesample.However, duetolowercoverageofobservationsatlaterages,thesimpleslope esti-mateswhenprobingtheinteractionattheselevelshavecorrespondingly largerstandarderrors.
3. Results
3.1. Descriptivesandage-onlygrowthmodels
Beforeformallyfittingmodelstothedata,weassesseddescriptives ofbothlearningrateandnetworkmodularityasafunctionofwaveand age(takingthemeanofwithin-sessiondata).Connecteddatapoints rep-resentthesameindividualacrosstimeandwavesarelabeledwith differ-entcolors(Fig.4A&B).Asreportedearlier(PetersandCrone,2017), learningratewashighoverallwithlateadolescentsperformingnear ceiling,andeitherlevelingoff or decliningforolderparticipants. Al-thoughnotconstrainedinthesamewayaslearningperformanceat up-pervalues,neuralnetworkmodularityappearstoincreaseatearlierages anddecliningatlaterages(seeSupplementalforformalregionsof sig-nificanceanalysisforallmodels).Asourfirstmodelbuildingstep,we fitarelativelysimplemodelbyincludinglinearandquadraticeffects ofagetobothlearningrateandnetworkmodularity.Predictedvalues ofeachmeasurewereobtainedfromthemixed-effectsmodel(Fig.4C &D;seeTable1),confirmingthesetrends.However,thissimplemodel neithercaptureswithin-sessionpractice-relatedchange,nordoesit dis-aggregatewithin-personchangesduetoexperience(i.e.,acrosswaves) andbetween-personchangesduetomaturation(i.e.,acrossage).We
Fig.4. Learning(A)andNetworkModularity(B)AcrossAgeandWave.Meansofwithin-sessiondatawereplottedagainstagetovisualizepotentialtime-related trends.Wavenumberwasindicatedbycolor,anddatafromthesameindividualwereconnectedbyasolidline.Ifindividualsonlycontributeddataatonetimepoint, datawasindicatedwithalonepoint.Linearmixed-effectsmodelsthatonlyconsiderthemixedeffectofagesuggestquadraticeffectspeakinginlateadolescencefor learningperformance(C),andfornetworkmodularity(D).Thesemodelsfailtoconsidereffectsofexperience(i.e.,wave)andwithin-sessionpracticeeffects.
Table1
Modeloutputfromage-onlymodels.
Learning Rate Network Modularity
Predictors Estimates SE P-Value df Estimates SE P-Value df
Intercept 0.937 0.004 < 0.001 359.604 0.994 0.011 < 0.001 361.688 Age std 0.277 0.023 < 0.001 1003.133 0.275 0.023 < 0.001 1992.711 Age 2 std − 0.140 0.021 < 0.001 1924.892 − 0.209 0.020 < 0.001 3523.698 Random Effects 𝜎2 0.010 0.055 𝜏00 0.002 ID 0.030 ID ICC 0.18 0.35 N 297 ID 297 ID Observations 4799 4799 Marginal R 2 / Conditional R 2 0.065 / 0.236 0.075 / 0.402
Note: Alleffectsroundedtothethirddecimalplacefordisplaypurposes.std=standardizedeffectsreported.𝜎2=level1 randomeffect.𝜏00=higherlevelrandomeffect(effectspecifiedbysubscript).ICC=Intraclasscorrelation.N =number ofunitsateachlevel.
nextformallytestedtheseeffectsusingthedevelopmentalmodel speci-fiedabove.
3.2. Behavioralimprovementsinlearningperformance
WebeganbyfittinganunconditionalrandomeffectsANOVAmodel (i.e.,arandominterceptateachlevel)todeterminethedistributionof varianceacrossthethreelevelsofthemodel(level1=within-session; level2=waves;level3=individual).Resultsindicatedthatthe
major-ityofvarianceinlearningratewasbetweentrialswithinthesamescan session(68.7%),anadditional19.3%ofthevariancewasbetweenscan sessionswithinthesameindividual(i.e.,changeacrosswaves),andthe remaining11.9%variancewasaccountedforbybetween-individual dif-ferencesinoveralllearningperformance(seeTable2 forfulldetails). However,toestablish abaselineforfuturemodels,wealsofitatwo levelrandomeffectsANOVAmodelwherelevel1and2arecollapsed (seeTable2).
Table2
ModeloutputfromrandomeffectsANOVAmodels.
3-Level Model Learning Rate Network Modularity
Predictors Estimates SE P-Value df Estimates SE P-Value df
Intercept 0.927 0.003 < 0.001 256.681 0.949 0.012 < 0.001 284.150 Random Effects 𝜎2 0.009 0.048 𝜏00 0.002 ID:Wave 0.014 ID:Wave 0.002 ID 0.030 ID ICC 0.31 0.49 N 297 ID 297 ID 3 Wave 3 Wave Observations 4799 4799 Marginal R 2 / Conditional R 2 0.000 / 0.313 0.000 / 0.486
2-Level Model Learning Rate Network Modularity
Predictors Estimates SE P-Value df Estimates SE P-Value df
Intercept 0.925 0.003 < 0.001 271.214 0.946 0.012 < 0.001 289.668 Random Effects 𝜎2 0.010 0.056 𝜏00 0.003 ID 0.037 ID ICC 0.22 0.40 N 297 ID 297 ID Observations 4799 4799 Marginal R 2 / Conditional R 2 0.000 / 0.218 0.000 / 0.398 Note: Alleffectsroundedtothethirddecimalplacefordisplaypurposes.𝜎2=level1randomeffect.𝜏
00=higherlevel randomeffect(effectspecifiedbysubscript).ICC=Intraclasscorrelation.N =numberofunitsateachlevel.
Table3
Modeloutputfrommaineffects-onlymodels.
