SocialCognitiveandAffectiveNeuroscience,2016,569–579 doi:10.1093/scan/nsv139 AdvanceAccessPublicationDate:19November2015 Originalarticle Emotion-induced loss aversion and striatal-amygdala coupling in low-anxious individuals D o w Caroline J. Charpentier,1,2,* Benedetto De Martino,3 Alena L. Sim,1 nlo a d Tali Sharot,2 and Jonathan P. Roiser1 ed fro m 1InstituteofCognitiveNeuroscience,UniversityCollegeLondon,LondonWC1N3AZ,UK,2AffectiveBrainLab, h DepartmentofExperimentalPsychology,UniversityCollegeLondon,LondonWC1H0AP,UK,and ttps 3DepartmentofPsychology,UniversityofCambridge,CambridgeCB23EB,UK ://a c a d *CorrespondenceshouldbeaddressedtoCarolineCharpentier,InstituteofCognitiveNeuroscience,17QueenSquare,LondonWC1N3AZ,UK. e m E-mail:[email protected]. ic .o u p .c o m Abstract /s c a Adaptingbehaviortochangesintheenvironmentisacrucialabilityforsurvivalbutsuchadaptationvarieswidelyacrossin- n /a dividuals.Here,weaskedhowhumansaltertheireconomicdecision-makinginresponsetoemotionalcues,andwhether rtic thisisrelatedtotraitanxiety.Developinganemotionaldecision-makingtaskforfunctionalmagneticresonanceimaging, le -a inwhichgamblingdecisionswereprecededbyemotionalandnon-emotionalprimes,weassessedemotionalinfluenceson b s lossaversion,thetendencytooverweighpotentialmonetarylossesrelativetogains.Ourbehavioralresultsrevealedthat tra c onlylow-anxiousindividualsexhibitedincreasedlossaversionunderemotionalconditions.Thisemotionalmodulationof t/1 decision-makingwasaccompaniedbyacorrespondingemotion-elicitedincreaseinamygdala-striatalfunctionalconnectiv- 1/4 ity,whichcorrelatedwiththebehavioraleffectacrossparticipants.Consistentwithpriorreportsof‘neurallossaversion’, /5 6 bothamygdalaandventralstriatumtrackedlossesmorestronglythangains,andamygdalalossaversionsignalswere 9 /2 exaggeratedbyemotion,suggestingapotentialroleforthisstructureinintegratingvalueandemotioncues.Increasedloss 3 7 5 aversionandstriatal-amygdalacouplinginducedbyemotionalcuesmayreflecttheengagementofadaptiveharm- 0 6 avoidancemechanismsinlow-anxiousindividuals,possiblypromotingresiliencetopsychopathology. 5 b y Keywords:decision-making;lossaversion;emotion;anxiety;functionalmagneticresonanceimaging(fMRI) gu e s t o n 2 9 Introduction M Trait anxietyislikely tobean importantfactor inpeople’s a Detecting and processing changes in our environment and tendencytoaltertheirdecisionsinresponsetoemotionalcues. rc h adapting our decisions in response to those changes, are im- Rodent studies have reported that low levels of anxiety are 2 0 portantfeaturesofhumanbehavior.Forexample,wearelikely associated with adaptive stress-coping and learning behavior 19 tobehaveandmakechoicesdifferentlyifwereceivepositiveor (LandgrafandWigger,2002;Herreroetal.,2006).Highlyanxious negativesocialfeedback(Sipetal.,2015),orifthreateningand humansexhibitdifficultyinmodulatinglearninginvolatileen- emotionally arousing cues appear in our surroundings (Mobbs vironments(Browningetal.,2015),cognitivecontrol(Derryberry et al., 2007, 2009). However, such behavioral adaptation varies and Reed, 2002; Bishop, 2007, 2009) and emotion regulation substantiallyacrossindividuals,andthefactorsthatinfluence (Etkinetal.,2010;FarmerandKashdan,2012);andithasbeen howpeoplealtertheirdecision-makinginemotionalsituations suggestedthattheflexiblemodulationofbehaviorinresponse remainpoorlyunderstood. to anxiogenic environmental changes may be an important Received:12June2015;Revised:18October2015;Accepted:12November2015 VCTheAuthor(2015).PublishedbyOxfordUniversityPress. ThisisanOpenAccessarticledistributedunderthetermsoftheCreativeCommonsAttributionLicense(http://creativecommons.org/licenses/by/4.0/), whichpermitsunrestrictedreuse,distribution,andreproductioninanymedium,providedtheoriginalworkisproperlycited. 569 570 | SocialCognitiveandAffectiveNeuroscience,2016,Vol.11,No.4 mechanismbywhichfurtherexposuretostresscanbeavoided Participantsweretoldthattheaimofthestudywastoinvesti- (Mathews and Mackintosh, 1998; Robinson et al., 2013, 2015). gate how memory was affected by emotions. This cover story Thereforeitispossiblethathighlyanxiousindividualsmayfail wasusedtoavoidparticipantsdeducingthetruegoaloftheex- toadapttheirdecision-makingunderemotionalconditions.On periment(demandcharacteristics:Orne,2009),i.e.themanipu- theotherhand,highanxietyisalsoassociatedwithexaggerated lation of their gambling behavior by emotion. They first responsestoemotionalstimuli(Etkinetal.,2004;Foxetal.,2007; practiced the memory task (see Supplementary Methods for a Steinetal.,2007;Sehlmeyeretal.,2011),raisingthe possibility full description). Then they were instructed that to make this thatdecision-makinginhighlyanxiousindividualsmaybedis- memorytaskmorechallenging,theywouldperformadistract- proportionatelyinfluencedbyemotion. ing gambling task while holding the stimuli in memory. They Therefore,weaskedwhetherdecisionmakingisinfluencedby next completed a training block of gamble-only trials. Finally, emotionalcuestoagreaterextentinlow-anxiousindividuals(po- they completed a training block of the combined emotional tentiallydrivenbygreaterbehavioralflexibility)orinhigh-anxious decision-makingtask(Figure1A). individuals(potentiallydrivenbygreateremotionalreactivity).To Participants returned to the laboratory for Day 2 (scanning D disambiguate between these hypotheses, we developed a func- session) after the screening session (mean delay¼17.92 days, o w tional magnetic resonance imaging (fMRI) paradigm where each range¼1–44days).Duringthissession,theyinitiallycompleted n lo decision(acceptingorrejectingagamble)wasprecededbyemo- one block of trials of the combined emotional decision-making a d tionalornon-emotionalprimes.Weexaminedpeople’sdecisions task(Figure1A)beforeenteringthescanner,andfourfurther11- e d in the framework of Prospect Theory (Kahneman and Tversky, minblocksduringfMRIscanning.Sincethereislargevariability fro 1979)andmodeledtheirlossaversion[thetendencytooverweigh in loss aversion across individuals, each participant’s indiffer- m potential losses relative to gains (Kahneman et al., 1991; Hardie encepointonthelossaversiontaskfromDay1wasusedonDay http etal.,1993)]underemotionalrelativetonon-emotionalconditions. 