Article

Methamphetamine-Induced Disruption ofFrontostriatal Reward Learning Signals:

Relation to Psychotic Symptoms

Javier Bernacer, Ph.D.

Philip R. Corlett, Ph.D.

Pranathi Ramachandra,M.R.C.Psych.

Brady McFarlane, M.R.C.Psych.

Danielle C. Turner, Ph.D.

Luke Clark, Ph.D.

Trevor W. Robbins, Ph.D., F.R.S.

Paul C. Fletcher, M.R.C.Psych.,Ph.D.

Graham K. Murray, M.D., Ph.D.

Objective: Frontostriatal circuitry is criti-cal to learning processes, and its disruptionmay underlie maladaptive decision mak-ing and the generation of psychotic symp-toms in schizophrenia. However, there isa paucity of evidence directly examiningthe role of modulatory neurotransmitterson frontostriatal function in humans. Inorder to probe the effects of modulationon frontostriatal circuitry during learningand to test whether disruptions in learningprocesses may be related to the pathogen-esis of psychosis, the authors exploredthe brain representations of reward predic-tion error and incentive value, two keyreinforcement learning parameters, beforeand after methamphetamine challenge.

Method: Healthy volunteers (N=18) un-derwent functional MRI (fMRI) scanningwhile performing a reward learning taskon three occasions: after placebo, aftermethamphetamine infusion (0.3 mg/kgbody weight), and after pretreatment with400 mg of amisulpride and then metham-phetamine infusion. Brain fMRI represen-tations of learning signals, calculated using

a reinforcement Q-learning algorithm, werecompared across drug conditions.

Results: In the placebo condition, rewardprediction error was coded in the ventralstriatum bilaterally and incentive value inthe ventromedial prefrontal cortex bilater-ally. Reward prediction error and incentivevalue signals were disrupted by metham-phetamine in the left nucleus accumbensand left ventromedial prefrontal cortex,respectively. Psychotic symptoms weresignificantly correlated with incentivevalue disruption in the ventromedial pre-frontal and posterior cingulate cortex.Amisulpride pretreatment did not signifi-cantly alter methamphetamine-inducedeffects.

Conclusions: The results demonstratethat methamphetamine impairs brain rep-resentations of computational parametersthat underpin learning. They also demon-strate a significant link between psychosisand abnormal monoamine-regulated learn-ing signals in the prefrontal and cingulatecortices.

Am J Psychiatry Bernacer et al.; AiA:1–9

The processes of making a prediction, acting on theprediction, registering anymismatch between expectationand outcome, and then updating future predictions in thelight of any mismatch are critical factors underpinninglearning and decision making. Considerable evidencedemonstrates that the striatum is a key structure involvedin registering mismatches between expectation and out-come (prediction errors) (1). Representations of the valueof stimuli and actions (incentive values) are encoded inventral and medial parts of the frontal cortex (2). Availableevidence, mainly from experimental animal research,suggests that the monoamine neuromodulators norepi-nephrine, serotonin, and especially dopamine play animportant role in neural computations of incentive valueand reward prediction error (3–6). These findings frombasic neuroscience have considerable implications forour understanding of psychiatric illness, given that manypsychiatric disorders have been shown to involve frontal orstriatal pathology, monoamine neurochemical imbalances,and abnormalities in learning and decision making (7).

It has been suggested that in schizophrenia, abnormallearning of associations and inappropriate attribution ofincentive salience and value could be key processesunderpinning the development of psychotic symptoms(8, 9). Functional MRI (fMRI) studies in unmedicatedpatients and in patients with active psychotic symptomshave shown abnormal frontal, striatal, and limbic markersof prediction error or stimulus value during or afterassociative learning, with some evidence of a correlationbetween abnormal brain learning indices and symptomseverity (10–15). In this study, we sought to explore furtherthe link between prediction error-based learning, brainmechanisms of valuation, and psychotic symptom forma-tion by examining the ability of methamphetamine toinduce transient psychotic symptoms in healthy volun-teers and relating this to the drug’s effect on neurallearning signals.Amphetamines can induce psychotic-like symptoms

even after a single administration, especially at high doses(16–18). Experimental administration of amphetamines

AJP in Advance ajp.psychiatryonline.org 1

and related stimulants, whether in humans or animals, hasproved a long-standing useful preclinical model of aspectsof the pathophysiology of schizophrenia. Themechanismsthrough which amphetamines cause psychotic symptomsare unknown, but through their effects on release ofdopamine (and/or other monoamines) they may inducea disruption of frontostriatal reinforcement value andprediction error learning signals, which in turn maycontribute to the generation of symptoms.

With this in mind, we adopted an approach characteriz-ing 1) the impact on fMRI measures of administration ofan amphetamine (methamphetamine) on neural computa-tions of incentive value and prediction error during learn-ing in healthy volunteers, and 2) the degree to whichmethamphetamine-induced disruption of these neuralcomputations is related to drug-induced changes in mentalstate. In a third session, participants received pretreatmentwith the second-generation antipsychotic amisulpride, a po-tent dopamine D2 receptor antagonist, before methamphet-amine was administered. We included amisulpride in thestudy in the hope of gaining insight into the mechanism ofaction of antipsychotic medication and clarifying the neuro-chemistry of any changes in mental state or brain learningsignals induced by methamphetamine.