Learning Rate Network Modularity
Predictors Estimates SE P-Value df Estimates SE P-Value df
Intercept 0.927 0.004 < 0.001 717.508 0.987 0.013 < 0.001 520.937 First Half std 0.025 0.016 0.113 4473.435 0.088 0.013 < 0.001 4491.297 Second Half std 0.022 0.016 0.151 4473.468 0.121 0.013 < 0.001 4491.319 Wave std 0.023 0.018 0.198 1205.547 − 0.004 0.019 0.845 694.861 Age std 0.250 0.31 < 0.001 299.628 0.285 0.039 < 0.001 311.080 Age 2 std − 0.136 0.021 < 0.001 1941.031 − 0.209 0.020 < 0.001 3747.423 Random Effects 𝜎2 0.010 0.051 𝜏00 0.002 ID 0.030 ID ICC 0.18 0.37 N 297 ID 297 ID Observations 4799 4799 Marginal R 2 / Conditional R 2 0.060 / 0.232 0.112 / 0.441
Note: Alleffectsroundedtothethirddecimalplacefordisplaypurposes.std=standardizedeffectsreported.𝜎2=level1 randomeffect.𝜏00=higherlevelrandomeffect(effectspecifiedbysubscript).ICC=Intraclasscorrelation.N =number ofunitsateachlevel.
3.2.1. Separatingeffectsofageandwave
Next,wefita mixedeffectsmodelwithfixed predictorsof learn-ingrateateachleveltoassesstheseparableeffectsofageandwave onlearningrate.Importantly,waveandageweresufficiently decou-pledinthis model(r =0.397,varianceinflationfactor= 1.19,SE in-flation=1.08times),andallpredictorswerecentered.Inthismodel neithereffectofthewithin-sessionpracticepredictorsweresignificant. Thissuggeststhattherewasnototalsystematicchangeinlearningrate withinascansessionnettheeffectsofwaveandage,norwastherea significantwithin-personeffectofwave.Inotherwords,neither within-sessionpracticeorbetween-sessionexperiencerelatedtoincreased per-formancewhenaccountingfortheage-relatedchange.However, indi-vidualsshowedsignificant linear(𝛽 = 0.250,SE=0.031,p< .001) andquadratic(𝛽 =−0.136,SE= 0.021,p<.001)effects ofage on learningrate.Allresultsarereportedasstandardizedeffects(Table3). Alikelihoodratiotestsuggeststhatthismodeloffersanimprovement overtheunconditionalmodel(ΔAIC=−158.7;𝜒𝑑𝑖𝑓𝑓 2 =168.71,df=5,
p<.001).
3.2.2. Learningrateimprovementsshowinteractionsacrosslevelsoftime
Finally,wetestedtheinteractivemodeloflearninganddevelopment forparticipants’learningrates.Todoso,weaddedtwo-waycross-level interactiontermstothepreviousmodel.Three-wayinteractionswere exploredbutwerenotfoundtobesignificantandsothemodelwith two-wayinteractionswasretained.Allpredictorswerecenteredto cre-ate interactiontermswhich wereuncorrelatedwiththe maineffects andtofacilitatetheinterpretationof maineffectsinthepresenceof interactionterms(AikenandWest,1991).Therewasasignificant posi-tiveinteractionofwaveandthequadraticeffectofageonlearningrate (𝛽 =0.083,SE=0.021,p<.001)suchthatateachlaterwaves,the quadraticdecreaseslessen.Toprobethisinteraction,weplottedmean within-sessionlevelincreasesinlearningrateacrossageforeachwave (Fig.5).Resultssuggestthatwithoutrepeatedexposuretothetask(i.e., experience) therearepredicteddecreasesin learningperformanceat youngerages(redtrajectory),butthatpracticehelpscompensateand causeperformancetoleveloff instead(greenandbluetrajectories).The quadraticeffectwherewaveiscodedaszero(i.e.,wave2)reflectsthe
Fig. 5.Increased Experience Impacts
Learn-ing Trajectories. Compared with first
expo-sure(red),accumulatingexperience(greenand blue)betweenwavestendedtopredictbetter learningperformanceatlaterages, compensat-ingforexpecteddeclinesinperformance dur-ingyoungadulthood.
Table4
Modeloutputfrominteractionsmodels.
Learning Rate Network Modularity
Predictors Estimates SE P-Value df Estimates SE P-Value df
Intercept 0.935 0.005 < 0.001 768.901 0.997 0.014 < 0.001 513.739 First Half std − 0.007 0.020 0.751 4473.838 0.094 0.017 < 0.001 4495.715 Second Half std 0.021 0.020 0.297 4473.787 0.115 0.017 < 0.001 4495.776 Wave std 0.011 0.031 0.722 3143.488 − 0.044 0.029 0.131 1734.974 Age std 0.160 0.044 < 0.001 650.191 0.327 0.049 < 0.001 471.938 Age 2 std − 0.080 0.042 0.053 615.163 − 0.261 0.047 < 0.001 513.285 First Half ∗ Wave
std − 0.016 0.022 0.469 4484.108 − 0.036 0.019 0.058 4508.580 Second Half ∗ Wave
std − 0.044 0.022 0.048 4484.488 − 0.022 0.019 0.244 4508.591 First Half ∗ Age
std − 0.049 0.022 0.027 4466.136 0.068 0.019 < 0.001 4482.736 Second Half ∗ Age
std 0.025 0.022 0.252 4466.143 − 0.004 0.019 0.818 4482.736 Wave ∗ Age
std − 0.038 0.024 0.111 870.145 0.002 0.026 0.947 648.001 First Half ∗ Age 2
std 0.047 0.024 0.048 4464.637 − 0.027 0.020 0.182 4481.364 Second Half ∗ Age 2
std − 0.025 0.024 0.292 4464.630 − 0.002 0.020 0.906 4481.360 Wave ∗ Age 2 std 0.083 0.021 < 0.001 4746.733 0.057 0.018 0.001 4674.686 Random Effects 𝜎2 0.010 0.051 𝜏00 0.002 ID 0.030 ID ICC 0.19 0.37 N 297 ID 297 ID Observations 4799 4799 Marginal R 2 / Conditional R 2 0.066 / 0.239 0.126 / 0.449
Note: Alleffectsroundedtothethirddecimalplacefordisplaypurposes.std=standardizedeffectsreported.𝜎2=level1 randomeffect.𝜏00=higherlevelrandomeffect(effectspecifiedbysubscript).ICC=Intraclasscorrelation.N =number ofunitsateachlevel.
effectsseenintheage-onlymodelandisconsistentwithpriorresearch (Petersetal.,2016;PetersandCrone,2017),however,byincluding in-teractionswithexperience,weshowhowthattotaleffectisinfluenced byrepeatedexposuretothetask(seeTable4 forfulldetails).A likeli-hoodratiotestsuggeststhatthismodeloffersanimprovementoverthe main-effectsonlymodel(ΔAIC=−22;𝜒𝑑𝑖𝑓𝑓 2 =38.03,df=8,p<.001). Inafollow-upsensitivityanalysis,wefoundthatthispatternofeffects heldwhenincludingIQasacovariate.