2toindividuallytailorthegamblematrix(Figure1BandC). s Toinvestigatewhetheravoidanceofpotentiallossesisalteredspe- Afterthescan,participantscompletedtheBDIasecondtime ://a c cificallyunder threat,or under emotionalarousalingeneral, we and the state-trait anxiety inventory (Spielberger et al., 1983). ad e usedbothnegativeandpositiveemotionalcues. Noneofthe28participantsscoredabove15ontheBDIatthis m Based on prior work implicating the amygdala and ventral time (mean¼1.68, s.d.62.09, range 0–10). Mean trait anxiety ic .o striatuminbothlossaversion(Tometal.,2007;DeMartinoetal., was 31.2 (s.d.66.34). Trait anxiety was used as a covariate in u p 2010;Canessaetal.,2013;Sokol-Hessneretal.,2013)andthepro- theanalyses;inaddition,amediansplitwasperformed,with14 .c o cessing of emotional cues (Adolphs, 2002; Glascher and participants in a ‘low’ trait anxiety group (mean¼25.9, m /s Adolphs,2003;Dalgleish,2004;Mobbsetal.,2006;Phelps,2006; s.d.63.03, range 20–30) and 14 in a ‘high’ trait anxiety group c a PessoaandAdolphs,2010;Wangetal.,2014),wehypothesized (mean¼36.5,s.d.63.65,range33–44). n/a thattheseregionswoulddrivetheinfluenceofemotiononeco- rtic nomic decisions. Specifically, we tested two mechanistic le hypotheses:(i)thatenhancedamygdalaandstriatumresponses Emotionaldecision-makingtask -ab s topotentiallossesrelativetogains(‘neurallossaversion’)may Each trial started with the presentation of either two or four tra be directly modulated by emotion in a manner that drives primestimulifromthesamecondition(happy/fearful/neutral/ ct/1 changesinbehaviorand(ii)thattheamygdalaandventralstri- object)for3s(prime:Figure1A).Participantswereinstructedto 1 atumplaycomplementaryrolesinthismodulationofdecision- memorizetheirlocation.Afterajittereddelayof2–6s,thegam- /4/5 making, and it is their functional integration (as opposed to bleappearedfor2sandparticipantsdecidedwhethertoaccept 69 theiractivation)thatunderlieschangesinlossaversion. orrejectthisgamble.Therewasanother2–6sjittereddelaybe- /23 7 foretheprobeface/objectappeared.Participantshad2stoindi- 5 0 cate the location where the probe had been displayed in the 6 5 Materialsandmethods first screen, followed by a 1s fixation cross between trials. b y Participants Gambleoutcomeswerenotrevealed.Thetwodelayswerejit- gu teredtodecorrelatetheprimestimulifromthegamblepresen- es Thirtyhealthyvolunteerswererecruitedbyadvertisement.Data tationtime,butalwayssummedto8s,suchthattheintertrial t o n fromtwoparticipantswereexcludedbecauseofalackofbehav- intervalwasmaintainedataconstant16sthroughoutthetask. 2 9 ioralconsistencyinthegamblingtask,makinglossaversionim- Participantscompleted196trialsofthiscombinedtask(49trials M possible to model. Final analyses included 28 participants (15 of each of the four conditions: happy, fearful, neutral, object). arc males, 13 females, age range 19–47 years, mean 26.5 years). Gambles were randomly sampled from a 7*7 gain–loss matrix h 2 Participants gave written informed consent and were paid for centeredoneachparticipant’sownindifferencepoint(example 0 1 their participation in an incentive-compatible manner. See matrices: Figure 1B and C). This was done to ensure that the 9 SupplementaryMethodsforexclusioncriteriaandpaymentde- samerangesofwinsandlosseswerepresentedforeachemo- tails.Thestudywasapprovedbythe localdepartmentalethics tion condition, and to optimize sensitivity to detect emotion- committee. drivenchangesinlossaversionwithamajorityofgamblesclose totheparticipant’sindifferencepoint. Procedure Behavioraldataanalysis Participantsattendedthelaboratoryon2differentdays.OnDay 1(screeningsession), participants wereadministeredthe Mini Behavioral datawereanalyzedusingIBMSPSS Statistics (v.21) InternationalNeuropsychiatricInterview(Sheehanetal.,1998), andMatlab.Missedgambletrialswereexcluded.Foreachpar- Beck depression inventory (BDI; Beck et al., 1961) and an MRI ticipant,theprobabilityofacceptingthegamble(P ),mean accept safetyquestionnaire.Eligibility forthestudy required:nopast reaction time (RT) to accept or reject the gamble, number of or present psychiatric disorders, including alcohol/substance missed trials and working memory accuracy were calculated dependence/abuse, BDI<15 and no MRI contraindications. separatelyfor thedifferentemotion conditions andsubmitted C.J.Charpentieretal. | 571 D o w n lo a d e d fro m h ttp s ://a c a d e m ic .o u p .c o m /s c a n /a rtic le -a Fig.1.Experimentaldesign.(A)Oneachtrial,participantswerefirstpresentedwithanarray(prime)of2or4faces(allhappy,allfearfulorallneutral)orobjects(light b s bulbs)andhad3stomemorizeit.Theythenhadtodecidewhethertoacceptorrejectamixedgambleinwhichtherewasa50%chanceofwinningtheamountin tra green,anda50%chanceoflosingtheamountinred.Finally,aprobefromthefirstarraywaspresentedandparticipantshadtoreportitsposition.(B,C)Tooptimize c modelfittingandsensitivitytoemotionalcontext,anestimateofeachparticipant’sindifferencepoint(IP)wasobtainedfromthepracticesessionandusedtodefine t/1 1 thegamblegain/lossmatrix.Eachmatrixwasformedbycombiningsevenpotentialgainswithsevenpotentiallosses,leadingto49gambles,repeatedacrosseachof /4 thefourconditions.Examplematricesareshownwiththeresultinggambleexpectedvalue(EV¼0.5*gainþ0.5*loss),centeredonanIPof0(B,non-lossaversepartici- /5 6 pant)or4.5(C,highlylossaverseparticipant). 9 /2 3 7 5 to repeated-measures analysis of variance (ANOVA) to assess Differentmodelswererunusingthisprocedure,whereloss 0 6 the impact of emotion on behavior. Trait anxiety scores were aversion (k) and choice consistency (m) parameters were esti- 5 b addedascovariatesintheanalyses(SupplementaryTableS1). mated: (i) across all trials; (ii) separately for emotional (happy y g Toassesslossaversionatwo-parametermodelwasestimated andfearfulfaces)andnon-emotional(neutralfacesandobjects) u e basedonProspectTheory’ssubjectiveutilityfunctionusingamax- trialsor(iii)separatelyforeachofthefouremotionconditions. st o imumlikelihoodestimationprocedureinMatlab(Kahnemanand Model comparison analyses using Bayesian Information n 2 Tversky,1979;Sokol-Hessneretal.,2009;Chibetal.,2012).Foreach Criteron(BIC)(Schwartz,1978)revealedthatthetwo-condition 9 M trial,thesubjectiveutility(u)ofeachgamblewasestimatedusing model wasmoreparsimoniousthan the four-condition model a thefollowingequation(withlossescodedasnegativevalues): (see Supplementary Methods for details). Therefore, the two- rc h conditionmodelwasusedpreferentiallyinallanalyses,except 2 uðgambleÞ¼0:5(cid:2)gainþ0:5(cid:2)k(cid:2)loss (1) toverifythattheeffectsobtainedwereindependentofvalence 