We hypothesized that administering methamphetaminewould impair reinforcement learning anddisrupt frontal andstriatal learning signals. We hypothesized furthermore thatindividual differences in the degree of frontal and striatalreinforcement learning signal disruption would be associ-ated with individual differences in the degree to which thedrug induced psychotic symptoms. We reasoned that ifthese hypotheses were to be confirmed, it would strengthenthe evidence supporting a disruption in frontostriatallearning processes in the generation of psychotic symp-toms in schizophrenia and other psychiatric disorders. Athird hypothesis, investigating the mechanism of action ofantipsychotic medication and the precise neurochemicalbasis of methamphetamine-induced changes, was that pre-treatment with amisulpride prior to methamphetamineadministration would mitigate any abnormalities inducedby methamphetamine.

Method

Participants and Pharmacological Conditions

The study was approved by the Cambridgeshire 2 NationalHealth Service research ethics committee. Eighteen healthyvolunteers (11 of them men; mean age, 25.3 years [SD=4.9])without psychiatric or neurological disorders or contraindica-tions for MRI gave written informed consent and were includedin the study. Participants attended on three visits, separated byat least 1 week. In one visit, they received an infusion over 10minutes with a methamphetamine solution (0.3 mg/kg of bodyweight), approximately 1 hour before the scan, and a placebotablet. In another visit, participants received the intravenousmethamphetamine as described above, and they were given anamisulpride tablet (400 mg) approximately 1 hour before the

infusion. In the third visit, they received a saline infusion anda placebo tablet. The order of the visits was pseudorandomizedfor each participant in a counterbalanced manner. Participants,researchers who administered fMRI, and psychiatrists who mea-sured mental state were all blind to the pharmacological con-dition of the visit. One of the male participants was excludedbecause of an error during drug administration.

Reinforcement Learning Task

During the fMRI scan, participants carried out an instrumentaldiscrimination learning task with probabilistic feedback thatrequired making choices to maximize wins and minimize losses(Figure 1; see also the data supplement that accompanies theonline edition of this article). In each trial, one of three possiblepairs of abstract pictures was randomly presented: rewarding,punishing, or neutral. There were 90 trials per visit in total.Selection of one of the pictures (by button press) would lead toa particular outcome (a picture of a £1 coin in rewarding trials,a red cross over a £1 coin in punishing trials, and a purple circlethe same size of the coin in neutral trials) with a 70% probability,whereas selection of the other picture led to the outcome with30% probability.

Rating Scales and Behavioral Analyses

Immediately after the fMRI scan, participants were inter-viewed by an experienced psychiatrist who had passed themembership examination of the Royal College of Psychiatriststo measure the severity of any mild (prodromal) psychotic symp-toms (Comprehensive Assessment of At-Risk Mental States, sub-scales 1.1, 1.2, and 1.3) (19).

Computational Model

We estimated reward prediction error and incentive valueparameters for each trial by following a basic Q-learningalgorithm (20), as described elsewhere (11, 21; see also theonline data supplement).

fMRI Data Acquisition and Analysis

Brain imaging data were collected using a 3-T Siemens TIMTrio system and analyzed in the FMRIB Software Library (FSL;http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/). (Details of data acquisitionand preprocessing are provided in the online data supplement.)Ten explanatory variables were defined in our statistical model:1) incentive value of the chosen cue during reward trials asestimated by our computational model; 2) incentive value of thechosen cue during punishment trials; 3) incentive value of thechosen cue during neutral trials; 4) onset of reward cue; 5) onsetof punishment cue; 6) onset of neutral cue (no parametricmodulator was used for these three onset variables); 7) rewardprediction error during valenced outcomes (reward or punish-ment trials); 8) prediction error during neutral outcomes; 9)outcomes on reward or punishment trials; and 10) outcomes onneutral trials. We decided to include both valences in the samereward prediction error regressor because of evidence that themesostriatal reward prediction error signal is coded in the sameway in reward and punishment trials (22). Incentive values werecoded separately for reward and punishment trials because ofprevious evidence that expected values are coded in differentfrontal regions according to valence (23). All these regressorswere modeled as 2-second events and were convolved witha canonical double gamma function. Temporal derivatives of theevents were added to the model. We focused on examining drugeffects on brain representations of the reward prediction errorsignal (i.e., significant parameter estimate for explanatory vari-able 7 with respect to the residuals) or of the incentive value ofthe selected action in reward trials (significant parameterestimate for explanatory variable 1 with respect to the residuals).