3.3. Changesinnetworkmodularity
Similartothebehavioralanalysis,wefirstfitanunconditional ran-domeffectsANOVAmodeltoparticipants’neuralnetworkmodularity data.Themajorityofvarianceinnetworkmodularitywasbetweentrials withinthesamescansession(51.4%),relativelyless(15.6%)ofthe
vari-ancewasbetweenscansessionswithinthesameindividual(i.e.,change acrosswaves),withtheremainder(33.0%)accountedforby between-individualdifferencesinnetworkmodularity(Table2).
3.3.1. Separatingeffectsofageandwave
Next,wefitamain-effectsonlymodelwithpredictorsincludingtask block,wave,andage.Therewerelinear(𝛽 =0.285,SE=0.0.39,p<
.001)andquadratic(𝛽 =−0.136,SE=0.021,p<.001)effectsofage, suchthatnetworkmodularitytendedtoincreaseearlyinadolescence andlevel off and decrease acrosslateadolescence andyoung adult-hood. Additionally,modularityincreasedacross blockswithin waves (i.e., practice) across both halves of the task (first half:𝛽 = 0.088,
SE=0.013,p<.001;secondhalf:𝛽 =0.121,SE=0.013,p<.001). Therewasnoindependenteffectofwaveonnetworkmodularity.This suggeststhatwhilemodularitytendstoincreasewithin-sessionacross
Fig.6. ExperienceandAgeImpactBrainNetworkOrganization.A)Accumulatingexperience(greenandblue)acrosswavespredictsincreasednetworkmodularity comparedwithfirstexposure(red),howeverthesedifferencesonlyemergeduringthetransitionfromadolescencetoyoungadulthood.B)Increasedagepredicts greaterpositivegainsinnetworkmodularityacrossblocksinthefirsthalfofthetask.
thetaskandwitholderindividuals,thereisnoindependenteffectof repeatedexposuretothesametaskenvironment(Table3).Asexpected, thismodelofferedimprovementsovertherandomeffectsANOVAmodel (ΔAIC=−476.3;𝜒𝑑𝑖𝑓𝑓 2 =486.3,df=5,p<.001).
3.3.2. Networkmodularityshowsinteractionsacrosslevelsoftime
Wenextfittheinteractivemodelof development tothenetwork modularitydata.Similartothemodelwithlearningperformance,there wasasignificantinteractionofwaveandthequadraticeffectofageon networkmodularity(𝛽 =0.057,SE=0.018,p=.001;Table4).When probed(Fig.6A),therewasasimilarcompensatorypatterntotheone seeninlearningperformance,suchthatexperienceacrosswaves(green andblue)predictedpositiveshiftsinmodularityatlateragescompared withthefirstexposuretothetask(red).Interestingly,thesedifferences appeartoonlyemergeduringthetransitionfromadolescencetoyoung adulthood,whereasincreasedexperiencedoesnotimpactmodularityat youngerages.Furthermore,therewasasignificantpositiveinteraction ofageandpracticeinthefirsthalfofthetask(𝛽 =0.068,SE=0.019,p <.001),suchthatolderindividualsshowedmorerapidgainsin modu-larityacrossthefirsthalfofthetask(Fig.6B).Thismodeloffered con-tinuedimprovementsoverthemaineffectsonlymodel(ΔAIC=−18.18;
𝜒2
𝑑𝑖𝑓𝑓 =34.18,df=8,p<.001).
3.3.3. Predictinglearningwithnetworkmodularity
Finally,wetestedwhethernetworkmodularitypredictslearning per-formanceaboveandbeyondtheeffectsoftime.Todoso,weentered net-workmodularityandinteractiontermsbetweenmodularityandthetime predictorsintothemodel.Inadditiontosimilareffectsofthetime pre-dictors,thismodelrevealedasignificantpositiveinteractionbetween networkmodularityandage(𝛽 =0.062,SE=0.023,p=.007;Table5). Probingthisinteraction(Fig.7),weplottedthemodel-implied trajec-toryoflearningrateacrossageforindividualswhoshowedrelatively low(−1SD),mean,andrelativelyhigh(+1SD)networkmodularity. Thisshowedthatindividualswhoevincedrelativelyhighnetwork mod-ularity(bluetrajectory)tendedtoshowhigherlearningrates,butonly afterthetransitionfromadolescencetoyoungadulthood(seeregion ofsignificance analysisinSupplementfordetails).Alikelihoodratio
testsuggeststhatincludingthebrainasapredictorincreasedmodelfit (ΔAIC=−4.1;𝜒𝑑𝑖𝑓𝑓 2 =20.10,df=8,p=.010).