01 9 orspecifictoemotionratherthanfacesingeneral.Thepercent- wherekisthe‘lossaversion’parameter:k>1indicatesanover- agechangeinkandinlbetweenemotionalandnon-emotional weighing of gains relative to losses and k<1 the converse. conditionswascalculatedfromthistwo-conditionmodel.Both Thesesubjectiveutilityvalueswereusedinasoftmaxfunction variables were normally distributed with Skewness values toestimatetheprobabilityofacceptingeachgamble(codedas0 smallerthan1andKurtosisvaluessmallerthan3(percentage or1foreachrejectedoracceptedgamble,respectively): change in k: Skewness¼(cid:3)0.252, Kurtosis¼0.669; percentage change in m: Skewness¼0.672, Kurtosis¼2.033). However, the Pðgamble acceptanceÞ¼ð1þexpð(cid:3)l(cid:2)uðgambleÞÞÞ(cid:3)1 (2) distribution of the loss aversion parameter k was positively skewed,sowhenanalyseswhererunonthisparameterperse wheremisthelogitsensitivityor‘inversetemperature’param- (e.g.correlationbetweenlossaversionandtraitanxiety),kval- eter,anindexofchoiceconsistencyforrepeatedidenticalgam- ueswerelog-transformedbeforerunningstatisticaltests. bles, equivalent to the maximal slope of a logistic regression Toestimateriskaversionweusedaprocedurereportedpre- curve:highermvaluesindicatemoreconsistentchoices. viously (De Martino et al., 2010), based on the behavioral 572 | SocialCognitiveandAffectiveNeuroscience,2016,Vol.11,No.4 sensitivitytogamblevariance(O’NeillandSchultz,2010).When To assess whether these responses were modulated by gamblevarianceishigh(e.g.win£10/lose£10relativetowin£2/ emotion, another model contained the same regressors as lose£2),theriskishigh;therefore,riskaverseindividualswill above but separately for emotional trials (happy and fearful exhibit a stronger reduction in gamble acceptance as gamble faces) and non-emotional trials (neutral faces and objects). varianceincreases.Tocalculatethissensitivitytogamblevari- Thisconstitutedourprimaryanalysisbasedonthebehavioral ance,alinearregressionwasrunbetweengamblevariance[cal- results suggesting that grouping trials into emotional and culated for each gamble as (0.5*gain(cid:3)0.5*loss)2] and the non-emotionaloneswasmostparsimonious.However,toen- probabilityofgambleacceptance(calculatedforgroupsofgam- surethatoureffectswerenotdrivenbyfaceprocessingperse bles with the same variance). For each subject, risk aversion andtocontrastemotionalwithneutralfaces,weestimateda was approximated by the negative value of this regression further model in which neutral face and object trials were slope,separatelyforemotionalandnon-emotionalconditions. modeledseparately. Notethatthedesignofthetaskdidnotallowustoconcurrently First-level contrasts were created through linear combin- estimatelossandriskaversioninthesameutilitymodel.Todo ations oftheresulting betaimages and analyzed at the group D so,thetaskshouldhaveincludedasubsetoftrialswhereriskis level with one-sample t-tests, using the standard summary- o w present,butlossesdonotneedtobeweighedagainstgains,so statistics approach to random-effects analysis in SPM. A clus- n thatthe model can distinguish between risk and loss aversion. ter-formingthresholdofP<0.001uncorrectedwasapplied,fol- loa d However,becauseoffMRItimeconstraints,wewerenotableto lowed by family-wise error (FWE) correction at P<0.05, using e d addthesetrialstothetask.Riskaversionwasthereforeestimated small-volumecorrection(SVC)inouraprioriregionsofinterest fro separatelyandaddedasacovariateintheanalysestoensureit (ROIs).Thesewerebilateralventralstriatum(caudateandputa- m didnotaffecttheresults.Inparticular,toensurethattheinflu- men, left and right combined) given its role in loss aversion http enceoftraitanxietywasspecifictothechangeinlossaversion, (Tom et al., 2007; Canessa et al., 2013) and bilateral amygdala, s partial correlations were conducted, where emotion-driven given its role in processing emotion (Adolphs, 2002; Glascher ://a c change in loss aversion was correlated with trait anxietywhile and Adolphs, 2003; Dalgleish, 2004; Phelps, 2006; Pessoa and ad e controllingforchangesinriskaversionandchoiceconsistency. Adolphs, 2010; Wang et al., 2014). ROIs were anatomically m defined using the automated anatomical labeling atlas in the ic .o SPMWfuPickAtlastoolbox(Tzourio-Mazoyeretal.,2002). u MRIdataacquisitionandanalysis p .c o Acquisition parameters. Neuroimaging data were collected ona Functional connectivity analyses. Amygdala-striatum functional m/s SiemensAvanto1.5TMRIscannerusinga32-channelheadcoil. connectivity was analyzed using psychophysiological inter- ca To correct for inhomogeneities of the static magnetic field, action(PPI)inSPM8.First,theventralstriatumclusterfoundto n/a fieldmaps were acquired and used in the unwarping stage of track value (using a P<0.001 (uncorrected) threshold and rtic dpaotsaedporefp4rodcuemssminyg.anFodu2r03fufnucntciotinoanlalscvaonlunminegs,swesesrieonasc,qucoirmed- mBOaLsDketdimweitshetrhieesawnaastoemxticraacltReOdI)acwroasssuasleldvaosxetlhseinsetehdisremgiaosnk. le-ab s usingapre-scannormalizedgradientecho-planarimagingse- andadjustedforalleffectsofinterest.Afirst-levelmodelwas tra tqtriumixeen¼¼c6e34.(cid:5)1w36i2t4hs,,vothxeeeclhsofoizlelot¼wim3ine(cid:5)g¼35(cid:5)p0a3mrmasmm,e3t,ef3lris6p:axvaioanllugsmleliec¼e9sr0es(cid:4)pa,emtipmtiloeand- cBterOexLatDt(ecdtoinmfotreraseseatricbehestpw(paerhteyincsiipoealmongotitciinaolcnlarueldgairnnegsdstonhroe),nd-teehcmeoonetmviooonlvtaieoldntarsitlarlcisaotnaa-tl ct/11/4/5 forwhole-braincoverage,tilt¼(cid:3)30(cid:4).AT1-weightedmagnetiza- thetimeofthegamble:psychologicalregressor)andtheircross- 69 tion-prepared rapid gradient-echo (MP-RAGE) anatomical scan product(PPIregressor).Thismodelalsoincludedthefollowing /23 7 wasacquiredattheendofthesession(176sagittalslices,repe- regressors: prime onset, memory probe onset, parametric 5 0 tition time¼2.73s, echo time¼3.57ms, flip angle¼7(cid:4), ma- modulators for gains and losses at gamble onset, all split by 65 trix¼224(cid:5)256,voxelsize¼1(cid:5)1(cid:5)1mm3). emotioncondition,aswellasmissedgambleonset(ifany)and by movement regressors. Two nuisance timeseries were also gu Statistical analyses. MRI data preprocessing and analysis were added, from a white-matter voxel (corpus callosum body, es performedusingStatisticalParametricMapping(SPM8)software MontrealNeurologicalInstitute(MNI)coordinates:0,14,19)and t o n (Wellcome Trust Centre for