2 ajp.psychiatryonline.org AJP in Advance

METHAMPHETAMINE-INDUCED DISRUPTION OF FRONTOSTRIATAL REWARD LEARNING SIGNALS

We also focused our analyses on two brain regions of interest: theventral striatum and the ventromedial prefrontal cortex (some-times referred to as the ventromedial prefrontal cortex/orbitofrontalcortex; see the online data supplement for details on the regions ofinterest).

For group analyses, we used the “randomise” tool from FSL (www.fmrib.ox.ac.uk/fsl/randomise), a permutation-based method (24).We performed 5,000 permutations and smoothed each voxelvariance (3 mm), as recommended for experiments with fewdegrees of freedom (24). All results were thresholded at p,0.05,family-wise error corrected, after threshold-free cluster en-hancement (25), except as otherwise specified. Although ourfocus was on our hypothesized regions of interest, weperformed additional (secondary) whole-brain analyses (cor-rected for multiple comparisons).

Results

Behavioral Results

Psychotic symptoms. Methamphetamine induced mildpsychotic symptoms (as rated by the ComprehensiveAssessment of At-Risk Mental States scale), even when itwas administered together with amisulpride (Figure 2)(placebo condition: score, 0.18 [SD=0.39]; methamphet-amine condition: score, 1.94 [SD=2.44]; methamphet-amine plus amisulpride condition: score, 3.12 [SD=3.28];placebo compared with methamphetamine, t=22.87,df=16, p=0.011; methamphetamine compared with meth-amphetamine plus amisulpride, t=21.57, df=16, p=0.138).

Task performance. There was no effect of drug on theaccuracy of performance. The number of correct choices(“high-likelihood” in reward, “low-likelihood” in punishment)

when decisions on reward and punishment trials weresummed was similar in all visits (placebo condition:total=43.8 [SD=8.9]; methamphetamine condition:total=45.3 [SD=6.1]; methamphetamine plus amisulpridecondition: total=41.4 [SD=9.1]). Participants made ahigher number of correct decisions in win trials than inloss trials, as demonstrated by the effect of valence in the332 (drug-by-valence) repeated-measures analysis ofvariance (ANOVA) (F=7.9, df=1, 16, p=0.013). No drugor drug-by-valence interaction effects were observed.

FIGURE 1. Discrimination Learning Task Used in All Three Pharmacological Conditionsa

Interval(0–4 s)

Fixation(0.5–4.5 s)

Cues(3 s)

Outcome (1.5 s)

Incorrectchoice

Correctchoice

70%

30%

30%

70%

a After a variable intertrial interval, participants were presented one of three pairs of stimuli, each corresponding to a gain, loss, or neutral trial(only gain trial displayed). The high-probability cue led to the main outcome (in this case, a £1 win) seven out of 10 times, whereas the low-probability cue provided the main outcome only 30% of the time.

FIGURE 2. Main Behavioral Differences Between DrugConditionsa

4

5

6

3

2

1

0

CA

AR

MS

sco

re

Placebo

Methamphetamine

Methamphetamineplus amisulpride

a Methamphetamine significantly induced psychotic symptoms involunteers (p=0.011), even with amisulpride pretreatment(p=0.003). Error bars indicate standard deviation. CAARMS=Com-prehensive Assessment of At-Risk Mental States.

AJP in Advance ajp.psychiatryonline.org 3

BERNACER, CORLETT, RAMACHANDRA, ET AL.

Learning parameters to calculate reward prediction error

and incentive values. After testing 100 different values ofthe learning rate constant and of the exploration/exploitation constant, the set of learning parameters thatbest explained participants’ performance was a learningrate constant of 0.43 and an exploration/exploitationconstant of 0.3. These constants were used for all pa-rticipants and pharmacological conditions to calculatereward prediction error and incentive values for the fMRIanalysis for reward, punishment, and neutral trials. Themodel fitted to the behavior was equally good for all threepharmacological conditions (drug effect, F=2.36, df=2, 32,p=0.11; drug-by-valence interaction, F=1.31, df=4, 64,p=0.28) and was significantly better than chance for allconditions (p,0.005 for each condition).

Individual learning parameters. For the behavioral analy-sis, we calculated the individual pair of values that bestexplained the performance of each participant during thetask for the reward and punishment conditions.

Learning rate. A 332 (drug-by-valence) repeated-measures ANOVA with the learning rate as the outcomevariable revealed no effect of any of the factors, nor ofany interaction between them. When considering re-ward and punishment trials as a single learning condi-tion, we observed no effect of drug. In an analysisexamining only reward trials, there was evidence oflearning impairment induced by methamphetamine(F=6.90, df=2, 32, p=0.003). Post hoc tests demonstrated

that the learning rate was reduced under methamphet-amine and restored by amisulpride (see Figure S1 in theonline data supplement).

