• Aucun résultat trouvé

Dynamic functional brain networks underlying the temporal inertia of negative emotions

N/A
N/A
Protected

Academic year: 2022

Partager "Dynamic functional brain networks underlying the temporal inertia of negative emotions"

Copied!
13
0
0

Texte intégral

(1)

Article

Reference

Dynamic functional brain networks underlying the temporal inertia of negative emotions

GAVIRIA LOPEZ, Julian, et al.

Abstract

Affective inertia represents the lasting impact of transient emotions at one time point on affective state at a subsequent time point. Here we describe the neural underpinnings of inertia following negative emotions elicited by sad events in movies. Using a co-activation pattern analysis of dynamic functional connectivity, we examined the temporal expression and reciprocal interactions among brain-wide networks during movies and subsequent resting periods in twenty healthy subjects. Our findings revealed distinctive spatiotemporal expression of visual (VIS), default mode (DMN), central executive (CEN), and frontoparietal control (FPCN) networks both in negative movies and in rest periods following these movies. We also identified different reciprocal relationships among these networks, in transitions from movie to rest. While FPCN and DMN expression increased during and after negative movies, respectively, FPCN occurrences during the movie predicted lower DMN and higher CEN ex- pression during subsequent rest after neutral movies, but this relationship was reversed after the elicitation of negative emotions. Changes in [...]

GAVIRIA LOPEZ, Julian, et al . Dynamic functional brain networks underlying the temporal inertia of negative emotions. NeuroImage , 2021, vol. 240, p. 118377

PMID : 34256139

DOI : 10.1016/j.neuroimage.2021.118377

Available at:

http://archive-ouverte.unige.ch/unige:154624

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

(2)

ContentslistsavailableatScienceDirect

NeuroImage

journalhomepage:www.elsevier.com/locate/neuroimage

Dynamic functional brain networks underlying the temporal inertia of negative emotions

Julian Gaviria

a,b,

, Gwladys Rey

b

, Thomas Bolton

c,d

, Dimitri Van De Ville

c,e

, Patrik Vuilleumier

a,b

aLaboratory for Behavioral Neurology and Imaging of Cognition, Department of Fundamental Neurosciences, University of Geneva, Geneva, Switzerland

bSwiss Center for Affective Sciences, University of Geneva, Geneva, Switzerland

cMedical Image Processing Lab, Institute of Bioengineering/Center for Neuroprosthetics, École Polytechnique Fédérale de Lausanne (EPFL), Geneva, Switzerland

dLausanne University Hospital (CHUV), Lausanne, Switzerland

eDepartment of Radiology and Medical Informatics, University of Geneva, Geneva, Switzerland

a r t i c le i n f o

Keywords:

Brain networks Emotions

Dynamic functional connectivity (dFC) Task-rest interaction

Affective inertia

a b s t r a ct

Affectiveinertiarepresentsthelastingimpactoftransientemotionsatonetimepointonaffectivestateatasub- sequenttimepoint.Herewedescribetheneuralunderpinningsofinertiafollowingnegativeemotionselicited bysadeventsinmovies.Usingaco-activationpatternanalysisofdynamicfunctionalconnectivity,weexamined thetemporalexpressionandreciprocalinteractionsamongbrain-widenetworksduringmoviesandsubsequent restingperiodsintwentyhealthysubjects.Ourfindingsrevealeddistinctivespatiotemporalexpressionofvisual (VIS),defaultmode(DMN),centralexecutive(CEN),andfrontoparietalcontrol(FPCN)networksbothinnegative moviesandinrestperiodsfollowingthesemovies.Wealsoidentifieddifferentreciprocalrelationshipsamong thesenetworks,intransitionsfrommovietorest.WhileFPCNandDMNexpressionincreasedduringandafter negativemovies,respectively,FPCNoccurrencesduringthemoviepredictedlowerDMNandhigherCENex- pressionduringsubsequentrestafterneutralmovies,butthisrelationshipwasreversedaftertheelicitationof negativeemotions.ChangesinFPCNandDMNactivitycorrelatedwithmorenegativesubjectiveaffect.These findingsprovidenewinsightsintothetransientinteractionsofintrinsicbrainnetworksunderpinningtheinertia ofnegativeemotions.Morespecifically,theydescribeamajorroleofFPCNinemotionelicitationprocesses,with prolongedimpactonDMNactivityinsubsequentrest,presumablyinvolvedinemotionregulationandrestoration ofhomeostaticbalanceafternegativeevents.

1. Introduction

Abundantworkinneuroscienceshowsthatfunctionalbrainactivity andconnectivityisdynamicallyorganizedacrossseveraldistributednet- workswithsimultaneousongoingfluctuationsandselectivereciprocal interactions(e.g.,correlatedoranticorrelated)accordingtocurrentbe- havioraldemands(Foxetal.,2005).Thesespatiotemporalfluctuations inintrinsicfunctionalnetworks(IFNs)canbeobservedduringparticu- lartasksaswellasduringrest.AmongIFNs,thedefaultmodenetwork (DMN)encompassesasetofmidlinecorticalareas,includingprecuneus andmedialprefrontalcortices(MPFC) thatareusuallyactiveduring mind-wanderingin“task-off” conditionsatrest(Christoff etal.,2016).

However,thisnetworkmightalsoplayamorespecificandactiverole inintegrativefunctionsrelatedtohigher-levelaspectsofself-awareness andintrospection(Fairetal., 2008).TheDMNactivityis associated withself-reflectiveandmemory-relatedtaskconditions(Sprengetal., 2009)andtypicallyanticorrelateswithactivityofthefronto-parietal

Correspondingauthor.

controlnetwork(FPCN),whichisrecruitedbyexternallydirectedatten- tion(Dixonetal.,2018a).DMNactivityalsofluctuatesbetweenperiods ofpositiveandnegativecorrelationwiththecentralexecutivenetwork (CEN)andsaliencynetworks(SN),indicatingswitchesbetweendiffer- entcognitiveprocessingmodes(Dixonetal.,2017).Inaddition,thereis evidencethattheDMNexpressionismodulatedbyemotionalinforma- tionandcurrentmood(Gaviriaetal.,2021;Nummenmaaetal.,2012; SatputeandLindquist,2019).However,itsexactroleinaffectivepro- cessesisunclear.

Inthepresentstudy,weinvestigatedhowIFNsdynamicallyrecon- figureandreciprocallyinteractwitheachotherinresponsetonegative emotionalepisodes,bothduringemotionalstimulationitselfandduring therecoveryperiodfollowingsuchstimulation.Dynamicreciprocalin- teractionsamongIFNsdovetailswithageneraltheoreticalframework ofaffectivephenomena(Scherer,2009)accordingtowhichemotions emergefromacoordinatedrecruitmentofcomponentprocesses,each underpinnedbydistinctneuralsystems,whosesynchronizationistrig-

https://doi.org/10.1016/j.neuroimage.2021.118377. Received18May2021;Accepted7July2021 Availableonline10July2021.