Neuroimaging, London, UK, http:// fromacerebrospinalfluidvoxel(centerofrightlateralventricle, 2 9 www.fil.ion.ucl.ac.uk/spm)inMatlab.Preprocessingincludedfield MNIcoordinates:4,14,18);bothinthesamey-planeastheven- M map correction, realignment, unwarping, coregistration, spatial tralstriatumpeakvoxel.Finally,contrastsweredefinedonthe arc normalizationand smoothing (see Supplementary Materials for striataltimeseries(physiological)regressor,modeling‘mainef- h 2 details).Foreachparticipant,thegenerallinearmodelwasused fect’functionalconnectivityandonthePPIregressor,modeling 01 9 tomodelbloodoxygenlevel-dependent(BOLD)signalsduringthe themodulationoffunctionalconnectivitybyemotionalcontext, task,incorporatinganautoregressive[AR(1)]modelofserialcor- whichwereanalyzedatthesecondlevel.Again,tomakesure relationsandahigh-passfilterat1/128s. that our effects were driven by emotional faces rather than Twofirst-levelmodelsweredefined.Thefirstmodelidenti- facesingeneral,thePPIanalysiswasrepeatedforemotionalvs fiedbrainregionstrackinggainandlossvalueindependentof neutralfacestrialsatthetimeofthegamble(excludingtheob- emotion (similar toTom et al., 2007). It included the following jectcondition). regressors(andassociateddurations),collapsedacrossallmem- ory conditions and convolved with the SPM synthetic hemo- dynamicresponsefunction:primeonset(3s);gambleonset(2s) Results withgainvalue,lossvalue(codedasnegativevalues)andchoice Emotionalmodulationoflossaversiondependsontrait difficulty (distance between gamble expected value and partici- anxiety pant’sindifferencepoint)asparametricmodulators;memoryprobe onset(stickfunction);missedgambleonset(ifany:stickfunction). Emotionalstimuliincreasedlossaversion(k)inindividualswith Thesixmovementparameterswerealsoincludedinthemodel. low trait anxiety. Thepercentage change inloss aversion was C.J.Charpentieretal. | 573 emotional relative to non-emotional stimuli (þ5.49%, s.d.¼6.38%,t ¼3.22,P¼0.007),therewasnosignificanteffect (13) of emotion on loss aversion in the ‘high’ trait anxiety group ((cid:3)2.81%,s.d.¼7.52%,t ¼(cid:3)1.40,P¼0.18). (13) Tocheckthatcollapsingemotional(happy/fearful)andnon- emotional(neutral/object)conditionstogetherdidnotalterthe results,weexaminedthechangeinlossaversionseparatelyfor each condition (happy, fearful and neutral) relative to object (Figure2B).First,therewasnodifferencebetweenpositiveand negative emotion: the valence (happy/fearful) * trait anxiety group (low/high) interaction was non-significant (F ¼0.24, (1,26) P¼0.63) and the effect of trait anxiety group on emotionally drivenchangesinloss aversionwassignificant forhappyand D fearful faces separately (happy: t(26)¼2.04, P¼0.05; fearful: ow t(26)¼2.75,P¼0.01).Second,theeffectofneutralfacesrelativeto nlo objectstimulionlossaversiondidnotdifferbetweentraitanx- a d iety groups (t ¼0.6, P¼0.55). Third, model comparison ana- e (26) d lyses(seeSupplementaryMethods)showedthatestimatingloss fro aversionforemotionalandnon-emotionaltrials(i.e.collapsing m happy and fearful together, and neutral and object together) http wasmoreparsimoniousthanestimatinglossaversionseparately s forallconditions. ://a c Analyses of the choice consistency parameter (l) revealed ad e that participants were more consistent in their gambling m Fig.2.Emotionalcuesmodulatelossaversioninlow-anxiousindividuals.(A) choicesonemotionalthannon-emotionaltrials,thoughthisef- ic The change in loss aversion following emotional relative to non-emotional .o fect did not correlate significantly with trait anxiety (see u primes was negatively correlated with trait anxiety across participants. (B) p SupplementaryResultsandFigureS1).Similarly,onlytheemo- .c Participantswithlowtraitanxiety(median-split,N¼14pergroup)showedasig- o nificant increase in loss aversion following both happy and fearful stimuli. tionalmodulationofgambleacceptanceandlossaversioncor- m/s Collapsingfearfulandhappytrialsintoanemotionalconditionandneutraland relatedsignificantlywithtraitanxiety,rulingoutthepossibility c a objecttrialsintoanon-emotionalconditionwasjustifiedbythefactthatthere that differences in memory, RTs or missed trials may have n/a wjeactssntiomvuallie.nTcweoe-ftfaeicletd,aPn-vdanluoeds:if*fPe<re0n.c0e5s.EbrertowrebeanrsnreepurteraslenfatcSeEsMre.lativetoob- drivTeanktehnetoobgseetrhveerd,eofuferctbse(hSauvpiporleaml reenstualrtysTsaubglgeeSs1t).that emo- rticle tionalcuestriggerchangesinlossaversionasafunctionoftrait -ab s calculated for each participant between emotional and non- anxiety, such that low-anxious individuals show the greatest tra etomlootsisonavaelrtsriioanls,vathluuessrpeemrsoev,ianngdinretelaritnedditvoidturaailtvaanrxiaiebtiylitsycodruees eremveoatilotnhaaltlyt-hinisdeufcfeecdtiinscnreoatsderiivnenlobsysraivsekrasivoenrs.iIonnaodrdbityiocnh,owicee ct/11 across participants. We identified a significant negative rela- consistency. Finally, and surprisingly, both positive and nega- /4/5 tionship between emotion-driven change in loss aversion and tiveemotionalstimulihaveasimilareffect. 69 traitanxiety(r ¼(cid:3)0.524,P¼0.004,Figure2A),suchthatlow- /2 (28) 3 anxious individuals showed the greatest increase in loss 75 Neuralresponsestodecreasinglossesaregreaterthan 0 aversioninducedbyemotionalcues.Importantly,baselineloss 6 toincreasinggains 5 aversion(modeledacrossalltrialsindependentofemotioncon- b y ditionandlog-transformedbecausepositivelyskewed)wasnot ThefirststepinourfMRIdataanalysiswastoverifythatexpected g u correlatedwithtraitanxiety(r ¼(cid:3)0.031,P¼0.88;afterremov- value signals were observed in the brain at the time of gamble, e (28) s ingoneoutlierwithveryhighk:r(28)¼0.043,P¼0.83),suggesting withexaggeratedresponsestolosses(‘neurallossaversion’)inthe t o n that the effect of trait anxiety on loss aversion change is not ventral striatum. Next, we examined responses to emotional 2 9 simplydrivenbyregressiontothemean. primes as well as emotional modulation of value signals in the M Thepercentagechangeinriskaversionbetweenemotional amygdala.Finally,weinvestigatedtheinteractionbetweenthese a rc and non-emotional trials, although highly correlated with two regions using functional connectivity. For all analyses, we h 2 change in loss aversion (r(28)¼0.65, P<0.001; expected given additionallyassessedtherelationshipwithtraitanxiety. 