Exploration/exploitation parameter. There was no effect ofdrug, valence, or their interaction on the exploration/exploitation parameter.

fMRI Results

Placebo. The reward prediction error signal was repre-sented bilaterally in clusters encompassing the nucleusaccumbens and ventral aspects of the caudate nucleusand putamen (p,0.05, family-wise error corrected, withinlimbic striatal region of interest) (Figure 3, Table 1). Voxelswithin the left and right ventromedial prefrontal cortexrepresented the incentive value of the chosen action(p,0.05, family-wise error corrected, within the ventro-medial prefrontal cortex region of interest) (Figure 3,Table 1). No voxels within the ventral striatum repre-sented incentive value, and no voxels within the ventro-medial prefrontal cortex represented reward predictionerror at our chosen statistical threshold. Secondary analysesat the whole-brain level are reported in the online datasupplement (see Figure S3 and Table S1).

Placebo compared with methamphetamine. A cluster lo-cated in the left ventral striatum showed a significantlydisrupted reward prediction error signal after metham-phetamine challenge (placebo . methamphetamine,p,0.05, family-wise error corrected, within the ventral

FIGURE 3. fMRI Results in Placebo Condition for the Reward Prediction Error and Incentive Value Signalsa

Rewardprediction

error

Incentivevalue

Y=10 Z=–8 X=12

Y=36 Z=–10 X=–4

a Thresholded at p,0.05, family-wise error corrected. Left hemisphere is shown in the right side of the image. Coordinates are expressed instandard space, in millimeters.

4 ajp.psychiatryonline.org AJP in Advance

METHAMPHETAMINE-INDUCED DISRUPTION OF FRONTOSTRIATAL REWARD LEARNING SIGNALS

striatal region of interest) (Figure 4, Table 1). In addition,the incentive value signal was attenuated by metham-phetamine in the ventromedial prefrontal cortex (pla-cebo . methamphetamine, p,0.05, family-wise error

corrected, within the ventromedial prefrontal cortexregion of interest) (Figure 4, Table 1). The reversecontrasts (methamphetamine . placebo) showed nosignificant voxels within our region-of-interest analyses.

TABLE 1. Summary of fMRI Results in the Regions of Interest for the Placebo Visit and Drug Effect

Contrast and Analysisa Region Voxels Coordinates (x, y, z) (mm) Peak p

PlaceboReward prediction error analysis Right nucleus accumbens 121 14, 8, –12 0.008

Left ventral caudate nucleus 46 –10, 16, –6 ,0.001Left ventral putamen 24 –26, 6, –4 0.015

Incentive value analysis Left ventromedial prefrontal cortex 227 –4, 34, –14 0.001Placebo . methamphetamineReward prediction error analysis Left ventral caudate nucleus 37 –10, 16, –6 0.007Incentive value analysis Left ventromedial prefrontal cortex 9 –4, 34, –12 0.034a Reward prediction error analyses are restricted to a region of interest including the limbic striatum. Incentive value analyses are restricted toa region of interest including the ventromedial prefrontal cortex.

FIGURE 4. fMRI Differences Between Placebo and Methamphetamine and Correlation With Psychotic Symptomsa

Placebo Methamphetamine

Rewardprediction

error

Incentivevalue

Incentivevalue

Y=14

Y=34

X=0

–0.1

0.1

0.2

0.3

0

0.2

–0.2

0.4

0

0.2

–0.2

–0.4

0.4

0

% S

ign

al Ch

an

ge

% S

ign

al Ch

an

ge

% S

ign

al Ch

an

ge

Symptom severity0 2 4 6 8

Symptom severity0 2 4 6 8

a The top image shows a cluster in the left ventral caudate nucleus indicating a significant difference for reward prediction error signal(p,0.05, family-wise error corrected); the bar graph indicates the drug effect on the signal change extracted for that particular cluster. Errorbars indicate standard deviation. The middle image shows a methamphetamine-disrupted incentive value signal in the ventromedialprefrontal cortex (p,0.05, family-wise error corrected); as shown in the graph, the signal change extracted from this cluster in themethamphetamine visit correlated with symptom severity (Spearman’s rank-order correlation [rs]=–0.54, p=0.025). In the bottom image andgraph, a large significant cluster was found in the posterior cingulate when the incentive value signal in the methamphetamine condition wascorrelated with symptom severity at a whole brain level (p,0.05, family-wise error corrected) (rs=–0.73, p=0.001).

AJP in Advance ajp.psychiatryonline.org 5

BERNACER, CORLETT, RAMACHANDRA, ET AL.

Amisulpride plus methamphetamine compared with meth-

amphetamine alone. Amisulpride did not protect againstthe effects of methamphetamine. No significant clusterswere observed when comparing reward prediction errorand incentive value signal in the methamphetamine con-dition relative to the methamphetamine plus amisulpridecondition.