1053-8119/© 2021TheAuthor(s).PublishedbyElsevierInc.ThisisanopenaccessarticleundertheCCBY-NC-NDlicense (http://creativecommons.org/licenses/by-nc-nd/4.0/)

(3)

geredbybehaviorallyrelevanteventsandpromotesadaptivechanges intheorganism(Leitãoetal.,2020;MeulemanandRudrauf,2018).Ev- idenceforaffectiveinfluencesonDMNactivityanditsinteractionwith othernetworks comesfrom several observations(Buckner andDiNi- cola,2019; Kaiseret al., 2015),buttheirfunctional significance re- mainsunresolved.Adown-regulationofwithin-DMNconnectivitywas reportedinhealthyparticipantsduringsadmoodinductionthroughself- generatedmemories(Harrisonetal.,2008)ormovies(Borchardtetal., 2018),whiledisturbancesofDMN connectivityareobservedinclini- calmooddisorderssuchasdepressionandbipolardisease(Reyetal., 2014;Zovettietal.,2020).DynamicshiftsinthebalancebetweenDMN andSNarealsothoughttomediateadaptiveresponsestoacutestres- sors,promotinghighervigilanceandfear(Zhangetal.,2019).Further- more,changes intheactivationandspatialconfigurationofDMN at restcan be modifiedbypreceding cognitive(Barneset al.,2009) or affective(Eryilmazetal.,2014,2011;Gaviriaetal.,2021)taskcondi- tions.Inparticular,exposuretoemotionalstimuliorrewardshasbeen foundtoinducelong-lastingchangesinbrainactivityandconnectiv- ity, overlappingwith DMN,SN, andFPCN,which persist evenafter terminationof theelicitingevent itself,e.g.,during subsequentrest- ingstate(Borchardtetal.,2018;Eryilmazetal.,2011;Harrisonetal., 2008;vanMarleetal.,2010).Suchcarry-overeffectsofemotionson neuralactivityatrestmightreflectspontaneousself-regulationmech- anismsrestoringahomeostaticbalanceinbrainstate(Eryilmazetal., 2011;Gaviriaetal.,2021),andisenhancedbyindividualanxietylevels (Pichonetal.,2015).Thisisconsistentwithbehavioralevidencefora temporal“inertia” ofaffectfollowingtransientemotions,exacerbated bymaladaptiveemotionregulation(KuppensandVerduyn,2017),and contributingtoruminativeprocessestriggeredbynegativeeventsinpsy- chopathologicalconditions(Apazoglouetal.,2019).Inaccordancewith thisidea,cognitivereappraisalofnegativeaudiovisualstimuli(com- paredtofreelywatching)mayproducenotonlyareducedneuralre- sponsetothesestimuliinlimbicareas(e.g.,amygdala),butalsoanin- creasedactivationofcorticalnetworksoverlappingwithDMNandFPCN duringbothstimuluspresentationandsubsequentrest(Lamkeetal., 2014).

However, previous studies on emotion-responsive networks and emotioncarry-overgenerally focusedon static connectivitypatterns, temporally summarized atthe whole-brainlevel, eitherduringemo- tionalstimulationorduringrestingconditionsfollowingemotionalstim- ulation(comparedtoneutral).Theydidnotcharacterizethedynamics ofchangesinspecificIFNsduringthesedifferentperiods;nordidthey examinehowthesechangesmayemergefrominteractionsbetweendif- ferentIFNssimultaneouslymodulatedbyemotionalstate.Hence,these studiesdonotofferanymechanisticinsightsintohowIFNsdynamically unfoldduringandafteremotions,andhowtheiracuteengagementbyan emotion-elicitingeventmayinfluencethesubsequentreturntoanormal restingbaselineafterterminationofthisevent(Hermansetal.,2014).

Todirectlyaddresstheseissues,wedesignedanovelemotionelicita- tionparadigmwhereparticipantsfreelyviewedsad(vs.neutral)movies followedbyarestingstateperiod.TouncoverIFNsdifferentiallyen- gagedduringemotionalmoviesandtheirsubsequentcarry-overatrest, aswell astheir reciprocalrelationships,we leveraged aninnovative methodologyallowingustotrackdynamicfunctionalconnectivity(dFC) withco-activationpattern(CAP) analysis(Liuetal.,2018b;Liuand Duyn,2013).Bydelineatingbrain-wideCAPsmodulatedbyemotionin movieandrestperiods,weaskedthefollowingquestions:(1)whatis thecarry-overimpactofnegativeeventsonthedynamicsandreciprocal interactionsofbrainnetworksandhowdotheyunfoldintimebeyond theseevents,i.e.,intheaftermathofemotionduringsubsequentrest(see Fig1A);(2)whetherchangesinbrainstatessubsequenttonegativeemo- tions,particularlyinDMNandinterconnectednetworks,areassociated withchangesinsubjectiveaffectivestate;and(3)howisthecarry-over effectemotiononrestingIFNsrelatedtotheoccurrenceofspecificac- tivitypatternsduringtheprecedingemotionalevents.Bycharacterizing thespatiotemporalfeaturesofIFNsduringandafternegativeaffectin-

duction,ourstudyprovidesnewinsightsonthebraindynamicsofemo- tionalresponsesandmayhelpidentifynovelneurobiologicalmarkers fornegativeaffectexperienceandregulation,possiblyalteredinclini- calpopulationswithanxietyandmooddisorders.

2. Materialsandmethods 2.1. Participants

Twenty right-handed, French-speaking female volunteers (mean age=23.2±4.3)werecontactedviapostersandonlineadvertisement.

They allprovidedwritteninformed consentaccordingtotheGeneva UniversityHospitalEthicsCommittee. Onlyfemale participantswere recruitedbecausepilottestingsuggestedstrongeremotionalinduction in women,comparedtomale,particularlywiththemovieclipsused here(seeSupplementarymethods).Controlledinclusioncriteriawere:

nohistoryofneurologicalandpsychiatricdiseases;knownmenstrual phaseandnocontraceptivemethodtoruleouthormonaleffectsonemo- tionalprocessing(Protopopescuetal.,2005)orfunctionalconnectivity (Petersenetal.,2014;Pletzeretal.,2016);nomajorheadmovement duringthescanningsessions(<1.5mminallaxes).Scanningwassched- uledforeachparticipantintheearlystageofthemenstrualfollicular phase,whenthelevelsofestradiolandprogesteronehormonesaremod- erate(Andreanoetal.,2018;Endicott,1993).Nicotineandcaffeinecon- sumptionwasprohibited10hoursbeforescanning.

2.2. Behavioralparadigm

Initially, theparticipants filledthe French versionof thePANAS questionnaires. Thiswas followedbyanfMRI experimentwithare- peatedmeasuredesign.Thescanningsessioncomprisedtwoexperimen- tal“contexts”.Onecontextincludedanaudio-visualclip(5min)with sademotionalcontent(“negativemovie” condition),followedbyarest- ingperiod(5min)(“negativerest” condition).Asacontrolforaffec- tivevalence,thesecondexperimentalcontextcomprisednon-emotional videoclip(5min)(“neutralmovie”),followedbyanotherrestingpe- riod(“neutral” restcondition)(5min).Participantswereinstructedto keeptheireyesopenedandtostayawakeacross therestingperiods.

Thiswascontrolledbyaneye-trackingdevicebothonlineandoffline.

Importantly,thepresentationorderofthetwocontextswascounterbal- ancedacrossparticipants(seeFig.1A, forschematicoverviewof the paradigm design).At theend of each experimentalscanningsession (i.e.,“neutral” and“negative”),furtherPANASscoresandemotionques- tionnairesaboutthemovieswerealsofilled.Approximately20minutes elapsedbetweentheendofoneexperimentalcontextandthestartingof thenextone.Additionalinformationonmoviesandsubjectiveratings isprovidedintheSImaterial.

2.3. MRIdataacquisition

Anatomicalandfunctionalwhole-brainMRIdatawereacquiredwith a3Tscanner(SiemensTIMTrio)attheBrain&BehaviorLaboratory, UniversityofGeneva.Amultiband-sequencewasimplemented,witha voxelsizeof2mmisometricandaTRof1.3s(seeSIforfurtherdescrip- tionofbothdataacquisitionandpreprocessing).

2.4. Braindataanalysis

Thepresentstudyfocusedonthedynamicexpressionandinterac- tionof IFNsover timeboth duringstimulationwithnegativemovies andatrestfollowingstimulation(i.e.,emotionalinertia).Tothisaim, ourapproachcomprisedtwosuccessivestageswithdifferentgoals:(1) PreliminaryGLM-basedidentificationofbrainregionsofinterest(ROIs) withfunctionallyrelevantactivityduringourparadigmtobeusedas seeds;(2)In-depthCAPanalysisofthetime-varyingdynamicfunctional connectivityoftheseseedstocharacterizebrainnetworksdifferentially

(4)

Fig.1. Behavioralparadigm.A)Paradigmde- signillustratingthesequenceofexperimental events.Negatively-valencedorneutralmovies werefollowed bya resting period.Thepre- sentationof the two affectivecontexts (i.e.,

“neutral” and“negative”)wascounterbalanced acrossparticipants.B)Subjectiveaffectiverat- ingsofmoviesobtainedaftereachexperimen- talcontext.Valencewasratedfromverynega- tiveonthelowestscalevalues(=0)toverypos- itiveonthehighestscalevalue(=9).Whiskers standforstandarderrorsofmeans.PFDRadjust- mentformultiplecomparisons(p<.01).

engagedbyexperimentalconditions,aswellastheirinteractionsattwo differenttimepoints(i.e.,movieandrest).Thisapproachentailsatwo- stepdata-drivenapproach thatintegratestwodifferent andindepen- dentlevelsofinterrogationofbrainactivityinourtask-restparadigm (Shine andPoldrack,2018).Namely,anassessmentof functionalac- tivationlevel(GLM)alongamagnitudescale,anddynamicfunctional connectivitylevel(CAPs)alongatemporaldomain.