01 that loss and risk aversion were estimated separately; see De Awhole-brainanalysiswasfirstconductedtoidentifyclus- 9 Martinoetal.,2010;Canessaetal.,2013),wasnotcorrelatedwith ters with a parametric response to decreasing losses and trait anxiety (r ¼(cid:3)0.23, P¼0.24). Finally, the correlation be- increasinggains,time-lockedtothepresentationofthedecision (28) tweenemotion-inducedchangeinlossaversionandtraitanx- andindependentofemotioncondition.Thisgainandlosscon- iety was unchanged when controlling for change in risk trastisequivalenttoasingleparametricmodulatorrepresent- aversionandchangeinchoiceconsistency(partialcorrelation, ing the expected value of the gamble (0.5*gainþ0.5*loss, with r ¼(cid:3)0.51,P¼0.008). losses coded as negative values). Three clusters surviving (28) Performingamediansplitontraitanxietyscoresconfirmed whole-braincorrectionformultiplecomparisonswerefoundto theemotionalmodulationoflossaversion,revealingasignifi- track gamble value (Supplementary Table S2A), located in the cant condition (emotion/no emotion) * trait anxiety (low/high) rightventralstriatum(Figure3A),rightamygdala/hippocampus interaction (F ¼6.96, P¼0.014). Although the ‘low’ anxious (Figure 3B) and anterior cingulate/orbitofrontal cortex (ACC/ (1,26) groupshowedasignificantincreaseinlossaversionfollowing OFC) (Figure 3C), confirming previous reports of generic value 574 | SocialCognitiveandAffectiveNeuroscience,2016,Vol.11,No.4 D o w n lo a d e d fro m h ttp s ://a c a d e m ic .o u p .c o m /s c a n /a rtic le -a b s tra c t/1 1 /4 /5 6 9 /2 3 7 5 0 6 5 b y g Fig.3.Brainregionstrackinggambleexpectedvalues.Awhole-brainanalysiswasconductedtoidentifyregionsshowingaparametricresponsetodecreasinglosses u e andtoincreasinggains.Clusterssurvivingwhole-brainFWEcorrectionwerefoundintheventralstriatum(A),theamygdalaextendingintothehippocampus(B)and s theACC/OFC(C).ActivationsaredisplayedatP<0.001(uncorrected)ontheaverageanatomicalscanfromall28participants.ColorbarsrepresentT-values.(D–F) t o n Parameterestimates(betas)extractedfromtheparametricresponsetolosses(redbars)andtogains(greenbars)separatelyrevealedgreatertrackingoflossesrelative 2 togainsintheseregions(attrendlevelintheACC).Notethatthelattercontrastsareorthogonaltothatusedforvoxelidentificationandthereforedonotrequirecor- 9 M rectionforavoxel-wisesearch.Two-tailedP-values:*P<0.05,†P<0.1.ErrorbarsrepresentSEM. a rc h 2 signalsintheseregions (Gottfriedet al.,2003;Tometal.,2007; earlier.Thisresponsecorrelatedwithtraitanxietyacrossindivid- 0 1 Kable and Glimcher, 2009; Morrison and Salzman, 2010). uals,aresultpresentedanddiscussedinSupplementaryResults. 9 Parameter estimates (betas) were extracted for each region, separately for losses and gains. We identified a greater para- metricresponsetodecreasinglossesrelativetoincreasinggains Amygdalaresponsetoemotionalcuescorrelates in each of these three regions (significant in the ventral stri- positivelywithtraitanxiety atum, t ¼3.52, P¼0.002; and amygdala, t ¼3.16, P¼0.004; (27) (27) marginally significant in the ACC/OFC, t ¼1.95, P¼0.06; In a second analysis, time-locked to the presentation of the (27) Figure 3D–F), consistent with previous reports of loss-biased prime, we contrasted emotion (happy and fearful faces) with value signals in these regions (Tom et al., 2007; Canessa et al., non-emotion (neutral faces and objects) trials. Consistent with 2013;Sokol-Hessneretal.,2013).Weconfirmedtheresultinthe priorreports(Sabatinellietal.,2011),thisrevealedawidespread ventralstriatumbycreatingalossminusgaincontrast(similarto patternofactivation,withwhole-braincorrectedresultsinthefu- Tometal.,2007).Thisvoxel-wisesearchyieldedaclusterinthe siformgyrus,occipitalgyrus,OFC/medialprefrontalcortex,pos- ventral striatum that responded more strongly to decreasing terior cingulate cortex, middle temporal gyrus, anterior insula, losses than increasing gains, overlapping with that reported inferior temporal gyrus extending into frontal gyrus and C.J.Charpentieretal. | 575 D o w n lo a d e d fro m h ttp s ://a c a d e m ic .o u p .c o m /s c a n /a rtic le -a b s Fig.4.Modulationofamygdalaresponsesbyemotionalcues.(A)Aclusterintherightamygdalashowedgreaterresponsetoemotionalvsnon-emotionalprimes(i.e.at tra ainmityiagldsatliamRuOluI.sTphreesceonlotratbiaornr).epArcetsiveanttisonT-ivsalduisepslaanydedvoaxtePls<a0r.e00o1ve(urlnacidororenctthede)a,vbeurtagsueravnivaetodmFWicaElvscoaxnelf-rleovmelaSllV2C8(pPaSrVtCic<ip0a.0n5t)s.in(B)thAemayngadtaolmairceaslplyondseefitnoedemboiltaitoenraall ct/11 primeswaspositivelycorrelatedwithtraitanxiety.(C)Extractingparametricresponsetolossesandtogainsinthisamygdalaclusteratthetimeofgamble,separately /4 foremotionandnoemotiontrials,revealedthattheamygdalaonlytracksdecreasinglossesfollowingemotionalcues.Two-tailedP-values:*P<0.05.Errorbarsrepre- /56 sentSEM. 9/2 3 7 5 amygdala extending into bilateral hippocampus (Supplementary repeated-measuresANOVA.Therewasasignificantinteraction 0 6 TableS3).Givenpreviousliteraturesuggestingthatamygdalare- (F(1,27)¼7.998, P¼0.009, Figure 4C), driven by a positive amyg- 5 b sponses vary with anxiety (Etkin et al., 2004; Stein et al., 2007; dala parametric modulation, on emotional trials only, by y g Sehlmeyeretal.,2011),weextractedsignalfromtherightamygdala decreasinglosses(t ¼3.61,P¼0.001)butnotincreasinggains u (27) e cluster (peak voxel MNI coordinates: 33,(cid:3)1,(cid:3)26; Figure 4A) and (t(27)¼(cid:3)0.88, P¼0.39). However, there were no relationships st o identified a significant positive relationship with trait anxiety with trait anxiety or emotion-elicited change in loss aversion n 2 (r(28)¼0.417,P¼0.027;Figure4B).Thisrelationshipwasnotdriven (addedascovariates:allP>0.25). 