Neural mechanisms of motivated forgetting

Neural mechanisms of motivated forgetting

Michael C. Anderson

and Simon Hanslmayr

1MRCCognitionandBrainSciencesUnit,UniversityofCambridge,Cambridge,UK

2BehaviouralandClinicalNeuroscienceInstitute,UniversityofCambridge,Cambridge,UK

3SchoolofPsychology,UniversityofBirmingham,Birmingham,UK

4DepartmentofPsychology–Zukunftskolleg,UniversityofKonstanz,Konstanz,Germany

Not all memories are equally welcome in awareness.

Peoplelimitthetimetheyspendthinkingaboutunpleas- antexperiences,aprocessthatbeginsduringencoding, butthatcontinueswhencueslaterremindsomeoneof thememory.Here,wereviewtheemergingbehavioural andneuroimagingevidencethatsuppressingawareness of anunwelcome memory,at encodingor retrieval,is achieved by inhibitory control processes mediated by thelateralprefrontalcortex.Thesemechanismsinteract with neural structures that represent experiences in memory,disruptingtracesthatsupportretention.Thus, mechanisms engaged to regulate momentary aware- nessintroduce lastingbiases inwhich experiences re- main accessible. We argue that theories of forgetting thatneglectthemotivatedcontrolofawarenessomita powerfulforceshapingtheretentionofourpast.

Aneglectedforcethatshapesretention

Overthe past century, memory research has focused on passivefactorsthat makeusforget.Forgetting hasbeen proposedtoresultfromthedecayofmemoriesovertime, the accumulation of similar interfering experiences in memory, and changes in physical context that make it hardertorecallthepast[1].This historicalemphasison passivefactorsfitsthecommonassumptionthatforgetting isanegativeoutcomeand,thus,anyprocessunderlyingit must happen involuntarily. Although forgetting is often negative,thisemphasisneglectsafundamentalfeatureof humanexistence:notallexperiencesarepleasant.When reminded of negative events, we are not well disposed towards them andwe deliberately limit their tenure in awareness. This process is familiar to most people; a reminder evokes a brief flash of memory and feeling, abruptlyfollowedbyeffortstoexcludetheunwantedmem- oryfromawareness.Wedothistopreserveouremotional state,toprotectoursenseofself,andsometimessimplyto concentrate on what needs to be done. Therefore, any scientific theoryofforgettingmust include anaccount of theconsiderablemotivationalforcesthatshaperetention.

Here,wereviewthegrowingresearchonneuralmech- anisms underlying motivated forgetting. The term

‘motivatedforgetting’hererefers toincreased forgetting arising from active processes that down-prioritise un- wanted experiencesin serviceof creating or sustaining anemotionalorcognitivestate.Forexample,tosustain positiveemotionsorconcentration,beliefinsomestateof affairs, confidence, or optimism, it may be necessaryto reduceaccessibilityofexperiencesthatunderminethose states.Here, wefocusonneuralevidence forthe roleof inhibitorycontrolprocessesinthevoluntaryinterruption ofmnemonicprocessing.Acoreclaimisthattheseinhibi- tory control processes, widely studied in psychology and cognitive neuroscience, can be targeted flexibly at differentstagesofmnemonicprocessingandat different typesofrepresentationtomodulatethestateoftracesin memory.

Insupportofthisview,wereviewevidencethatinhibi- tion can be engaged either during memory encoding or retrieval tolimit retentionofunwanted memories. Stop- ping encoding may disrupt the consolidation of traces already formed, and also prevent further reflection on the experiencethat wouldenhanceits longevity.Bycon- trast, stopping retrieval disrupts the automatic progres- sion from cues to an associated memory, the persisting effectsofwhichinfluencewhethertheexperienceremains accessible.Bothencodingandretrievalstoppingterminate anunfoldingmnemonicprocesssothatanexperiencecan be excluded from conscious awareness. Through these effortstoterminateawareness, attentionalcontrol inter- actswithtracesinepisodicmemorytoshapewhatwedo anddonotrememberofourpast.

Inhibitorycontrolatencoding

An effective way of keeping an unwanted memory from beingretrievedinthefutureistodisruptandtruncateits encoding. Theseprocesses areinvestigatedwith directed forgettingparadigms,inwhichparticipantsreceiveacue toforgetinformationthattheyjustacquired[2].Hundreds of studies conducted over the past 50 years reveal that humans can readily implement such forgetting instruc- tions,demonstratingthatmotivationindeedshapesencod- ing.Inhibitionhasbeenproposedtohavearoleinstopping encodingprocessesintheseprocedures,althoughpassive factorsalsoarelikelytohavearole(e.g.,[3]).Wefocushere onevidenceindicatingadistinctcontributionofinhibitory control in actively limiting the encoding of unwanted experience. This evidence has been collected with the Feature Review

1364-6613/

ß2014TheAuthors.PublishedbyElsevierLtd.Thisisanopenaccessarticleunder theCCBYlicense(http://creativecommons.org/licenses/by/3.0/).http://dx.doi.org/

10.1016/j.tics.2014.03.002

Correspondingauthor:Anderson,M.C. (michael.anderson@mrc-cbu.cam.ac.uk).

TrendsinCognitiveSciences,June2014,Vol.18,No.6 279

Konstanzer Online-Publikations-System (KOPS) URL: http://nbn-resolving.de/urn:nbn:de:bsz:352-0-263779

item-method [4] and list-method [5] directed forgetting procedures(Box1).

