Neuropsychologia 44 (2006) 1315–1324

Decision-making in amnesia: Do advantageous decisions requireconscious knowledge of previous behavioural choices?

Klemens Gutbrod a, Claudine Krouzel a, Helene Hofer a, Rene Muri a,Walter Perrig b, Radek Ptak c,∗

a Department of Neurology, Division of Neuropsychological Rehabilitation, University Hospital, Bern, Switzerlandb Department of Psychology, University of Bern, Bern, Switzerland

c Division of Rehabilitation, University Hospital Geneva, Switzerland

Received 8 April 2005; received in revised form 12 January 2006; accepted 22 January 2006Available online 2 March 2006

Abstract

Previous work has reported that in the Iowa gambling task (IGT) advantageous decisions may be taken before the advantageous strategy isknown [Bechara, A., Damasio, H., Tranel, D., & Damasio, A. R. (1997). Deciding advantageously before knowing the advantageous strategy.Sctshtatrp©

K

1

novttiID

Sf

0d

cience, 275, 1293–1295]. In order to test whether explicit memory is essential for the acquisition of a behavioural preference for advantageoushoices, we measured behavioural performance and skin conductance responses (SCRs) in five patients with dense amnesia following damage tohe basal forebrain and orbitofrontal cortex, six amnesic patients with damage to the medial temporal lobe or the diencephalon, and eight controlubjects performing the IGT. Across 100 trials healthy participants acquired a preference for advantageous choices and generated large SCRs toigh levels of punishment. In addition, their anticipatory SCRs to disadvantageous choices were larger than to advantageous choices. However,his dissociation occurred much later than the behavioural preference for advantageous alternatives. In contrast, though exhibiting discriminatoryutonomic SCRs to different levels of punishment, 9 of 11 amnesic patients performed at chance and did not show differential anticipatory SCRso advantageous and disadvantageous choices. Further, the magnitude of anticipatory SCRs did not correlate with behavioural performance. Theseesults suggest that the acquisition of a behavioural preference – be it for advantageous or disadvantageous choices – depends on the memory ofrevious reinforcements encountered in the task, a capacity requiring intact explicit memory.

2006 Elsevier Ltd. All rights reserved.

eywords: Amnesia; Decision-making; Explicit memory; Implicit memory; Reinforcement; Somatic marker

. Introduction

Decision-making is commonly perceived as a highly ratio-al and conscious process during which arguments in favourr against a behavioural choice are compared. Evidence for thisiew is our ability to make decisions that are unpleasant momen-arily, but in the long term may be advantageous. However,he assumption that decision-making is always conscious andndependent of emotional factors has recently been challenged.n their somatic-marker hypothesis, Damasio (1994, 1998) andamasio, Tranel, and Damasio (1991) propose that rational cog-

∗ Corresponding author at: HUG-Service de Reeducation 26, av. de Beau-ejour, 1211 Geneva 14, Switzerland. Tel.: +41 22 382 35 24;ax: +41 22 382 83 38.

E-mail address: radek.ptak@hcuge.ch (R. Ptak).

nitive processes involved in decision-making are assisted byemotion-related signals (‘somatic markers’). According to thetheory, somatic markers may be present before full consciousknowledge of the ‘goodness’ of a choice is available and theymay bias decisions away from a potentially disadvantageouschoice whether we are consciously aware of them or not. Thehypothesis further proposes that the critical region for the acqui-sition of somatic markers is the ventromedial prefrontal cortex(PFC).

Support for these assumptions comes from studies with the‘Iowa gambling task’ (IGT, see Bechara, Damasio, Damasio, &Anderson, 1994). In this task participants are shown four decksof cards and asked to select cards from these decks. On eachtrial they win some money, but on some trials they are punishedand loose money. Unbeknownst to them, the immediate gainon decks C and D is small, but the possible loss is even smaller,

028-3932/$ – see front matter © 2006 Elsevier Ltd. All rights reserved.oi:10.1016/j.neuropsychologia.2006.01.014

1316 K. Gutbrod et al. / Neuropsychologia 44 (2006) 1315–1324

resulting in the long run in a small benefit. On decks A and B thegain is large, but the possible loss is even larger, which in the longrun results in a net loss. Healthy participants begin the IGT withselecting cards from all four decks, but quickly learn to preferthe advantageous decks. In addition, they show larger antici-patory skin conductance responses (SCRs) prior to selecting acard from the disadvantageous compared to the advantageousdecks, which is in line with the notion that somatic markers assistthe decision-making process in the IGT. Further, several studieshave found that patients with ventromedial PFC damage makemore disadvantageous than advantageous choices in the IGT(Bechara, Damasio, Tranel, & Damasio, 1997; Bechara, Tranel,& Damasio, 2000; Fellows & Farah, 2005; but see Manes et al.,2002), a finding taken as evidence for an insensitivity of thesepatients to future consequences of current actions (Bechara,Tranel, Damasio, & Damasio, 1996).

However, some conclusions from studies using the IGT haverecently been questioned. One source of disagreement concernsthe role of ventromedial PFC in decision-making. Some authorspropose that a reversal-learning deficit may better explain thecard-choosing behaviour of ventromedial PFC patients thanan insensitivity to future outcomes, since patients must shiftin responding away from the risky decks that initially seemrewarding, but are gradually associated with high punishment(Clark, Cools, & Robbins, 2004; Fellows & Farah, 2003; Maia& McClelland, 2004). The second problem concerns the roleoBtcaShmpaalctetisdeMuwoapebkta

Nevertheless, the question remains whether conscious knowl-edge is a by-product of participants’ involvement in the task, orwhether it is a necessary component of successful performance.

