Haloperidol Impairs Learning and Error-related Negativity in Humans

Patrick J. Zirnheld, Christine A. Carroll, Paul D. Kieffaber, Brian F. O’Donnell, Anantha Shekhar, and William P. Hetrick

Abstract & Humans are able to monitor their actions for behavioral Performance Task, the Eriksen Flanker Task, and a learning- conflicts and performance errors. Growing evidence suggests dependent Time Estimation Task. Haloperidol significantly that the error-related negativity (ERN) of the event-related attenuated ERN amplitudes recorded during the flanker task, cortical brain potential (ERP) may index the functioning of impaired learning of time intervals, and tended to cause this response monitoring system and that the ERN may de- more errors of commission, compared to placebo, which pend on dopaminergic mechanisms. We examined the role did not significantly differ from diphenhydramine. Drugs of dopamine in ERN and behavioral indices of learning by had no significant effects on the stimulus-locked P1 and N2 administering either 3 mg of the dopamine antagonist (DA) ERPs or on behavioral response latencies, but tended to af- haloperidol (n = 17); 25 mg of diphenhydramine (n = 16), fect post-error reaction time (RT) latencies in opposite ways which has a similar CNS profile but without DA proper- (haloperidol decreased and diphenhydramine increased ties; or placebo (n = 18) in a randomized, double-blind RTs). These findings support the hypothesis that the DA manner to healthy volunteers. Three hours after drug ad- system is involved in learning and the generation of the ministration, participants performed a go/no-go Continuous ERN. &

INTRODUCTION subjects need to have learned a representation of the Humans are able to monitor their actions for behavioral goal behavior for this negativity to be observed conflicts and performance errors, and then modify (Holroyd & Coles, 2002; Dehaene et al., 1994). Increased subsequent behavior accordingly. Growing evidence amplitude of the ERN may index the occurrence of a suggests that the negativity error (Ne; Falkenstein, discrepancy, or mismatch, between the neural represen- Hohnsbein, Hoormann, & Blanke, 1990) or error-related tations of a given response (e.g., an erroneous response) negativity (ERN; Gehring, Coles, Meyer, & Donchin, compared to the neural representations of the required 1990) of the event-related cortical brain potential (ERP) response on a given experimental trial (Falkenstein, may index the functioning of such a monitoring system. Hoorman, et al., 2000; Falkenstein et al., 1990). There is The Ne/ERN is a negative deflection of the ERP that is evidence that the ERN indicates when the consequences more pronounced on incorrect compared to correct of a response are worse than expected (Holroyd & Coles, trials. This negative-going, frontocentral component (De- 2002). Although emotional engagement in a task may haene, Ponser, & Tucker, 1994) peaks about 50 to facilitate ERN amplitude (Dikman & Allen, 2000), con- 80 msec after a response (Falkenstein et al., 1990; scious awareness of a response error is not necessary Gehring et al., 1990) and may be obtained by averaging for the manifestation of an ERN (Nieuwenhuis, Ridder- the response-locked ERPs for incorrect trials alone (Fal- hinkhof, Bloom, Band, & Kok, 2001; Falkenstein, Hoor- kenstein, Hoormann, Christ, & Hohnsbein, 2000; man, et al., 2000) and faked mistakes do not elicit the ERN Scheffers, Coles, Bernstein, Gehring, & Donchin, 1996; (Stemmer, Witzke, & Scho¨nle, 2001). However, whether Gehring, Goss, Coles, Meyer, & Donchin, 1993). Inter- the ERN represents an error detection process (Gehring estingly, the Ne/ERN appears to index the subjective et al., 1993; Falkenstein et al., 1990), a response evalua- sense that a mistake has been made and is not necessarily tion process (Vidal, Hasbroucq, Grapperon, & Bonnet, indicative of an objectively quantified error (Falkenstein, 2000), or a conflict detection process (see Botvinick, Hoorman, et al., 2000; Scheffers & Coles, 2000). However, Braver, Barch, Carter, & Cohen, 2001, for a review) is still unresolved. Nevertheless, several factors appear critical to the elicitation of the ERN, including knowl- Indiana University edge of the goal behavior (Holroyd & Coles, 2002;

D 2004 Massachusetts Institute of Technology Journal of Cognitive Neuroscience 16:6, pp. 1098–1112 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/0898929041502779 by guest on 26 September 2021 Dehaene et al., 1994) and a subjective sense of inaccu- the outcome, with greater negative valence yielding rate performance (Scheffers & Coles, 2000). larger ERN amplitudes, and (2) ERN amplitude is larger Investigations of the anatomical origin of the ERN are to infrequent negative outcomes than to frequent neg- informative about candidate neural systems involved ative outcomes, because infrequent negative outcomes in response monitoring, including neural pathways and tend to be less predictable. Thus, ERN appears to reflect neurotransmitters. The anterior cingulate cortex and the a valence-sensitive updating process associated with supplementary motor area have been repeatedly identi- learning. Based on these cumulative findings, as well fied as possible sources of the ERN (Mathalon, Faust- as animal data (see Schultz, Dayan, & Montague, 1997, man, Gray, Askari, & Ford, 2002; Dikman & Allen, 2000; for a review) and results from previous computer mod- Luu, Collins, & Tucker, 2000; Holroyd, Dien, & Coles, els of reinforcement learning using the temporal differ- 1998; Dehaene et al., 1994). Consistent with these ob- ence algorithm (Sutton & Barto, 1998), Holroyd and servations, studies of patients with lesions in either the Coles (2002) compellingly articulated a model in which anterior cingulate cortex (Stemmer, Segalowitz, Witzke, they proposed ‘‘(i) that the ERN reflects the transmis- Lacher, & Scho¨nle, 2000) or the dorsolateral prefrontal sion of a reinforcement signal to the anterior cingulate, cortex (Gehring & Knight, 2000) indicate that both (ii) that this error signal is carried by the mesencephalic cortices are necessary for the generation of a normal dopamine system, and (iii) that it is used to train the ERN. Holroyd, Praamstra, Plat, and Coles (2002) rea- anterior cingulate cortex to optimize performance on soned that because the cingulate sulcus is the only the task at hand’’ (p. 686). Indeed, they reviewed that horizontal part of the medial frontal lobe and ERPs the putative ERN source generators, such as the cingu- are primarily generated by postsynaptic potentials, the late cortex and the medial prefrontal cortex, are densely vertical orientation and large size of apical dendrites innervated by dopamine terminals coming from the in this region make the cingulate sulcus a primary can- ventral tegmental area (VTA) (Carr & Sesack, 2000; didate as the ERN generator. Luu and Tucker (2001) Haber & Fudge, 1997; Heimer et al., 1997). In addition, attributed the genesis of the ERN to an asynchrony the medial prefrontal cortex directly projects to the between a midline theta wave oscillation arising from VTA (Au-Young, Shen, & Yang, 1999; Sesack & Pickel, the centro-medial frontal cortex and a lateral theta 1992; Phillipson, 1979) and the cingulate cortex projects wave oscillation arising from the sensory motor cortices. to the VTA via the nucleus accumbens (Haber & Fudge, They argue in favor of the involvement of the anterior 1997; Heimer et al., 1997). cingulate and supplementary motor cortices in the origin Not only do the putative neuronal source generators of the centro-medial oscillation, which would reflect of the ERN and the neural pathways among related ‘‘response checking’’ (prefrontal) and ‘‘error output’’ structures implicate the dopamine system, but findings (anterior cingulate). from behavioral studies of learning and con- In addition to the ‘‘classical’’ ERN, which appears verge on the role of dopamine in attention and learning when subjects experience the subjective sense of having (Margolin, 1978). For example, deficits in procedural made a mistake but have not yet received feedback learning have been observed in both Parkinsonian pa- about their performance, a feedback-related negativity tients (Wallesch et al., 1990) and in normal controls after (FRN) is observed after feedback to incorrect perfor- a small dose of haloperidol (Kumari et al., 1997). Dex- mance (Miltner, Braun, & Coles, 1997). Interestingly, troamphetamine, which increases dopamine synaptic source localization analyses also place the FRN in the release, improves measures of attention and procedural area of the anterior cingulate (Miltner et al., 1997). This learning (Kumari et al., 1997). Furthermore, the dopa- negative-going ERP component, which appears after mine reuptake inhibitors dextroamphetamine and meth- negative feedback, has also been reported by several ylphenidate have both been found to improve attention others (Gehring & Willoughby, 2002; Holroyd & Coles, in patients with attention deficit hyperactivity disorder 2002; Nieuwenhuis et al., 2002; Masaki, Tanaka, Takasa- (Spencer et al., 1995; Arnold, Christopher, Huestis, & wa, & Yamazaki, 2001; Tucker, Hartry-Speicer, McDou- Smeltzer, 1978), and bromocriptine, a direct dopamine gal, Luu, & deGrandpre, 1999; Takasawa, Takino, & agonist, improved verbal learning in a patient who Yamazaki, 1990). Taken together, the evidence indicates sustained a lesion that interrupted dopaminergic path- that the ERN and the FRN index the extent to which a ways (Dopkin & Hanlon, 1993). given response or outcome differs from an expected Moreover, animal research demonstrates that dopa- response or outcome, rather than behavioral inaccuracy mine neurons convey a reward signal used by the brain per se. in learning. Animals treated with dopamine receptor Consistent with this interpretation that ERN and FRN blockers learn less rapidly to press a bar for a reward are manifestations of a response monitoring system, pellet (Beninger, 1989), and although rats with lesions Holroyd and Coles (2002) showed that learning was in the dopamine-rich VTAs could still learn simple associated with progressive increases in ERN amplitude motor behaviors, learning was impaired when task across an experimental session. Their work also demon- complexity increased (Oades, 1982). In addition, a vari- strated that (1) ERN depends critically on the valence of ety of experiments using microelectrode measurements