Correlation with psychotic symptoms. In order to testwhether reward prediction error and incentive valuesignaling were associated with psychotic symptoms in-duced by methamphetamine, we extracted the meanparameter estimate of both learning signals within the twoclusters found in the placebo versus methamphetamineanalysis (in the ventral striatum and ventromedial pre-frontal cortex) and correlated these with psychoticsymptoms as measured in the methamphetamine condi-tion. No correlation was found between the striatal rewardprediction error signal in methamphetamine and psy-chotic symptoms (Spearman’s rank-order correlation[rs]=0.22, p=0.39). However, the incentive value extractedfrom the ventromedial prefrontal cortex cluster showeda negative correlation with symptom severity (rs=20.54,p=0.025), demonstrating that participants with a poorerincentive value signal in the ventromedial prefrontalcortex experienced more severe psychotic symptoms(Figure 4). Amisulpride significantly reduced the strengthof the correlation between the ventromedial prefrontalcortex incentive value signal and psychotic symptoms(Steiger’s z-test=2.1, p=0.04).

Our region-of-interest analyses were supplementedwith a regression analysis at the whole-brain level (correct-ing for multiple comparisons across all brain voxels) toexamine whether any additional regions showed associa-tions between drug-induced psychotic symptoms andlearning signals. This analysis revealed a cluster centeredin the posterior cingulate cortex in which the incentivevalue parameter estimates were negatively associated withthe severity of psychotic symptoms (p,0.05, family-wiseerror corrected) (Figure 4). Amisulpride did not signifi-cantly alter the strength of this association (Steiger’sz-test=1.41). Reward prediction error correlations withsymptom severity were statistically nonsignificant at theselected threshold. There were no correlations with manicsymptoms within either region of interest or at the whole-brain level, suggesting a degree of specificity for therelationship between ventromedial prefrontal cortex andposterior cingulate incentive value representation andmethamphetamine-induced psychotic experience.

Discussion

Our study yielded the following findings: 1) intravenousmethamphetamine induced mild psychotic symptomsin healthy volunteers; 2) methamphetamine significantlyattenuated the reward prediction error signal in the limbicstriatum and significantly attenuated the incentive value

signal in the ventromedial prefrontal cortex; 3) metham-phetamine induced behavioral changes in learning,leading to lower learning rates during reward-relatedreinforcement learning; 4) the degree to which metham-phetamine disrupted the encoding of incentive values inthe ventromedial prefrontal cortex correlated with thedegree to which the drug induced mild psychoticsymptoms; 5) the degree to which methamphetaminedisrupted the encoding of incentive values in the posteriorcingulate correlated with the degree to which the druginduced mild psychotic symptoms; and 6) pretreatmentwith amisulpride did not alter symptoms or the ventrome-dial prefrontal cortex incentive value signal, but it didalter the relationship between the ventromedial prefrontalcortex incentive value signal andmild psychotic symptoms.According to an influential account of psychotic symp-

tom formation, a disturbance in the ways that affectedindividuals evaluate stimuli and learn associations leads tomistaken evaluation of irrelevant phenomena as motiva-tionally salient and to faulty association of unconnectedideas and events, ultimately leading to the emergence ofcharacteristic alterations in perceptions and beliefs (8, 9,26). In this study, we show that a drug intervention thatinduces psychotic symptoms is also associated withdisruption of frontal and striatal neural learning signals.Moreover, the degree to which methamphetamine disrup-ted representations of incentive value in the ventromedialprefrontal cortex was correlated with the degree to which itinduced psychotic symptoms, shedding light on themechanisms of how amphetamines cause psychosis andincreasing support for the argument that brainmechanismsof learning about incentive value and motivational im-portance are involved in the pathogenesis of psychoticsymptoms in schizophrenia.Considerable evidence has implicated frontal lobe func-

tion as being critical in the pathophysiology of schizophre-nia (27). Previous research has extensively demonstratedthe importance of the ventromedial prefrontal cortex in therepresentation of action value in rewarding events (23, 28,29). Here, we suggest an implication of neurochemicaldisruption of prefrontal value computation: the genera-tion of psychotic symptoms. This is in keeping withprevious evidence demonstrating an association betweenmedial frontal lobe function during learning and psy-chotic symptoms in schizophrenia, although as far as weare aware, a specific disruption of cortical incentive valuesignaling has yet to be described in schizophrenia (30).An additional novel finding of our study is to implicatea link between a disruption of incentive value signaling inthe posterior cingulate with the psychotogenic effects ofmethamphetamine. Although we did not have a specifichypothesis about the effects of methamphetamine in thisregion, we note that this region has previously beenshown to encode information about reward value (31)and that fMRI studies of memory and learning havedocumented posterior cingulate dysfunction in actively