4. Discussion
Thecontributionsofexperienceversusdevelopmentcanbedifficult toteaseapart,especiallywithinalongitudinalsample(JollesandCrone, 2012; Telzeret al.,2018) becauseage,experience, andpractice are confounded.Totestaninteractionmodelofexperienceand develop-ment,weutilizedanovellongitudinalapproachtoseparateout vari-ability in learningperformanceandbrainnetworkmodularityalong different timescales:1)within-sessionpracticeacrossblocksof learn-ing,2)within-person,across-waves,and3)acrossage.Briefly,wefound thatlearningperformancetendstoimprovethroughoutadolescenceand leveloff intoadulthood.Incontrast,networkmodularityappearstopeak aroundmiddleadolescenceandthenshowsdeclinesacrossyoung adult-hood.However,bothoftheseeffectsaremoderatedbytheamountof experienceindividualshaveaccumulatedacrosswaves,withmore ex-periencerelatingtohigherperformanceandnetworkmodularity dur-inglearning.Whenconsideringthebrain-behaviorrelationships, rela-tivelyhighermodularitypredictsbetterlearningperformancebutonly atolderages,suggestingthattheimportanceofindividualdifferences arounddevelopmentaltrendsfordeterminingbehavioraloutcomesmay dependontimingindevelopment.Wediscusseachofthesefindingsin greaterdetailbelow.
4.1. Multi-Levelchangesinlearningrateandbrainnetworks
Evenwithoutprobingcross-levelinteractions,amajoradvantageof themodelsemployedhereistheseparationofvarianceintheoutcome acrosslevelsoftime.Inthecurrentstudy,wefoundthatbehavioral im-provementsinlearningoccuracrossage,butdonotchange systemati-callyacrosswavesorblockswithinthetask.Incontrast,network mod-ularityshowedsystematicwithin-sessionincreasesaswellaspositive ageeffects,whileshowingnosystematicchangesacrosswaves.While these totaleffectsshouldbe interpretedwithcaution giventhe pres-enceofhigher-orderinteractioneffects,theyneverthelesshighlightan
Fig.7. DifferentialImpactsofNetwork Modu-larityonLearningRateByAge.Atearlierages, increasedmodularitydidnotpredictincreased learning,butthisrelationshipemerges follow-ingadolescence.
Table5
Modeloutputfrombrainaspredictoroflearningperformancemodel. Learning Rate
Predictors Estimates SE P-Value df
Intercept 0.935 0.005 < 0.001 794.679 First Half std − 0.003 0.021 0.886 4476.805 Second Half std 0.006 0.021 0.775 4482.941 Wave std − 0.056 0.067 0.401 4670.922 Age std 0.174 0.045 < 0.001 718.911 Age 2 std − 0.086 0.043 0.046 679.282
First Half ∗ Wave
std − 0.018 0.022 0.415 4481.722 Second Half ∗ Wave
std − 0.040 0.022 0.072 4485.072 First Half ∗ Age
std − 0.054 0.023 0.018 4472.145 Second Half ∗ Age
std 0.005 0.023 0.822 4502.070 Wave ∗ Age
std − 0.034 0.025 0.172 987.428
First Half ∗ Age 2std 0.044 0.025 0.073 4469.835
Second Half ∗ Age 2std − 0.013 0.025 0.604 4483.426
Wave ∗ Age 2std 0.076 0.022 0.001 4754.347
Modularity std − 0.014 0.029 0.627 4764.879 Modularity ∗ First Half
std 0.006 0.021 0.786 4532.995 Modularity ∗ Second Half
std 0.044 0.021 0.044 4545.020 Modularity ∗ Wave std 0.061 0.061 0.317 4774.206 Modularity ∗ Age std 0.062 0.023 0.007 4222.697 Modularity ∗ Age 2 std − 0.004 0.025 0.888 4389.408 Modularity ∗ Wave ∗ Age
std 0.035 0.021 0.093 4776.952 Modularity ∗ Wave ∗ Age 2
std − 0.043 0.021 0.041 4724.872 Random Effects 𝜎2 0.010 𝜏00ID 0.002 ICC 0.19 N ID 297 Observations 4799 Marginal R 2 / Conditional R 2 0.071 / 0.243
Note: Alleffectsroundedtothethirddecimalplacefordisplaypurposes. std=standardizedeffectsreported.𝜎2=level1randomeffect.𝜏
00=higher levelrandomeffect(effectspecifiedbysubscript).ICC=Intraclass correla-tion.N =numberofunitsateachlevel.
advantageofthegrowthmodelemployedhere(McCormick,preprint). Controllingfortheeffectsofrepeatedexposuretothetaskenvironment allowustobemoreconfidentthattheeffectsofagearedueto matura-tionalforcesratherthancomfortorfamiliaritywiththetaskorscanner environment(Bell,1953;JollesandCrone,2012).Eveniftheseeffects arenotsignificant,asisthecasehere,thesemodelsallowustocheck ourassumptionsabouttheprocessesunderlyinglongitudinalchange.
4.2. Cross-levelinteractionsbetweengrowthatdifferentscales
Moreinterestingly,thereweresignificantcross-levelinteractions be-tweentimepredictorsinthemodelsforbothbehaviorandbrain tra-jectories.Forbothoutcomesofinterest,resultssupportedtheideaof aninteractivemodelofdevelopment,wheretheimpactsofexperience changeacrossage.Forlearningrate,themodelrecapitulatedprevious findings in thistask (Peters andCrone,2017) showingthat learning ratereachespeaklevelsaroundlateadolescence,andisconsistentwith other findingsshowing improvements inlearningacross adolescence (vanDuijvenvoordeetal.,2008; Petersetal.,2016;McCormick and Telzer,2017a).This highlightsthatthemodelwithgrowthat multi-plelevelsaccommodatesthesameinferencesmadewiththeage-only models.Inthisway,adoptingthecurrentapproachoffersadvantages withoutlimitingtheinferencesmadeaboutdevelopmentaltrajectories. Indeed,theinteractionofwaveandthequadraticeffectofageon learn-ingperformancerevealsthatchangesinbehavioraredrivenbycomplex interplays betweendevelopmentandexperience,withexperience ap-pearingtoplaydifferentrolesinsupportingoverallperformanceacross adolescenceandyoungadulthood.Specifically,probingtheinteraction showsthatincreasedexperiencecompensatesforthemaineffectofage, boostingbehavioralperformancegainsfurtherintotoyoungadulthood. Extensive experience(blue trajectoryinFig.5)bluntsotherwise pre-dicteddeclinesandstabilizesperformanceatthepeakachievedduring middletolateadolescence.However,thisexperience-related improve-mentisnotuniversal acrossthedevelopmentalperiodconsidered,as mid-adolescentperformanceisheightenedevenatfirstwave.Thismight suggestthatduringadolescence,learningisuniversallyimproved,and thenindividualdifferencesbecomemorerelevantasindividuals tran-sition out ofthis period(see Pattwellet al.,2011 forasimilar idea in theareaofcontextualfearduringadolescence).Noneof the mod-els testedshowed within-sessioneffectson learningperformance, al-thoughwemayhavebeenunder-poweredtodetectsucheffectsgiven therestraintsof thetask (i.e.,therewasamaximumof12 trialsper blockregardlessoflearning).Overall,theseresultsareapowerful val-idation ofthemodel,convergingwithpreviousfindings onthesame datawhilesimultaneouslyallowingfortheadditionofmorecomplex time-dependentrelationships.