In order to distinguish between these two possibilities andto further examine the role of somatic markers in the IGT, wetested eleven severely amnesic patients on the task. There areseveral reasons why performance of amnesic patients in theIGT is of particular interest. First, amnesic patients should beunable to acquire sufficient conceptual knowledge to guide theirchoices. Second, amnesic patients demonstrate intact learningin tasks that do not require conscious remembering (Cohen &Squire, 1980; Schacter, 1993; Shimamura, 1986), some of whichare emotion-based (Hamann, Cahill, & Squire, 1997; Tranel& Damasio, 1993). A few tasks on which amnesic patientsshow preserved learning bare some resemblance to the IGT(Knowlton, Squire, & Gluck, 1994; Knowlton, Mangels, &Squire, 1996), and amnesics show intact autonomic responsesto material they had previously been exposed to despite severelyimpaired conscious recollection (Bauer & Verfaellie, 1992;Tranel & Damasio, 1993). Thus, the performance of amnesicpatients in the IGT allows conclusions about the role of con-scious awareness in the task. A further question concerns theeffect of damage to specific limbic regions on performancein the IGT. Amnesia may not only result from damage to thelimbic system and the diencephalon, but also the basal fore-brain and posterior orbitofrontal cortex, both adjacent to brainrIpop

2

2

pe

cribrDag3rgadcefo2

pe

f awareness and its relation to anticipatory SCRs in the IGT.echara et al. (1997) found that only about 70% of those par-

icipants who exhibit normal learning on the IGT are able toomment accurately on the nature of the task and are fullyware of the differences between decks. Crucially, anticipatoryCRs in healthy subjects could be measured before participantsad conscious knowledge about whether they were about toake a ‘good’ or a ‘bad’ choice. These findings were inter-

reted as evidence that somatic signals biasing decisions towardsdvantageous choices are present before participants know thedvantageous strategy (Bechara et al., 1997). However, a prob-em with the interpretation of SCR data is that the nature of skinonductance responses does not permit direct conclusions aboutheir significance: high SCRs may reflect positive or negativemotions as well as general arousal signals. A large anticipa-ory SCR may therefore be a correlate of the expectation of anmmediate high loss rather than an indicator of long-term deci-ions (Tomb, Hauser, Deldin, & Caramazza, 2002). There is alsoisagreement concerning the question of what has to be consid-red as conceptual knowledge and how it could be established.aia and McClelland (2004) note that open questioning such as

sed by Bechara et al. (1997) (e.g. ‘Tell me all you know abouthat is going on in this game’) may fail to reliably cue recallf all relevant knowledge. Using more detailed questions theseuthors found that card selecting behaviour of healthy partici-ants is highly correlated with their immediate conscious knowl-dge about the task (Maia & McClelland, 2004). When theyehave advantageously, participants show sufficient quantitativenowledge of the contingencies of the chosen cards, suggestinghat performance in the IGT may be explained by consciouslyccessible knowledge rather than unconscious somatic signals.

egions damaged in patients with a decision-making deficit in theGT (Bechara et al., 1996). We therefore separated our amnesicatients into those who had damage to the basal forebrain andrbitofrontal cortex (‘anterior’ group) and those who had moreosterior damage (‘posterior’ group).

. Methods

.1. Participants

Table 1 summarizes the demographic characteristics of the participants. Allarticipants gave written informed consent. The study was approved by the localthical committee of the University Hospital of Bern.

The criterion for involvement in the study was the presence of a severe,ircumscribed and chronic memory impairment. The five patients of the ‘ante-ior’ group all had damage to the basal forebrain and orbitofrontal cortex, thats brain regions in the vicinity of ventromedial PFC. Patients 1, 2, 4 and 5 hadrain damage following the bleeding from a ruptured aneurysm of the ante-ior communicating artery, and patient 3 had suffered from herpes encephalitis.ue to vasospasms, patient 5 additionally had lesions in the territory of the

nterior and middle cerebral arteries. The ‘posterior’ group was more hetero-eneous, both with respect to aetiology and lesion localisation. Patients 1 andshowed amnesia following an episode of hypoxia. Although MRI did not

eveal any visible brain damage, both patients were included in the posteriorroup since hypoxia is known to cause primarily damage to the hippocampus ordjacent regions (Zola-Morgan, Squire, & Amaral, 1986). Patients 2 and 6 hadamage to temporal structures due to encephalitis. After resection of a colloidyst in the 3rd ventricle and a right fornix lesion patient 4 had suffered bilat-ral thalamic infarction. Finally, patient 5 became amnesic following bleedingrom an aneurysm of the middle cerebral artery. Fig. 1 shows overlap mapsf brain lesions drawn on standard templates using MRIcro (Rorden & Brett,000).

The control group consisted of eight healthy persons matched to the amnesicatients with regard to age and years of education (see Table 1). Patients werexamined 1–14 years after the incident leading to amnesia. Table 2 summarizes

K. Gutbrod et al. / Neuropsychologia 44 (2006) 1315–1324 1317

Table 1Demographic characteristics of the patient and control group (BF: basal forebrain, MTL: medial temporal lobe)

Patient Age (years) Education (years) Sex Etiology Lesion localization

Anterior group1 49 13 M Haemorrhage BF, left gyrus rectus, left middle orbital gyrus, left anterior cingulate, left

putamen2 54 19 M Haemorrhage BF, bilateral gyrus rectus, bilateral middle orbital gyrus, bilateral anterior

cingulate, left superior frontal gyrus (medial part)3 41 13 F Herpes encephalitis BF, left gyrus rectus, left middle orbital gyrus, bilateral hippocampus,

bilateral parahippocampal gyrus, left fusiform gyrus, bilateral amygdala,left temporal pole, left insula

4 54 19 M Haemorrhage BF, right gyrus rectus, right middle orbital gyrus, right anterior cingulate,right hippocampus, right amygdala, right temporal pole, right caudatenucleus

5 35 13 M Haemorrhage BF, bilateral gyrus rectus, bilateral middle orbital gyrus, bilateral cingulate,bilateral superior frontal gyrus medial part, bilateral amygdala, bilateraltemporal pole, bilateral insula, bilateral frontal operculum