Item-methoddirectedforgetting

Item-method directed forgetting has a long tradition in cognitivepsychology[4].Thiseffectisrobust,asreflected by the range of conditions under which it has been reported,includingbothexplicitandimplicitmemorytests [6,7]. Item-method directed forgetting usually has been explained in terms of selective rehearsal (see Glossary) accordingtowhichto-be-forgottenitemsaresparedfrom further processing andare subject to passive forgetting, whereas to-be-remembered items are actively rehearsed [2].Interestingly,theoccurrenceofitem-methoddirected forgettinginrecognitiontestshasbeenused asan argu- ment for passive, noninhibitory explanations, because somehavearguedthatinhibitionshouldonlytemporarily reducetheaccessibilityoftheaffecteditemsand,therefore, itshouldbepossibletoreleasetheseitemsfrominhibition later[2].

Although selective rehearsal is a common interpreta- tionof item-method directed forgetting [2], recent beha- vioural and neural evidence indicates that inhibitory controloverepisodicencodingmayhaveabiggerrolethan has been acknowledged. For example, the selective re- hearsal accountemphasisesprocesses actingonto-be-re- membered items, which are rehearsed more extensively andelaboratelywhenthecuetorememberisgiven.There- fore,thesystemshouldexperiencemorecognitiveloadin theremembercomparedtothe‘forget’condition,inwhich peoplecansimplydroptheto-be-forgottenitemfromwork- ingmemory.Thispredictionwastestedinseveralexperi- mentsin whichparticipantsperformed asecondarytask after the remember and/or forget cue was given [8,9].

However,contrarytotheselectiverehearsalaccount,the forget condition was more effortful than the remember condition,asreflectedbyslowerreactiontimestoperform thesecondarytaskduringexecutionoftheforgetinstruc- tion.Moreover,stoppingamotorresponseafterthecueis moresuccessfulintheforgetcomparedwiththeremember condition [9], suggestingthat forget cues trigger similar inhibitorymechanismstothoseengagedwhenstoppinga motor action [10]. However, further clarification of this possibilityisneeded[9].Theseresultsclearlyimplythatan active process contributes to item-method directed for- getting[11],andraisethepossibilitythatitisinhibitory innature.Thispossibilityisconsistentwithevidencethat directedforgettingcuesleadtotheremovalofitemsfrom workingmemoryandnotmerelytopassivedecay[12,13].

Several recent functional (f)MRI studies support the hypothesisthat item-method directedforgettingengages anactiveprocessthatinhibits ongoingencoding[14–18].

These studies consistently indicate that attempting to forget a recent item engages prefrontal and parietal regions,suggestingthat forgettingiseffortful, consistent with behavioural findings [15,16,18]. The right superior andmiddlefrontalgyrus(approximatelyBA9/10),andthe right inferior parietal lobe (approximately BA 40) are consistently more active during intentional forgetting (to-be-forgotten items that are actually forgotten) com- paredwithincidentalforgetting(to-be-remembereditems Glossary

Accessibilityversusavailability::atheoreticaldistinctiononwhy memory retrievalcanfail.Wemayfailtoretrievememoriesbecausewedonotaccessa storedmemory(i.e.,accessibility)orbecausethememoryisnotavailable anymoreinthesystem(i.e.,availability).

Brainoscillations::regularfluctuationsvisibleintheEEGand/ormagnetoen- cephalogram(MEG),mostlikelyreflectingsummatedexcitatoryandinhibitory postsynapticpotentials.Brainoscillationsoccuratdifferentdistinctfrequen- cies (up to 150Hz) and have an important rolein synchronising neural assemblies[104]andshapingsynapticplasticity[35].

Cue independence: : the tendency for suppression-induced forgetting to generalisetonoveltestcuesotherthantheoneoriginallyusedasacueduring retrievalsuppression.

Directsuppression::amethodoflimitingawarenessofanunwantedmemory whenareminderappearsinwhichapersondisengagestheretrievalprocessto either prevent the memory for coming to mind, or to limit its time in awareness.Inhibitionisthoughttobeakeyprocessinoverridingthenatural operationoftheretrievalmechanism.

Effectiveconnectivityanalysis::aformofconnectivityanalysisthatallowsone toinfernotonlythatneuralactivityintwodistinctregionsisrelated(statistically), butalsothedirectionalnatureofthatrelation.Effectiveconnectivityanalyses, suchasdynamiccausalmodelling,permitcausalinferencesabouttheinfluence ofonebrainregiononanotherinconditionsofinterest.

Episodiccontext: :thespatiotemporal environmentinwhich astimulusis encountered.Therepresentationofthiscontextanditsassociationtoastimulus formafundamentalfeatureofepisodicmemoryofthestimulus.Contextcanalso refertointernalstatesthatgetassociatedtoastimulus(e.g.,moodorincidental thoughts),whichissometimesreferredtoas‘mentalcontext’.

Event-relatedpotential(ERP)::atime-varyingbrainsignalwithpositiveand negative deflections (so-called‘components’), obtainedby averagingover severalEEGsegmentscorrespondingtoataskorstimulus.

Fading affect bias: : the documented tendency for negative emotions associated with personal experiences to decline more quickly over time comparedwithpositiveemotions.

Inhibitorycontrol::acontrolprocessthatdownregulatesactivityofinterfering orotherwiseunwantedrepresentationsintheserviceofacurrenttaskorgoal, reducingtheirinfluenceoncognitionandbehaviour.