Findings withthebrainshowedadistinct patternof effectsfrom growthinlearningperformance.Previousworkinthisareahasshown developmental changes duringlearningin regional activation ofthe fronto-parietal (Peters et al., 2016) and striatal regions (Peters and
Crone,2017;McCormickandTelzer,2017a),aswellasseed-based con-nectivitywiththeorbitofrontalcortex(McCormickandTelzer,2017a). Inthe current study, we insteadexamine changes in brainnetwork organizationduringlearning.Learningovershortperiodsoftimehas beenshowntoalterneuralnetworkorganization(Bassettetal.,2011; Telesfordetal.,2017;Gerratyetal.,2018),buttheseprocesseshavenot beencomparedwithlong-termchangesduetomaturationpreviously. Inthecurrentstudy,weshowchangesinnetworkmodularitybothat theshort-termwithin-sessionandlong-termdevelopmentallevel.Inthe shortterm,playing repeatedblocks within-sessionis associatedwith heightenedmodularitybetween brainnetworksinvolvedin learning, andthiseffectacrossthefirsthalfofthetaskincreaseswithage.This suggeststhatolderindividualsareabletosegregate(i.e.,high within-networkandlowbetween-networkconnectivity)therelevantnetworks ofregionstoagreaterextentthanyoungerparticipants.Thequadratic totaleffectofagesuggeststhatratherthanalinearincreasein modu-larityacrossexperienceacrosswavesandmaturation,network modu-larityinsteaddecreasesatlaterages(Fig.6A),withlateadolescentsand youngadultsshowingdecreasinglymodularnetworkscomparedwith middle-adolescents(alsoconsistentwiththeage-onlymodel(Fig.4D). Thispatternofnetworkconnectivityevolvessimilarlyacross develop-menttostriatalactivationduringlearningversusapplication(Petersand Crone,2017).However,theinteractionof this ageeffect withwave showsananalogouseffectasseeninlearningperformance,where ex-perienceappearstobluntexpecteddecreasesinnetworkmodularityin youngadulthood.Whilethiscompensatoryeffectdoesnotcounteract theoveralldecreasesasseeninthebehavioralresults,itdoesserveto shiftnetworkorganizationtowardamoreadolescent-typicalphenotype. Forbothbehavioralperformanceandnetworkmodularity,thepattern ofresultsseenhereclearlysupportaninteractiveviewofexperience anddevelopment,wherematurationconstrainstheeffectofexperience ratherthansimplyservingasaformofextendedpractice.
Regardlessoftheeffectsofexperience(i.e.,wave),thereappearto bemoresubstantialdecreasesinmodularitycomparedwithbehavioral performance.Inthecontextoflearninganddevelopment,thispattern oflong-termchangeinneuralnetworksmighthavetwopotential ex-planations.First,itmightbethatbrainnetworksshowthegreatest ca-pacityfor modularityduring adolescence,andthis capacitysupports theheightenedflexiblelearning(JohnsonandWilbrecht,2011;Casey, 2015)andfeedbacksensitivity(Petersetal.,2016;vanDuijvenvoorde etal.,2014;McCormickandTelzer,2017a;b;2018b)thatcharacterize thisdevelopmentalperiodregardlessofpracticeorexposureeffects. Al-ternatively,thesechangesmightreflectdifferencesinhowthebrain per-formssimilaractionsacrossdevelopment.Forinstance,highmodularity mightbenecessaryforadolescentstoachievehighperformance,while adultsmightnotrequirethistothesamedegree.Forinstance, stabiliz-inglearningperformancewithexperienceisstillassociatedwithoverall (albeitblunted)decreasesinmodularityduringyoungadulthood.This mightalsobeconsistentwithanexpansion-normalizationtheoryof de-velopmentandlearning(Wengeretal.,2017)wherebrainsshowinitial changesinstructureandfunction(e.g.,increasedsynapticformation, in-creasedactivation)thatthenreturntobaselinewithoutcompromising behavioralperformance.Thesehypothesesarenotmutuallyexclusive, andtheeffectsprobedheremightsuggestboththatadolescentbrains supportimprovedperformanceregardlessofotherinfluences(e.g., ex-perience)andthatreturnstonetworkphenotypesseeninyounger ado-lescentsdoesnotleadtoacompletecollapseofbehavioralperformance.