Mean 46.6 ± 8.4 15.4 ± 3.3

Posterior group1 38 15 M Hypoxia No damage visible on MRI2 53 20 M Herpes encephalitis Bilateral hippocampus, bilateral parahippocampal gyrus, bilateral

amygdala, bilateral temporal pole, right insula3 37 10 F Hypoxia No damage visible on MRI4 64 10 M Infarction Bilateral anterior/medial thalamus, right fornix5 36 19 F Haemorrhage Right hippocampus, right parahippocampal gyrus, right amygdala, right

temporal pole, right insula, right caudate nucleus, right lenticular nucleus6 46 13 M Limbic encephalitis Bilateral hippocampus

Mean 45.7 ± 11.1 14.5 ± 4.3

ControlsMean 42.3 ± 6.9 16.4 ± 2.3

Fig. 1. Overlap maps showing the region of highest overlap of amnesic patients with damage involving the basal forebrain and/or orbitofrontal cortex (anterior group,A) and amnesic patients without such damage (posterior group, B). Lesions of the anterior group were drawn on templates with Z-values −24, −16, and −8 inTalairach space (Talairach & Tournoux, 1988) while templates with Z-values of −24, −16 and 8 were used for the posterior group (the position of these sections isshown on the saggital templates). The color scale indicates the absolute number of shared lesions for every damaged area. Note that the lesions of two patients in theposterior group with hypoxic brain damage were not drawn because no damage was visible on MRI. The left hemisphere is shown on the right side and vice versa.

1318 K. Gutbrod et al. / Neuropsychologia 44 (2006) 1315–1324

Table 2Neuropsychological results of amnesic patients in the Wechsler Adult Intelligence Scale-Revised (WAIS-R), the Wechsler Memory Scale-Revised (WMS-R) andthe California Verbal Learning Test (CVLT)

Patient group WAIS-R WMS-R CVLT

Full-scale IQ General memory Delayed memory Word list learning Long delay free recall

Anterior1 126 105 79 36 (<1) 5 (<1)2 115 82 63 21 (<1) 4 (<1)3 94 58 <50a 25 (<1) 2 (<1)4 119 69 58 27 (<1) 1 (<1)5 104b 62 60 34 (<1) 5 (<1)

Mean 111.6 ± 12.7 75.2 ± 19 62 ± 10.7 28.6 ± 6.3 3.4 ± 1.8

Posterior1 126 82 79 29 (<1) 2 (<1)2 118 60 <50a 32 (<1) 5 (<1)3 89 63 52 26 (<1) 1 (<1)4 102 80 61 30 (<1) 3 (<1)5 128 56 <50a 35 (<1) 2 (<1)6 97 55 <50a 36 (<1) 1 (<1)

Mean 110 ± 16.2 66 ± 12 57 ± 11.6 31.3 ± 3.8 2.3 ± 1.5

In parentheses age-adjusted percentile ranks. The WAIS-R full-scale IQ and WMS-R memory indices have a mean of 100 and a standard deviation of 15.a The WMS-R does not provide scores below 50.b Because of mild aphasic problems, only performance IQ was calculated.

their results in neuropsychological tests probing intelligence and memory andTable 3 shows their performance in tests of executive functioning. A standardneuropsychological examination revealed that all amnesic patients had normalintelligence and (except patient 4 of the posterior group) normal or only slightlyimpaired performance in rule deduction, control of interference, or verbal anddesign fluency tests. In contrast, all had severe impairments in tests of verbaland visual learning and long-term recall.

2.2. Procedure

2.2.1. Behavioural measuresThe IGT was administered as described in detail by Bechara et al. (1994).

The subjects were asked to select on each trial a card of one of four decks,labelled A–D. Participants were free to switch between decks at any time and as

often as they wished. On every card, they won a small amount of play money.A choice from the ‘advantageous’ decks C and D resulted in a gain (reward) of50 units of play money. For a choice from the ‘disadvantageous’ decks A andB 100 units were given. However, in addition to winning money, on some trialsparticipants encountered losses. Choices from decks A and C were punishedfive times every 10 choices (A with a loss of 150–300, C with a loss of 25–75)while decks B and D were punished only once every 10 choices (B with a loss of1250, D with a loss of 250). The schedule of gains and losses was such that 10choices from the advantageous decks resulted in an average loss (punishment)of 250 units, while the average loss on the disadvantageous decks was muchhigher (1250 units). Thus, a strategy to select cards exclusively from decks Aand B led in the long term to a loss of all money, whereas a strategy to selectonly from decks C and D led to a net gain. Participants were not informed aboutthe differences between decks, nor that the task would be discontinued after

Table 3Neuropsychological characteristics of the amnesic patients in the Stroop test (interference condition), Wisconsin Card Sorting Test (WCST), word and design fluency

Patient Stroop WCST Fluency

Time Errors No. categories Perserverative errors Verbal Design

Anterior group1 23 (68) 1 (42) 6 (69) 11 (46) 41 (98) 31 (54)2 26 (60) 0 (73) 4 (7) 14 (34) 17 (24) 15 (10)3 17 (93) 0 (80) 4 (7) 29 (1) 27 (69) 20 (16)4 20 (86) 0 (73) 6 (62) 10 (58) 29 (82) 26 (50)5 23 (57) 0 (76) 6 (69) 21 (10) 28 (69) 34 (69)

Mean 21.8 ± 3.4 0.2 ± 0.4 5.2 ± 1.1 17 ± 8 28.4 ± 8.5 25.2 ± 7.8

Posterior group

I

1 21 (65) 0 (76) 6 (69)2 25 (65) 0 (73) 6 (62)3 35 (14) 2 (11) 6 (69)4 58 (<1) 0 (67) 1 (<1)5 26 (47) 1 (43) 6 (69)6 52 (<1) 1 (80) 6 (69)

Mean 36.2 ± 15.4 0.7 ± 0.8 5.2 ± 2

n parentheses age-adjusted percentile rank.