Zirnheld et al. 1099 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/0898929041502779 by guest on 26 September 2021 have helped explain how the firing rate of dopamine Wieringa, Nager, Dengler, & Mu¨nte, 2001), diphenhy- neurons is coded within the midbrain (Schultz et al., dramine, a muscarinic and histamine blocker, was given 1997; Schultz, 1996; Mirenowicz & Schultz, 1994; Schultz, to a second group to rule out the possibility that these Apicella, & Ljungberg, 1993; Ljungberg, Apicella, & side effects, rather than the dopamine-blocking effects Schultz, 1992; Romo & Schultz, 1990; Schultz, 1986). of haloperidol, accounted for the observed effects. In In the macaque monkey, these neurons increase their addition, a placebo group was included in this double- firing rate when an unexpected reward occurs in the blind randomized design. Three hours after drug admin- environment or when a conditioned cue that is predic- istration, participants performed a go/no-go Continuous tive of a reward is presented. However, if the reward is Performance Task (CPT; adapted from Harvey et al., omitted following conditioning of the cue, a phasic 1990), the Eriksen Flanker Task (Eriksen & Eriksen, decrease in the firing rate of the dopamine neurons 1974), and a learning-dependent Time Estimation Task occurs at the time of reward expectation (Schultz et al., (TET) developed specifically for this study. 1997). These results suggest a role for the VTA projec- tions to the cingulate, the prefrontal cortex and the accumbens nucleus as the conveyor of an error signal, RESULTS which in turn would be used by the frontal and basal Behavioral Data forebrain structures to learn new goals and behaviors. Taken together, the human and animal literatures offer Continuous Performance Task compelling reasons to directly examining the role of the There were no significant medication effects on CPT dopamine system in ERN. accuracy measures, including hits, F(2,47) = 0.54, p = The purpose of the present study was to examine the .586, and false alarms, F(2,47) = 1.96, p = .152. Medi- relationship between the ERN and the dopamine sys- cations did not affect response latency across all trials, tem. If the ERN is the manifestation of frontal gener- F(2,47) = 0.09, p = .913, hits, F(2,47) = 0.04, p = .961, ators, whose activity is either modulated or updated by or false alarms, F(2,23) = 0.484, p = .623 (see Table 1). the signal conveyed by VTA dopamine projections, then Because 27 subjects did not commit any false alarm dopamine-blocking drugs should impair both the ERN errors, false alarm response latencies were only mea- and learning but spare other ERP components such as sured for the 23 remaining subjects (haloperidol: n =5; the stimulus-locked P1 and N2 (Neuwenhuis et al., 2002; diphenhydramine: n = 9; placebo: n = 9). Ridderinkhof et al., 2002; Davies, Segalowitz, Dywan, & Pailing, 2001). This hypothesis is consistent with Hol- Eriksen Flanker Task royd and Coles’ (2002) recent theories and provided a direct test of their model. However, if the putative A significant main effect of medication was found for the frontal generators are not influenced by dopamine, the percentage of errors (i.e., wrong button presses) com- ERN should remain unaffected by dopamine-blocking mitted on the Eriksen Flanker Task, F(2,48) = 3.22, p = drugs, even though learning may still be impaired by .049. Post hoc analyses revealed that the largest differ- dopamine blockade by involvement of other projection ences in percentage of errors were between the haloper- areas of dopamine neurons such as the hippocampus or idol, mean [SD] = 10.03 [5.67], and placebo, mean [SD] the prefrontal cortex. To answer these questions, we = 6.24 [5.13], groups ( p = .095) and between the compared how behavioral and event-related brain po- haloperidol and diphenhydramine, mean [SD] = 5.89 tential measures of attention and learning were affected [4.94], groups ( p = .072). However, the percentage of by a small dose of the potent dopamine D1 and D2 correct responses (i.e., total number of trials, n = 640, blocker haloperidol, compared to placebo and diphen- minus the number of erroneous and nonresponse trials hydramine. Given that haloperidol also has antihista- divided by the total number of trials) did not significantly mine blockade properties and sedatives have been differ between medication conditions: F(2,48) = 1.91, associated with diminished ERN amplitudes (Johannes, p = ns; placebo: mean [SD] = 90.03 [6.02]; diphenhy-

Table 1. Behavioral Measures from Continuous Performance Task

Hits False Alarms No. of Hits RT No. of False alarms RT Mean Response Latency Drug Condition (M ± SD) (M ± SD) (M ± SD) (M ± SD) (M ± SD)

Haloperidol 19.23 ± 2.19 398.16 ± 122.24 .41 ± .62 416.90 ± 334.41 399.30 ± 123.66 Diphenhydramine 18.69 ± 2.87 391.48 ± 90.17 2.25 ± 3.96 304.37 ± 154.88 386.39 ± 79.24 Placebo 18.29 ± 2.84 387.76 ± 109.29 3.00 ± 5.48 338.05 ± 160.53 385.79 ± 102.16

1100 Journal of Cognitive Neuroscience Volume 16, Number 6 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/0898929041502779 by guest on 26 September 2021 Table 2. Reactions Times in Eriksen Flanker Task in Correct and Incorrect Trials, Averaged across Accuracy, and on Trials Immediately Following an Error (i e., Post-error Trials)

Correct RT Incorrect RT Average RT Post-error RT (msec; M ± SD) (M ± SD) (M ± SD) (M ± SD) Haloperidol (n = 17) 515.59 ± 60.18 490.58 ± 85.98 512.94 ± 62.42 557.11 ± 63.24 Diphenhydramine (n = 16) 564.72 ± 92.55 560.32 ± 125.64 565.13 ± 94.20 622.48 ± 99.49 Placebo (n = 18) 521.35 ± 62.24 491.37 ± 79.75 519.99 ± 62.93 579.38 ± 62.29

dramine: mean [SD] = 82.84 [17.15]; haloperidol: mean diphenhydramine groups ( p = .499). Baseline perfor- [SD] = 81.63 [15.89]. The main effect of medication on mance (i.e., initial task performance more than 3 hr after response latencies was not significant (see Table 2). drug administration), F(2,47) = 0.15, p = .865, and end However, there was a trend for medication to affect performance, F(2,47) = 1.31, p = .278, did not signifi- post-error reaction time (RT) slowing (i.e., response cantly differ between medication groups: baseline per- slowing on trials immediately after error trials), F(2,48) = formance, haloperidol: mean [SD] = 7.62 [3.48], 3.12, p = .053. Post hoc analyses revealed that diphen- diphenhydramine: mean [SD] = 7.25 [3.75], placebo: hydramine delayed post-error RTs after stimulus onset, mean [SD] = 6.94 [3.78]; end performance, haloperidol: mean [SD] = 622.48 msec [99.49], compared to halo- mean [SD] = 141.37 [49.74], diphenhydramine: mean peridol, mean [SD] = 557.11 msec [63.24] ( p = .045). A [SD] = 158.81 [77.66], placebo: mean [SD] = 181.67 trend was also observed for incorrect response latencies, [84.39]. Although not significant, the baseline perfor- F(2,48) = 2.73, p = .075: Although the observed mean mance of the haloperidol group was slightly better than latencies were later in the diphenhydramine group, the diphenhydramine group, F(2,47) = 0.14, p = .65 mean [SD] = 560.32 [125.64], compared to placebo, (post hoc: p = .955). Interestingly, the percentage of mean [SD] = 491.37 [79.75], and haloperidol, mean mistakes on the Eriksen Flanker Task correlated with [SD] = 490.58 [85.98], this was not significant ( p = differential performance (r = À .32, p< .05) and with .113, p = .114, respectively). end performance (r = À .40, p< .01) on the Time Estimation Task, indicating that fewer mistakes on the Eriksen task were associated with higher rates of learn- Time Estimation Task ing from the beginning to the end of the Time Estima- A main effect of medication was found for the learning tion Task and better overall performance by the end of measure (i.e., difference between accuracy at baseline this task. and at the end of this task) of the Time Estimation Task, There were no significant drug effects on behavioral F(2,47) = 3.31, p = .045, such that subjects in the response latencies in the Time Estimation Task (see haloperidol group showed significantly less learning Table 3). When response times were averaged across compared to the placebo group, p = .036; differential stimulus conditions, a trend, similar to that observed in performance, haloperidol: mean [SD] = 203.62 [209.02], the Eriksen Flanker Task on post-error slowing, was diphenhydramine: mean [SD] = 315.56 [360.51], place- found for diphenhydramine, mean [SD] = 695.00 bo: mean [SD] = 450.00 [250.99]. In contrast, compared [33.50], to slow overall response latencies compared to placebo, diphenhydramine did not significantly im- to haloperidol, mean [SD] = 589.69 [33.50], F(2,47) = pair learning, and there was also no significant difference 2.93, p = .063 (pairwise comparison using Tukey’s HSD: in differential performance between the haloperidol and p = .078).