4.3. Changingbrain-behaviorrelationships
Whilecharacterizingtrajectoriesofbrainandbehaviorseparatelyis informative,wealsoexamined whetherindividualdifferencesin net-workorganizationmightpredictlearningperformanceaboveand be-yonddevelopmental(PetersandCrone,2017)andexperientialeffects, aswellas whetherthat relationshipchangedacross time.Consistent withtheinteractionviewofdevelopmentandexperience,significant
in-teractionsbetweennetworkmodularityandtimepredictorsinthemodel revealedasignificantmoderationofbrain-behaviorrelationshipsacross age.Thismeansthatinearlyadolescence,thereisnoeffectofincreased modularityonlearningperformance,butthatincreasedmodularity pre-dictsenhancedlearningratesforolderindividuals.Thesefindingsare consistentwithpreviousworkinyoungadultsshowingpositive associa-tionsbetweennetworkmodularityandsuccessfullearning(Bassettetal., 2011;Ellefsenetal.,2015)andhigher-ordercognitiveprocessing gener-ally(Kitzbichleretal.,2011;Braunetal.,2015).However,thesepositive associationsatlateragesareparticularlyinterestinginthecontextof thedevelopmentaltrendsdetected,whereolderindividualsonaverage showdecreasedmodularityacrosswaves.Despitenormativedecreases innetworkmodularityduringlateadolescenceandyoungadulthood, individualswithhighermodularityshowgreaterlearningperformance. Thismightlendsupporttothefirstexplanationoftheinteraction ef-fectofageandwave– thatadolescent-typicalneuralphenotypesoffer advantagesinperformance(e.g.,JohnsonandWilbrecht,2011;Jones etal.,2014;vanvanDuijvenvoordeetal.,2016).Ifyoungadultssimply didnotneedhighlymodularnetworkstoperformwhatisarelatively easytaskforthem(i.e.,theexpansion-renormalizationhypothesis),then wewouldexpectthattherelationshipbetweenmodularityandlearning performancewoulddecrease.However,thepositiveinteractioneffect (Fig.7)suggeststhatolderindividualsshowanevengreaterdependence onhighmodularityforsuccessfullearningperformancecomparedwith youngerparticipants.Thispatternofresultspresentsacompellingcase foradolescent-specificadvantagesinlearning.Forolderindividuals, re-taining“immature,” butapparentlymore-optimal,network configura-tionshelpsboostbehavioralperformance.
4.4. Conclusions
Insummary, wetook anovelmodelingapproachtodisaggregate change in learning and neuralnetworks across different timescales. Whilewefocusonlearninghere,theseresultshighlightthepotential flexibilityofmixed-effectsmodelsforprobingcomplexdevelopmental trajectories in otherdomains.Wefoundnonlinearpatternsof devel-opment for both behaviorandbrain, which were moderatedby ex-perience. Specifically,greater experience withthetask across waves supportedincreasedlearningandnetworkmodularityrelativeto com-paratively naïvesubjectsat laterages,highlightingthattheseeffects canpotentiallybiasage-relatedinferencesunlessexplicitlyincludedin longitudinalmodels.Futureresearchusingacceleratedlongitudinal de-signs(orseeMcCormick,preprint foralternatives)shouldtakecareto model practice/exposure-relatedeffectstoremove this confound. Fi-nally,weshowedchangingbrain-behaviorrelationshipsacross adoles-cence, wherehigher networkmodularitypredicts increased learning performanceonlyfollowingthetransitionintoyoung-adulthood.These results presentcompellingsupport foraninteractiveview of experi-ence anddevelopment,wherechanges in thebrainimpact behavior in context-specific fashionbased ondevelopmentalgoals (Crone and Dahl,2012;Romeretal.,2017).
Credit authorship contribution statement
Ethan M. McCormick: Conceptualization, Methodology, Formal analysis, Writing - original draft, Visualization. Sabine Peters: In-vestigation, Data curation, Writing - review & editing, Supervision. Eveline A. Crone: Investigation,Datacuration,Writing-review& edit-ing, Supervision,Project administration,Funding acquisition. Eva H. Telzer: Conceptualization,Writing-review&editing,Supervision. Acknowledgments
AuthorContributions:S.P.andE.A.C.designedresearch,and per-formedresearch;E.M.M.,analyzeddata;E.M.M.,S.P.,E.A.C.&E.H.T. wrotethepaper.
Thisresearchwasfundedbyastartinggrantof theEuropean Re-searchCouncil(ERC-2010-StG-263234awardedtoE.A.C.)andagrant fromtheNetherlandsOrganizationforScientificResearch(NWO-VICI 453–14–001awardedtoE.A.C.).WewouldliketothankLauravander Aar,SibelAltikulaç,NeeltjeBlankenstein,BarbaraBraams,Suzannevan deGroep,JulietteCassé,DiannevanderHeide,JorienvanHoorn, Cé-dricKoolschijn,BabetteLangeveld,KyraLubbers,BatshevaMannheim, MaravanderMeulen,RosaMeuwese,SandyOvergaauw,JiskaPeper, Elisabeth Schreuders,Merel Schrijver,JochemSpaans, MarijeStolte, ErikdeWater,andBiancaWesthoff fortheirhelpwithdatacollection.. Finallywewouldliketothankallparticipantsandtheirparentsfortheir collaboration.
E.M.M.wassupportedinthisresearchbygrantsfromtheNational InstitutesofHealth(R01DA039923,R01EB022904awardedtoE.H.T.) andgenerousfundsfromtheUniversityofNorthCarolinaatChapelHill.
Theauthorsdeclarenocompetingfinancialinterests. Data and code availability
ProcessedbehavioralandfMRIdata,aswellasallcodeusedinthe analysessupportingtheprimaryfindingsofthisstudyareavailableat OpenScienceFramework;projecthttps://osf.io/62gwz/.Investigators interestedin obtainingrawbehavioralandfMRIdatashouldcontact E.A.C..
Supplementary materials
Supplementarymaterialassociatedwiththisarticlecanbefound,in theonlineversion,atdoi:10.1016/j.neuroimage.2021.117784. References
Aiken, L.S. , West, S.G. , 1991. Multiple Regression: Testing and Interpreting Interactions. Sage Publications, Inc .
Bassett, D.S. , Wymbsb, N.F. , Porterc, M.A. , Muchae, P.J. , Carlsona, J.M. , Graftonb, S.T. , 2011. Dynamic reconfiguration of human brain networks during learning. Proc. Natl. Acad. Sci. 108 (18), 7641–7646 .
Bassett, D.S. , Yang, M. , Wymbs, N.F. , Grafton, S.T. , 2015. Learning-induced autonomy of sensorimotor systems. Nat. Neurosci. 18 (5), 744–751 .