2.5. Functionalseeddefinitionbyunivariateanalysisofbrainactivity

WefirstperformedaGLManalysistodefineROIsasseedsforthe mainCAPanalysis.Changesinglobalneuralactivitywereassessedbe- tweensadandneutralconditionsduringmovies(“negativemovie” vs.

“neutralmovie” conditions)andduringrest(“negativerest” vs.“neutral rest” conditions)usingastandardprocedureforblockdesigninSPM12 (Frackowiaketal.,1997).Inadditiontoremovingsignalsatverylow frequencies,temporalautocorrelationacrossthefunctionaldatawasac- countedfor byusingthe“FAST” algorithmin SPM12(Corbinet al., 2018;Olszowyetal.,2019).FourGLMboxcarregressorsrepresented eachexperimentalcondition(i.e.,neutralmovie,neutralrest,negative movie,andnegativerestrespectively).Theseregressorswereconvolved withastandardhemodynamicresponsefunction(HRF)accordingtoa blockeddesign, whichwasthensubmittedtoaunivariateregression analysis.Realignmentparameterswereaddedtothedesignmatricesof bothmodels,toaccountforanyresidualmovementconfounds.Inall cases,thedesignmatrixincludedlow-frequencydriftswithacutoff fre- quencyat1/128Hz(othercutoff frequenciesdidnotchangetheGLM results).Flexiblefactorialanalysesofvariance(ANOVAs)werethenper- formedonthemaincontrastsofinterest(i.e.,“negativemovie>neutral movie”,and“negativerest>neutralrest”),todefinetwodifferentseed ROIsforourmainconnectivityanalyses.Onewithstimulus-relatedac- tivityandanotherwithrest-relatedactivity.Allstatisticalanalyseswere carriedoutatthewhole-brainlevel,withathresholdatPFWE<.05,cor- rectedatwholebrainlevel(unlessspecifiedotherwise).

2.6. Dynamicfunctionalconnectivityanalysis(CAPsidentification)

Toexaminechangesin IFNsasafunctionofexperimental condi- tions(neutralmovie,neutralrest,negativemovie,andnegativerest),we implementedaco-activationpattern(CAP)methodology(Changetal.,

2016;LiuandDuyn,2013;Pretietal.,2017).Ithasbeenpreviously shown (Liuetal.,2013;LiuandDuyn,2013) howtransientassocia- tionsbetweenfMRIsignalscapturedbyspecificco-activationpatterns (CAPs),correspondtotheintrinsicneuralarchitectureconfiguredfor supportingspecializedfunctions(IFNs).Remarkably,theCAPapproach providesameasureofdynamicfunctionalconnectivitywitharelevant seedbydecomposingfluctuationsofthefMRIBOLDsignalsintomulti- plespatialpatternsthatreflecttheinstantaneousorganizationandreor- ganizationofnetworksco-activatedwiththeseedovertime(Liuetal., 2018b;Pretietal.,2017).Thetwoemotion-responsiveROIsidentified byourGLManalysis(insulaandprecuneusfrom“negativemovie” and

“negativerest” conditions,respectively)wereusedasseedsfortheCAPs generation.ByusingseedsdefinedbyanindependentGLM-basedanal- ysis,wecouldensureto“anchor” theobservedCAPstofunctionallyrel- evantnetworksinthecurrentexperimentalparadigm,ratherthanhigh- lightotherlessspecificIFNs.Unlikestaticconnectivityanalysisbased oncorrelationsbyaveragingoverlongperiods,thedFCapproachwith CAPsaccountsfortheBOLDtemporalvariabilitybyrepresentinginstan- taneousbrainconfigurationsatsingletimepoints(fMRIvolumes)and thensortingtheminafewdominantpatternsthroughclusteringanaly- sis(Liuetal.,2018b).CAPswerecomputedwiththe“tbCAPs” toolbox (Boltonetal., 2020),followingastepwisepipelineillustratedin Fig.

S1Amoredetaileddescriptionofourmethodologicalprocedureispro- videdelsewhere(Boltonetal.,2020).Importantly,thisapproachyields atemporalmetrictoquantifydFCvariabilityovertimebycomputing theoccurrencesofeachCAPineachcondition.Occurrencesaredefined asthesumofframes(time-points)assignedtoagivenCAPamongall theretainedframes,acrosstheentirescanningduration.

FollowingthegenerationofCAPsforeachseed,wecomparedtheir expressionacrosstheaffectivecontextswithgeneralizedlinearmixed models[gLMM(Brooksetal., 2017)]usingtheRsoftware.Weper- formeda2×2gLMM-basedfactorialanalysiscomprisingtwofixedfac- torsdenotedbythefollowingR-basedformulasyntax:

occurrences∼ context∗period+(1|subject),

wherethefixedfactor“context” comprisedtwolevelscorrespondingto theaffectivevalenceofmovies(i.e.,negativeorneutral)andthesec- ondfactor“period” representedthetypeofcondition(movieorrest).

Individualparticipants(subject)weremodeledasarandomfactor,and

“occurrences” wasthedependentvariable,computedforeachexperi-

(5)

mentalconditionandeachCAP.This2×2analysisallowedustoevalu- ateseparatelythemaineffectsofstimulusexposure(movievs.rest)and emotionalvalence(negativevs.neutral),aswellastheirinteractions, foreach of theCAPs.Subsequently,only thesignificantCAPs show- ingeithermaineffectsorinteractionsofthefactors“context” (neutral, negative)andperiod(movie,rest),wereconsideredasexperimentally relevantandselectedformoredetailedevaluation(seenextsection).

2.7. AssociationofaffectivescoreswithCAPs

Wealsotestedwhetherdifferencesinthetemporalmetricsofthese relevantCAPswereassociatedwithbehavioralmeasuresofsubjective emotionalstate(PANAS).Indoingso,weverifiedthatallthePANAS datawasnormallydistributed,exceptforthepositiveaffectivescores (PA)following the neutralcontext[Shapiro-Wilk test (p=.03)].This wasalso thecase forall metricfromthe differentCAPs in bothex- perimentalcontexts,exceptfortheprec-VISin“neutralrest” (p=.04);

andprec-CENin“neutralmovie” (p=.02).Inordertoproperlymodel thesedata,weimplementedourregressionanalyseswithgeneralizedes- timatedequations(GEEs),astatisticalframeworkforflexiblymodelling non-independentandcorrelateddata(i.e.,repeatedmeasures)withboth normalandnon-normaldistributions(PekárandBrabec,2018).Inthese analyses,theCAPsoccurrenceswereintroducedasmultipleoutcomes, andaffectiveindices(PANASscores)werethepredictorvariables.Pval- ueswereadjustedwithfalsediscoveryrate(FDR)formultipletesting (BenjaminiandHochberg,1995),whichisalsooptimaltocontrolfor dependency(BenjaminiandYekutieli,2001).