13 (46) 28 (69) 31 (54)13 (38) 26 (73) 23 (34)20 (12) 23 (38) 22 (21)29 (5) 27 (73) 20 (24)6 (79) 29 (76) 39 (82)19 (14) 19 (24) 22 (21)

16.7 ± 7.8 25.3 ± 3.7 26.2 ± 7.4

K. Gutbrod et al. / Neuropsychologia 44 (2006) 1315–1324 1319

completion of 100 card selections. They were asked to indicate their selectionby naming one of the four decks, and the experimenter announced verbally theheight of the gain (50 or 100 units) or – when the occasion arose – the loss.

To evaluate the degree of conceptual knowledge participants had of the task,the task was briefly interrupted after every 20 card turns and participants wereinvited to tell what was going on in the task and how they felt about it. As in thestudy of Bechara et al. (1997), the “pre-punishment” period covered the startof the task when subjects sampled the decks and before they encountered thefirst loss. The “pre-hunch” period consisted of the next series of cards whensubjects continued to choose cards from various decks, but professed no notionof what was happening. The “hunch” period corresponded to the period whenthey reported “liking” or “disliking” certain decks, and “guessed” which deckswere risky or safe, but were not sure of their answers. The “conceptual” periodcorresponded to the period when subjects were able to articulate accurately thenature of the task and tell for certain which were the ‘good’ and ‘bad’ decks,and why they were good or bad. In addition to measuring the card selectionbehaviour of the participants, we administered two explicit memory tasks imme-diately after finishing the gambling task. In the recognition task, participantswere shown four possible sequences of profits and losses encountered during 20card selections from decks A–D, without being told from which deck these rein-forcement schedules were chosen. They were asked to assign each reinforcementschedule to one of the four decks. Following the recognition task, participantswere asked to indicate for each of the four reinforcement schedules whether itwas advantageous or disadvantageous in the long run (assignment task). Thedependent variable was the number of cards recognized or assigned correctly(max. four).

2.2.2. Skin conductance responsesSkin conductance responses (SCRs) were sampled at 20 Hz with two elec-

trodes attached to the middle phalanges of the index and the middle finger oftcouwwdg

2

eapctdpt

swct

3

3

gpcfr

Fig. 2. Decision-making performance of control participants and amnesicpatients in the gambling task. The figure shows the difference score (±S.E.M.)of advantageous selections minus disadvantageous selections as a function oftrial block.

4, where it was comparably high to control participants. How-ever, this good performance in block 4 was essentially due topatient 2, who exclusively made advantageous choices in thisblock. The score of the posterior group always remained closeto zero, suggesting that they exhibited neither a preference foradvantageous nor for disadvantageous decks.

A two-way mixed ANOVA with between-subject factorgroup and a within-subject facture block revealed a significanteffect of group (F2,16 = 6.69, p < 0.01) and a significant inter-action of group with block (F8,64 = 3.77, p < 0.01). Post hocNewman–Keuls tests revealed that this interaction reflected anincrease of performance in healthy participants between block1 and 5, whereas, no reliable difference between blocks couldbe found in both amnesic groups. Interestingly, healthy partici-pants started sampling more often from disadvantageous decks,but very quickly shifted towards advantageous decks, as exem-plified by a significant increase of performance from block 1 toblock 2 (t7 = 2.2, p < 0.05).

3.2. Degree of conceptual knowledge

All eight healthy controls reached the pre-punishment(100%), the pre-hunch (100%) and the hunch period (100%),and six of them reached the conceptual period (75%). This pro-portion is comparable to the study of Bechara et al. (1997), whoupoot

fttp(r

he non-dominant hand. SCRs were measured continuously during the task onhannel 1 of the amplifier. On every card choice (i.e. when the participant namedne of the four decks) the experimenter set a mark in channel 2, which was latersed to define the data window for anticipatory SCRs. A second mark was sethen the experimenter announced the gain/loss, and was used to define the dataindow for reinforcement SCRs. In order to leave enough time for the SCR toecrease to baseline level following the card choice, the reward/punishment wasiven 10 s after the card choice.

.2.3. Data analysisThe 100 card selections were divided into five blocks of 20 trials and for

ach block the total number of selections from advantageous decks (C and D)nd the total number from disadvantageous decks (A and B) were counted. Thereference for advantageous cards in comparison to disadvantageous cards wasalculated as the difference between the choice frequency of cards from advan-ageous decks and the choice frequency of cards from disadvantageous decks:ifference score = (C + D) − (A + B). Scores below zero indicated that partici-ants were selecting disadvantageously, whereas, scores above zero indicatedhat they were selecting advantageously.

For the analysis of anticipatory SCRs a data window of 5 s prior to each cardelection was defined. A data window of 5 s following the reward/punishmentas defined for the analysis of reinforcement SCRs. The event-related SCR was

alculated as the difference of the highest amplitude value and the first value inhe window (baseline).

. Results

.1. Behavioural performance in the gambling task

Fig. 2 shows the acquisition of a preference for advanta-eous cards (C and D) across the five learning blocks. Controlarticipants showed a steadily increasing difference score, indi-ating that they gradually acquired and retained a preferenceor the advantageous decks. The difference score of the ante-ior amnesic group was near zero in all five blocks except block

sed the same methodology. In contrast to controls, all amnesicatients reached the pre-punishment and pre-hunch period, butnly three (patients 2 and 3 from the anterior group, patient 4f the posterior group) reached the hunch, and only patient 4 ofhe posterior group reached the conceptual period.

A difference in the degree of explicit knowledge was alsoound in the recognition and assignment tasks. Healthy par-icipants had a mean of 3.75 ± 0.4 correct responses in bothasks. Amnesic patients were able to indicate whether the pro-osed reinforcement schedules would result in a gain or a lossmean, anterior group: 3.4 ± 0.4; posterior group: 3 ± 0.4 cor-ect responses). This was not different from healthy controls

1320 K. Gutbrod et al. / Neuropsychologia 44 (2006) 1315–1324

(Mann–Whitney test: Z < 1.1, p > 0.2). In contrast, they wereimpaired when asked to indicate to what deck each schedulebelonged (anterior group: 1 ± 0.6; posterior group: 2.5 ± 0.6correct responses). The difference to controls reached signifi-cance for the anterior group (Z = 2.61, p < 0.01), but due to lackof statistical power only approached significance in the posteriorgroup (Z = 1.62, p < 0.11).