Table 3. Reaction Times from the Time-Estimation Task in which Subjects Were Required to Respond at Different Latencies Depending on the Angle of Tilt of the Airplane Wing (308 tilt = 851 to 1200 msec; 458 tilt = 501 to 850 msec; 608 tilt = 0 to 500 msec)

308Tilt Stimuli 458 Tilt Stimuli Requiring 608 Tilt Stimuli Average RT across Requiring Slow RT Medium-Speed RT Requiring Fast All Stimuli (M ± SD) (M ± SD) RT (M ± SD) (M ± SD) Haloperidol (n = 16) 654.40 ± 176.79 593.22 ± 136.83 521.45 ± 85.24 589.69 ± 33.50 Diphenhydramine (n = 16) 805.06 ± 175.02 709.63 ± 151.99 570.31 ± 137.62 695.00 ± 33.50 Placebo (n = 18) 808.21 ± 178.34 692.07 ± 151.06 540.35 ± 154.63 680.21 ± 31.58

Zirnheld et al. 1101 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/0898929041502779 by guest on 26 September 2021 Event-Related Brain Potentials 80, 150, and 200 msec postresponse. As expected for an ERN effect, the plots showed that the error-related effects were largest at the frontocentral sites FCZ and Response-locked ERN CZ at around 60 to 80 msec after the response. More- The response-locked, grand averaged ERP waveforms over, a comparison of scalp voltage distributions shown and scalp voltage maps that were derived separately in Figures 1 and 2 indicates that the group difference from Eriksen Flanker trials on which participants made waveforms exhibited the same scalp distribution as the correct and incorrect responses are presented in Figure 1. ERN on error trials. As expected, the ERN was maximal at frontocentral A repeated measures ANOVA, with electrode (FCZ & recording sites on error trials and peaked 60 to 80 msec CZ) as a repeated measure, showed that there was no after the motor response. On the error trials, a signifi- significant effect of medication on ERN-like activity in cant main effect of medication on ERN amplitude was the correct trials, F(2,48) = 0.849, p = .434, mean [SE] observed in a repeated measures ANOVA, with electrode ERN-like amplitude under haloperidol = À3.10 [1.19]; (FCZ & CZ) as the repeated measure, F(2,48) = 3.79, p = diphenhydramine = À5.87 [1.22]; placebo = À5.62 .029, mean [SE] square root normalized ERN amplitude [1.15]. None of the pairwise group comparisons were under haloperidol = À 2.35 [.20]; diphenhydramine = significant either. À 2.89 [.22]; placebo = À 3.10 [.19] (see Figure 1). The effect size, as indicated by a partial h2 of .14, was ‘‘large’’ Stimulus-locked P1 and N2 (Cohen, 1988, pp. 283–287) and indicated that 14% of variance in ERN amplitude was due to the medication There was no significant effect of medication on the P1 condition. This main effect remained significant when amplitude at OZ where P1 was largest (see Figure 3, top each electrode was examined separately: FCZ, F(2,48) = panel), F(2,45) = 0.342, p = .712, haloperidol mean 3.30, p = .045 and CZ, F(2,48) = 4.15, p = .022. Post hoc [SD] = 6.63 [2.68]; diphenhydramine = 5.66 [3.06]; analyses revealed that ERN amplitudes were significantly placebo = 6.45 [3.91]. P1 latency was not significantly attenuated by haloperidol compared to placebo ( p = affected by medication, F(2,45) = .274, p = .72, halo- .010) and that the difference between haloperidol and peridol mean [SD] = 109.97 [15.03]; diphenhydramine = diphenhydramine approached significance ( p = .066). 114.13 [17.33]; placebo = 111.29 [13,81]. Similarly, no ERN amplitudes in the diphenhydramine group did not effect of drug condition was observed on N2 amplitude differ from placebo ( p = .468). at CPZ (see Figure 3, bottom panel) where N2 was In order to determine that the observed group differ- largest, F(2,48) = 1.00, p = .374, haloperidol: mean ences in ERN amplitude in fact reflected ERN-like activ- [SD]=À 2.28 [3.31]; diphenhydramine = À .78 [3.06]; ity, the scalp voltage maps of difference waveforms were placebo = À 2.68 [5.03]. examined (see Figure 2). For example, the top row of Importantly, the main effect of drug condition on ERN Figure 2 shows the scalp distribution of the difference in amplitude (as measured at CZ and FCZ) remained voltage between the placebo and haloperidol group at 0, significant after entering P1 amplitude at OZ as a cova-

Figure 1. Left: average re- sponse-locked ERP for correct (top) and incorrect (bottom) trials for each of the medication conditions (haloperidol [HAL], diphenhydramine [DIPH], and placebo [PLAC]) at FCZ. The dopamine blocker, haloperidol, significantly attenuated the amplitude of the ERN compared to placebo. Right panel: scalp voltage plots at 80 msec postresponse for correct (top) and incorrect (bottom) trials across each medication condition (left to right: haloperidol, diphenhydramine and placebo). The scalp voltage plots show the characteristic frontocentral activity of the ERN event-related brain potential.

1102 Journal of Cognitive Neuroscience Volume 16, Number 6 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/0898929041502779 by guest on 26 September 2021 Figure 2. The waveforms at the bottom of the figure show the time course of the response- locked, difference waveforms created by subtracting the grand average ERN waveform associated with one medication condition from the ERN waveform associated with another medication condition (derived from error trials). The differences between the placebo and haloperidol groups and the diphenhydramine and haloperidol groups were maximal between 60 and 80 msec postresponse, as expected for an ERN effect. The scalp voltage plots indicate that these group differences were maximal at frontocentral recording sites around 80 msec postresponse.

riate, F(2,45) = 4.97, p = .036. The same was true when .01 at CZ), poorer end performance (n = 50, r = À .44, controlling for N2 amplitude at CPZ, F(2,48) = 3.81, p = p< .01 at FCZ; n = 50, r = À .34, p< .05 at CZ), and .029, and when controlling for both the P1 and N2 longer response latencies (n = 50, r = À .47, p< .01 at amplitudes, F(2,45) = 4.91, p = .038. FCZ and n = 50, r = À .43, p< .05 at CZ for the required slow response; n = 50, r = À .34, p< .05 at FCZ; n = 50, r = À .33, p< .05 at CZ, for the required midspeed Relationships between ERPs and Behavioral response) in the Time Estimation Task. The correlation Task Performance between ERN amplitude and poorer learning remained In the entire sample, smaller amplitude ERNs (i.e., less significant when statistically controlling for P1 amplitude negative amplitude values) were correlated with in- at OZ and N2 amplitude at CPZ (n = 51, r = 52, p = .02 creased errors of commission in the Eriksen Flanker at FCZ; n = 51, r = 56, p< .01 at CZ), even though the Task (n = 51, r = .51, p< .01 at FCZ; n = 51, r = 57, p< total error rate on the flanker task and false alarm .01 at CZ) and increased total errors (including trials on latencies in the CPT task correlated with decreases in which participants did not respond), (n = 51, r = 48, P1 amplitude in OZ (n = 47, r = 42, p< .01; n = 23, r = p< .01 at FCZ; n = 51, r = 47, p< .01 at CZ). Smaller 50, p< .05, respectively). Interestingly, increases in ERNs were correlated with longer response latencies on correct, incorrect, and overall latencies in the flanker correct trials (n = 51, r = .31, p< .03 at FCZ; n = 51, r = task were all associated with increases in N2 latencies 28, p< .05 at CZ) but not incorrect trials (n = 51, r = 23, too (n = 501, r = 33, p< .02; n = 51, r = 29, p< .04; n = p = .09 at FCZ; n = 51, r = 21, p = .14 at CZ) in the 51, r = 33, p< .02. Finally, shorter N2 peak latencies flanker task. Larger ERN amplitudes were associated were correlated with larger amplitude ERNs (n = 51, r = with decreased numbers of false alarms (n = 50, r = 33, p< .05 at FCZ; n = 51, r = 33, p< .05 at CZ). À .32, p< .05 at FCZ; n =50,r = À .40, p< .01 at CZ) and fewer errors in the CPT (n = 50, r = .31, p< .05 at À DISCUSSION FCZ; n = 50, r = À .38, p< .01 at CZ). Smaller ERN amplitudes were also associated with poorer learning To our knowledge, this is the first direct examination of (n = 50, r = À .41, p< .01 at FCZ; n = 50, r = À .36, p< the effects of dopamine system manipulations on the