9 M by responses to faces in general: while amygdala responses to The earlier modulation of amygdala value signal was spe- a emotional relative to neutral faces correlated with trait anxiety cifictoemotionalcues,ratherthanfacesingeneral.Whenex- rc h (r ¼0.39,P¼0.04),amygdalaresponsestoneutralfacesrelative tractingtheresponsetolossesseparatelyforemotion,neutral 2 (28) 0 toobjectsdidnot(r(28)¼(cid:3)0.042,P¼0.83;marginallysignificantdif- and object conditions, and submitting the resulting betasto a 19 ferencebetweenthetwocorrelations:Steiger’sZ¼1.83,P¼0.067). one-way ANOVA (emotion/neutral/object), there was a signifi- cantmaineffectofemotion(F ¼3.49,P¼0.038).Theamyg- (2,54) dala parametric response to losses was higher on emotional Emotionalcuesmodulatelossaversionsignalsinthe facerelative toneutralfacetrials(t(27)¼3.08,P¼0.005)butnot amygdala onneutralfacerelativetoobjecttrials(t(27)¼(cid:3)0.6,P¼0.55). To test whether the earlier amygdala responses play a role in the observed emotion-driven changes in loss aversion, para- Striatal-amygdalafunctionalconnectivityisassociated metricresponsestolossesandtogains(atthetimeofthegam- withchangesinlossaversion ble)wereextractedseparatelyforemotionalandnon-emotional trials, from the right amygdala cluster identified earlier (re- Are emotionally induced changes in loss aversion driven by spondingtoemotionalprimes—MNI:33,(cid:3)1,(cid:3)26).Theresulting ventralstriatum-amygdalainteractionsduringemotionaldeci- parameter estimates (betas) were submitted to a 2 (cid:3) (gamble sion-making?Totestthishypothesis,weconductedaPPI,with component:loss/gain)by(cid:3)2(prime:emotional/non-emotional) emotion vs non-emotion (at the time of the decision) as the 576 | SocialCognitiveandAffectiveNeuroscience,2016,Vol.11,No.4 modulates functional connectivity changes specifically in re- sponsetoemotionalstimuli. Discussion Howpeoplealtertheirdecisionsinresponsetoemotionalcues, and the neural mechanisms underlying such changes, vary withtheirleveloftraitanxiety.Specifically,werevealthatlow- anxious individuals exhibit increased loss aversion when primed with emotional cues. This was accompanied by and associatedwithincreasedfunctionalcouplingbetweenthestri- atumandamygdala,regionsthathavebeenimplicatedinloss aversion(Tometal.,2007;DeMartinoetal.,2010;Canessaetal., D 2013;Sokol-Hessneretal.,2013). o w Oneofourmainaimswastoestablishwhetherlossaversion n lo would be modulated by emotional cues to a greater extent in a d low-anxious individuals (which would be predicted by greater ed behavioral flexibility) or in high-anxious individuals (which fro would be predicted by emotional hypersensitivity). Our data m h support the first hypothesis. This is consistent with a recent ttp study in which only low-anxious individuals decreased risk- s takingunderstress(Robinsonetal.,2015).Wesuggestthatthis ://a c findingmayreflect anadaptiveabilityofindividualswithlow ad e anxietytodeployharm-avoidancestrategies(avoidingpotential m harm from monetary losses) in response to emotionally ic.o arousingcues.Thiscouldbelinkedwiththereducedsensitivity up to pathological anxiety disorders in this low-anxiety group .co m (Robinson et al., 2013, 2015) and with previous reports of anx- /s iety-related impairments in the ability to adapt behavior to c a changes in the environment (Blanchette and Richards, 2003; n/a FarmerandKashdan,2012;Robinsonetal.,2013,2015;Browning rtic etal.,2015). le -a An alternative interpretation of our findings could be that b s high-anxiousindividualsmayinfactexhibitgreaterattentional tra controlthanlow-anxiousindividualsandbebetteratignoring ct/1 the emotional primes, which are irrelevant to the gambling 1 /4 task.Althoughthisinterpretationisinconsistentwiththethe- /5 Fig.5.Emotionalmodulationofstriatum-amygdalafunctionalconnectivityis ory of impaired attentional control in anxiety (Eysenck et al., 69 relatedtotraitanxietyandlossaversionchange.(A)PPIanalysiswasconducted 2007;Bishop,2009),itremainspossiblethatsuchsuperioratten- /23 toassesshowventralstriatum-amygdalafunctionalconnectivitywasmodu- 7 tionalcontrolisafeatureofnon-clinicalanxiety(i.e.hightrait 5 latedbyemotionalrelativetonon-emotionalcues,usingtheventralstriatum 0 clusterasaseed(Figure3A).(B)ThePPIeffect(i.e.emotion-drivenincreased anxietyinhealthyindividuals;seeRobinsonetal.,2013forare- 65 connectivity)intheamygdalawasnegativelycorrelatedwithtraitanxiety.In view),andthatdysfunctionalattentionalcontrolonlyemerges by other words, low-anxious individuals exhibited increased ventral striatum- in clinical anxiety. Further work is needed to distinguish be- g u amygdalaconnectivityfollowingemotionalrelativetonon-emotionalstimuli. tweentheseexplanations. es (C)Thisincreasedfunctionalconnectivitywasassociatedwithemotion-elicited Recent literature has shown a growing interest in the link t o increaseinlossaversionacrossparticipants. betweenanxietyanddecision-making(forareviewseeHartley n 2 9 andPhelps,2012)andprovidedevidenceforheightenedsensi- M psychological factor (Figure 5A). The ventral striatum cluster tivity to uncertainty and ambiguity in high anxiety. A arc (showninFigure3A)wasdefinedastheseedregion,andbeta recentstudydemonstratedanincreasedframingeffectinhigh- h 2 estimatesforthephysiological(seeSupplementaryResults)and anxious individuals (Xu et al., 2013). According to Prospect 0 1 PPIeffectswereextractedfromtherightamygdalaclusterthat Theory (Kahneman and Tversky, 1979), framing effects on 9 respondedtoemotionalcues(thetargetregion:showninFigure choicecouldbedrivenbothbylossaversionandbydiminishing 4A).Theincreaseinventralstriatum-amygdalaconnectivitybe- sensitivitytochangesinvalueasvalue increases(resultingin tweennon-emotionalandemotionaltrialswasnegativelycor- riskavoidanceinthegaindomainandriskseekingintheloss relatedwithtraitanxiety(r(28)¼(cid:3)0.47,P¼0.012,Figure5B),and domain).Inourdata, wedidnotfindadirectrelationship be- positively correlated with emotion-elicited change in loss tween trait anxiety and loss aversion, suggesting that the aversion (r(28)¼0.42, P¼0.025, Figure 5C). In other words, low- increased framing effect observed in Xu et al. (2013) may be anxiousindividualsexhibitedanincreaseinstriatal-amygdala drivenby stronger diminishingsensitivitytovalue changes in functionalconnectivityonemotionaltrials,whichinturnwas high trait anxious individuals, rather than by increased loss associatedwithemotion-elicitedlossaversion. aversion. Similarly, our data is in line with a recent study in Again,thispatternofresultsheldwhenexaminingtheemo- adolescentsshowingthatclinicallyanxiousandhealthyadoles- tion-drivenchangeinstriatal-amygdalaconnectivity,excluding cents did not differ in their level of loss aversion (Ernst et al., theobjectcondition(negativecorrelationbetweenPPIandtrait 2014).Wenotethatoursampleofhealthyvolunteersincludeda anxiety: r(28)¼(cid:3)0.38, P¼0.047), suggesting that trait anxiety relatively constrained range of anxiety scores; it would C.J.Charpentieretal. | 577 thereforebeinterestingtoexaminelossaversion(anditsmodu- reward-related processes (Everitt and Robbins, 1992; Camara