2.8. CAPsinteractionanalysis

2.8.1. Context-dependentwithin-CAPsinteractions

Toaddressthe maingoal of our studyconcerning functional re- lationships between networks in different affective states, we exam- inedhowchangesintheexpressionofagivennetworkcorrelatedwith changesinothernetworks acrosssuccessiveperiods(e.g.,moviesvs.

post-movie rest). First, we computed a set of within-CAP pair-wise Pearsoncoefficients (r)to assessthe occurrence of each affectively- relevantCAPacrossdifferentconditions.Thisanalysisthusprobedfor context-dependent“movie-rest” transitions whereoccurrencesduring the“movie” periodwerecorrelatedwithoccurrencesduringthesub- sequent“rest” periodof thesameCAPin thesameaffectivevalence [e.g., r(CAP(A)neutralmovie, CAP(A)neutralrest) vs. r(CAP(A)negativemovie, CAP(A)negativerest)]. As a control comparison, these within-CAP re- lationships were compared to movie-rest correlations between dif- ferent contexts, when movies and rest conditions did not follow in direct succession [e.g., r(CAP(A)negativemovie, CAP(A)neutralrest) vs.

r(CAP(A)neutralmovie, CAP(A)negativerest). Fig.4]. Because only report- ingdifferences insignificance fordifferent conditions[e.g., onecor- relationsignificant(p<.05) fortransitionsin onecontextbutnotthe other (p>.05)] would lead to statistical fallacy (Nieuwenhuis et al., 2011),wecould thus directlyassessdifferences in themagnitude of context/valence-specific correlations relative to non-specific control correlationsbycomparingtheconfidenceintervals(CI)oftheirPearson coefficients,whichallowsconsideringboththemagnitudeandpreci- sionoftheestimatedeffects(Zou,2007).Allwithin-CAPcomparisons wereimplementedwiththeR-basedCocorpackage(Diedenhofenand Musch,2015).Dependencyofwithin-CAPcorrelationswereaccounted followingZuoetal.,(Zou,2007),whentheircomparisonsincludedover- lappingvariables[e.g.,r(CAP-Aneutralmovie,CAP-Aneutralrest)vs.r(CAP- Aneutralmovie,CAP-Anegativerest).Fig.4].

2.8.2. Context-dependentbetween-CAPsinteractions

Asecond setof analysesconcerned thebetween-CAPs of interest associations,andallowedustoexaminetheirreciprocalrelationships as a function of emotional context. The correlation between occur- rences of differentCAPs wasagain computedcontext-wise, toprobe

howexpressionofoneCAPduringthemoviewatchingperiodwasas- sociatedtotheexpressionofotherCAPsinthejustfollowingresting stateperiod[e.g.,r(CAP(A)neutral,CAP(B)postneutral)].Thecorrelations foreachpairofCAPswerethencomparedbetweenthetwoaffective contexts [e.g., r(CAP(A)neutral, CAP(B)postneutral) vs. r(CAP(A)negative, CAP(B)postnegative).Fig.5B].Dependencyofnon-overlappingvariables was accounted for these between-CAPs comparisons as appropriate (Zou,2007).Further descriptionofstatisticaltests includingR-based mathematicalformulaeforbothwithin-CAPsandbetween-CAPscom- parisonsaredescribed in“SI.TestsforcomparingCAPscorrelations” section.

3. Results

3.1. Behavioralindicesofemotionalinductionbymovies

Wefirstverifiedthatmovieswithnegativecontentinduceddistinc- tivepatternsinbothbehaviorandbrainmeasures,comparedtoneutral movies.Affectiveratingsofmovieclipsconfirmedareliabledifference inemotionalexperiencewithmorenegativevalenceandhigherarousal elicitedbythenegativecomparedtoneutralmovies(Fig.1B).PANAS scores assessingsubjective affect also differed between thepre- and post-contextmeasuresasafunctionofemotionalcondition,withsignif- icantincreasesinthenegativeaffect(NA)scoresfollowingnegatively- valencedmovies (Fig.3A).Furtherbehavioralresults concerning the moviesaredescribedin “SI.Psychologicalindicesof emotionelicita- tion” section.

3.2. PreliminaryGLM-baseddefinitionofregionsofinterest

PriortotheCAPanalysis,wedefinedfunctionallyresponsiveseeds withstimulus-relatedandrest-relatedactivitybycomputinglinearcon- trastsbetweenexperimentalblocks.Movieswithnegativecomparedto neutralvalenceproducedselectiveincreasesinseverallimbicregions (PFWE<.05,wholebrain-correctedatclusterlevel;clustersizeestimated at Puncorrected<.001) predominantlyin theleft insulacortex (includ- ingbothposteriorandanteriorparts,withpeakMNI:-40,4,10;121 voxels),togetherwiththebilateralanteriorcingulatecortex,bilateral middlefrontalgyrus,andrighttemporalpole.Astheinsulahasbeen consistentlyassociatedtoemotionalprocessingacrossmanyparadigms (Benellietal., 2012;CritchleyandGarfinkel,2017;Liuetal.,2012; Paulusetal.,2005;Somervilleetal.,2013),mostofteninrelationto negative valence (Knutson et al., 2014),this region was selected as ourfirstROI (stimulus-related).Conversely, “negativerest” vs.“neu- tral rest” activated widespread areas associated withthe DMN,pre- dominantlyinbilateralprecuneus(peakMNI:4,-54,42;541voxels.

PFWE<.05,wholebrain-correctedatclusterlevel;clustersizeestimated atPuncorrected<.001),andtoalesserdegreeinoccipitalcortexandleft medialsuperiorfrontalgyrus.Thisaccordswithpreviousworkreport- ingactivationoftheprecuneusandotherDMNregionsatrestandtheir modulationbyemotionalorcognitivefactors(Foxetal.,2018;Hoetal., 2015;Northoff,2016;Sprengetal.,2010).Wethereforeselectedthe precuneusasthesecondseed(rest-related)foroursubsequentconnec- tivityanalysis.Notably,ourtwoseedROIslargelyoverlappedwiththe insula(87%ofvoxels)andprecuneus(78%ofvoxels)fromafunctional brainparcellationatlas(Schaeferetal.,2018),andconvergedwithinde- pendentmeta-analyticdatashowingconsistentactivationoftheseareas inemotionandrestconditions,respectively.

3.3. CAPsexpressedacrossstimulationperiodsandaffectivecontexts

TwoCAPanalyseswereconductedusingtheseedROIsidentifiedby ourpreliminaryGLManalysis(see“Methods” sectionandFig.S1A),one reflectingfunctionalcouplingwiththeleftinsulaandanotherreflecting couplingwiththeprecuneus.Intotal14CAPswereidentified,basedon adata-driven“consensus” procedure(Montietal.,2003)indicatingK=7

(6)

Table1

ResultsofGLMMsanalysesontemporaloccurrencesofrelevant/emotion-sensitiveCAPs.Statisticalresultsofa2×2factorial assessmentofthe“context” (neutralvs.negative),and“period” (moviesvs.rest)effectsonCAPsoccurrences.Significantposthoc pairwisecontrastsdrivingmaineffectsandinteractionsarelistedintheright-handcolumns.,thisfactorialanalysisrevealedthat 4outofthe14brainnetworksidentifiedinourCAPanalysis(seefulldescriptioninFig.S2)exhibitedadistinctiveactivityprofile accordingtoourexperimentalconditions.ThesefourrelevantCAPswerethereforeselectedforallmainanalysesinourstudy.

PFDR(falsediscoveryrate)adjustmentformultiplecomparisons(p<.05;p<.01;p<.001).

GLMMs FACTORIAL ANOVA LS MEANS

CAP

CONTEXT neu.-neg.