Zirnheld et al. 1103 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/0898929041502779 by guest on 26 September 2021 the Eriksen Flanker Task compared to both placebo and diphenhydramine. ERN amplitude attenuation was sig- nificantly and positively correlated with poorer behav- ioral performance, as indicated by increased error rates on the Eriksen Flanker Task and impaired learning on the Time Estimation Task. Furthermore, ERN amplitude attenuation was associated with increased latencies in the Time Estimation Task, suggesting that when subjects had impaired error detection or excessive conflict, their response was slowed. Finally, smaller ERN amplitudes were associated with increased errors and poorer learn- ing on the CPT and Time Estimation Task. Taken togeth- er, these results demonstrate a selective effect of the dopamine blocker, haloperidol, on ERN and associated measures of learning but not stimulus-related ERPs re- lated to stimulus perception and early selective attention. Although the present findings provide the first direct evidence for the role of dopamine neurotransmission in the ERN, they are consistent with a variety of existing, albeit indirect, evidence and models of the role of dopamine in learning and response monitoring. Our Figure 3. Stimulus-locked ERPs for each medication condition. The findings support Holroyd and Coles’s (2002) model, top panel shows the time course of the P1 ERP measured at Oz, where which asserts ‘‘(i) that the ERN reflects the transmission it was maximal. There were no medication effects on the P1. The of a reinforcement signal to the anterior cingulate, (ii) bottom panel shows the time course of the N2 ERP at CPZ, were it was that this error signal is carried by the mesencephalic maximal. There were no medication effects on the N2. dopamine system, and (iii) that it is used to train the anterior cingulate cortex to optimize performance on ERN in humans. The findings provide strong support for the task at hand’’ (p. 686). This model ties together a the role of the dopamine system in attention (go/no-go number of observations that are consistent with our Continuous Performance Task), procedural learning findings. For example, human behavioral studies show (Time Estimation Task), and the ERN (Eriksen Flanker that dopamine blockers impair procedural learning Task). The dopamine blocker haloperidol significantly whereas dopamine agonists enhance it (Kumari et al., attenuated ERN amplitudes compared to placebo and 1997). Deficits in procedural learning have been ob- this attenuation effect approached significance ( p = served in Parkinsonian patients (Wallesch et al., 1990), .066) when compared to the active CNS control drug, whose ERN amplitude is also diminished (Falkenstein, diphenhydramine, which did not significantly differ from Heilscher, et al., 2000; but see Holroyd et al., 2002). placebo (PLC). This effect cannot be attributed to drug- Furthermore, the dopamine reuptake inhibitors dextro- induced impairments of early visual perception of the amphetamine and methylphenidate have both been stimuli or attention allocation, since medications did not found to improve attention in patients with attention significantly affect P1 and N2 amplitudes, respectively. deficit hyperactivity disorder (Spencer et al., 1995; Ar- Moreover, the ERN attenuation by haloperidol remained nold et al., 1978), and bromocriptine, a direct dopamine significant even when controlling for P1 and N2 ampli- agonist, improved verbal learning in a patient who tudes. Perception of the visual stimuli as measured by sustained a lesion that interrupted dopaminergic path- the P1 amplitude did affect performance in the CPT and ways (Dopkin & Hanlon, 1993). the flanker tasks, since P1 amplitude attenuation was More specifically, our findings are consistent with associated with increased false alarms in the CPT task previous indications that the ERN originates from the and more errors in the Eriksen Flanker Task; however, dopamine terminal dense anterior cingulate or supple- these effects were not specific to haloperidol. Similarly, mentary motor cortices (Mathalon et al., 2002; Miltner ERN amplitudes were attenuated when less attention et al., 1997; Dehaene et al., 1994). Similarly, the present was allocated to the visual stimuli, as indicated by smaller findings are in line with recent models of reinforcement N2 amplitudes. In addition, when attention allocation learning conceptualizing VTA firing as a provider of an was delayed, as indicated by N2 latencies, responses on error signal necessary for appetitive learning to occur the flanker task also were delayed. However, neither of (Contreras-Vidal & Schultz, 1999; Suri & Schultz, 1999; these two effects was specific to haloperidol. Sutton & Barto, 1998) Such models are based on Haloperidol significantly impaired learning in a time observations that VTA dopamine neurons spike only estimation task compared to placebo and diphenhydra- when reward is unexpected and that the firing frequency mine, and showed a trend toward increasing errors in shows a transient decrease when an expected reward is

1104 Journal of Cognitive Neuroscience Volume 16, Number 6 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/0898929041502779 by guest on 26 September 2021 not delivered (Schultz et al., 1997) In addition, dopa- ERN-related conflict detection/error-prevention system mine affects long-term potentiation via D1 receptors in the brain. (Gurden, Takita, & Jay, 2000), although D1 and D2 Several methodological issues warrant discussion. receptors interact to allow long-term depression to First, the observed deleterious effects of haloperidol develop in the striatum (Centonze, Picconi, Gubellini, on the ERN and related measures were produced by a Bernardi, & Calabresi, 2001) For these reasons, although relatively small, but apparently sufficiently potent, dose D2 receptors appear to play a role in synaptic weight (3 mg) of this dopamine blocker. Therapeutic daily change, D1 receptor blockade could possibly attenuate doses in schizophrenia usually range between 5 and the ERN amplitude by acting at the triadic synapses in 30 mg (Schatzberg & Nemeroff, 1995) In Tourette’s the cingulate cortex, as reviewed by Holroyd and Coles disorder, haloperidol is typically effective at daily doses (2002), by dampening a disinhibition of cingulate den- ranging from 0.5 to 6 mg (Schatzberg & Nemeroff, 1995, drites triggered by a dopamine drop. However, haloper- p. 691). Therefore, the dose of 3 mg of haloperidol idol is essentially a D2 blocker with some D1 blockade should have significantly disrupted dopamine transmis- properties (Seeman, 1990) and whether it acts as an sion, as we have assumed. inhibitory neurotransmitter at the cingulate level re- Second, the administration of the active control drug, mains unclear (see Holroyd & Coles, 2002, p. 698, for diphenhydramine, which was intended to mimic halo- a review). On the other hand, as outlined above, dopa- peridol’s sedative effects but not disrupt dopamine mine plays a role in promoting synaptic weight change. transmission, allowed for more specific determination In any case, blocking D1 and D2 receptors will have the of the role of dopamine in the ERN and related meas- predominant effect of interrupting dopamine transmis- ures of attention and learning. Indeed, it has been sion between the mesocortical, mesolimbic or nigros- established that diphenhydramine antagonizes hista- triatal terminals and their effectors, which include the mine at H1 receptors in the central nervous system prefrontal and cingulate cortices as well as the nucleus and has anticholinergic muscarinic properties (Schatz- accumbens and the dorsal striatum. Blocking dopamine berg & Nemeroff, 1995), just as haloperidol (Seeman, autoreceptors in the midbrain also increases dopami- 1990; Richelson, 1989). Muscarinic blockade is usually nergic firing to the point of ‘‘depolarization blockade,’’ associated with side effects such as blurry vision and which would dampen the dopamine signal (Janicak, altered alertness, and histamine blockade is often asso- Davis, Preskorn, & Ayd, 1993) even though the net ciated with sedation. However, haloperidol shows a effect—as evidenced by extrapyramidal symptoms relatively low affinity for the muscarinic receptor com- caused by neuroleptics—is a dopamine transmission pared to diphenhydramine (Seeman, 1990; Richelson, blockade. Taken together, the present conclusion that 1989). With respect to the histamine system, we are not dopamine neurotransmission plays a direct and specific aware of studies comparing the affinities of diphenhy- role in the ERN and related measures of learning is dramine and haloperidol for these receptors. Taken consistent with a variety of indirect evidence and models together, the available evidence indicates that the ob- of response monitoring. served effects on the ERN and learning are likely due to Our findings are also consistent with a more general dopamine blockade rather than cholinergic blockade, an literature suggesting that the dopamine system may be interpretation that is consistent with the strong evidence broadly involved in attention and cognition. For exam- implicating dopamine in learning and ERN generation. ple, haloperidol has been shown to suppress a 40-Hz In addition, our active control drug, diphenhydra- transient evoked potential associated with attended mine, readily crosses the blood–brain barrier and has stimuli (Ahveninen et al., 2000); attenuate processing been previously shown to have a dose- and age-related negativity, a marker of selective attention (Kahkonen negative impact on measures of attention and learning et al., 2001); increase the mismatch negativity, indicating (Bender, McCormick, & Milgrom, 2001; Taga, Sugimoto, attention to irrelevant stimuli (Kahkonen et al., 2001); Nishiga, Fujii, & Kamei, 2001; Weiler et al., 2000; Vurr- impair digit symbol substitution (Lynch, King, Green, man, van Vegel, Sanders, Muntjewerff, & O’Hanlon, Byth, & Wilson-Davis, 1997), and pimozide, a dopamine 1996; Decker & McGaugh, 1991) That we administered D2 antagonist, diminishes attention shifting (Mehta, a sufficiently high dose of diphenhydramine is suggested Sahakian, McKenna, & Robbins, 1999). Haloperidol also by the trend effects it had on RT measures. On the worsens latent inhibition in humans (Williams et al., Eriksen Flanker Task, diphenhydramine tended to delay 1997), impairs working memory (Didriksen, 1995), and RTs on trials immediately following an error, compared alters spatial recognition (Mehta et al., 1999). These to haloperidol, but RTs in the placebo group did not findings support the role of dopamine as a marker of significantly differ from RTs in either the haloperidol the salience of the information being processed, which group or the DPH group. Thus, diphenhydramine did in turn may be used by the nervous system as a signal in not appear to be entirely devoid of behavioral influence, learning. Nevertheless, whereas these previous reports despite the fact that attention, learning, and ERN meas- implicate dopamine in attention and learning, the pres- ures were not affected. Taken together, the results of ent findings argue for a specific role of dopamine in the present study suggest a selective effect of dopamine