Bates, D. , Mächler, M. , Bolker, B. , Walker, S. , 2015. Fitting linear mixed-effects models using lme4. J. Stat. Softw. 67 (1), 1–48 .
Bell, R.Q. , 1953. Convergence: an accelerated longitudinal approach. Child Dev. 145–152 . Blakemore, S.J. , Mills, K.L. , 2014. Is adolescence a sensitive period for sociocultural pro-
cessing? Annu. Rev. Psychol. 65, 187–207 .
Braun, U. , Schäfer, A. , Walter, H. , Erk, S. , Romanczuk-Seiferth, N. , Haddad, L. , Meyer-Lin- denberg, A. , 2015. Dynamic reconfiguration of frontal brain networks during execu- tive cognition in humans. Proc. Natl. Acad. Sci. 112 (37), 11678–11683 .
Bullmore, E.T. , Bassett, D.S. , 2011. Brain graphs: graphical models of the human brain connectome. Annual Review of Clinical Psychology 7, 113–140 .
Casey, B.J. , 2015. Beyond simple models of self-control to circuit-based accounts of ado- lescent behavior. Annual Review of Psychology 66, 295–319 .
Casey, B.J. , Jones, R.M. , Hare, T.A. , 2008. The adolescent brain. Ann. N. Y. Acad. Sci. 1124 (1), 111–126 .
Casey, B.J. , Tottenham, N. , Liston, C. , Durston, S. , 2005. Imaging the developing brain: what have we learned about cognitive development? Trends Cognit. Sci. 9 (3), 104–110 .
Ciric, R. , Wolf, D.H. , Power, J.D. , Roalf, D.R. , Baum, G.L. , Ruparel, K. , Gur, R.C. , 2017. Benchmarking of participant-level confound regression strategies for the control of motion artifact in studies of functional connectivity. Neuroimage 154, 174–187 . Cohen, J.R. , D’Esposito, M. , 2016. The segregation and integration of distinct brain
networks and their relationship to cognition. Journal of Neuroscience 36 (48), 12083–12094 .
Crone, E.A. , Dahl, R.E. , 2012. Understanding adolescence as a period of social–affective engagement and goal flexibility. Nat. Rev. Neurosci. 13 (9), 636–650 .
Crone, E.A. , Steinbeis, N. , 2017. Neural perspectives on cognitive control development during childhood and adolescence. Trends in Cognitive Sciences 21 (3), 205–215 . Dale, A.M. , 1999. Optimal experimental design for event-related fMRI. Hum. Brain Mapp.
8 (2–3), 109–114 .
Daw, N.D. , O’doherty, J.P. , Dayan, P. , Seymour, B. , Dolan, R.J. , 2006. Cortical substrates for exploratory decisions in humans. Nature 441 (7095), 876–879 .
Daw, N.D. , Shohamy, D. , 2008. The cognitive neuroscience of motivation and learning. Social Cognition 26 (5), 593–620 .
Ellefsen, K.O. , Mouret, J.B. , Clune, J. , 2015. Neural modularity helps organisms evolve to learn new skills without forgetting old skills. PLoS Comput. Biol. 11 (4), e1004128 . Flora, D.B. , 2008. Specifying piecewise latent trajectory models for longitudinal data.
Struct. Equ. Model. 15 (3), 513–533 .
Galván, A. , 2010. Neural plasticity of development and learning. Hum. Brain Mapp. 31 (6), 879–890 .
Gerraty, R.T. , Davidow, J.Y. , Foerde, K. , Galvan, A. , Bassett, D.S. , Shohamy, D. , 2018. Dy- namic flexibility in striatal-cortical circuits supports reinforcement learning. J. Neu- rosci. 2017–2084 .
Gerraty, R.T. , Davidow, J.Y. , Wimmer, G.E. , Kahn, I. , Shohamy, D. , 2014. Transfer of learning relates to intrinsic connectivity between hippocampus, ventromedial pre- frontal cortex, and large-scale networks. J. Neurosci. 34 (34), 11297–11303 . Gonzalez-Castillo, J. , Hoy, C.W. , Handwerker, D.A. , Robinson, M.E. , Buchanan, L.C. ,
Saad, Z.S. , Bandettini, P.A. , 2015. Tracking ongoing cognition in individuals using brief, whole-brain functional connectivity patterns. Proc. Natl. Acad. Sci. 112 (28), 8762–8767 .
Johnson, C. , Wilbrecht, L. , 2011. Juvenile mice show greater flexibility in multiple choice reversal learning than adults. Dev. Cognit. Neurosci. 1 (4), 540–551 .
Jolles, D. , Crone, E.A. , 2012. Training the developing brain: a neurocognitive perspective. Frontiers in Human Neuroscience 6, 76 .
Jones, R.M. , Somerville, L.H. , Li, J. , Ruberry, E.J. , Powers, A. , Mehta, N. , ... , Casey, B.J. , 2014. Adolescent-specific patterns of behavior and neural activity during social rein- forcement learning. Cognitive, Affective, & Behavioral Neuroscience 14 (2), 683–697 . Kitzbichler, M.G. , Henson, R.N. , Smith, M.L. , Nathan, P.J. , Bullmore, E.T. , 2011. Cog- nitive effort drives workspace configuration of human brain functional networks. J. Neurosci. 31 (22), 8259–8270 .
Li, F. , Duncan, T.E. , Duncan, S.C. , Hops, H. , 2001. Piecewise growth mixture modeling of adolescent alcohol use data. Struct. Equ. Model. 8 (2), 175–204 .
Luna, B. , Padmanabhan, A. , O’Hearn, K. , 2010. What has fMRI told us about the develop- ment of cognitive control through adolescence? Brain Cognit. 72 (1), 101–113 . McCormick, E.M. , Telzer, E.H. , 2017a. Adaptive adolescent flexibility: neurodevelop-
mental of decision-making and learning in a risky context. J. Cognit. Neurosci. 29, 413–423 .