PERIOD movie-rest

INTERACTION

event ×context CONTRASTS P P FDR

insuFPCN F = (1,54) = 15.8 P = .0002

F = (1,54) = 46.6

P < .0001

F = (1,54) = 4.5 P = .03

neg. movie > neu. movie < .0001 .0003 neg. movie > neu. rest < .0001 < .0001 neg. movie > neg. rest < .0001 < .0001 precVIS F = (1,54) = 7.9

P = .006

F = (1,54) = 4.4 P = .03

F = (1,54) = 2.1 P = .15

neg. movie > neu. movie .003 .01 neg. movie > neu. rest .0009 .005 neg. movie > neg. rest .009 .05 precDMN F = (1,54) = 4.9

P = .03

F = (1,54) = 2.2 P = .13

F = (1,54) = 0.1 P = .60

neg. rest > neu. movie .0008 .001 neg. rest > neu. rest .01 .03 neg. rest > neg. movie .02 .05 precCEN F = (1,54) = 4.5

P = .03

F = (1,54) = 7.0 P = .01

F = (1,54) = 10.2 P = .002

neu. movie > neu. rest .0001 .0006 neu. movie > neg. movie .0003 .001 neu. movie > neg. rest .001 .002

foreachseedastheoptimalnumberofdistinctbrainmapsintermsof replicabilityforourdataset(seepipelineinFig.S1).Thespatialconfig- urationoftheseCAPsisdepictedin“SI.Fig.S2.2” andgloballyaccords withpreviousworkonIFNs(Christoff et al.,2016;Coleetal.,2014; Goldman-Rakic,1995;Krageletal.,2019;Uddinetal.,2019).

3.4. ChangesintemporaldynamicsofCAPsacrossconditions

Critically,we thencomparedthe temporaloccurrencesofall the 14 CAPsbetween experimental conditions. Usinga 2(affectivecon- text)x2(stimulationperiod)factorialanalysisofCAPoccurrences,we foundthatonly4outofthese14networksweredifferentiallyexpressed acrosstheaffectivecontexts(i.e.,negativevs.neutral.Table1),includ- ingtheinsula-connectedCAP5(insuFPCN)andtheprecuneus-connected CAP1,CAP3,andCAP5[precVIS,precDMN,andprecCEN,respectively (Fig.2)].Peakcoordinatesforthesefournetworksarelistedin“SI.Table S1”.BelowwedescribeeachoftheseCAPsinturn.

Amongnetworksco-activatedwiththeinsula,insuCAP5wastheonly oneshowingasignificantmodulationbyaffectivevalence.Itcomprised alargesetoffrontoparietalregionsassociatedwithcognitivecontroland saliencedetection(Coleetal.,2014;Dixonetal.,2018a),and,therefore, consideredtooverlapwiththeFPCNreportedinotherstudies(Fig.2A).

ThisCAPexhibitedaninteractionbetween“affectivecontext” and“stim- ulationperiod” [gLMMinteraction:F(54)=4.5;P.03],indicatingmore frequentoccurrencesofthisnetworkduringnegativemoviescompared tootherconditions,inadditiontomaineffectsofbothperiod[gLMM maineffect“period”:F(54)= 46.6;P.<0001. 𝛽movie=19.30,95%CI (16.32;22.30)],andcontext[gLMMmaineffect“context”:F(54)=15.8;

P.0002.𝛽negative=16.83,95%CI(13.86;19.80).Remarkably,48%of theoccurrencesofthisinsuFPCNCAPacrossalltheconditionswereseen inthe“negativemovie” condition[incontrastto24%during“neutral movie” (seepost-hocresultsintable1)].

Theprecuneus-based precCAP1overlapped with a classic“visual network” centeredon occipitalandposteriorparietalareas(precVIS, Fig.2B)(Göttlichetal., 2017; Kang etal., 2016; Liuet al., 2018a; O’Connor et al., 2002; Riedel et al., 2018; Saarimäki et al., 2018; Sripada etal., 2014).It showeda distinctiveincrease of occurrence rates during the “negative movie” condition, with main effects of both“affectivecontext” [gLMMmaineffect“context”:F(54)=7.9;P

=.006.𝛽negative=18.40,95%CI (15.94;20.90)]and“stimulation pe- riod” [gLMMmaineffect“period”:F(54)=4.4;P=.03.𝛽movie=17.79, 95%CI(15.33;20.30)],reflectinggenerallyhigheroccurrencesduring moviesthanrestandduringnegativethanneutralconditions,respec-

tively.However,therewasnosignificant“period” x“context” interac- tion[gLMMinteraction:F(54)=2.1;P=.15]despiteitspredominance duringthenegativemovies,inwhichtheprecVISexhibited35%ofto- taloccurrencesacrossconditions[vs.21%duringneutralmovies(see post-hocresultsinTable1)].

Two further emotion-sensitive CAPs were also derived from the precuneus-basedanalysis.Mostnotably,theprecCAP3showedavery similarspatialconfigurationtoDMN(Christoff etal.,2016),encompass- ingmedialprefrontal,posteriorcingulate,andinferiorparietalcortices (Fig.2C).Thefactorialanalysisindicatedamaineffectof“affectivecon- text” [gLMMmaineffect“context”:F(54)=4.9;P=.03;𝛽negative=16.7, 95%CI(13.85;19.60)],buttherewasnosignificanteffectofstimulation period[gLMMmaineffect“period”:F(54)=2.2;P=.13;𝛽rest=16.74, 95%CI(13.18;19.01)],norany“context” x“period” interaction[gLMM interaction:F(54)=0.27;P=.60].However,thisprecDMNCAPdomi- natedinthe“negativerest” condition”,whichrepresented32%ofits occurrencesacrossconditions[vs.22%duringneutralrest(Table1)].

Incontrast,theprecCAP5involveddorso-lateralfronto-parietalareas (Fig. 2D) commonly associated with the central executive network (CEN)(Goldman-Rakic,1995;MenonandUddin,2010).Thisnetwork alsoshowedasignificantinteractionbetween“affectivecontext” and

“stimulationperiod” [gLMMinteraction:F(54)=10.2;P=.002],elicited byacombinationofincreasedoccurrencesinthe“neutral” conditions [gLMMmaineffect“context”:F(54)=4.5;P=.03.𝛽neutral=15.58,95%CI (12.45;18.70)]andinthe“movie” conditions[gLMMmaineffect“pe- riod”:F(54)=7.0;P=.01;𝛽movie=16.11,95%CI(12.98;19.20)].Thus, unliketheprecDMNCAP3,theprecCENCAP5showedthehighestnumber ofoccurrencesduringthe“neutralmovie” condition[39%ofitstotal occurrencevs.21%duringnegativemovies(seestatisticalcontrastsin Table1)].

Altogether, these dataconvergewithprevious studiestoindicate that emotionaleventsproducedistinctiveandcoordinatedeffectson specificIFNs,especiallytheDMNandhigherregulatorysystemsasso- ciatedwithFPCNandCEN,withsignificantcarry-overeffectslingering beyondemotionaleventsthemselvesandextendingintosubsequentrest periods.

3.5. RelationshipofbrainCAPswithbehavioralmeasuresofaffect

Toprobeforthebehavioralsignificanceofchangesobservedinbrain networksacrossconditions,weexaminedwhethertheindividualoccur- rences of eachrelevant CAPcould predictsubjectiveaffective rating scores(fromPANAS)thatwereobtainedattheend ofscanningruns

(7)

Fig. 2. Spatiotemporal CAPs modulated by affective context. Functionalbrain networks were identified using co-activation pattern analysis(CAP)basedontime-dependentcou- pling with the insula [A. insu-FPCN (fron- toparietalcontrolnetwork)]orwiththepre- cuneus[B.prec-VIS(visualnetwork),C.prec- DMN (default mode network), D. prec-CEN (centralexecutivenetwork)].Onlythesefour CAPsshowedasignifcantmodulationoftheir occurrencerateasafunctionofexperimental condition(see suppl.Fig.S2).Thebrainto- pographymapsofeachCAPisillustratedwith athresholdofz>2.58[equivalenttop<.01 (Liuetal.,2018)].Furtheranatomicalinforma- tiononpeakregionsisprovidedinTables1. Temporaloccurrenceratesareplottedtoshow theirexpression(%)acrossexperimentalcon- ditions.StatisticalassesmentsofCAPsvariance andinteractionsarereportedinTable1.,and Fig.S2.PFDRadjustmentformultiplecompar- isons(p<.05;p<.01;p<.001).