6 ajp.psychiatryonline.org AJP in Advance

METHAMPHETAMINE-INDUCED DISRUPTION OF FRONTOSTRIATAL REWARD LEARNING SIGNALS

psychotic patients and ketamine-induced psychoticstates (11, 32).Behavioral analyses confirmed that methamphetamine

mildly disrupted learning. While the number of correctdecisions was not affected bymethamphetamine, we founda deleterious effect of methamphetamine on learning ratesin reward trials. We speculate that methamphetamine-induced disruption of monoamine signaling led toincreasing levels of uncertainty and a consequent delete-rious effect on learning. Impairments in the precision ofthe brain computations underlying learning introduceuncertainty into environmental appraisal, which, accord-ing to Bayesian accounts of belief updating, may lead tosmall prediction errors being given undue weight andconsequent false inference (26, 33). Abnormal decisionmaking under uncertainty is an important factor predis-posing individuals to delusion formation in psychoticillness (34). Recent accounts of dopamine function inlearning emphasize that dopamine neuron firing encodesnot only information about expected value and predictionerror but also information about their precision (35), suchas the variance.Our study shows that methamphetamine, a drug

known to increase synaptic dopamine levels, affects bothbehavioral and brain representations of learning param-eters along with a link to mental state changes that aretypical of the early stages of psychosis. Thus, our resultsare consistent with the theory that a hyperdopaminergicstate leads to psychotic phenomenology because of dis-ruption in dopamine’s role in evaluation and learning ofassociations of stimuli. However, as methamphetamineaffects not only dopamine but also norepinephrine and, toa lesser extent, serotonin, our study allows us to drawdirect inferences about how amphetamines may impairdecision making and induce psychotic symptoms via a“hypermonoaminergic” state, but not about a hyperdopa-minergic state specifically. Pretreatment with 400 mgamisulpride before administration of methamphetaminehad no effect on symptoms or brain response, and thuswe cannot conclusively demonstrate that methamphet-amine’s effects on brain reward learning signals weremediated by stimulation of dopamine D2 receptors.Indeed, the lack of effect of amisulpride may suggest analternative explanation, that methamphetamine’s effectson symptoms and brain learning signals may have beenmediated by nondopaminergic mechanisms, such as itseffects on norepinephrine release, or possibly throughdopaminergic effects on dopamine D1 receptors, whichwould not have been blocked by amisulpride. Amisulpridepretreatment did modulate the relationship betweenmethamphetamine-induced alterations in prefrontal func-tion and psychotic symptoms. This suggests that althoughamisulpride does not normalize methamphetamine’sdeleterious effects on ventromedial prefrontal incentivevalue signaling, it may reduce the tendency of these ven-tromedial prefrontal disruptions to manifest in psychotic

symptoms. The subtle effects of amisulpride in this studymake it hard to draw conclusive interpretations concerningthe relative contributions of norepinephrine and dopamineto the results. If we had used a higher dose of amisulpride,it might have resulted in clearer results, though at the riskof possibly inducing Parkinsonian side effects in thevolunteers.To our knowledge, only one previous study has ex-

amined the effect of a prodopaminergic drug on brainrepresentations of reward learning parameters in healthyhumans; that study found that levodopa improved striatalrepresentations of reward prediction error during learning(21). We now show that an amphetamine-induced hyper-monoaminergic state can be associated with impairments inboth frontal and striatal representations of incentive valueand reward prediction error. The divergence between ourresults and the previous study can be resolved when con-sidering that levodopa, being a dopamine precursor,facilitates stimulus-locked dopamine release. Methamphet-amine, however, can cause stimulus-independent release ofmonoamines through its actions on the dopamine, norepi-nephrine, and serotonin transporters and the vesicularmonoamine transporter-2, changing the balance of phasicand tonic release of monoamines, thus reducing the signal-to-noise ratio during learning and potentially reducingfrontostriatal transmission via stimulation of presynapticdopamine D2 receptors (36–38). This finding is highlyrelevant in understanding both how stimulant intoxicationmay lead to maladaptive learning and induce psychiatricsymptoms (16) and how a hyperdopaminergic state inpsychotic illness could be accompanied by impairedassociative learning (39, 40).In summary, our study provides new evidence of the

role of monoaminergic frontostriatal function in under-pinning feedback-based learning in humans. We show, forthe first time, that a drug-induced state can disrupt neuralrepresentations of computational parameters necessaryfor reinforcement learning, while causing a reducedlearning rate; these findings have implications for un-derstanding decision-making impairments seen in am-phetamine intoxication. The finding that the degree towhich methamphetamine induces psychotic symptomsis related to the degree to which the drug affects theencoding of incentive value in the frontal and cingulatecortices is consistent with theories linking abnormalmechanisms of learning and incentive valuation in thepathogenesis of psychosis.

Received July 27, 2012; revisions received Dec. 26, 2012, and Jan.21, 2013; accepted Feb. 14, 2013 (doi: 10.1176/appi.ajp.2013.12070978). From the Department of Psychiatry and the Behaviouraland Clinical Neuroscience Institute, University of Cambridge, Cam-bridge, U.K.; the Institute for Culture and Society, University ofNavarra, Navarra, Spain; Cambridgeshire and Peterborough NHSFoundation Trust; and the Department of Psychiatry, Yale University,New Haven, Conn. Address correspondence to Dr. Murray (gm285@cam.ac.uk).

AJP in Advance ajp.psychiatryonline.org 7

BERNACER, CORLETT, RAMACHANDRA, ET AL.