Zirnheld et al. 1105 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/0898929041502779 by guest on 26 September 2021 blockade on learning, behavioral indices of perfor- to explain ERN amplitude deficits in Parkinson’s patients mance, and on the ERN amplitude. (Holroyd et al., 2002). However, Falkenstein, Hoorman, A third methodological point concerns the apparent et al. (2000, p. 94) showed that ERN amplitude was not discrepancy between the lack of findings in the CPT and significantly different between groups of subjects who the results in the Eriksen Flanker Task. Whereas subjects had high (20%) versus low (6%) error rates. Moreover, taking haloperidol committed more errors than those they reported that short response windows, which taking either diphenhydramine or placebo, this differ- produce ‘‘high time pressure’’ to respond, rather than ence was statistically significant for the Eriksen Flanker the errors rates per se, accounted for attenuated ERN Task but not for the CPT. This may have resulted from a amplitudes. This is consistent with Gehring et al.’s difference in sensitivity between the two tasks, since (1990, 1993) reports that the amplitude of the ERN there were more trials (640) in the Eriksen Flanker Task was larger when accuracy was emphasized by instruction than in the CPT (100), or it may be that haloperidol has than when speed was emphasized. Taken together, relatively little effect on vigilance as measured by the these data suggest that ERN amplitudes may be dimin- CPT. ished when short response windows preclude the re- Fourth, it is unlikely for several reasons that the lack sponse evaluation process that the ERN is believed to of impaired learning in the Time Estimation Task was index (i.e., the comparison of the given response to the due to extrapyramidal symptoms or slowed motor re- cognitive representation of the ‘‘correct’’ response on sponses. Foremost, subjects reported no such symp- that trial). In the version of the Eriksen Task imple- toms and none were observed by the study physicians. mented in the present study, accuracy was emphasized In addition, no differences were noted in initial perfor- and the participants were allowed 1000 msec to re- mance on the Time Estimation Task between haloperi- spond, which is longer compared to commonly used dol and the other two drugs. Furthermore, RTs did not response windows (i.e.,  500 msec; Holroyd & Coles, differ significantly across groups in any of the tests. 2002; Falkenstein, Hoorman, et al., 2000; Luu et al., Fifth, Rammsayer (1999) found that haloperidol dimin- 2000). Thus, given that (1) heightened pressure to ished performance in a time estimation task in healthy respond during short response windows, rather than adults. Although consistent with our results, Ramm- errors rates per se, appears to underlie attenuated ERN sayer’s findings raise the possibility that our subjects amplitudes; (2) our procedures minimized the potential may have been unable to learn to properly time their adverse effects of shortened response windows on the responses in the Time Estimation Task, not because of ERN amplitude; and (3) different error rates are not decreased learning per se, but because of an impaired necessarily associated with different ERN amplitudes, it ability to evaluate time. This seems unlikely given the seems unlikely that the observed ERN attenuation was absence of significant overall RT differences across med- driven by high error rates rather than the dopaminergic ication groups in the present study. In addition, Ramm- effect of haloperidol. Nevertheless, subsequent studies sayer’s experiments involved reproducing time intervals will be required to address this potential confound. (i.e., tone durations) rather than estimating appropriate However, it should be noted that even if the relationship motor timing based on the feedback provided, as was the between ERN amplitude and error rates were unequiv- case in the present study. In the same experiment, ocal, there is the remaining problem of determining Rammsayer administered a motor skills task in which causal direction. For example, if a primary function of subjects had to time their movements to avoid touching the anterior cingulate is to monitor behavior, it is logical the walls of a computer-generated maze. Interestingly, to expect that a functional impairment or an anatomical there were no group differences in motor skills or task ablation of the anterior cingulate would result in in- completion time; however, when the task was repeated, creased error rates (Cohen, Kaplan, Moser, Jenkins, & haloperidol selectively impaired motor skills acquisition Wilkinson, 1999) Thus, error rates may not be a nuisance compared to placebo, suggesting that the drug selective- artifact but instead reflect the functionality of the mech- ly impairs practice-mediated motor learning. Our time anism of primary interest. estimation task reproduced these results, since we found In summary, given (1) the role of dopamine neurons as no difference between drug groups in baseline perfor- conveyors of an error signal used in animal learning mance on the time estimation task, nor in RTs, but we (Schultz et al., 1997; Schultz, 1996; Schultz et al., 1993; showed that haloperidol impaired the learning of the Mirenowicz & Schultz, 1994; Ljungberg et al., 1992; Romo timed motor response across the experiment. Therefore, & Schultz, 1990; Schultz, 1986), (2) the evidence linking if haloperidol impairs RT estimation, it may do so by dopamine to learning (Rammsayer, 1999; Kumari et al., directly impairing learning. 1997; Dobkin & Hanlon, 1993; Wallesch et al., 1990), (3) Finally, ERN attenuation has been previously shown to the likely emergence of the ERN and the FRN from correlate with error rates (Gehring et al., 1993), and anterior cingulate dendrites (Holroyd & Coles, 2002), therefore, one might argue that the observed ERN (4) the ERN amplitude changes reflecting an updating amplitude attenuation was due to higher error rates in process following a temporal difference rule (see Hol- the haloperidol group. This possibility has been offered royd & Coles, 2002), and (5) the present findings