Fig.3. PANAS scoresandtheirassociations withCAPsacrossconditions.A)Meanscores onPANAS self-reports obtainedbefore (pre) andafter(post)eachoftheexperimentalcon- texts(neutralandnegative).B)Beta(𝛽)values of GEE-based associations between affective scores[positive(PA)andnegative(NA)]and occurrenceratesoftheCAPsofinterest.InsuF- PCNoccurrencesobservedduringthe“negative movie”,aswellasinsuFPCNandprecDMNoc- currencesduringthe“negativerest” conditions weresignificantlycorrelatedwiththenegative affectscores(NA)measuredattheend(post)of thenegativecontext.PFDRadjustmentformul- tiplecomparisons(p<.05;p<.01).

ineachemotionalcontext.ResultsfromourGEEanalysisshowedthat morenegativeaffect (NA)scorespost-scanning wereassociatedwith higheroccurrencesofinsuFPCNCAPduringthe“negativemovie” pe- riod(𝛽=.24,95%CI(0.15;0.33);PFDR<01;Fig.3B.Up-right),and inverselytoloweroccurrencesofinsuFPCNduringthe“negativerest” period(𝛽 =-.32, 95%CI(-0.30; 0.13); PFDR < 05,Fig.3B.Bottom- right). Ontheother hand,occurrences of theprecDMN CAP during thesame“negativerest” conditionalsoshowedapositivecorrelation withthepost-scanningNAscores(𝛽=.20,95%CI(0.20;0.46),PFDR<

01).Fig.3B.Bottom-right).Thus,higherNAscoresfollowingexposure tothe“negativecontext” werelinkednotonlytogreateroccurrences ofinsuFPCNduringnegativemovies,butalsotolesseroccurrencesof insuFPCN accompaniedwithgreater occurrencesofprecDMN during thesubsequent“negativerest” period.Bycontrast,positiveaffect(PA) scoreswerenotrelatedtotheexpressionofanyofthesenetworks(see Fig.3B.Left).

3.6. FunctionalinteractionswithinCAPs

Finally,weturnedtothekeyquestionofourstudy.Namely,whether the reciprocalrelationshipsbetween brain networksduring andafter movieepisodesweremodifiedbytheiraffectivevalence(i.e.,neutral or negative).Todo so,we firstcomputed movie-restcorrelationsin individual occurrencesfor each CAP of interest(precVIS, precDMN, precCEN, and insuFPCN)across all participants,and thenexamined how these relationships differed between the twoaffective contexts [within-CAPtransitions;e.g.,r(CAP(A)neutralmovie;CAP(A)neutralrest)vs.

r(CAP(A)negativemovie; CAP(A)negativerest)].Remarkably,weobserveda consistentanticorrelationbetweentheexpressionofagivenCAPdur- ingmovieperiodsanditsexpressionduringsubsequentrest,forboth affectivecontextsandfortheprecVIS,precDMN,andinsuFPCNCAPs, butnotforprecCEN(seeFig.5A).Althoughtheseanticorrelationswere numericallyhigherinthenegativecontext(rvalues-.46to-.67)than

(8)

Fig.4. Within-DMNCAPinteractionsacross.

Comparisonsofthemovie-restanticorrelation fortheDMN(temporaloccurrencespercon- dition),illustratingsignificantdifferencesfor transitionsbetweensuccessive movie-restpe- riodsinthenegativecontext(yellowcolors), relativetocontrolcomparisonswithconditions fromthe“neutral” context(bluecolors).The otherthreeCAPsof interest(prec-VIS,prec- CEN,insu-FPCN)showednosucheffectofneg- ativeemotiononthesametransitions.p<.05.

Dependencyofcorrelations acrossconditions was considered when comparisons included overlappingvariables(Zou,2007).

theneutralcontext(-.33to-.39),adirectcomparisonofanticorrelation magnitudes(seemethods)showednosignificantdifferenceforanyof thefourCAPs.

However,thewithin-CAPcorrelationsfortransitionsfrommovieto restconditionsweresignificantlymodulatedbyaffectivecontextincom- parisontocontrolconditionswhenthemoviesandrestdidnotfollowin directsuccession[e.g.,r(CAP(A)neutralmovie;CAP(A)negativerest],butse- lectivelyfortheprecDMN(Fig.4)andnotforthethreeothernetworks.

Thus,fortheprecDMNCAP,theanticorrelationpatternobserveddur- ingthepost“negativemovie” transitiontothesubsequent“negative rest” conditionacrossallparticipants[r(19)=-.67;P=0009;PFDR=.01]

was significantlyhigherrelative totheanticorrelation between non- successive “negative movie” and “neutral rest” periods [r(19)=-.19;

P=.38PFDR=.38.P=.02,CI=(-1.16;-.03)forthedifference],orrelative totheanticorrelationobservedbetweennon-successiveneutralmovie andnegativerestperiods[r(19)=-.05;PFDR<1.P=.04,CI=(-1.14-.01) forthedifference].NoneoftheotherCAPsshowedanydifferencein thesecomparisons(TableS2).

Altogether,thesedatahighlightrobustantagonisticrelationshipsbe- tweentheexpressionof particularfunctionalnetworksduringactive moviewatchingandtheirsubsequentexpressionduringrest,regardless ofaffectivestate(exceptfortheprecCENCAP),butwithadistinctive impactofnegativeemotiononthesetransitionsfortheprecDMNCAP (Fig.4).

3.7. FunctionalinteractionsbetweenCAPs

Lastly,weexaminedreciprocalinteractionsamongnetworksbyde- terminingwhethertheoccurrencerateofagivenCAPofinterestduring moviewatchingcouldpredicttheoccurrenceof otherCAPsduringthe followingrestperiod,andtestedwhetherthisrelationshipvariedina context-dependentmannerfornegativecomparedtoneutralconditions (between-CAP analysis; e.g., r(CAP(A)negativemovie; CAP(B)negativerest) vs. r(CAP(A)neutralmovie;CAP(B)neutralrest)]. Resultsfrom this analysis (Fig.5B)revealedapositiveassociationbetweentheinsuFPCNinthe movie-watchingperiod andprecCEN insubsequent restingperiodin the“neutral” context[r(19)= .54;P=.02;PFDR=.04],which was sig- nificantlydifferent[P=.02;CI=(.02;1.23).Fig.5B.]fromthesuppres- sionofsuchinteractioninthe“negative” context[r(19)=.-19;P=.44;

PFDR=.44].Conversely,theinsuFPCNduringmoviewatchingexhibited apositivecorrelationwiththeprecDMNduringsubsequentrestexclu- sivelyinthe“negative” context[r(19)=.28;P=.23;PFDR=.23],absent inthe“neutral” context [r(19)=-.34;P=14;PFDR=.14],withasignif- icantdifferencebetweenthetwocontexts[P=.04;CI=(-1.17;-0.10).

Fig.5B].Ontheotherhand,theprecDMNduringmoviewatchingwas stronglyanticorrelatedwiththeprecCENduringsubsequentrestexclu- sivelyinthe“negative” context[r(19)=.-61;P=.003PFDR=.005],asig- nificantdifference[P=.02;CI=(0.02; 1.20)]compared totheabsence

ofsuchcorrelationintheneutralcontext[r(19)=-.11;P=.64;PFDR=.64 (Fig.5B)].Otherrelationshipsbetweenpairsofnetworkswerenotsig- nificantandnotmodulatedbyaffectivecontext.

4. Discussion

Ourstudyidentifiedasetofdistributedbrainnetworkswithdynam- icallyfluctuatingco-activationpatterns(Boltonetal.,2020;Liuetal., 2018b;Pretietal.,2017) thatweremodulatedbynegativeemotion, with differential expression(i.e., occurrences) and interactions (i.e., correlations) observed both during emotional stimulation itself (sad movies)andduringitsaftermath(subsequentrest).Specifically,weun- coveredfournetworkswhosetemporalprofiledifferedasafunctionof emotionalconditions,overlappingwiththeDMN,CEN,FPCN,andVIS networksreportedinotherstudies.Theseresultsshow,first,thatneg- ativeemotionaleventsengageaselectsetofbrain-widesystems,with anatomicallyandtemporallyspecificconnectivityarisingnotonlydur- ingemotionalevents,butalsoextendingduringsubsequentrest(Fig.2).