Latepositivecomponent(LPC)::apositiveERPcomponentrelatedtoepisodic retrieval.Duringaretrievaltask,theLPCemergesapproximately400–800ms afterstimulusonset,ismaximaloverparietalrecordingsites,andisassumed toreflectretrievalofcontextualdetailsofthestudyepisode(i.e.,recollection [105]).

Long-rangesynchrony: :synchronisationbetween distant cellpopulations separated by several centimetres (e.g., frontal and parietal). Long-range synchronyisusuallyestimatedbasedontheco-variationofoscillatoryphase betweentworecordingcites.

Mnemicneglect::thetendencyforpeopletohaveahigherrateofforgetting for negative feedback about themselves and their performance, than for neutralorpositivefeedback,evenwhenencodingtimeismatched.

N2: :a negative ERP component related to cognitive control, and often associated with motor response inhibition. The N2 refers to enhanced frontocentralnegativitytypicallyapproximately150–400ms.

Repetitionpriming::improvedperformanceinprocessingastimulusarising frompriorexposuretothestimulus.

Repetitionsuppression::thefindingthatrepetitionsofastimuluselicitless neuralactivityinareasinvolvedinprocessingthestimulus,comparedwith nonrepeatedstimuli,takentobeamarkerofmemoryforthestimulus.

Repetitivetranscranialmagneticstimulation(rTMS)::atechniquecommonly usedtostimulateaspecificbrainareabyapplyingatime-varyingmagnetic fieldthatinduceselectriccurrentflowinthebrain.

Selectiverehearsal::apassive,noninhibitoryaccountusedto explainthe reduced memoryperformancefor to-be-forgottenitems,relative to to-be- remembereditems.

Sociallysharedretrieval-inducedforgetting::whenapersonisrecountingan experiencesharedbylisteners,thetendencyforthelistenerstolaterforget(at a higher rate) details not recounted by the speaker. The higherrate of forgettingisthoughttoarisefromlistenerscovertlyretrievingtheexperience as it is being recounted and, consequently, inducing retrieval-induced forgettingonnonretrievedknowledge.

Suppression-inducedforgetting::intheTNTprocedure,impairedrecallofno- thinkitems,comparedwithbaselinememoriesthatareneitherretrievednor suppressed.

Think/no-thinkprocedure(TNT)::themainprocedureusedtostudyretrieval suppression,wherebypeoplearerepeatedlypromptedwithcuestomemories andaskedtoeitherretrieve(think)thememory,ortostopitsretrieval(no- think),withtheresultthatsuppresseditemsaremorepoorlyrecalledonlater tests.

Thoughtsubstitution::a methodof preventingretrievalof an unwanted memorywhenareminderappearsinwhichapersongeneratesalternative thoughtsassociatedtotheremindertooccupyawareness.

280

thatareforgotten (Figure 1A)[15,16,18].Although these findingssuggestthatintentionalforgettingrecruitsaddi- tionalprocesses beyond thoseassociated with incidental forgetting,these activations do notspecify the nature of those processes. For example, activations during forget trialsmightreflectengagement ofthe defaultmode net- work,whichischaracterisedbypositivebloodoxygenation level-dependent (BOLD) correlations between superior prefrontalandparietalcortexduringrest[19].Thus,these

findingsmaysimplyreflectagreaterincidenceofpassive rest during forgettrialscompared withremembertrials.

However, speakingagainstthisview,connectivityanaly- sesshowthatactivityintherightdorsolateralprefrontal cortex (DLPFC) during forget trials predicts decreased activityinthelefthippocampus,especiallyduringsuccess- fulintentionalforgetting[18].Thislatterresultisincom- patiblewith thedefaultmodenetworkhypothesis,which predicts the opposite (positive) connectivity pattern Box1.Itemandlist-methoddirectedforgetting

Instudiesofdirectedforgetting,twoproceduresaregenerallyused:the item-method and the list-method (see [1]). These paradigms are illustratedinFigureI.AsillustratedinFigureIA,initem-methoddirected forgetting,participants studyitemsoneata time,andeach item is followedbyaforget(F)orremember(R)instruction.Later,memoryfor all items is tested.As shown in Figure IB,in list-method directed forgetting,aentirelistofitemsisfirststudied,followedbyaForR instruction. A second list is then studied, usually followed by a R

instruction.Attheend,arecalltestoccurs.FigureICcomparesthetypical behaviouralresultsobtainedinitem-method(FigureICi)andlist-method (FigureICii)directedforgetting.FigureICiishowsthetwofoldeffectof the forget cueon the recalltest,withforgettingoflist-1 itemsand enhancementoflist-2items.Inbothparadigms,participantsdonot knowinadvancewhethertheyshouldforgetorremembertherespective item.Thus,thecontrolprocessesmediatingtheseeffectsmustacton memoryrepresentationsandnottheinitialperceptionofanevent.

(A)

Encoding Leaf

Test

List 1 List 2 Memory Memory

Item-method List-method Behavioural results

Item-method Test

Cue Leaf

Leaf Arm

Arm new?^Old/

Old/

new?

Old/

new?

Arm

Beard Beard

Knee Winter

Gold ...

... ...

F F

R R

Encoding (list 2) Encoding

(list 1)

List-method

(B) (C)

(i)

(ii)

TRENDS in Cognitive Sciences FigureI.Theitemandlist-methodsforstudyingdirectedforgetting,alongwiththetypicalpatternoffindings(forrealexamples,see[8]and[5,25],respectively).