1106 Journal of Cognitive Neuroscience Volume 16, Number 6 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/0898929041502779 by guest on 26 September 2021 showing a negative impact of dopamine blockade on Design and Procedure ERN amplitude and on learning time intervals, we argue In a double-blind fashion, subjects were randomly as- that the ERN is likely to result from a dopamine-mediated signed to one of three medication conditions: haloper- learning mechanism that indicates the discrepancy be- idol (3 mg orally; n = 17), diphenhydramine (25 mg tweenanexpectedoutcomeandanactualoutcome orally; n = 16), or placebo (orally; n = 18). Three hours attributed to an action-imperative stimulus pair. Howev- after drug administration, each participant completed er, further experiments will have to determine if dopa- three behavioral tasks: a Continuous Performance Task mine blockers selectively slow the appearance of the ERN (adapted from Harvey et al., 1990), the Eriksen Flanker with learning, as would be predicted if a VTA signal is Task (Eriksen & Eriksen, 1974) and a Time Estimation used as an error signal by cingulate neurons to learn to Task (see below) that consisted of a learning compo- predict response-outcome mismatches. nent. The EEG was recorded during the last two tasks, though we report on only the former. A 3-hr interval METHODS between drug administration and testing was chosen Participants based on the known pharmacokinetics of diphenhydra- mine and haloperidol (Kaplan & Sadock, 1998). Both Sixty-two healthy subjects were recruited through news- drugs are highly lipophilic and readily cross the blood– paper advertisement, the Internet, posters, and word of brain barrier. mouth. After giving informed consent (IUPUI protocol #0011-11), participants were determined to be free of DSM-IIIR Axis I (Mini International Neuropsychiatric Go/No-go Continuous Performance Task Interview; Sheehan et al., 1997) and Axis II disorders (Personality Diagnostic Questionnaire; Hyler, Skodol, Using Neuroscan’s STIM software (Sterling, VA), two Kellman, Oldham, & Rosnick, 1990). Each subject’s series of fifty 8 by 5 cm capital letters were presented medical history was taken, and a physical examination at 1-sec intervals on the center of a computer monitor and an electrocardiogram were administered to (1) for 50 msec. The two series contained 21 target sequen- prevent possible adverse reactions to the study drugs, ces (two Ts in a row) and six isolated Ts. Subjects were (2) rule out medical conditions likely to influence re- instructed to press a button following two consecutive sults, and (3) exclude subjects with recent alcohol, drug, presentations of the target letter T. If several pairs of Ts or medication use (with the exception of contraceptives were presented consecutively, subjects had to respond and postmenopause hormone substitution). A urine only after the second T of each pair. Responses that screen was used to rule out pregnancy, and nursing occurred at the appropriate time were counted as ‘‘hits’’ mothers were also excluded. Participants reported nor- and incorrect responses were counted as ‘‘false alarms.’’ mal or corrected-to-normal vision and hearing. Of the initial 62 participants, one subject showed signs Eriksen Flanker Task of akathesia and was asked to discontinue the experi- ment. A second subject did not complete the Continu- The Eriksen Flanker Task consisted of four possible ous Performance Task due to a computer malfunction, letter strings: ‘‘HHHHH,’’ ‘‘SSSSS,’’ ‘‘SSHSS,’’ ‘‘HHSHH.’’ and another wished to discontinue testing in the middle Each string was presented an equal number of times in of the Time Estimation Task due to a headache caused randomized order across eight blocks of 80 trials each. by wearing the EEG cap. Furthermore, 10 subjects made Subjects were required to respond only to the central fewer than 10 errors on the Eriksen Flanker Task (see letter of each string by pressing either the left or right below), which precluded the extraction of a reliable button on a response pad. Button assignment to the error-related ERP. These subjects were therefore re- target letters was counterbalanced across the eight moved from all subsequent analyses, leaving 51 subjects blocks. Each letter string was presented for 52 msec, in the analyses (20 men and 31 women, mean age = 0.5 cm above an ever-present 6 Â 6 mm fixation cross 34.24 years, SD = 12.60) for the Eriksen Flanker Task, presented in the middle of the computer monitor. If a and 50 subjects were included in the analyses for response was not made within 1 sec of the letter string both the Continuous Performance Task and the Time presentation, participants received a 1 Â 1 cm ‘‘!’’ as a Estimation Task. Finally, four subjects were excluded feedback stimulus to indicate they did not respond on in the P1 analyses (one from the haloperidol group, time. Responses made within the allotted time were three from the diphenhydramine group), because the followed 700 msec later by a feedback stimulus in the P1 could not be scored reliably due to unstable baseline. form of either a 1 Â 1 cm plus or minus sign above the The treatment groups did not significantly differ in cross, depending on whether the response was correct either age, F(2,49) = 0.54, p = .586, haloperidol: or incorrect, respectively. The feedback stimulus re- mean age = 32.69 years; placebo: mean age = 36.72 mained for 500 msec after which another string was years; diphenhydramine: mean age = 33.00 years, or sex presented. Subjects were not allowed to practice. This x2(2) = 0.01. particular task was chosen because flanker tasks have

Zirnheld et al. 1107 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/0898929041502779 by guest on 26 September 2021 been shown to reliably elicit the ERN with a minimum zontal line segment, which represented the appropriate of trials (Dikman et al., 2000; Luu et al., 2000; Gehring restabilization of the aircraft’s wing to the horizontal et al., 1993). Feedback was provided to help subjects position. Feedback following incorrect responses con- enhance their accuracy. sisted of the presentation of a red, cartoon-like crash symbol bearing the word ‘‘CRASH’’ on the CRT for 800 msec. Feedback in the form of an exclamation Time Estimation Task point was presented if no response was made within The TET was included in this study for two reasons. 1400 msec of the banked line presentation. The next First, the reinforcement theory on which the present trial began with the presentation of a central fixation study is, in part, based, predicts that the dopamine crosshair for 200 msec. To increase motivation, subjects neurons deliver an error signal that is used to adjust were awarded 5 cents for each correct response, in the values of the possible actions in a choice RT task addition to an hourly rate of $10 an hour for about such as the TET, and dampening this error signal should 8 hr and a bonus of $40 for showing up for the result in impairment in learning (Sutton & Barto, 1998; experiment. The reward for correct responses in the Schultz et al., 1997). Second, both feedback (Miltner Time Estimation Task allowed for a possible supplemen- et al., 1997) and imperative visual stimuli signaling an tal gain of $33 over the 660 trials ($15 on average). error (Falkenstein et al., 1990; Gehring et al., 1990) have Subjects were verbally and visually informed about the the potential to elicit the FRN or the ERN, with the durations of each time window, but had to learn to time appearance of the ERN coinciding with learning (Hol- their responses by trial and error. They were not allowed royd & Coles, 2002) Accordingly, the TET was used to to practice beforehand. assess learning here. The electrophysiological data, which are still being analyzed, will be reported else- where. Hence, the TET task was designed as a game in EEG Recording which subjects learned to fly an airplane. The plane was The EEG recordings (5K gain) were made using a 32- symbolized by a 6-cm white line centered above an ever- channel Ag–AgCl electrode cap (NeuroMedical Supplies, present 6 Â 6 mm fixation cross. The orientation of the Sterling, VA), and horizontal and vertical electrooculo- segment varied from trial to trial to mimic the way that graphic (EOG) activity (2.5K gain) was monitored with an airplane’s wing might tilt, or bank, during flight. From electrodes placed at the outer canthus of both eyes, and the initial horizontal position, the ‘‘wing’’ segment tilted above and below the left eye, respectively. All EEG 8 8 at six different angles: (1) 30 to the right, (2) 45 to the electrodes were referenced to the tip of the nose. The 8 8 8 right, (3) 60 to the right, (4) 30 to the left, (5) 45 to data were sampled using a SynAmps (NeuroScan, Inc., 8 the left, and (6) 60 to the left. If the plane tilted to the Sterling, VA) 16-bit analog-to-digital converter at the rate right, subjects were instructed to press the left button of 1000 points per second, with a 0.05-Hz high-pass on a response pad to return the wing to the stable (single pole, RC, roll-off high-pass filter of À6 dB/octave) horizontal position. If it tilted to the left, they were and 200-Hz low-pass filter (IIR implementation of a instructed to press the right button. The goal of the Butterworth low-pass filter). For all but one subject, ‘‘game’’ was to keep the plane’s wing in the horizontal impedances were below 10 k . The EEG was recorded position. Depending on the angle, subjects had to during both the Eriksen Flanker Task and the TET, but 8 respond more or less rapidly: A 60 tilt required a only the results of the former are presented here. The 8 response within 500 msec; a 45 tilt required a response amplitudes and latencies of the ERN were measured at 8 between 500 and 850 msec; and a 30 tilt required a FCZ and CZ. response between 850 and 1200 msec. Responses occur- ring outside the respective windows were considered errors, and trials for which responses occurred after Analyses 1400 msec were counted as nonresponses and excluded from the analyses. EEG quantification The six possible line orientations appeared with equal The EEG data were epoched separately for stimulus- probability across five blocks of 132 trials. Each block of locked (P1 and N2) and response-locked (ERN) analyses, 132 trials began with the horizontal line below a fixation then digitally low-pass filtered at 30 Hz (24 dB/octave). cross for 800 msec and was followed by the first trial. For the response-locked waveforms, trials were visually Each trial began with the presentation of the fixation inspected to remove epochs with gross EMG artifact, cross for 200 msec, followed by one of the six tilted line then horizontal and vertical eye blinks were removed segments. This imperative ‘‘wing’’ stimulus was pre- with an independent components analytic method ( Jung sented until a response occurred, or until 1400 msec et al., 2000) Response-locked ERP averages were con- elapsed, whichever came first. The feedback was pre- structed using the error trials for each subject, and did sented 200 msec after response for 800 msec. Feedback not include trials when the subject did not respond. The on correct response consisted of a return of the hori- ERNs were measured at FCZ and CZ by first searching for