3.3. Anticipatory SCRs

In a few trial blocks, some participants did exclusively selectcards from advantageous or from disadvantageous decks. Con-sequently, some anticipatory SCRs were missing and had to beimputed using a linear regression algorithm. This concerned 7(8.8%) values of healthy participants and 4 (3.6%) values ofamnesic patients. Fig. 3 shows the development of anticipatorySCRs across blocks prior to the selection of a card from theadvantageous or disadvantageous decks. There was an increas-ing discrepancy between anticipatory SCRs before disadvanta-geous compared to advantageous choices in healthy participants,but not in the anterior amnesic group. The posterior amnesics dif-fered slightly from the anterior group with respect to anticipatoryresponses to disadvantageous choices. These remained stableacross blocks in the anterior group, but decreased across blocksin the posterior group. A mixed ANOVA with group as between-saI(psatdttpct

significant changes in anticipatory SCRs across blocks, or sig-nificant differences between advantageous and disadvantageouschoices.

Thus, the main finding of this analysis was an increase ofanticipatory SCRs of healthy participants to disadvantageouschoices from block 1 to block 5, which was not observed in bothamnesic groups.

3.4. Reinforcement SCRs

Fig. 4 shows the mean SCR amplitudes to different levelsof reward and punishment. Remember that the average loss inevery 10 choices from deck A and B was comparable (1250), butthat the frequency of loss was much greater on deck A (5× 250)than on deck B (1× 1250). This difference in reinforcement wasreflected in the punishment SCRs, as all three groups showedlarger punishment SCRs to cards from deck B compared to allother decks (Fig. 4). A three-way mixed ANOVA with groupas between-subject factor and deck and reinforcement (reward,punishment) as repeated measures revealed a significant maineffect of deck (F3,48 = 3.74, p < 0.05) and a significant interac-tion between deck and reinforcement (F3,48 = 4.59, p < 0.01).This interaction reflected the fact that a punished choice fromdeck B resulted in larger SCRs than the punishments in all otherconditions. Although this effect did not interact with the factorgttttcolrjdpit

F patiend

ubject, and block and choice (advantageous, disadvantageous)s repeated factors did not reveal any significant main effects.mportantly, the main effect of group fell far from significanceF2,16 = 1.2, p = 0.33), indicating that the overall level of antici-atory SCR did not differ between groups. However, there was aignificant interaction of group by choice (F2,16 = 6.71, p < 0.01)nd of group by choice by block (F8,64 = 2.2, p < 0.05). Post hocests of this three-way interaction revealed only two significantifferences. First, healthy participants showed larger anticipa-ory SCRs for disadvantageous choices in the fifth comparedo the first block. Second, in the fifth block healthy partici-ants exhibited higher SCRs prior to making a disadvantageoushoice compared to an advantageous choice. In contrast to con-rol participants none of the two amnesic groups did show any

ig. 3. Anticipatory SCRs (±S.E.M.) of control participants and both amnesicecks.

roup (three-way interaction: F6,48 = 0.74, p = 0.62), its magni-ude appeared to be much smaller in the anterior group comparedo both other groups. The absence of an interaction includinghe factor group could have been due to the great variability ofhe SCR data, or to the fact that the anterior group showed aomparable pattern, but a smaller overall amplitude of SCRs. Inrder to distinguish between these two possibilities, we calcu-ated punishment SCRs for each participant and for each deckelative to his mean punishment SCR across decks A–D and sub-ected these values to a two-way ANOVA with factors group andeck. This analysis revealed a main effect of deck (F3,48 = 11.8,< 0.001), but no effect of group (F2,16 = 1.24) and no signif-

cant interaction (F6,48 = 1.29). The relative punishment SCRso deck B were larger than to all other decks, and this was true

t groups prior to the selection of cards from advantageous or disadvantageous

K. Gutbrod et al. / Neuropsychologia 44 (2006) 1315–1324 1321

Fig. 4. Reinforcement SCRs (±S.E.M.) of control participants and amnesic patients following gain (reward) or loss (punishment) of money as a function of the carddeck (A and B: disadvantageous; C and D: advantageous).

for each group (controls: 198 ± 35%; anterior: 222 ± 44%; pos-terior: 158 ± 41%). Thus, even when the relative magnitude ofpunishment SCRs was considered, the anterior group did notshow a different response pattern than the other two groups.Finally, we asked whether patients with damage involving theamygdala would show particularly low punishment SCRs. Thiswas not the case, as the five patients with unilateral or bilat-eral amygdala damage had comparable mean punishment SCRs(0.67 ± 1.2) with healthy controls (0.69 ± 0.39; patients withoutamygdala involvement: 0.28 ± 0.27).

3.5. Individual performance

In order to evaluate individual performance, we compared thefrequency of each participant to make a disadvantageous choicein block 5 with a performance predicted by chance using thebinomial test. In the anterior amnesic group, patients 1–4 didscore at chance and patient 5 had a significant preference for thedisadvantageous decks. In the posterior group, patients 1–5 didscore at chance and patient 6 had a significant preference forthe advantageous decks. In comparison, seven control partici-pants showed a significant preference for advantageous decks,while the results of one control did not differ from chance. Theseindividual results confirm the finding of the group comparisons(Section 3.1) that most amnesic patients score at chance in theIGT while most healthy participants exhibit significant learninge

fifhsccaaselO

Fig. 5. The frequency of a choice from disadvantageous decks (black bars,referred to left ordinate) and anticipatory SCRs (white bars, referred to rightordinate) to these choices in block 5. Results of individual patients as well asthe mean (±S.E.M.) of the control group are shown. The stippled line indicateschance performance in the behavioural task.

preference for advantageous decks (patient 6 of the posteriorgroup) did not exhibit particularly large anticipatory SCRs todisadvantageous decks. It must be specified, however, that thispatient showed a learning pattern different from healthy partic-ipants, since his number of advantageous decisions was belowthe 5th percentile of control values in blocks 3 and 4, and heonly switched to advantageous choices in block 5. Across thepatient groups the correlation between the frequency of disad-vantageous choices and the magnitude of anticipatory SCR wasnot significant (Spearman’s ρ = 0.39, p > 0.2).