(A) (B)

(i)

(ii)

Item-method DF forgeng success

fMRI EEG rTMS

DLPFC (fMRI)

Neural sync. (EEG)

Forgeng

% Difference

DLPFC 40

30 20 10 0

Key:

Control

Enhancement DLPFC

(∼BA 9)

(C)

TRENDS in Cognitive Sciences

Figure1.Neuralcorrelatesofdirectedforgetting(DF).(A)Anactivationmapofarecentitem-methoddirectedforgettingfunctional(f)MRIstudy[18].Redareasillustrate significantvoxels(P<0.005)indicatinggreateractivityforto-be-forgottenitemsthatareactuallyforgottencomparedwithto-be-remembereditemsthatareremembered.

(B,C)Theresultsofamultimodallist-methodDFexperiment[32].(B)Forgetinstructionswereassociatedwithincreasedbloodoxygenationlevel-dependent(BOLD)signal intheleftdorsolateralprefrontalcortex(DLPFC)andreducedalpha/betalong-rangesynchrony[11–18Hz(i)],whichwerenegativelycorrelatedonasingletriallevel(ii).(C) StimulatingtheDLPFCwithrepetitivetranscranialmagneticstimulation(rTMS;1Hz)selectivelyincreasedlist-1forgetting,withoutaffectinglist-2enhancement.Adapted, withpermission,from[32](B,C).Abbreviation:EEG,electroencephalogram.

281

betweenDLPFCandthemedialtemporallobe(MTL)[20].

Rather, negativeconnectivitybetween rightDLPFC and hippocampus suggests that the right prefrontal cortex exerts inhibitory control over the encoding activity in the MTL [21], similar to that observed during retrieval (see‘Neural BasisofRetrievalSuppression’).Oneplausi- ble hypothesis is that the active forgetting mechanism implicated by behavioural studies [15,16,18]may reflect theactionofthisfrontohippocampalmodulatorysystem.

Theneuralcorrelatesofitem-methoddirectedforgetting have also been studied with intracranial event-related potentials, providing information about the temporal dy- namicsoftheforgettingmechanismwithintheMTL[22].

Thisstudyfoundthatforgetcuesthatcauselaterforgetting eliciteddecreasednegativityintheanterior hippocampus comparedwithremembercuesthatledtoforgetting.Nota- bly, enhanced negativity in the hippocampus at approxi- mately 500ms is usually related tosuccessful encoding.

These authors further found that forget cues triggered sustained positivity in the rhinal cortex, an interfacing structurebetweenthe cortexand hippocampus.Together with scalpevent-relatedpotential(ERP)studies,showing sustained prefrontal positivity after the forget cue [23], localisedtotherightDLPFC[24],thesestudiesconverge with fMRIdatatosuggestthatitem-method directedfor- gettingrecruitsarightprefrontal–MTLnetworktotermi- nate episodicencoding processes.Together,thesestudies questionapurelypassivebasedviewofitem-methoddirect- edforgetting,which hasbeen thepopularaccount among experimentalpsychologists.

List-methoddirectedforgetting

Sometimes, we may wish to forgeta set of events that is extendedintime(e.g.,arecentdoctor’svisit,ordisputewith anunpleasantacquaintance).Thissituationismodelledby thelist-methodofdirectedforgetting(Box1).Atypicalexper- imentcomprisestwolists(e.g.,10–20itemsineachlist),with aforgetorremembercuegivenafterthefirstlist[2,25].After encodingthesecondlist,abriefdistractingtaskfollowsand thenrecallistested.Onthisfinaltest,peopletypicallyrecall the first list more poorly when it is followed by a forget, comparedwitharemember,instruction.Interestingly,people recall theitems followingaforgetcue betterthantheydo itemsstudiedafteraremembercue(Box1,Figure1C)(e.g., [25]).Thesecomplementaryeffectsarereferredtoaslist-1 forgettingandlist-2enhancement.Theseeffectsariseonfree recall, cued recall, and recognition tests,although, in the latter case, deficits areoften restricted to source memory.

List-method directed forgetting effects have also been ob- servedinautobiographicalmemory[26,27].Poorerrecallof theto-be-forgottenlistisbelievedtoreflectreducedaccessi- bilityratherthanreducedavailabilityoftheforgottenmate- rial[2,28,29].Bothactive[5,30]andpassivemechanismshave alsobeenproposedforthisphenomenon[3,31].

As with item-method directed forgetting, imaging re- search withthelist method indicatesthat instructions to forget trigger an active process that disrupts mnemonic activity.Forexample,twostudiesexaminedtheneurophys- iologicalmechanismsofdirectedforgettingbyfocussingon brainoscillations[30,32].Priorworkestablishedthatmem- oryformationistypicallyaccompaniedbyincreasedlarge

scalesynchrony,aneuralmarkerthoughttoreflectupregu- latedsynapticplasticity[33–35].Strikingly,cuingpeopleto forgetajust-studiedlistdecreasedthelarge-scalesynchrony inawidespreadcorticalnetworkinthealpha/betafrequency range. Individual differences in this effect predicted for- gettingofto-be-forgottenitems[30],suggestingthatdecreas- ingsynchronydisruptsneuralprocessesthatwouldimprove retention. This finding was replicated in a multimodal electroencephalography (EEG)–fMRI study[32], inwhich itwasfoundtobeassociatedwithincreasedBOLDsignalin the left DLPFC (Figure 1B). Following this discovery, a combined EEG–repetitive transcranial magnetic stimula- tion (rTMS) experiment demonstrated that stimulating theleftDLPFCwithrTMSduringaforgetinstructionalso reducedneuralsynchrony,significantlyincreasingdirected forgetting(Figure1C).EnhancedforgettingfollowingrTMS indicates that stimulation facilitated processes needed to implement directed forgetting, consistent with findings showingthatrTMScanenhance,ratherthandisruptpro- cessing(see[32]foradetaileddiscussionoftheenhancing effectsofslowrTMSonforgetting).Thisfindingsupportsa causal role of frontally driven processes in inducing for- gettingeffectsandcomplementsworkshowingthatprefron- tallesionsdisruptlist-methoddirectedforgetting[36].