1108 Journal of Cognitive Neuroscience Volume 16, Number 6 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/0898929041502779 by guest on 26 September 2021 the most negative peak in a 200-msec postresponse and that a plateau was reached after the 44th trial. Thus, window (or 150-msec postresponse for analysis of the sum of the accuracy score over the first five pre- ERN-like activity on correct trials), and then subtracting sentations of each tilt angle was calculated to reflect from this value the amplitude of the most positive baseline performance and the sum of the accuracy score point located in the preceding 150 msec (Scheffers & over the last 66 tilt angles was calculated to indicate end Coles, 2000) performance. The baseline performance was then mul- For the stimulus-locked waveforms, trials were base- tiplied by 66 and the end performance was similarly line corrected according to a 100-msec prestimulus multiplied by 5 to ensure that each of these performance period, then visually inspected for unstable baseline measures were similarly weighted integers. The differ- activity (i.e., fluctuations greater than 100 AV) during ential performance, our index of the magnitude of this interval. Subsequently, an ocular artifact correction learning, was given by the difference between the base- algorithm was applied (Semlitsch, Anderer, Schuster, & line and end scores. Presslich, 1986) and then the data were averaged, base- line-corrected, and run through a 0.5-Hz high-pass fil- ter (24 Db/octave) to further stabilize the prestimulus Statistical Analyses baseline activity. P1 amplitude and latency at Oz was Repeated measures ANOVAs using medication condition scored by measuring the peak positivity between 70 and (haloperidol, diphenhydramine, and placebo) as a be- 150 msec poststimulus relative to the prestimulus base- tween-subject variable and electrode (ERN: FCZ & CZ; line. N2 amplitude and latency was calculated by mea- P1: OZ; N2: CZ, CPZ, & PZ) as a within-subject variable suring the peak negativity between 140 and 325 msec were separately conducted on the ERP amplitudes and postresponse relative to the preresponse baseline. latencies, and on the behavioral measures. ANCOVA using P1 amplitude at Oz and N2 amplitude at CPZ as Behavioral Analyses covariate was also conducted with the ERN data. Tukey’s conservative honestly significant difference (HSD) test Percentages of response types (hits, miss, false alarms) was used for all relevant post hoc analyses. Pearson and the total percentage of correct responses were product–moment correlations were also computed to calculated for the CPT. Response latencies were also evaluate the relationships among the behavioral and determined. The number of erroneous responses, the EEG data. number of missing responses, and the number of cor- rect responses, whose sum was always 640 in the Eriksen Flanker Task, were obtained. The error rate was defined Acknowledgments as the number of erroneous responses divided by the We are grateful to Jennifer Vohs for her time and patience and total number of trials (640) multiplied by 100. The rate to Drs. Shekhar Pushpahla, Rezwan Khan, and other psychiatry of correct responses was computed by dividing the residents for the coverage they provided. Thanks to Virginia number of correct responses (reflecting both nonres- Blevins for her assistance with patient testing and Kim Sunblad for the practical experience she brought to this study. We ponses and erroneous ones) by 640 and multiplying by appreciate Dr. Marten K. Scheffers’s insightful comments on an 100. Average response latencies for correct trials, error early draft of this article, as well as the careful reviews provided trials, all trials, and trials immediately following error by Dr. Clay Holroyd and two anonymous reviewers. trials in each task were also calculated. To quantify the Reprint requests should be sent to William P. Hetrick, rate of learning for the TET, stimulus-specific tilt trials Department of Psychology, Indiana University—Bloomington, (308,458,608 to the right or to the left) were first sorted 1101 East Tenth Street, Bloomington, IN 47405, or via e-mail: by presentation order. The first presentation of a 308 [email protected]. right tilt was scored zero if the response was inaccurate and one if it was correct. This procedure was repeated for each of the stimulus types (i.e., 308 to the left, 458 to REFERENCES the right, 458 to the left, 608 to the right, and 608 to the Ahveninen, J., Kahkonen, S., Tiitinen, H., Pekkonen, E., left). By adding the scores, we obtained a ‘‘first presen- Huttunen, J., Kaakkola, S., Ilimoniemi, R. J., & Jaaskelainen, tation accuracy score’’ (ranging from 0 to 6). The same I. P. (2000). Suppression of transient 40-Hz auditory scoring system was used for the second, third, and response by haloperidol suggests modulation of human selective attention by dopamine D2 receptors. Neuroscience fourth presentations and so on until 22 accuracy scores Letters, 292, 29–32. were obtained from one block of 132 trials. The same Arnold, L. E., Christopher, J., Huestis, R., & Smeltzer, D. J. procedure was used for the remaining four blocks, (1978). Methylphenidate versus dextroamphetamine vs which yielded a total of 110 accuracy scores ranging caffeine in minimal brain dysfunction: Controlled from 0 to 6, ranked by order of imperative stimulus comparison by placebo and washout design with Bayes’ analysis. Archives of General Psychiatry, 35, 463–483. presentation. Inspection of the grand average accuracy Au-Young, S. M. W., Shen, H., & Yang, C. H. (1999). scores across all subjects by presentation order indicated Medial prefrontal cortical output neurons to the ventral that most of the learning took place in the first 22 trials tegmental area (VTA) and their responses to burst-patterned