Second,theoccurrencesofsomeoftheseCAPsisdirectlyrelatedtosub- jectiveemotionalstate,withmorenegativeaffect(NA)reportedbypar- ticipantswhoshowmorefrequentoccurrencesoftheinsuFPCNandthe precDMNduringandafternegativemovies,respectively(Fig.3).Third, we observed specifictemporal relationships“within” and“between” these networks,suchthattheirexpressionrateduringmovieperiods predictedtheirexpressioninsubsequentrest.Critically,thesefunctional dynamicsamongnetworksdifferedsignificantlyinthenegativeaffec- tivecomparedtotheneutralconditions(Figs..4and5).Together,these dataprovidenewevidenceonhowbrain-widesystemsaredynamically andinteractivelyrecruitedbyemotionsandtheirregulation,withrobust carry-overeffectsbeyondemotionaleventsthemselves.Theseneuralre- sultsaddtobehavioralevidenceof“emotionalinertia” (Kuppensetal., 2010;KuppensandVerduyn,2017)andsuggestthatemotionregulation processesmayextendoverprotractedperiodsoftimefollowingemotion elicitation.Moregenerally,ourstudyprovidesnewinsightsonaffective braindynamicsanduncoversusefulneuralmarkersforadaptive reg- ulationmechanismsthatmediatearestorationofhomeostaticbalance afterstressfulevents,possiblyalteredinpsychopathologicalconditions suchasdepression,anxiety,orPTSD.

4.1. Emotion-responsiveCAPswithdifferenttemporalprofilesanddifferent functionalroles

Amongnetworksdifferentiallymodulatedbynegativeemotion,the precVIS CAP comprisedvisualareas (Orbanet al., 2004) mainlylo- catedindorsalextrastriatecortexandmostlyengagedduringthe“neg- ative movie” periods. This accords with abundant evidence for en- hanced activity of the visual system in responseto affective stimu- lation(Krageletal., 2019;Vuilleumier etal.,2004),presumablyre- flectingmodulatorytop-downsignalsfromemotion-responsivelimbic

(9)

Fig.5. Between-CAPsinteractionsacrosstime.

A. Pairwise correlations in occurrences be- tweenCAPsofinterestobservedduringmovies andrest periods, respectively, computed for the“neutral” (left)andthe“negative” context (right),PFDR adjustmentformultiplecompar- isons(p<0.05;p<0.01).B.Comparisons ofbetween-CAPscorrelationsthatshowedsig- nificantdifferencesbetweenthetwoemotion contexts.Dependencyofnon-overlappingvari- ableswereconsideredfollowing(Zou,2007).

areas (McAlonan et al., 2008; Vuilleumier and Driver, 2007). Nev- ertheless,precVISactivityshowed no directinteraction withtheex- pressionof other CAPs of interest,consistent withthese sensoryar- eashaving no directlink with emotional experienceper se. Onthe other hand, the insuFPCN CAP also predominated during emotion elicitation periods (negative movie), but its occurrences correlated withnegativeaffect scoresreportedafterscanning.Anatomically,in- suFPCN comprised components of the “salience” and “dorsal atten- tion” systems,typicallyengagedintheappraisalofarousinginforma- tion(Sanderet al.,2018). Thispattern fits wellwiththeoretical ac- counts(FeldmanB.andSatpute,2013;SchererandMoors,2019) ac- cordingtowhichemotioninvolvesanintegrationofprocessesassoci- atedwithselectiveattention(Touroutoglouetal.,2012),interoception (CritchleyandGarfinkel,2017),anddetectionofbehavioralrelevance (Roy etal., 2012). Higher insuFPCN expressionduring thenegative movieindicatesthisnetworkisanimportantcomponentofneuralpro- cessesunderlyingemotionelicitation(MeauxandVuilleumier,2015; Sanderetal.,2018).

Incontrast,theprecDMNCAPshowedthegreatestnumberofoccur- rencesduringrestperiodsfollowingnegativemovies,anditspresence alsocorrelatedwithnegativeaffectscores.Theseresultsaddtorecent findingsof similaraftermathsof emotionallyarousingstimuliduring restingstate,e.g.,followingfearfulorjoyfulmovies(Eryilmazetal., 2011)aswellasrewardorpunishmentoutcomes(Eryilmazetal.,2014; Pichonetal.,2015).Emotionalcarry-overeffectsinbrainconnectiv- ity may also underpin changes in affective andcognitive responses tofuture events (Qiao-Tasserit et al., 2017),and thus contributeto adaptivebehavioralfunctionsof emotions.EnhancedprecDMNactiv- ityatresthaspreviouslybeenassociatedwithself-referentialmental activityandintrospection(Christoff etal.,2016;Foxetal.,2018),au- tobiographicalmemory(Engenetal., 2017),aswell asmood disor-

ders (Liemburg etal.,2012; Shelineetal., 2009;Songetal., 2016).

PrecDMNupregulationisoftenregardedasabiomarkerfordepressive rumination(Hamiltonetal.,2015;Sambataroetal.,2014;Whitfield- gabrieliandFord,2012),possiblyresultingfromemotiondysregulation (Aubryetal.,2016).Moreover,bothspontaneousemotionregulation strategies(Ableretal.,2010;JoormannandGotlib,2010)andvolun- tarycognitivereappraisal(Ertletal.,2013;Uchidaetal.,2015)enhance DMNconnectivity, suggestingalink withfunctionalrecoverymecha- nismsactingtodownregulatenegativeemotionsasobservedinthepost sadmovierestperiodshere.

Additionally,inourstudy,precDMNwastheonlyCAPexhibitinga significantimpactofthenegativeaffectivecontextonfunctionaltran- sitionsbetweenitsexpressionrateduringrestanditsexpressionduring thejustprecedingmovie(within-CAPdynamics).Althoughaconsistent anticorrelationofCAPoccurrencesbetweenmovieandrestperiods(i.e., morefrequentoccurrencesduringmoviespredictinglessfrequentoccur- rences duringrest)wasobservedforallemotion-responsivenetworks (exceptprecCEN),onlytheprecDMNshowedastatisticallysignificant amplificationofthisantagonisticrelationshipfornegativerestperiods thatdirectlyfollowednegativemovieperiods.Inotherwords,thelower theexpressionofprecDMNduringthesadmovie,thehigheritsexpres- sionduringsubsequentaftermathatrest(seeFig.4).Thisstrongand valence-dependentshiftindFCprovidesnewevidenceforakeyroleof theDMNinhomeostaticadjustmentafteracutestressors(Hermansetal., 2014),possiblythroughcoordinatedinteractionswithotherbrainnet- works(seebelow).Anti-correlatedactivityamongnetworksmayrepre- sentageneralandbiologicallymeaningfulprocessinfunctionalbrain dynamics acrossvariousconditions(Lietal., 2021).Nevertheless, it remainstofullydeterminethefactorsaccountingforsuch“push-pull” DMNanticorrelationsinoccurrencesbetweenemotionalepisodesand subsequentrecoveryperiodsatrest.Importantly,theabsenceofsuch

(10)

behaviorinmovie-resttransitionsfortheprecCENindicatesthatthis functionalshiftinconnectivitystateisnotcausedbynon-specific re- boundorhemodynamiceffectsbutappearbothnetwork-specificand context-sensitive.

Finally,theprecCEN showeda distinctive profilein occurrences, withpredominantexpressionduringneutralmoviesrelativetoother conditions.ThisCAPcompriseddorsalACCanddorsolateralprefrontal areasassociatedwitheffortfulexecutivecontrol(Coleetal.,2013),pre- sumablyactivatedbymoreabstractcognitivedemandstoprocessthe contentofneutralmovies(unlikemorenaturalabsorptionbyemotional scenes).Nevertheless,theprecCENexhibitedsignificantchangesinits interactionwithotherCAPsasafunctionofaffectivecontext,asdemon- stratedbyouranalysisofbetween-CAPsrelationshipanddiscussedbe- low.