Increased activation of the DLPFC together with de- creasedneural synchronysuggests that anactive control processcontributes todirectedforgetting.However,these findingsdonotspecifythatthisactiveprocessnecessarily engagesinhibitorycontrol.Forexample,priorstudieshave highlightedtheimportanceoftheDLPFCintaskswitching (reviewedin[37]).Therefore, DLPFCinvolvementduring the forget instruction might simply reflect a voluntarily inducedtaskswitchthatstopsrehearsal,andnotinhibition.

However,thisoverlapbetweendirectedforgettingandtask- switchingactivationsmaybedrivenbyinhibitoryprocesses involvedduringtaskswitching,asnumerousstudiesindi- cate [38]. Furthermore, in EEG studies, task switching typically inducesa pronounced increase infrontoparietal theta long-rangesynchrony [39],in stark contrastto the decreasesinalpha/betalong-rangesynchronyobservedin list-method directed forgetting. Although these findings suggest that directed forgetting activations are unlikely to arise from task switching, the relations between the processesinthetwotasksmeritsfurtherexploration.

Theimprovedrecalloflist-2thataccompaniesthefor- gettingoflist-1itemsmightsuggest asinglemechanism that enhances list-2 encoding by reducing interference fromlist-1items[2].However,therearereasonstodoubt this.First,forgettingandenhancementareoftenuncorre- lated(e.g.,[30,40,41]).Second,list-1forgettingoftencanbe modulated independently of list-2 enhancement [32,42].

Third,whereasalllist-1itemssufferforgetting,irrespec- tive of their serial position, the enhancement of list-2 appearstobedrivenbythefirstfewitems[42].Thelatter result fits electrophysiological data suggesting that the forgetcue enhancessubsequent encodingbecauseitacts like a ‘reset button’ that frees cognitive resources and allowsafreshencodingstart.Thisassumptionisreflected inEEGworkshowingthatoscillatorymarkersofencoding exhaustion,whichgraduallyincreasewiththenumberof encodeditems,areresetbytheforgetcue[42].Thislatter

282

effectismostevidentinalphaoscillatoryamplitude,which usuallydecreasesduringmemoryencoding[43,44].Thus, forgetting and enhancement in list-method directed for- getting appear toreflect different processes that canbe dissociatedonacognitiveandneurallevel.

The list anditem methods differ in the targetof for- getting. Whereas the item method targets individual items,thelistmethodtypicallydirectspeopletoforget a setofitemsdefinedbytemporalcontext(i.e.,‘theprevious list’). This broader targeting may be implemented by directinginhibitionatrepresentationsoftemporalcontext rather than individual items. Consistent with this, list- methoddirectedforgettinginducesashiftawayfromthe mentalcontextofthefirstlist,andthiscontextshiftmay makeithardertorecalllist-1items.Forexample,aforget instruction induces forgetting effects similar to those caused by other instructions designed merely to shift mentalcontextawayfromthefirstlist,withoutinstructing peopletoforget [3].Giventhat similar forgettingcanbe induced without instructing people toforget, some have arguedthatdirectedforgettingneednotreflectinhibitory control [3]. However, an alternative possibility is that directed forgetting instructions achieve context shifts in a mechanistically distinct way, by engaging inhibitory control to force a shift in context. Consistent with this possibility,directed forgetting and ‘mental context shift’

instructions appear to be mediated by different neural processes. Whereas mental context shift instructions mainlyaffectlocalalphaandthetasynchrony[45,46],only directedforgettingdisruptslong-range alpha/betaneural synchrony [30,32]. Combined with evidence for a causal roleofprefrontalcortexininducingthesechangesinboth long-range synchrony and forgetting [32], these findings suggest that an active inhibitory process disrupts list-1 contextindirectedforgetting.However,thesedissociations arebasedonbetween-studycomparisonsbecausenostudy has yet directly contrastedEEG synchronypatterns be- tweendirectedforgettingandmentalcontextchange.

Although inhibitionmaytypicallybe targetedat tem- poralcontextinlist-methoddirectedforgetting,othertar- gets are possible. For example, recent research has examined whether directed forgetting can be targeted selectivelyatsome,butnotall,ofthepre-cueinformation [47–51].Threeofthesestudiesdemonstratedthisispossi- ble [48–50].For instance,onestudy demonstratedselec- tivity of directed forgetting in three experiments using visual (colours) andauditory (words spokenby a female versusmalevoice)material[49].Inoneexperiment,list-1 items were spoken either by a male or a female voice, alternatingonanitem-by-itembasis,andtheparticipants wereabletoforgetitemsselectivelybasedonthegenderof thespeaker.However,somestudiesfailedtofindselective directedforgetting [47,51].The reasons for thesediscre- pancies are currently unknown. If directed forgetting is selective,itsuggeststhat theinhibitoryprocesses target dimensionsother thantemporalcontext.