Dr. Corlett has received research support from AstraZeneca, LP,and Pfizer and has served as a consultant for Pfizer. Dr. Turner andDr. Clark have served as consultants to Cambridge Cognition. Dr.Robbins has received research support from or served as a consultantto Cambridge Cognition, Chempartners, Eli Lilly, GlaxoSmithKline,Lundbeck, Merck, Shire, and Teva. Dr. Fletcher has served asa consultant for GlaxoSmithKline. The remaining authors report nofinancial relationships with commercial interests.Supported by a Clinical Scientist Award to Dr. Murray from the

Medical Research Council; by the University of Cambridge Behav-ioural and Clinical Neuroscience Institute, funded by a joint awardfrom the Medical Research Council and Wellcome Trust; by a CajaMadrid Foundation postdoctoral fellowship to Dr. Bernacer; byawards from the Wellcome Trust and the Bernard Wolfe HealthNeuroscience Fund to Dr. Fletcher; and by the Wellcome Trust ClinicalResearch Facility (WTCRF) at Addenbrooke’s Hospital. This work waspartly conducted at the Clinical Neuroscience Research Unit,Connecticut Mental Health Center. The authors recognize the supportof the Connecticut Department of Mental Health and AddictionServices. This publication was also made possible by Clinical andTranslational Science Awards grant UL1 RR024139 from the NationalCenter for Research Resources and the National Center for AdvancingTranslational Science, components of NIH, and the NIH Roadmap forMedical Research. Its contents are solely the responsibility of theauthors and do not necessarily represent the official view of NIH.The authors are grateful to the staff at WTCRF and the Wolfson

Brain Imaging Centre for help with data collection.

References

1. O’Doherty JP: Reward representations and reward-related learn-ing in the human brain: insights from neuroimaging. Curr OpinNeurobiol 2004; 14:769–776

2. Levy DJ, Glimcher PW: The root of all value: a neural commoncurrency for choice. Curr Opin Neurobiol 2012; 22:1027–1038

3. Schultz W: Dopamine signals for reward value and risk: basicand recent data. Behav Brain Funct 2010; 6:24

4. Jocham G, Ullsperger M: Neuropharmacology of performancemonitoring. Neurosci Biobehav Rev 2009; 33:48–60

5. Hikosaka O, Bromberg-Martin E, Hong S, Matsumoto M: Newinsights on the subcortical representation of reward. Curr OpinNeurobiol 2008; 18:203–208

6. Cools R, Robinson OJ, Sahakian B: Acute tryptophan depletionin healthy volunteers enhances punishment prediction butdoes not affect reward prediction. Neuropsychopharmacology2008; 33:2291–2299

7. Chudasama Y, Robbins TW: Functions of frontostriatal systemsin cognition: comparative neuropsychopharmacological studiesin rats, monkeys, and humans. Biol Psychol 2006; 73:19–38

8. Miller R: Schizophrenic psychology, associative learning, and therole of forebrain dopamine. Med Hypotheses 1976; 2:203–211

9. Kapur S: Psychosis as a state of aberrant salience: a frameworklinking biology, phenomenology, and pharmacology in schizo-phrenia. Am J Psychiatry 2003; 160:13–23

10. Corlett PR, Honey GD, Fletcher PC: From prediction error topsychosis: ketamine as a pharmacological model of delusions.J Psychopharmacol 2007; 21:238–252

11. Murray GK, Corlett PR, Clark L, Pessiglione M, Blackwell AD,Honey G, Jones PB, Bullmore ET, Robbins TW, Fletcher PC:Substantia nigra/ventral tegmental reward prediction errordisruption in psychosis. Mol Psychiatry 2008; 13:267–276

12. Gradin VB, Kumar P, Waiter G, Ahearn T, Stickle C, Milders M, Reid I,Hall J, Steele JD: Expected value and prediction error abnormalitiesin depression and schizophrenia. Brain 2011; 134:1751–1764

13. Morris RW, Vercammen A, Lenroot R, Moore L, Langton JM,Short B, Kulkarni J, Curtis J, O’Donnell M, Weickert CS, WeickertTW: Disambiguating ventral striatum fMRI-related bold signalduring reward prediction in schizophrenia. Mol Psychiatry2012; 17:280–289

14. Corlett PR, Murray GK, Honey GD, Aitken MR, Shanks DR, RobbinsTW, Bullmore ET, Dickinson A, Fletcher PC: Disrupted prediction-error signal in psychosis: evidence for an associative account ofdelusions. Brain 2007; 130:2387–2400

15. Corlett PR, Honey GD, Aitken MRF, Dickinson A, Shanks DR,Absalom AR, Lee M, Pomarol-Clotet E, Murray GK, McKenna PJ,Robbins TW, Bullmore ET, Fletcher PC: Frontal responses duringlearning predict vulnerability to the psychotogenic effects ofketamine: linking cognition, brain activity, and psychosis. ArchGen Psychiatry 2006; 63:611–621

16. Connell PH: Amphetamine Psychosis. London, MaudsleyMonograph, Chapman and Hill, 1958

17. Bell DS: The experimental reproduction of amphetamine psy-chosis. Arch Gen Psychiatry 1973; 29:35–40

18. Angrist B, Sathananthan G, Wilk S, Gershon S: Amphetaminepsychosis: behavioral and biochemical aspects. J Psychiatr Res1974; 11:13–23