Finally, we also calculated correlations between several neu-ropsychological test results and the frequency of disadvanta-geous choices in our patient group, but found no significantcorrelations.

4. Discussion

The aim of our study was to evaluate the importance of con-scious knowledge for successful performance in the IGT andto study the relation of this knowledge to anticipatory SCRsby examining the performance of amnesic patients in the task.

ffects.We further studied the relation between these behavioural

ndings and the magnitude of anticipatory SCR. We againocused on the last learning block since in this block nearly allealthy participants did show a significant learning effect. Fig. 5hows that control participants made only few disadvantageoushoices in the last block, but their anticipatory SCRs to thesehoices were high. This would be expected if anticipatory SCRscted as a biasing signal that prevented subjects from makingdisadvantageous choice. Most amnesic patients showed much

maller anticipatory SCRs to disadvantageous choices. How-ver, patients 2 and 4 in the anterior group showed relativelyarge anticipatory SCRs, yet scored at chance in the last block.n the other side, the only patient who showed a significant

1322 K. Gutbrod et al. / Neuropsychologia 44 (2006) 1315–1324

Similarly to findings of previous studies (Bechara et al., 1994;Bechara et al., 1996; Bechara et al., 1997), control participantsgradually acquired a behavioural preference for advantageouscards, and they were able to recognize after 100 trials whether areinforcement schedule belonged to deck A, B, C, or D. Thus, thebehavioural performance of healthy participants correspondedwell with their degree of conscious knowledge of underlyingcontingencies. We evaluated conscious knowledge in the sameway as Bechara et al. (1997), which may not be sufficiently sensi-tive to all task-relevant knowledge (Maia & McClelland, 2004).However, though the data of our healthy participants do not clar-ify to what extent conscious knowledge predicts behaviour, theyprovide an idea about the relation between anticipatory SCRsand behaviour. Healthy subjects showed a slight preference forthe ‘disadvantageous’ decks in the first 20 trials. This has beenreported in other studies (Denburg, Tranel, & Bechara, 2005;Evans, Bowman, & Turnbull, 2005) and can be explained by thefact that at this phase of the game no punishments are encoun-tered. As soon as the first punishments were given, subjectsswitched towards the more secure cards, which is evident in theirbehaviour by the end of the 40th trial (Fig. 2). This rapid switchin performance was significant in our study, and is similar to theresults of other studies (Denburg et al., 2005; Evans et al., 2005).The important finding is that this behavioural change precededa change in anticipatory SCRs, since SCRs to advantageous anddisadvantageous choices remained similar up to the fourth trialbeabtBrnpndf‘bTosaTtbamNvDfacat

ment SCRs of patients with ventromedial PFC damage examinedby Bechara et al. (1999) were not modulated by the magnitudeof punishment (note however that in a later study ventromedialPFC patients did show differential responses to high/low levelsof punishment, Bechara et al., 2000), and patients with circum-scribed amygdala damage exhibited nearly no punishment SCRs(Bechara et al., 1999). Thus, the failure of our amnesic patientsin the IGT cannot be explained by a lack of sensitivity to highlevels of punishment. However, in contrast to their reinforcementSCRs, anticipatory SCRs of our amnesic patients did not followthe same pattern as anticipatory SCRs of healthy participants.The patients showed no increase in anticipatory SCRs across trialblocks, and no difference between anticipatory SCRs to advan-tageous and disadvantageous choices. This pattern seems in linewith the predictions of the somatic marker hypothesis, whichproposes that overt behaviour is biased by a negative ‘feeling’preceding a behavioural choice (Bechara et al., 2005; Damasio,1994). However, in ventromedial PFC and amygdala patients afailure to generate large anticipatory SCRs to disadvantageouschoices accompanies the tendency to select disadvantageouscards more often than advantageous cards (Bechara et al., 1999;Bechara et al., 2000), while it was associated with randomperformance in our patients. More importantly, we found nocorrelation between the magnitude of anticipatory SCRs andthe behavioural preference in block 5. Further, the only patientwho exhibited normal performance in the last block did notgeilhvCoabet

tntfCifbpvretPtcib

lock (80 trials). Thus, the rapid change in behavioural pref-rence between block 1 and 2 must have been generated byvariable other than the type of somatic activation assessed

y SCR measures. The findings of recent studies suggest thathis variable is explicit task knowledge. Evans et al. (2005) andowman, Evans, and Turnbull (2005) reported that subjective

atings of the ‘goodness’ or ‘badness’ of cards improved sig-ificantly within the first 40 trials that is in the ‘pre-hunch’eriod in which the participants of Bechara et al. (1997) couldot express verbally what was going on. Using a much moreetailed questionnaire than we Maia and McClelland (2004)ound that participants selected cards that were nearly alwaysadvantageous’ in the sense that they represented the best choiceased on the series of gains and losses encountered previously.hese ‘advantageous’ choices correlated well with the degreef explicit task knowledge assessed with the questionnaire, anduch knowledge could well explain the rapid development ofpreference for advantageous choices in our control group.

he results of our amnesic patients provide further support forhis view. Only two of our 11 amnesic patients had learned aehavioural preference by the end of the last trial block, one fordvantageous and one for disadvantageous choices. The perfor-ance of the other nine patients was not different from chance.ote that this differentiates our patients from the patients withentromedial PFC or amygdala damage examined by Bechara,amasio, Damasio, and Lee (1999), who exhibited a preference

or disadvantageous choices. Since our patients had brain dam-ge including regions damaged in these patients it is of interest toompare the SCR pattern of these different patient groups. Justs healthy controls, both amnesic groups showed largest SCRso the highest levels of punishment (deck B). In contrast, punish-

enerate particularly large anticipatory SCRs. We previouslyxamined a patient with a cerebellar lesion and affective blunt-ng who – just as patients with ventromedial PFC damage –earned a preference for disadvantageous choices, yet – just asealthy controls – showed greater anticipatory SCRs to disad-antageous compared to advantageous choices (Annoni, Ptak,aldara-Schnetzer, Khateb, & Zei Pollermann, 2003). Thesebservations suggest that the relation between anticipatory SCRsnd overt behaviour may not be as straightforward as proposedy Bechara et al. (1997), and that a lack of explicit task knowl-dge may be sufficient to explain why most of our patients failedo acquire a behavioural preference in the IGT.