Toconclude,behaviouralandneurophysiologicalstudies indicatethatencodingcanbedisruptedortruncatedbyan activeinhibitorycontrolmechanismthatlimitstherepre- sentationofanexperienceinlong-termmemory.Similarto inhibitorycontrolinthemotorsystem,wherehigher-order

controlregionsintheprefrontalcortexsuppressactivityin lower-order motor regions to stop a movement [52], the prefrontal cortex targets memory-related structures in theMTL[18,22].Theseprocessesreduceencodingactivity anddownregulatelong-rangeneuralsynchrony[32]todis- rupttheformationofunwantedmemories.

Inhibitorycontrolatretrieval

Unwantedexperiencesareoftenstoredinmemory,despite efforts to limit encoding. When this happens, limiting awareness becomes a problem of controlling retrieval.

Retrieval can ofcoursebe prevented by avoidingremin- ders, which is acommon behaviour after an unpleasant event.However,whenunwelcomeremindersoccur,people often try to exclude the unwanted memory from aware- ness.Stoppingretrievalofanunwantedmemoryisknown as‘retrievalsuppression’,aprocessthatengagesresponse overridemechanismsformallysimilartostoppingareflex- ive motor action [53,54]. Retrieval suppression is often studied with the think/no-think paradigm (TNT) [53], whichmimicssituationswhenwetry tosuppressunwel- comeremindings(Box2).

Behaviouraleffectsofretrievalsuppression

The TNTprocedure consistentlyshows thatpeople can limit retrieval [53–55]. Two main findings support this conclusion. First, suppressing retrieval consistently abolishes the benefits of reminders on memory, as reflectedinthesizeabledifferenceinfinalrecallbetween ThinkandNo-Thinkitems.Thus,ataminimum,suppres- singretrievalreducesthefacilitationthatretrievedmem- ories usually enjoy. Second,suppressing retrieval often reducesrecallforNo-Thinkitemsbelowthatobservedfor baseline items, a phenomenon known as ‘suppression- induced forgetting’. Suppression-induced forgetting is

Box2.TheTNTparadigm

The TNTprocedure mimics situations in which weencounter a remindertoamemoryweprefernottothinkabout,andtrytokeep the memoryout ofmind [53]. Tocreate reminders, participants studycue–target pairs(e.g., wordpairs, orpicturepairs,such as

‘ordealroach’)andarethentrainedtorecalltheseconditem(roach) ofthepairwhenevertheyencounterthefirst(ordeal)asareminder.

Participantsthenparticipatein theTNTphase,inwhich theyare askedtoexertcontroloverretrieval.Oneachtrial,remindersfrom thepairs appearingreen orred;when thereminder appearsin green, participants are to recall the response, whereas, for red reminders,participantsareaskedtoavoidretrievingtheresponse, preventingitfromenteringawareness.Thekeyquestionconcerns whetherpeoplecanrecruitinhibitiontopreventthememoryfrom intrudingintoconsciousness,andwhetherdoingsodisruptslater retentionoftheunwantedmemory.

To measure the effects of retrieval suppression, participants receiveafinaltestinwhichtheyaregiveneachreminderandare asked torecalltheassociated response.Memoryperformance is compared between items that participants suppressed (no-think trials),itemsthattheyretrieved(thinktrials),anditems thatthey studied,butneithersuppressednorretrievedduringtheTNTphase (baselinetrials).Comparingfinalrecallofno-thinkitems toeither think or baseline items indicates whether retrieval suppression affects retention. Comparing no-think tobaseline items is more appropriate when trying to establish that suppression makes memoryworse,asopposedtomerelypreventingmemoryimprove- mentthatmightarisefromrepeatedreminders.

283

especially informative because it indicates that, during retrieval suppression, reminders do not merely fail to enhanceretention, buttriggerprocessesthatimpairac- cesstotheunwantedmemory.Thesefindingshighlighta centralthemeofthisarticle:thatone’sdispositiontowards a memory affects how well it is retained. Reminders enhanceretentionwhenapersoniswelldisposedtowards amemory,butwhenonehasmotivationsforexcludinga memory from awareness,retrieval canbe stopped, pre- ventingthebenefitsofretrievalandfurtherdisruptingthe memory.Thesesymmetricaleffectsofremindersindicate ahighlevelofcontrolover theretrievalprocess,control thatshapesaccessibility.

Much is now known about suppression-induced for- getting.First,forgettingincreaseswiththenumberoftimes amemoryissuppressed[53,55–60],indicatingthatsuppres- sionyieldscumulativeeffects.Theforgettingeffectcanbe furtherincreasedifparticipantsaregiventimetoprepare forsuppression[61],indicatingtheimportanceofanticipa- toryprocesses.Suppression-inducedforgettingariseswith manystimuli,includingwordpairs,face–scenepairs[62–

65], face–word pairs [58], word–object pairs [66,67], and pairscomprisingwordsandnonsenseshapes[68].Suppres- sion-inducedforgettinghasevenbeenobservedwithauto- biographical experiences [69–71], although suppression impairsmemoryforeventdetailsmorethanaccesstothe eventitself.Somestudieshavereportedalackofsuppres- sion-inducedforgettingwhenitmightotherwisebeexpected (see [55] fora detailed discussion with hypotheses). For- getting effectsoccurwhetherthe memoryisa neutralor negatively valenced word or scene [59,60,62–65,72–79], althoughitremainsunclear whether forgettingincreases [60,62], decreases [65,80], or is unaffected [74,75] with negative, compared with neutral valence. Although few studieshaveexaminedhowlongforgettinglasts,onestudy foundthatasinglesuppressionsessionproducesforgetting thatlastsatleast24h[81],withotherevidencesuggesting that it may dissipate after a week [70,80]. Suppression- induced forgetting is diminished in young children [82]

andolderadults[56],twopopulationshypothesisedtohave deficientinhibitorycontrolfunction.Interestingly,individ- ualdifferencesinparticipants’perceptionsoftheirabilityto controlunwantedthoughtsindailylifepredictsuppression- inducedforgettingofaversivescenes[77].