McCormick, E.M. , Telzer, E.H. , 2017b. Failure to retreat: blunted sensitivity to negative feedback supports risky behavior in adolescents. Neuroimage 147, 381–389 . McCormick, E.M. , Telzer, E.H. , 2018. Not doomed to repeat: enhanced medial prefrontal
cortex tracking of errors promotes adaptive behavior during adolescence. J. Cognit. Neurosci. 30 (3), 281–289 .
McCormick, E.M. , Gates, K.M. , Telzer, E.H. , 2019. Model-based network discovery of de- velopmental and performance-related differences during risky decision-making. Neu- roImage 188, 456–464 .
McCormick, E.M. ( preprint ). Multi-Level Multi-Growth Models: New Opportu- nities for Addressing Developmental Theory Using Longitudinal Designs. 10.1101/2020.10.21.349274
Mehta, P.D. , West, S.G. , 2000. Putting the individual back into individual growth curves. Psychol. Methods 5 (1), 23 .
Pattwell, S.S. , Bath, K.G. , Casey, B.J. , Ninan, I. , Lee, F.S. , 2011. Selective early-acquired fear memories undergo temporary suppression during adolescence. Proceedings of the National Academy of Sciences 108 (3), 1182–1187 .
Peters, S. , Crone, E.A. , 2017. Increased striatal activity in adolescence benefits learning. Nat. Commun. 8 (1), 1983 .
Peters, S. , Braams, B.R. , Raijmakers, M.E. , Koolschijn, P.C.M. , Crone, E.A , 2014. The neu- ral coding of feedback learning across child and adolescent development. J. Cognit. Neurosci. 26 (8), 1705–1720 .
Peters, S. , Van Duijvenvoorde, A.C. , Koolschijn, P.C.M. , Crone, E.A , 2016. Longitudi- nal development of frontoparietal activity during feedback learning: contributions of age, performance, working memory and cortical thickness. Dev. Cognit. Neurosci. 19, 211–222 .
Power, J.D. , Barnes, K.A. , Snyder, A.Z. , Schlaggar, B.L. , Petersen, S.E. , 2012. Spurious but systematic correlations in functional connectivity MRI networks arise from subject motion. NeuroImage 59 (3), 2142–2154 .
Romer, D. , Reyna, V.F. , Satterthwaite, T.D. , 2017. Beyond stereotypes of adolescent risk taking: Placing the adolescent brain in developmental context. Developmental Cogni- tive Neuroscience 27, 19–34 .
Rubinov, M. , Sporns, O. , 2011. Weight-conserving characterization of complex functional brain networks. NeuroImage 56 (4), 2068–2079 .
Sadaghiani, S. , D’Esposito, M. , 2015. Functional characterization of the cingulo-opercular network in the maintenance of tonic alertness. Cerebral Cortex 25 (9), 2763–2773 . Seitzman, B.A. , Gratton, C. , Marek, S. , Raut, R.V. , Dosenbach, N.U. , Schlaggar, B.L. , ... ,
Greene, D.J. , 2020. A set of functionally-defined brain regions with improved repre- sentation of the subcortex and cerebellum. NeuroImage 206, 116290 .
Shirer, W.R. , Ryali, S. , Rykhlevskaia, E. , Menon, V. , Greicius, M.D. , 2012. Decoding sub- ject-driven cognitive states with whole-brain connectivity patterns. Cereb. Cortex 22 (1), 158–165 .
Shulman, E.P. , Smith, A.R. , Silva, K. , Icenogle, G. , Duell, N. , Chein, J. , Steinberg, L. , 2016. The dual systems model: review, reappraisal, and reaffirmation. Dev. Cognit. Neu- rosci. 17, 103–117 .
Steinberg, L. , Albert, D. , Cauffman, E. , Banich, M. , Graham, S. , Woolard, J. , 2008. Age dif- ferences in sensation seeking and impulsivity as indexed by behavior and self-report: evidence for a dual systems model. Dev. Psychol. 44 (6), 1764 .
Telesford, Q.K. , Ashourvan, A. , Wymbs, N.F. , Grafton, S.T. , Vettel, J.M. , Bassett, D.S. , 2017. Cohesive network reconfiguration accompanies extended training. Hum. Brain Mapp. 38 (9), 4744–4759 .
Telzer, E.H. , McCormick, E.M. , Peters, S. , Cosme, D. , Pfeifer, J.H. , van Duijvenvo- orde, A.C. , 2018. Methodological considerations for developmental longitudinal fMRI research. Dev. Cognit. Neurosci. 33, 149–160 .
van Duijvenvoorde, A.C. , de Macks, Z.A.O. , Overgaauw, S. , Moor, B.G. , Dahl, R.E. , Crone, E.A. , 2014. A cross-sectional and longitudinal analysis of reward-related brain activation: effects of age, pubertal stage, and reward sensitivity. Brain and Cognition 89, 3–14 .
van Duijvenvoorde, A.C. , Peters, S. , Braams, B.R. , Crone, E.A. , 2016. What motivates ado- lescents? Neural responses to rewards and their influence on adolescents’ risk taking, learning, and cognitive control. Neuroscience & Biobehavioral Reviews 70, 135–147 . Van Duijvenvoorde, A.C. , Zanolie, K. , Rombouts, S.A. , Raijmakers, M.E. , Crone, E.A. , 2008. Evaluating the negative or valuing the positive? Neural mechanisms supporting feed- back-based learning across development. J. Neurosci. 28 (38), 9495–9503 .
Vigilant, L. , Roy, J. , Bradley, B.J. , Stoneking, C.J. , Robbins, M.M. , Stoinski, T.S. , 2015. Re- productive competition and inbreeding avoidance in a primate species with habitual female dispersal. Behav. Ecol. Sociobiol. 69 (7), 1163–1172 .
Wenger, E. , Brozzoli, C. , Lindenberger, U. , Lövdén, M. , 2017. Expansion and renormal- ization of human brain structure during skill acquisition. Trends Cognit. Sci. 21 (12), 930–939 .