Nonetheless, this conclusion appears to be inconsistent withhe fact that amnesic patients demonstrate intact or at least sig-ificant learning abilities in several tasks that are quite similar tohe IGT. For example, many amnesic patients show intact per-ormance in the Wisconsin Card Sorting Test (WCST, Heaton,helune, Talley, Kay, & Curtis, 1993), a test measuring categor-

cal learning. All but one of our 11 patients showed normal per-ormance in the WCST. In the weather forecast task developedy Knowlton et al. (1994, 1996) patients are asked to predictossible outcomes that are probabilistically linked to differentisual patterns. They show intact acquisition of probabilisticelations, at least in the first 50 trials. Amnesic patients also showntirely intact learning of probabilistic relations between spa-ial sequences and different choice alternatives (Ptak, Gutbrod,errig, & Schnider, 2001). All these are associative learning

asks in which participants acquire relations between severalhoice stimuli and associated reinforcement schedules. Also,n all these tasks single trials do not provide conclusive feed-ack and – just as in the IGT – information about inherent

K. Gutbrod et al. / Neuropsychologia 44 (2006) 1315–1324 1323

contingencies can only be acquired across several trials. Why,if amnesic patients show learning in such similar tasks, did theyscore at chance in the IGT? The reason for their failure maybe a slight, but significant methodological difference betweenour and other studies. In the WCST and the probabilistic learn-ing tasks discussed above, reinforcement follows immediately achoice made by the participant. In contrast, in our study rein-forcements were given 10 s after the participant had selected acard. This long delay was necessary to allow anticipatory SCRsto decrease to baseline level, but made it exceedingly difficult foramnesic patients to notice the contingency between their choiceand the following reward/punishment. Such detrimental effectsof a delay have been observed in other associative learning tasks.For example, amnesic patients show intact classical conditioningwhen the unconditioned stimulus is presented immediately afterthe conditioned stimulus, but are impaired when a small inter-val separates the two (Clark & Squire, 1998; Daum, Channon,Polkey, & Gray, 1991; McGlinchey-Berroth, Carrillo, Gabrieli,Brawn, & Disterhoft, 1997). The delay-effect may also explainthe conflicting finding of Turnbull and Evans (2006) whoseamnesic patient showed entirely intact learning in the IGT. Theauthors did use the same version of the task, but did not measureSCRs and consequently gave immediate rewards/punishments.Interestingly, a similar difficulty to build associations betweenchoice and outcome has been observed in healthy participantswho performed the IGT while being engaged in a secondary taskt2rd

riT&io

A

sidf

R

A

B

B

Bechara, A., Damasio, H., Damasio, A. R., & Lee, G. P. (1999). Dif-ferent contributions of the human amygdala and ventromedial pre-frontal cortex to decision-making. Journal of Neuroscience, 19(13), 5473–5481.

Bechara, A., Damasio, H., Tranel, D., & Damasio, A. R. (1997). Decidingadvantageously before knowing the advantageous strategy. Science, 275,1293–1295.

Bechara, A., Damasio, H., Tranel, D., & Damasio, A. R. (2005). The Iowagambling task and the somatic marker hypothesis: Some questions andanswers. Trends in Cognitive Sciences, 9(4), 159–162.

Bechara, A., Tranel, D., & Damasio, H. (2000). Characterization of thedecision-making deficit of patients with ventromedial prefrontal cortexlesions. Brain, 123, 2189–2202.

Bechara, A., Tranel, D., Damasio, H., & Damasio, A. R. (1996). Failure torespond autonomically to anticipated future outcomes following damageto prefrontal cortex. Cerebral Cortex, 6, 215–225.

Bowman, C. H., Evans, C. E. Y., & Turnbull, O. H. (2005). Artificialtime constraints on the Iowa gambling task: The effects on behaviouralperformance and subjective experience. Brain and Cognition, 57, 21–25.

Clark, L., Cools, R., & Robbins, T. W. (2004). The neuropsychology ofventral prefrontal cortex: Decision-making and reversal learning. Brainand Cognition, 55, 41–53.

Clark, R. E., & Squire, L. R. (1998). Classical conditioning and brain sys-tems: The role of awareness. Science, 280, 77–81.

Cohen, N. J., & Squire, L. R. (1980). Preserved learning and retentionof pattern-analysing skill in amnesia: Dissociation of knowing how andknowing that. Science, 210, 207–210.

Damasio, A. R. (1994). Descartes’ error. Emotion, reason and the humanbrain. New York: G.P. Putnam’s Sons.

Damasio, A. R. (1998). The somatic marker hypothesis and the possible

D

D

D

E

F

F

H

H

H

K

K

M

hat occupied working memory (Hinson, Jameson, & Whitney,002). These converging findings suggest that awareness of aelation between a stimulus and its outcome is a necessary con-ition for learning to occur.