lationbyemotion)inclinicallyanxiousindividuals. et al., 2008) and learning to avoid harmful negative outcomes Ourfindingsrevealedthatpositiveandnegativeemotional (Delgadoetal.,2009).Ourresultsprovideafurtherinsightintoa expressions induced similar changes in decision-making. This potential function of amygdala-striatum interactions, suggest- supports the hypothesis that increased avoidance of potential ingthatchangesinfunctionalconnectivitybetweenthesetwo lossesisrecruitedundergeneralemotionalarousal,ratherthan regions, as opposed to responses in each region per se, may specifically under incidental threat. Recent work, using a drive the tendency toward more conservative decisions under pharmacological manipulation of autonomic arousal, suggests emotionallyarousingconditions. thatarousalresponsesspecificallydrivelossaversion,butnot In summary, we show that incidental emotional cues can riskaversion(Sokol-Hessneretal.,2015).Althoughspeculative, modulatelossaversebehaviorandassociatedneuralresponses, thishypothesisofanarousal-drivenlossaversioncouldexplain sheddinglightonapotentialmechanisticaccountofemotional ourfindingsthat(i)manipulatingemotionalarousalinfluenced influences on economic decisions (Phelps et al., 2014). We lossaversioninthesamedirectionregardlessofvalenceand(ii) speculatethattheamygdalamayintegrateemotionalinforma- D thiseffectwasspecifictolossaversion,withriskaversion(esti- tionaboutexternalcuestogetherwithvalueinformationfrom o w matedseparately)notalteredbyemotionalmanipulation. the ventral striatum, to produce a decision signal. Individual n lo However,duetotimeconstraintsinthescanner,alimitation differences in amygdala-striatal coupling are related to trait a d of our task design was that we were not able to include add- anxiety,possiblyreflectingimprovedfunctionalintegrationbe- e d itionaltrialsnecessarytoestimateriskaversiontogetherwith tween these regions in low-anxious individuals and greater fro lossaversioninthesameutilitymodel.Typicallythiswouldbe flexibility to adapt decision-making in emotionally volatile m donebyaddingchoicesbetweenasuregainandagamble(fea- environments. http turingachanceofahighergainorzero);onsuchgain-onlytrials s only risk aversion (but not loss aversion) should contribute to ://a c safechoices.Withoutthesetrials,ourProspectTheory-derived Acknowledgements ad e model could not distinguish between risk and loss aversion. m Estimatingriskaversionseparately,usinganapproachthathas TheauthorsthankO.J.Robinsonforhelpfulcomments. ic .o beenusedbefore(DeMartinoetal.,2010;Canessaetal.,2013), u p andensuringitdidnotcontributetotheresultsbyaddingitasa .c o covariateinouranalyseswasthebestalternativetoovercome Funding m /s thislimitation. c This work was supported by a University College London a OurfMRIresultsshedlightonapotentialmechanismunder- Grand Challenge award to C.J.C. and J.P.R. and Wellcome n/a lying the emotional modulation of economic behavior, related TrustCareerDevelopmentFellowshiptoT.S. rtic toamygdala-striatum functional connectivity. Consistent with le previousstudies,wefoundthatbothamygdalaandventralstri- -ab s atumtrackedlossesmorestronglythangains(Tometal.,2007; tra DeMartinoetal.,2010;Canessaetal.,2013;Sokol-Hessneretal., Supplementarydata ct/1 2013); however, the modulation of these signals by emotional 1 cues was not associated with emotionally-driven changes in SupplementarydataareavailableatSCANonline. /4/5 lossaversion. Conflictofinterest.Nonedeclared. 69 Instead, we found that the interaction between amygdala /23 7 andventralstriatumwastheneuralmetricmostrelatedtothe 5 References 0 behavioral effects we observed. Emotionally-induced changes 6 5 infunctionalconnectivitybetweenventralstriatumandamyg- Adolphs, R. (2002). Neural systems for recognizing emotion. b y dala correlated negatively with trait anxiety and were associ- CurrentOpinioninNeurobiology,12,169–77. g u atedwithbehavioralchangesinlossaversion,withlow-anxious Beck,A.T.,Erbaugh,J.,Ward,C.H.,Mock,J.,Mendelson,M.(1961). e s individuals showing increased loss aversion together with An inventory for measuring depression. Archives of General t o n increased amygdala-striatum functional connectivity in re- Psychiatry,4,561–71. 2 9 sponse to emotional cues. We note a potential concern that Bishop,S.J.(2007).Neurocognitivemechanismsofanxiety:anin- M amygdala activations, according to a recent study, could be tegrativeaccount.TrendsinCognitiveSciences,11,307–16. a rc drivenbydrainagefromnearbyvesselssuchasthebasalveinof Bishop, S.J. (2009). Trait anxiety and impoverished prefrontal h 2 Rosenthal(BVR:Boubelaetal.,2015).However,inourfMRIdata controlofattention.NatureNeuroscience,12,92–8. 0 1 wedidnotobservepatternsconsistentwithBVRsignals,even Blanchette,I.,Richards,A.(2003).Anxietyandtheinterpretation 9 at a very liberal threshold (data not shown). In particular, our ofambiguousinformation:beyondtheemotion-congruentef- amygdalaclusterrespondingtoemotionalrelativetonon-emo- fect.JournalofExperimentalPsychologyGeneral,132,294–309. tional primes (MNI coordinates [30,(cid:3)1,(cid:3)24]) did not extend to Boubela,R.N.,Kalcher, K.,Huf, W.,et al.(2015).fMRI measure- theposterioramygdala,andinsteadwaslocatedinthelateral mentsofamygdalaactivationareconfoundedbystimuluscor- anterioramygdalaontheoppositesidetotheBVR(MNIcoord- related signal fluctuation in nearby veins draining distant inates [14.6,(cid:3)7.7,(cid:3)15.5] according to Boubela et al., 2015) with brainregions.ScientificReports,5,10499. voxelsadjacenttotheBVRnotactivated.Therefore,webelieve Browning,M.,Behrens,T.E.,Jocham,G.,Reilly,J.X.O.,Bishop,S.J. thatouramygdalaactivationsareunlikelytobeconfoundedby (2015).Anxiousindividualshavedifficultylearningthecausal acontributionfromtheBVR. statistics of aversive environments. Nature Neuroscience, 18, Amygdala-striatumconnectivityiswellestablishedinboth 590–6. animalandhumanfMRIworkandhasbeensuggestedtoplaya Camara,E.,Rodriguez-Fornells,A.,Mu¨nte,T.F.(2008).Functional roleinmotivatedbehavior(Price,2003;ZorrillaandKoob,2013), connectivity of reward processing in the brain. Frontiers in emotional memory (Ferreira et al., 2008; Paz and Pare, 2013), HumanNeuroscience,2,19. 578 | SocialCognitiveandAffectiveNeuroscience,2016,Vol.11,No.4 Canessa,N.,Crespi,C.,Motterlini,M.,etal.