19. Yung AR, Yuen HP, McGorry PD, Phillips LJ, Kelly D, Dell’Olio M,Francey SM, Cosgrave EM, Killackey E, Stanford C, Godfrey K,Buckby J: Mapping the onset of psychosis: the ComprehensiveAssessment of At-Risk Mental States. Aust N Z J Psychiatry 2005;39:964–971

20. Sutton R, Barto A: Reinforcement Learning. Cambridge, Mass,MIT Press, 1998

21. Pessiglione M, Seymour B, Flandin G, Dolan RJ, Frith CD:Dopamine-dependent prediction errors underpin reward-seekingbehaviour in humans. Nature 2006; 442:1042–1045

22. D’Ardenne K, McClure SM, Nystrom LE, Cohen JD: BOLDresponses reflecting dopaminergic signals in the human ventraltegmental area. Science 2008; 319:1264–1267

23. Blair K, Marsh AA, Morton J, Vythilingam M, Jones M, MondilloK, Pine DC, Drevets WC, Blair JR: Choosing the lesser of two evils,the better of two goods: specifying the roles of ventromedialprefrontal cortex and dorsal anterior cingulate in object choice.J Neurosci 2006; 26:11379–11386

24. Nichols TE, Holmes AP: Nonparametric permutation tests forfunctional neuroimaging: a primer with examples. Hum BrainMapp 2002; 15:1–25

25. Smith SM, Nichols TE: Threshold-free cluster enhancement:addressing problems of smoothing, threshold dependence, andlocalisation in cluster inference. Neuroimage 2009; 44:83–98

26. Fletcher PC, Frith CD: Perceiving is believing: a Bayesian ap-proach to explaining the positive symptoms of schizophrenia.Nat Rev Neurosci 2009; 10:48–58

27. Tan H-Y, Callicott JH, Weinberger DR: Dysfunctional and com-pensatory prefrontal cortical systems, genes, and the patho-genesis of schizophrenia. Cereb Cortex 2007; 17(suppl 1):i171–i181

28. Hare TA, O’Doherty J, Camerer CF, Schultz W, Rangel A: Disso-ciating the role of the orbitofrontal cortex and the striatum inthe computation of goal values and prediction errors. J Neu-rosci 2008; 28:5623–5630

29. Gläscher J, Hampton AN, O’Doherty JP: Determining a role forventromedial prefrontal cortex in encoding action-based valuesignals during reward-related decision making. Cereb Cortex2009; 19:483–495

30. Schlagenhauf F, Sterzer P, Schmack K, Ballmaier M, Rapp M,Wrase J, Juckel G, Gallinat J, Heinz A: Reward feedback alter-ations in unmedicated schizophrenia patients: relevance fordelusions. Biol Psychiatry 2009; 65:1032–1039

31. Kolling N, Behrens TE, Mars RB, Rushworth MF: Neural mech-anisms of foraging. Science 2012; 336:95–98

32. Northoff G, Richter A, Bermpohl F, Grimm S, Martin E, MarcarVL, Wahl C, Hell D, Boeker H: NMDA hypofunction in the pos-terior cingulate as a model for schizophrenia: an exploratoryketamine administration study in fMRI. Schizophr Res 2005; 72:235–248

8 ajp.psychiatryonline.org AJP in Advance

METHAMPHETAMINE-INDUCED DISRUPTION OF FRONTOSTRIATAL REWARD LEARNING SIGNALS

33. Mathys C, Daunizeau J, Friston KJ, Stephan KE: A Bayesianfoundation for individual learning under uncertainty. FrontHum Neurosci 2011; 5:39

34. Moritz S, Woodward T: Plausibility judgment in schizophrenicpatients: evidence for a liberal acceptance bias. Ger J Psychiatry2004; 7:66–74

35. Tobler PN, Fiorillo CD, Schultz W: Adaptive coding of rewardvalue by dopamine neurons. Science 2005; 307:1642–1645

36. Knutson B, Bjork JM, Fong GW, Hommer D, Mattay VS, WeinbergerDR: Amphetamine modulates human incentive processing. Neu-ron 2004; 43:261–269

37. Cruickshank CC, Dyer KR: A review of the clinical phar-macology of methamphetamine. Addiction 2009; 104:1085–1099

38. O’Donnell P, Grace AA: Dysfunctions in multiple interrelatedsystems as the neurobiological bases of schizophrenic symptomclusters. Schizophr Bull 1998; 24:267–283

39. Robbins TW: Relationship between reward-enhancing and ste-reotypical effects of psychomotor stimulant drugs. Nature 1976;264:57–59

40. Gray JA: Integrating schizophrenia. Schizophr Bull 1998; 24:249–266

AJP in Advance ajp.psychiatryonline.org 9

BERNACER, CORLETT, RAMACHANDRA, ET AL.