4.2. Emotion-relatedmodulationoffunctionalinteractionsbetweenCAPs

Akeyfindingofourstudyisthathigherorlowerexpressionofcer- tainCAPsduringthemovieperiodwasassociatedwithdifferentoccur- rencesof otherCAPsduringthesubsequentrestperiod(between-CAPs temporaldynamics),andtheserelationshipsamongnetworksweresig- nificantlymodifiedbytheemotionalcontext.Mostcritically,negative emotionproducedasignificantshiftinthefunctionalimpactofinsuF- PCNactivityonsubsequentprecDMNandprecCENatrest,suchthat astrongrelationshipofinsuFPCNwithprecCENintheneutralcontext wassuppressedinthenegativecontext,andreplacedbyapositivere- lationshipwithprecDMNinstead(Fig.5B).Thus,higheroccurrencesof insuFPCNduringsadmovies(whichcorrelatedwithmorenegativeaf- fect)predictedgreaterpresenceoftheprecDMNatrestfollowingthese movies(whichalsocorrelatedwithnegativeaffect).Thissuggeststhat higheraffectivesalienceencodedbyinsuFPCNmightenhancesubse- quentself-regulatoryprocessesandruminativethoughtssubservedby precDMN(Andrews-Hannaetal.,2014).Wenotethatthemagnitudeof correlationsbetweeninsuFPCNandsubsequentprecDMNactivitywas numericallymodestinbothaffectivecontextsbutshiftedfromanega- tiverelationshipintheneutralcontext(r=-.34)toapositiverelationship inthenegativecontext(r=.28),ahighlysignificantdifferencebetween thesetwoconditions.

Inparallel,precCENexpressionatrestalsoshiftedfromastrongpos- itivecorrelationwithinsuFPCNoccurrencesduringtheprecedingmovie intheneutralcontext,toastrongnegativecorrelationwithprecDMN occurrencesinthenegativecontext,bothrepresentinghighlysignificant changesincorrelationmagnitudeand/ordirection.Theseopposingef- fectson precCENmayaccord withbehavioralevidenceforimpaired executivecontrolabilitiesafteracute stressconditions(Chand etal., 2017),whichcouldbemediatedatleastpartlybyinhibitoryinterac- tionsfromprecDMNinthepost-negativecontext.Conversely,theposi- tiverelationshipbetweeninsuFPCNandsubsequentprecCENatrestin theneutralcontextsuggestsmoresynergicinteractionsbetweenthese twonetworks,commonlyimplicatedin externallyorientedcognition (CorbettaandShulman,2002;Sprengetal.,2016).Interestingly,such modulationofprecCENexpressionbyprecedingmovievalence,despite lowoccurrencesofprecCENatrestoverall(Fig.2D),underscoresthat ahighoccurrencerateisnotaprerequisiteformeasuringsignificant changesinCAPdynamics,andfurthershowsthatemotionmayimpact onthefunctionalintegrationofbrainnetworksratherthanjusttheir globalactivationlevel(Leitãoetal.,2020).

Insum,ourfindings shownotonlydistinctivepatternsofexpres- sionofinsuFPCNandprecDMNCAPsduringnegativeemotionepisodes andpost-emotion restperiods,respectively, butalsosignificantfunc- tionalrelationshipsbetweenthesenetworksduringthetransitionfrom emotionepisodestosubsequentrest.Together,thesedatapointtocom- plementaryrolesofFPCNintheintegrationofemotionelicitationpro- cessesandofDMNinself-regulationmechanismscontributingtoaffec- tivehomeostasis.

4.3. Outstandingissuesandfurtherdirections

Althoughwefoundrobustdifferencesintheexpressionandcorrela- tionofrelevantCAPsinourstudy,andcarefullycounterbalancedcon- ditionsbetweenparticipants,wecannotfullyruleoutthatothereffects relatedtomoviecontents ortimecontributed tochangesin network activityandtheirfunctionalrelationshipsasobservedhere.Furtherre- searchusingothernaturalisticdesignsandevokingdifferentemotions (e.g.,frustration,joy,etc.)willalsobevaluabletoconfirmandextend ourresults.Inaddition,ourgenderselectionwaslimitedtofemalepar- ticipants,basedonpilotdatashowinghigherandmoreconsistentemo- tionratingsthanmales,andpreviousstudiesshowingmoreintenseemo- tionexperiencesinfemales(Vrtičkaetal.,2012).Itwillbeimportantto replicateandextendthecurrentfindingstomalepopulations,because emotiondysregulationandresiliencetostressmayshowimportantgen- derdifferences.

5. Conclusions

UsingarecentdFCanalysismethodology(CAPs),weshednewlight onthespatio-temporalorganizationofbrainnetworksunderlyingthe dynamicsofnegativeaffectiveexperienceandsubsequentreturntorest.

Our data unveil specifictransitions in the connectivityand recipro- calrelationshipofbrain-widenetworksinvolvedinvisualperception, attention tosalient stimuli,executivecontrol, andintrospectiveself- monitoringprocessesduringbothemotionalepisodes(movies)andtheir aftermath(subsequentrest).Amongthesenetworks,theinsuFPCNand precDMNCAPswerefoundtoplayapivotalroleinthetemporaldy- namicsofemotionalexperience,fromelicitationtosubsequentrecov- ery,entertainingreciprocalrelationshipsnotonlywitheachotherbut alsowiththeprecCENCAP.Crucially,individualoccurrencesofthese CAPsduringmoviesandrestweredirectlylinkedtothesubjectiveaf- fectivestatereportedbyparticipants.Thesefindingsprovidenovelin- sights on brainmechanismsunderlying emotionexperience andreg- ulation (Dixonetal., 2018b;Panetal., 2018; Sripadaetal., 2014), resiliencetostress(Youngetal.,2017),andperseverativeruminative thinkinginnegativemoodstates(Eryilmazetal.,2011;Garfinkeletal., 2016; Harrison et al., 2008). In turn, they may also offer valuable biomarkers forunderstandingandassessingtheneuralbasisofclini- calpsychopathologyconditions.Nevertheless,becauseoursamplewas restrictedtofemaleparticipants,ourresultsconstituteapreliminaryex- plorationofbrainnetworksdynamicsinresponsetosademotions,and futureresearchinlarger,independentsamplesisneededtogeneralize ourfindings.

Acknowledgments

ThisworkwassupportedbytheSwissexcellenceScholarshippro- gram,theSchmidheinyfoundation,andtheColombianScienceMinistry (JG),aswellasbyaSinergiaGrantno.180319fromtheSwissNational Science Foundation (SNF),bythe SwissCenterof AffectiveSciences financedbyUNIGEandSNF(Grantno.51NF40_104897),andbythe Société AcadémiquedeGenève(SACAD).Thisstudywasconductedon theimagingplatformattheBrainandBehaviorLab(BBL)andbenefited fromsupportoftheBBLtechnicalstaff.Specialthankstodr.Frédéric Grouillerforhisusefulcommentsontheunivariateanalysis.Imaging analysiswascarriedoutwiththehigh-performancecomputing(HPC) facilities,attheUniversityofGeneva.

Dataavailability

Deriveddatasupportingthefindingsofthisstudyareavailableat https://yareta.unige.ch/#/home

Références

Documents relatifs

The crucial dynamic mechanical difference between all three of the PAMPS/PAAm DN-gels and normal cartilage is that all three had much lower loss angles (δ) than cartilage in both

Consistent with the idea that negative emotions such as ‘frustration’ (i.e. a negative affective state that is induced by a failure to obtain an incentive or the blockage of a

The proofs of these results are based on congruence properties and on a theorem of Weyl in the boundary case, and on new observability inequalities for the undamped wave

Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of

The predicted covariance structure depends on the readout parameters, and in particular on the temporal integration window w and typical number of neurons K used in the formation of

Using our novel network model of the mouse brain that permits systematic investigation of peripheral pathways, especially the ones for sensory data processing, we showed that

If this is suffi- ciently small, then a flow emerges within the manifold, which is approximately independent (again via timescale separation) from the shape of the manifold. In

The two-sample t test comparing the functional connectivity maps generated from the second seed (within the right middle cingulate gyrus) showed that this region had more