In sum, our study shows that a severe memory impairmentesults in a failure to acquire a behavioural preference – bet for advantageous or disadvantageous choices – in the IGT.hese findings provide support for several recent reports (MaiaMcClelland, 2004) that explicit knowledge may be of greater

mportance for successful performance in the IGT than previ-usly suggested.

cknowledgements

This work was part of the Masters degree of CK at the Univer-ity of Bern. We are grateful to all participants of this study. RPs supported by grants from the Swiss National Science Foun-ation (no. 3100-100 156) and the de Reuter and Schmidheinyoundations (Geneva).

eferences

nnoni, J.-M., Ptak, R., Caldara-Schnetzer, A. S., Khateb, A., & Zei Poller-mann, B. (2003). Decoupling of autonomic and cognitive emotionalreactions after cerebellar stroke. Annals of Neurology, 53, 654–658.

auer, R. M., & Verfaellie, M. (1992). Memory dissociations: A cognitivepsychophysiology perspective. In L. R. Squire & N. Butters (Eds.), Neu-ropsychology of memory (2nd ed., pp. 58–71). New York: The GuilfordPress.

echara, A., Damasio, A. R., Damasio, H., & Anderson, S. W. (1994). Insen-sitivity to future consequences following damage to human prefrontalcortex. Cognition, 50, 7–15.

functions of the prefrontal cortex. In A. C. Roberts, T. W. Robbins,& L. Weiskrantz (Eds.), The prefrontal cortex. Executive and cognitivefunctions (pp. 36–50). Oxford: Oxford University Press.

amasio, A. R., Tranel, D., & Damasio, H. (1991). Somatic markers and theguidance of behaviour. In H. Levin, H. Eisenberg, & A. Benton (Eds.),Frontal lobe function and dysfunction (pp. 217–228). New York: OxfordUniversity Press.

aum, I., Channon, S., Polkey, C. E., & Gray, J. A. (1991). Classical con-ditioning after temporal lobe lesions in man: Impairment in conditionaldiscrimination. Behavioural Neuroscience, 105(3), 396–408.

enburg, N. L., Tranel, D., & Bechara, A. (2005). The ability to decideadvantageously declines prematurely in some normal older persons. Neu-ropsychologia, 43(7), 1099–1106.

vans, C. Y., Bowman, C. H., & Turnbull, O. H. (2005). Subjective awarenesson the Iowa gambling task: The key role of emotional experience inschizophrenia. Journal of Clinical and Experimental Neuropsychology,27, 656–664.

ellows, L. K., & Farah, M. J. (2003). Ventromedial frontal cortex mediatesaffective shifting in humans: Evidence from a reversal-learning paradigm.Brain, 126, 1830–1837.

ellows, L. K., & Farah, M. J. (2005). Different underlying impairmentsin decision-making following ventromedial and dorsolateral frontal lobedamage in humans. Cerebral Cortex, 15, 58–63.

amann, S. B., Cahill, L., & Squire, L. R. (1997). Emotional perception andmemory in amnesia. Neuropsychology, 11(1), 104–113.

eaton, R. K., Chelune, G. J., Talley, J. L., Kay, G. G., & Curtis, G.(1993). Wisconsin card sorting test (WCST) manual revised and expanded.Odessa, FL: Psychological Assessment Resources Inc.

inson, J. M., Jameson, T. L., & Whitney, P. (2002). Somatic markers, work-ing memory, and decision making. Cognitive, Affective, and BehaviouralNeuroscience, 2(4), 341–353.

nowlton, B. J., Mangels, J. A., & Squire, L. R. (1996). A neostriatal habitlearning system in humans. Science, 273(5280), 1399–1402.

nowlton, B. J., Squire, L. R., & Gluck, M. (1994). Probabilistic classifica-tion learning in amnesia. Learning and Memory, 1, 106–120.

aia, T. V., & McClelland, J. L. (2004). A reexamination of the evidencefor the somatic marker hypothesis: What participants really know in the

1324 K. Gutbrod et al. / Neuropsychologia 44 (2006) 1315–1324

Iowa gambling task. Proceedings of the National Academy of Sciences,101(45), 16075–16080.

Manes, F., Sahakian, B., Clark, L., Rogers, R., Antoun, N., Aitken, M., etal. (2002). Decision-making processes following damage to the prefrontalcortex. Brain, 125, 624–639.

McGlinchey-Berroth, R., Carrillo, M. C., Gabrieli, J. D. E., Brawn, C. M.,& Disterhoft, J. F. (1997). Impaired trace eyeblink conditioning in bilat-eral, medial-temporal lobe amnesia. Behavioural Neuroscience, 111(5),873–882.

Ptak, R., Gutbrod, K., Perrig, W., & Schnider, A. (2001). Probabilistic learn-ing with limbic or prefrontal damage. Behavioural Neuroscience, 115(5),993–1001.

Rorden, C., & Brett, M. (2000). Stereotaxic display of brain lesions.Behavioural Neurology, 12, 191–200.

Schacter, D. L. (1993). Understanding implicit memory: A cognitive neuro-science approach. In A. F. Collins, S. E. Gathercole, M. A. Conway, &P. E. Morris (Eds.), Theories of memory (pp. 387–412). Hove: LawrenceErlbaum Associates.

Shimamura, A. P. (1986). Priming effects in amnesia: Evidence for a disso-ciable memory function. Quarterly Journal of Experimental Psychology,38A, 619–644.

Talairach, T., & Tournoux, P. (1988). Co-planar stereotaxic atlas of the humanbrain: Three-dimensional proportional system—An approach to cerebralimaging. New York: Thieme.

Tomb, I., Hauser, M., Deldin, P., & Caramazza, A. (2002). Do somatic mark-ers mediate decisions on the gambling task? Nature Neuroscience, 5(11),1103–1104.

Tranel, D., & Damasio, A. R. (1993). The covert learning of affective valencedoes not require structures in hippocampal system or amygdala. Journalof Cognitive Neuroscience, 5(1), 79–88.

Turnbull, O. H., & Evans, C. E. Y. (2006). Preserved complex emotion-basedlearning in amnesia. Neuropsychologia, 44(2), 300–306.

Zola-Morgan, S., Squire, L. R., & Amaral, D. G. (1986). Human amnesiaand the medial temporal region: Enduring memory impairment followinga bilateral lesion limited to field ca1 of the hippocampus. Journal ofNeuroscience, 6(10), 2950–2967.