Suppression-inducedforgettingexhibitspropertiescon- sistentwitharoleofinhibitorycontrol.Forexample,the forgettingoftengeneralisestonoveltestcues.Forinstance, after studying ordeal–roach, if participants suppress

‘roach’whenevertheyreceive‘ordeal’asacue,roachwill berecalledmorepoorly,regardlessofwhetheritistested with ‘Ordeal’ or ‘Insect’. Thus, suppressing a memory reducesitsaccessibilityfromavarietyofcues,aproperty knownas‘cueindependence’[53].Cueindependenceindi- catesthattheforgettingmostlikelyreflectsdisruptionof the suppressed trace itself rather than the particular pathway from the reminder to the trace (reviewed in [55];seealso[83]).Thisisusuallytakenasstrongevidence for an inhibitionprocess that suppresses thetrace ([53], althoughsee[84]foranalternative].Asadditionalsupport foranitem-specificinhibitionprocess,forgettinghasalso been foundonitemrecognition testsfor bothwordsand

abstractshapes[68,85].Moreover,theeffectevenoccurson indirectpriming tests, such as perceptual identification:

participants whosuppress retrieval ofvisual objects are lesslikelytoidentifycorrectlythoseobjectswhentheyare presented in visual noise[66,67]. Thus, suppressionnot onlyimpairsconsciousaccesstounwantedmemories,but alsoaffectstheirunconsciousinfluence,atleastontestsof objectperception.

Althoughresearchonretrievalsuppressionusuallyasks peopletorecallsuppresseditemsintentionally,thisargu- ably does not reflect real-world circumstances. In most cases,peopleareunlikelytotrytorecallexperiencesthey weremotivatedtosuppress.Amoreappropriatemeasureof theimpactofsuppressioninrealtermswouldmeasurethe tendencytoretrievethesuppressedcontent,ratherthanthe abilitytodoso[86].Forinstance,howlikelywouldpeoplebe torespondwiththesuppressedcontentonafreeassociation test?Interestingly, on suchtests,suppression effects are especiallypronounced[86].Thisraisesthepossibilitythat intentional recallmeasures underestimate thechange in spontaneous retrieval patterns that arise in real life.

Changes in retrieval patterns introduced by inhibition maybesustainedoverthelongtermbyalternativeassocia- tionsthatnaturallyariseinresponsetoreminders.Indeed, asking people to generate alternative associations to a reminder often increases forgetting, compared with not giving specific instructions. However, as noted shortly, thoughtsubstitutionisnotnecessarytoinduceforgetting, andseveralmechanismscontributetosuppression-induced forgetting.

Neuralbasisofretrievalsuppression

Similartodirectedforgetting,stoppingretrieval appears tobeachieved,inpart,bycontrolmechanismsmediatedby theprefrontalcortex.Retrievalsuppressionengageslater- al prefrontalcortex, includingDLPFC andventrolateral prefrontalcortex(VLPFC)often inthe righthemisphere [64,87–91].Theseregionsresembleareasinvolvedinstop- pingmotor actions,suggestingthat suppressionengages generalresponseoverridemechanismstostopretrieval(a pointtowhich wewill return).Critically, suppression is accompaniedbyreduced activityinbrainareas linkedto episodicrecollection[64,87–91].Forexample,suppression is associated with reduced hippocampal activity, some- timesalongwithothersubregionsoftheMTL.Giventhat single-unitelectrophysiologyandfunctionalneuroimaging have linked hippocampal activity to the presence of re- trieved memories in awareness, these findings suggest that inhibitory control interrupts hippocampal retrieval processes to suppress mnemonic awareness. Consistent withthishypothesis,frontohippocampalinteractionsdur- ingsuppressionhavebeenobservedwitharangeofmate- rials, including words [87–91], visual objects [67], and negativelyvalencedscenes[64],suggestingadomaingen- eralsuppressionprocess.

Althoughtheforegoingpatternsuggeststhatsuppres- sionengagestheprefrontalcortextoreducehippocampal activity,reducedactivityduringno-thinktrials(relativeto thinktrials)mightsimplyreflecthippocampalengagement during thinktrials.Thus,rather thanshowingthat sup- pression terminates retrieval, less hippocampal activity

284

mayreflectapassivefailuretoengageretrievalduringno- thinktrials.However,evidencehasgrownthatinhibitory controlreduceshippocampal activation.First,hippocam- palactivityisalsoreducedcomparedwithactivityduringa fixationbaselinecondition[64,91],suggestingthatreduc- tionsreflectmorethan anabsence ofpositiveactivation.

Second,DLPFCactivationduringno-thinktrialsisoften negatively correlated with hippocampal activity [63,64].

Indeed,themagnitudeofdownregulationandthecorrela- tionwithDLPFChasinsomestudiesincreasedoverblocks oftheTNTphase[64],suggestingprogressivelyimproved hippocampalregulationwithpractice.Third,reducedhip- pocampal activity predicts later forgetting of unwanted memories[64,91].Finally,effectiveconnectivityanalyses showa top-downmodulatoryinfluence ofDLPFC onthe hippocampus[67,88],withnegativecouplingfromDLPFC predicting the amount of suppression-induced forgetting [88].Althoughthepathwaysimplementingthistop-down influenceare unknown, some datasuggest thecingulum

bundle isa plausible candidate fora whitematter tract that could support the frontohippocampal interactions underlying suppression [90]. Together, these findings stronglysupportaroleofDLPFCinreducinghippocampal activity, interrupting recollection, and impairing reten- tion. Morebroadly, they specify a neurobiological model of memorycontrol that providesa frameworkfor under- standingdisorderedcontrolovermemory(Box3).