(2013).Thefunctional hippocampal expression of mineralocorticoid receptors. and structural neural basis of individual differences in loss NeurobiologyofLearningandMemory,86,150–9. aversion.JournalofNeuroscience,33,14307–17. Kable,J.W.,Glimcher,P.W.(2009).Theneurobiologyofdecision: Chib, V.S., De Martino, B., Shimojo, S., O’Doherty, J.P. (2012). consensusandcontroversy.Neuron,63,733–45. Neural mechanismsunderlying paradoxical performance for Kahneman,D.,Knetsch,J.L.,Thaler,R.H.(1991).Anomalies:the monetary incentives are driven by loss aversion. Neuron, 74, endowment effect, loss aversion, and status quo bias. The 582–94. JournalofEconomicPerspectives,5,193–206. Dalgleish, T. (2004). The emotional brain. Nature Reviews Kahneman,D.,Tversky,A.(1979).Prospecttheory:ananalysisof Neuroscience,5,583–9. decisionunderrisk.Econometrica,47,263–92. DeMartino,B.,Camerer,C.F.,Adolphs,R.(2010).Amygdaladam- Landgraf,R.,Wigger,A.(2002).Highvslowanxiety-relatedbe- age eliminates monetary loss aversion. Proceedings of the havior rats: an animal model of extremes in trait anxiety. NationalAcademyofSciencesoftheUnitedStatesofAmerica,107, BehaviorGenetics,32,301–4. 3788–92. Mathews,A.,Mackintosh,B.(1998).Acognitivemodelofselect- D Delgado,M.R.,Jou,R.L.,Ledoux,J.E.,Phelps,E.A.(2009).Avoiding ive processing in anxiety. Cognitive Therapy and Research, 22, o w negative outcomes: tracking the mechanisms of avoidance 539–60. n learning in humans during fear conditioning. Frontiers in Mobbs,D.,Marchant,J.L.,Hassabis,D.,etal.(2009).Fromthreat loa d BehavioralNeuroscience,3,33. to fear: the neural organization of defensive fear systems in e d Derryberry, D., Reed, M.A. (2002). Anxiety-related attentional humans.JournalofNeuroscience,29,12236–43. fro biases and their regulation by attentional control. Journal of Mobbs,D.,Petrovic,P.,Marchant,J.L.,etal.(2007).Whenfearis m AbnormalPsychology,111,225–36. near:threatimminenceelicitsprefrontal-periaqueductalgray http Ernst,M.,Plate,R.C.,Carlisi,C.O.,Gorodetsky,E.,Goldman,D., shiftsinhumans.Science,317,1079–83. s Pine, D.S. (2014). Loss aversion and 5HTT gene variants in Mobbs,D.,Weiskopf,N.,Lau,H.C.,Featherstone,E.,Dolan,R.J., ://a c adolescent anxiety. Developmental Cognitive Neuroscience, 8, Frith,C.D.(2006).TheKuleshovEffect:theinfluenceofcontext- a d 77–85. ual framing on emotional attributions. Social Cognitive and em Etkin, A., Klemenhagen, K.C., Dudman, J.T.,. (2004). Individual AffectiveNeuroscience,1,95–106. ic .o differencesintraitanxietypredicttheresponseofthebasolat- Morrison, S.E., Salzman, C.D. (2010). Re-valuing the amygdala. u p eral amygdala to unconsciously processed fearful faces. CurrentOpinioninNeurobiology,20,221–30. .c o Neuron,44,1043–55. O’Neill,M.,Schultz,W.(2010).Codingofrewardriskbyorbito- m /s Etkin,A.,Prater,K.E.,Hoeft,F.,Menon,V.,Schatzberg,A.F.(2010). frontalneuronsismostlydistinctfromcodingofrewardvalue. c a Failureofanteriorcingulateactivationandconnectivitywith Neuron,68,789–800. n/a theamygdaladuringimplicitregulationofemotionalprocess- Orne, M.T. (2009). Demand characteristics and the concept of rtic ing in generalized anxiety disorder. American Journal of quasi-controls.In:Rosnow,R.R.R.L.,editor.ArtifactinBehavioral le Psychiatry,167,545–54. Research,110–37.NewYork:OxfordUniversityPress. -ab s Everitt, B.J., Robbins, T.W. (1992). Amygdala-ventral striatal Paz,R.,Pare,D.(2013).Physiologicalbasisforemotionalmodula- tra interactions and reward-related processes. In: Aggleton, J.P., tion of memory circuits by the amygdala. Current Opinion in ct/1 editor.TheAmygdala:NeurobiologicalAspectsofEmotion,Memory, Neurobiology,23,381–6. 1 andMentalDysfunction,401–29.NewYork:Wiley. Pessoa,L.,Adolphs,R.(2010).Emotionprocessingandtheamyg- /4/5 Eysenck,M.W.,Santos,R.,Calvo,M.G.(2007).Anxietyandcogni- dala:froma“lowroad”to“manyroads”ofevaluatingbiolo- 69 tive performance: attentional control theory. Emotion, 7, gicalsignificance.NatureReviewsNeuroscience,11,773–82. /2 3 336–53. Phelps,E.A.(2006).Emotionandcognition:insightsfromstudies 75 0 Farmer, A.S., Kashdan, T.B. (2012). Social anxiety and emotion ofthehumanamygdala.AnnualReviewofPsychology,57,27–53. 6 5 regulationindailylife:spillovereffectsonpositiveandnega- Phelps, E.A., Lempert, K.M., Sokol-Hessner, P. (2014). Emotion b y tivesocialevents.CognitiveBehaviourTherapy,41,152–62. and decision making: multiple modulatory neural circuits. g u Ferreira, T.L., Shammah-Lagnado, S.J., Bueno, O.F.A., Moreira, AnnualReviewofNeuroscience,37,263–89. e s K.M.,Fornari,R.V.,Oliveira,M.G.M.(2008).Theindirectamyg- Price,J.L.(2003).Comparativeaspectsofamygdalaconnectivity. t o n dala-dorsalstriatumpathwaymediatesconditionedfreezing: AnnalsoftheNewYorkAcademyofSciences,985,50–8. 2 9 insights on emotional memory networks. Neuroscience, 153, Robinson,O.J.,Bond,R.L.,Roiser,J.P.(2015).Theimpactofstress M 84–94. on financial decision-making varies as a function of depres- a rc Fox, E., Mathews, A., Calder, A.J., Yiend, J. (2007). Anxiety and sionandanxietysymptoms.PeerJ,3,e770. h 2 sensitivity to gaze direction in emotionally expressive faces. Robinson,O.J.,Vytal,K.,Cornwell,B.R.,Grillon,C.(2013).Theim- 0 1 Emotion,7,478–86. pact of anxiety upon cognition: perspectives from human 9 Glascher,J.,Adolphs,R.(2003).Processingofthearousalofsub- threatofshockstudies.FrontiersinHumanNeuroscience,7,203. liminal and supraliminal emotional stimuli by the human Sabatinelli,D.,Fortune,E.E.,Li,Q.,etal.(2011).Emotionalpercep- amygdala.JournalofNeuroscience,23,10274–82. tion: meta-analyses of face and natural scene processing. Gottfried,J.A.,O’Doherty,J.,Dolan,R.J.(2003).Valueinhuman Neuroimage,54,2524–33. amygdalaandorbitofrontalcortex.Science,301,1104–7. Schwartz, G. (1978). Estimating the dimension of a model. HardieB.G.S.,Johnson,E.J.,Fader,P.S.(1993).Modelinglossaver- AnnualsofStatistics,5,461–4. sion and reference dependence effects on brand choice. Sehlmeyer,C.,Dannlowski,U.,Scho¨ning,S.,etal.(2011).Neural MarketingScience,12,378–94. correlates of trait anxiety in fear extinction. Psychological Hartley,C.A.,Phelps,E.A.(2012).Anxietyanddecision-making. Medicine,41,789–98. BiologicalPsychiatry,72,113–8. Sheehan, D.V., Lecrubier, Y., Sheehan, K.H., et al. (1998). The Herrero,A.I.,Sandi,C.,Venero,C.(2006).Individualdifferences Mini-International Neuropsychiatric Interview (M.I.N.I.): the in anxiety trait are related to spatial learning abilities and development and validation of a structured diagnostic