Opposing neural mechanisms underlie direct suppres- sion and thought substitution. Although hippocampal downregulationisafundamentaltoolofretrievalsuppres- sion,other mechanismsofcontrollingawarenessarepos- sible.Forexample,peoplemayredirectattentiontoother thoughts about a reminder. Such diversionary thoughts could either prevent the entrance of the memory into awareness, orreplacean intrudingmemory.Behavioural findings indicate that asking participants to generate thought substitutes for reminders can be effective in

Box3.Clinicalvariationinmotivatedforgetting

Individualdifferencesinmemorycontrolmaycauseeitherdeficientor exaggerated rates offorgetting of life events[55,106,107] that we mightbetterunderstandandremediatewithaneurobiologicalmodel ofmemorycontrol.

Deficientmemorycontrol

Intrusivememoriesandthoughtsarise inmanyclinicalconditions, such as post-traumatic stress disorder (intrusions), depression (rumination),attentiondeficitdisorder(distractingthoughts),obses- sive/compulsive disorder (obsessive thoughts), addiction (craving relatedthoughts),andanxiety(worries).Thissymptommayoriginate from deficient inhibitory control over memory. Supporting this, during retrieval suppression, adults with attention deficit disorder showimpairedsuppression-inducedforgetting,andalsodiminished modulationofhippocampalactivitybyDLPFC[63].Similarly,patients withpost-traumaticstressdisordershowimpairedresponseinhibi- tion [108], diminished engagement of lateral prefrontal cortex on response inhibition tasks [109], and, critically, reduced directed forgetting[110].Rumination,depression,andanxietyhavealsobeen linkedtoimpairedsuppression-inducedforgetting[79,111,112].

Deficientmemorycontrolmayreflectcompromisedfunctionofthe networksdiscussed inthisarticle. For example,disordered control may originate from diminished cortical volume or white matter connectivity between prefrontal cortex and sites of modulation.

Deficits inneurotransmitters relating toinhibitory control,such as dopamine, might also underpin disordered control, a possibility supportedbyworklinkinggeneticvariationindopaminemetabolism tomemoryinhibition[113].Alternatively,poormemorycontrolmay sometimesreflectlackofexperiencewithintrusivememoriesand, thus, inadequate development of thecontrol process, which may exhibitexperiencedependentplasticity.Forexample,afteratrauma, corticalthicknessinrightDLPFCincreasessignificantlyoverayear, withthesizeoftheincreasepredictingreducedpost-traumaticstress disordersymptoms[114].Takentogether,thesefindingssuggestthat bothpharmacologicalandtraininginterventionscouldbedesignedto bolstermemorycontrol.

Exaggeratedmemorycontrol

Onestrikingexampleofmotivatedforgettingispsychogenicamnesia, inwhichapersonexhibitsprofoundamnesiaforlargechunksoftheir personalexperiencesintheaftermathofanintenselystressfulperiod [115].Twostudiessuggestthatsuchcasesinpartreflectexceptionally effective memory control. One study examined two psychogenic amnesia patients, with amnesia extending years before scanning [116].Bothpatientswerescannedastheyidentifiedfaces.Somefaces wereofstrangers(novelfaces).Otherswereofpeoplethepatients knew,withhalfdrawnfrompeopletheymetbeforetheirwindowof

amnesia (identifiable faces), and the other half from during the windowoftimeaffectedbyamnesia(unidentifiablefaces).Unsurpris- ingly,patientsdidnotrecognisethenovelfaces,andcouldrecogniseall oftheidentifiablefaces.Intriguingly,althoughneitherpatientremem- bered anyoftheunidentifiable faces,thesefaceselicitedincreased activationinrightDLPFCandVLPFC,togetherwithreducedactivityin the hippocampus, as observed in laboratory studies of retrieval suppression (Figure I).Aftertreatment, one patient recoveredtheir memoriesand,uponrescanning,nolongerexhibitedthesuppression pattern.Thesefindingssuggest thatextreme psychologicaldistress leadsretrievalsuppressiontobeengagedinvoluntarilyinreactionto certain stimuli [116]. In an independent study, a patient with psychogenic amnesiawas showntoexhibitdramatically magnified suppression-inducedforgettingintheTNTprocedure[117],suggesting alinkbetweentheirconditionandsuppressionability.

(A) (B)

TRENDS in Cognitive Sciences

(ii) (ii)

(i) (i)

FigureI.Brain-imagingdatafromtwopatientswithdissociativeamnesia[116].

Patients1 (A)and 2(B)viewedimages offacesanddecided whetherthey recognisedthemfromtheirlife.Imageswereeitherstrangers(novel),facesthey knew,fromoutsidethewindowofamnesia(identifiablefaces)orfacesthey knewfromwithintheamnesicwindow(unidentifiablefaces).(Ai)and(Bi)depict brainareasthataremoreactiveforunidentifiablefacesthanforidentifiablefaces (rightdorsolateralprefrontalcortex).(Aii)and(Bii)depictbrainareasthatareless activeforunidentifiablefaces(hippocampus).

285