Zirnheld et al. 1109 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/0898929041502779 by guest on 26 September 2021 stimulation of the VTA: Neuroanatomical and in vivo Gehring, W. J., Coles, M. G. H., Meyer, D. E., & Donchin, E. electrophysiological analyses. Synapse, 34, 245–255. (1990). An event-related brain potential accompanying Bender, B. G., McCormick, D. R., & Milgrom, H. (2001). errors [abstract]. Psychophysiology, 27, 534. Children’s school performance is not impaired by short Gehring, W. J., Goss, B., Coles, M. G. H., Meyer, D. E., & term administration of diphenhydramine or loratatine. Donchin, E. (1993). A neural system for error detection Journal of Pediatrics, 138, 656–660. and compensation. Psychological Science, 4, 385–390. Beninger, R. J. (1989). Dissociating the effects of altered Gehring, W. J., & Knight, R. T. (2000). Prefrontal-cingulate dopaminergic function on performance and learning. interactions in action monitoring. Nature Neuroscience, Brain Research Bulletin, 23, 365. 3, 516–520. Botvinick, M. M., Braver, T. S., Barch, D. M., Carter, C. S., & Gehring, W. J., & Willoughby, A. R. (2002). The medial frontal Cohen, J. (2001). Conflict monitoring and cognitive control. cortex and the rapid processing of monetary gains and Psychological Review, 108, 624–652. losses. Science, 295, 2279–2282. Carr, D. B., & Sesack, S. R. (2000). Projections from the rat Gurden, H., Takita, M., & Jay, T. M. (2000). Essential role of prefrontal cortex to the ventral tegmental area: Target D1 but not D2 receptors in the NMDA receptor dependent specificity in the synaptic associations with mesoaccumbens long-term potentiation at hippocampal–prefrontal cortex and mesocortical neurons. Journal of Neuroscience, 20, synapses in vivo [rapid communication]. Journal of 3864–3873. Neuroscience, 20, 106. Centonze, D., Picconi, B., Gubellini, P., Bernardi, G., & Haber, S. N., & Fudge, J. L. (1997). The primate substantia Calabresi, P. (2001). Dopaminergic control of synaptic nigra and VTA: Integrative circuitry and function. Critical plasticity in the dorsal striatum. European Journal of Reviews in Neurobiology, 11, 323–342. Neuroscience, 13, 1071–1077. Harvey, P. D., Keefe, R. S. E., Moskowitz, J., Mohs, R. C., Cohen, J. (1988). Statistical power analysis for the behavioral Putnam, K. M., & Davis, K. L. (1990). Attentional markers of sciences (2nd ed.). Hillsdale, NJ: Erlbaum. vulnerability to schizophrenia: Performance of medicated Cohen, R. A., Kaplan, R. F., Moser, D. J., Jenkins, M. A., & and unmedicated patients and normals. Psychiatry Wilkinson, H. (1999). Impairments of attention after Research, 33, 179–188. cingulotomy. Neurology, 53, 819–824. Heimer, L., Alheid, G. F., de Olmos, J. S., Groenewegen, H. J., Contreras-Vidal, J. L., & Schultz, W. (1999). A predictive Haber, S. N., Harlan, R. E., & Zahm, D. S. (1997). The reinforcement model of dopamine neurons for learning accumbens. Beyond the core and shell. In M. Forbes, approach behavior. Journal of Computational S. Salloway, P. Malloy, & J. L. Cummings (Eds.), The Neuroscience, 6, 191–214. neuropsychiatry of limbic and subcortical disorders Davies, P. L., Segalowitz, S. J., Dywan, J., & Pailing, P. (2001). (pp. 58–61). Washington, DC: American Psychiatric Press. Error-negativity and positivity as they relate to other indices Holroyd, C. B., & Coles, M. G. H. (2002). The neural basis of of and stimulus processing. Biological human error processing: Reinforcement learning, dopamine Psychology, 56, 191–206. and the error-related negativity. Psychological Review, 109, Decker, M. H., & McGaugh, J. L. (1991). The role of 679–709. interactions between the cholinergic system and other Holroyd, C. B., Dien, J., & Coles, M. G. H. (1998) Error-related neuromodulatory systems in learning and memory. scalp potentials elicited by hand and foot movements: Synapse, 7, 151–168. Evidence for an output-independent error-processing Dehaene, S., Posner, M. I., & Tucker, D. M. (1994). Localization system in humans. Neuroscience Letter, 242, 65–68. of a neural system for error detection and compensation. Holroyd, C. B, Praamstra, P., Plat, E., & Coles, M. G. H. (2002). Psychological Science, 5, 303–305. Spared error-related potentials in mild to moderate Didriksen, M. (1995). Effects of antipsychotics on cognitive Parkinson’s disease. Neuropsychologia, 40, 2116–2124. behaviour in rats using the delayed non-match to position Hyler, S. E., Skodol, A. E., Kellman, H. D., Oldham, J. M., & paradigm. (1995). European Journal of Pharmacology, 281, Rosnick, L. (1990). Validity of the Personality Diagnostic 241–250. Questionnaire—revised: Comparison with two structured Dikman, Z. V., & Allen, J. J. B. (2000). Error monitoring during interviews. American Journal of Psychiatry, 147, 1043–1048. reward and avoidance learning in high- and low-socialized Janicak, P. G., Davis, J. M., Preskorn, S. H., & Ayd, F. Jr. (1993). individuals. Psychophysiology, 37, 44–54. In Principles and practice of psychopharmacology (2nd ed., Dobkin, B. H., & Hanlon, R. (1993). Dopamine agonist p. 102). Baltimore, MA: Williams & Wilkins. treatment of antegrade amnesia from a mediobasal forebrain Johannes, S., Wieringa, B. M., Nager, W., Dengler, R., & injury. Annals of Neurology, 33, 313–316. Mu¨nte, T. F. (2001). Oxazepam alters action monitoring. Eriksen, B. A., & Eriksen, C. W. (1974). Effects of noise letters Psychopharmacology, 155, 100–106. upon the identification of a target letter in a nonsearch task. Jung, T.-P., Makeig, S., Westerfield, M., Townsend, J., Perception and Psychophysics, 16, 143–149. Courchesne, E., & Sejnowski, T. J. (2000). Removal of Falkenstein, M., Heilscher, H., Dziobek, I., Schwartzenau, P., eye activity artifacts from visual event-related potentials in Hoormann, J., & Sundermann, B. (2000). Action monitoring, normal and clinical subjects. Clinical Neurophysiology, 111, error detection and the basal ganglia: An ERP study. 1745–1758. NeuroReport, 12, 157–161. Kahkonen, S., Aveneninen, J., Jaaskelainen, I. P., Kaakkola, Falkenstein, M., Hohnsbein, J., Hoormann, J., & Blanke, L. S., Naatanen, R., Huttunen, J., & Pekkonen, E. (2001). (1990). Effects of errors in choice reaction tasks on the Effects of haloperidol on selective attention: A combined ERP under focused and divided attention. In C.H.M. whole head MEG and high-resolution EEG study. Brunia, A.W.K. Gaillard, & A. Kok (Eds.), Psychological Neuropsychopharmacology, 25, 498–504. brain research (pp. 192–195). Tilburg, Germany: Tilburg Kaplan, H. I., & Sadock, B. J. (1998). and University Press. Dopamine receptor antagonists. In K. C. Millet (Ed,), Falkenstein, M., Hoormann, J., Christ, S., & Hohnsbein, J. Synopsis of psychiatry (pp. 983, 1024). Baltimore, MA: (2000). ERP components on reaction errors and their Williams & Wilkins. functional significance: A tutorial. Biological Psychology, Kumari, V., Corr, P. J., Mulligan, O. F., Cotter, P. A., Checkley, 51, 87–107. S. A., & Gray, J. A. (1997). Effects of acute administration of

1110 Journal of Cognitive Neuroscience Volume 16, Number 6 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/0898929041502779 by guest on 26 September 2021 d-Amphetamine and haloperidol on procedural learning in Romo, R., & Schultz, W. (1990). Dopamine neurons of the man. Psychopharmacology, 29, 271–276. monkey midbrain: Contingencies of responses to active Ljungberg, T., Apicella, P., & Schultz, W. (1992). Responses of touch during self-initiated arm movements. Journal of monkey dopamine neurons during learning of behavioral Neurophysiology, 63, 592–606. reactions. Journal of Neurophysiology, 67, 145–163. Schatzberg, A. F., & Nemeroff, C. B. (1995). The American Luu, P., Collins, P., & Tucker, D. M. (2000). Mood, personality Psychiatric Press textbook of psychopharmacology: and self-monitoring: Negative affect and emotionality in First edition (pp. 249, 620). Washington, DC: American relation to frontal lobe mechanisms of error-monitoring. Psychiatric Press. Journal of Experimental Psychology: General, 129, 43–60. Scheffers, M. K., & Coles, M. G. H. (2000). Performance Luu, P., & Tucker, D. M. (2001). Regulating action: Alternating monitoring in a confusing world: Error-related brain activity, activation of midline frontal and motor cortical networks. judgments of response accuracy, and types of errors. Clinical Neurophysiology, 112, 1295–1306. Journal of Experimental Psychology: Human Perception Lynch, G., King, D. J., Green, J. F., Byth, W., & Wilson-Davis, K. and Performance, 26, 141–151. (1997). The effects of haloperidol on , eye Scheffers, M. K., Coles, M. G. H., Bernstein, P., Gehring, W. J., & movements and psychomotor performance. Donchin, E. (1996). Event-related brain potentials and Psychopharmacology, 133, 233–239. error-related processing: An analysis of incorrect responses Margolin, D. I. (1978). The hyperkinetic child syndrome and to go and no-go stimuli. Psychophysiology, 33, 42–53. brain monoamines: Pharmacology and therapeutic Schultz, W. (1986). Responses of midbrain dopamine neurons implications. Journal of Clinical Psychiatry, 39, 1203, 7–30. to behavioral trigger stimuli in the monkey. Journal of Masaki, H., Tanaka, H., Takasawa, N., & Yamazaki, K. (2001). Neurophysiology, 56, 1439–1461. Negative component following incorrect feedback represents Schultz, W. (1996). Preferential activation of midbrain mismatch process between internal representation and dopamine neurons by appetitive rather than aversive stimuli. feedback information. Psychophysiology, 38, 64. Nature, 379, 449–451. Mathalon, D. H., Faustman, W. O., Gray, M., Askari, N., & Schultz, W., Apicella, P., & Ljungberg, T. (1993). Responses of Ford, J. M. (2002). Response monitoring dysfunction in monkey dopamine neurons to reward and conditioned schizophrenia: An event-related potential study. Journal stimuli during successive steps of learning a delayed of Abnormal Psychology, 111, 22–41. response task. Journal of Neuroscience, 13, 900–913. Mehta, M. A., Sahakian, B. J., McKenna, P. J., & Robbins, T. W. Schultz, W., Dayan, P., & Montague, P. R. (1997). A neural (1999). Systemic sulpride in young adult volunteers substrate of prediction and reward. Science, 275, simulates the profile of cognitive deficits in Parkinson’s 1593–1599. disease. Psychopharmacology, 146, 162–174. Seeman, P. (1990). Atypical neuroleptics: Role of multiple Miltner, W. H. R., Braun, C. H., & Coles, M. G. H. (1997). receptors, endogenous dopamine, and receptor linkage. Event-related brain potentials following incorrect feedback Acta Psychiatrica Scandinavia, 82, 14–21. in a time-estimation task: Evidence for a ‘‘generic’’ neural Semlitsch, H. V., Anderer, P., Schuster, P., & Presslich, O. system for error detection. Journal of Cognitive (1986). A solution for reliable and valid reduction of ocular Neuroscience, 9, 787–797. artifacts, applied to the P300 ERP. Psychophysiology, 23, Mirenowicz, J., & Schultz, W. (1994). Importance of 695–703. unpredictability for reward responses in primate dopamine Sesack, S. R., & Pickel, V. M. (1992). Prefrontal cortical efferents neurons. Journal of Neurophysiology, 72, 1024–1027. in the rat synapse on unlabelled neuronal targets of Nieuwenhuis, S., Ridderhinkhof, K. R, Bloom, J., Band, catecholaminergic terminals in the nucleus accumbens septi G. P. H., & Kok, A. (2001). Error-related brain potentials and on dopamine neurons in the ventral tegmental area. are differently related to awareness of response errors: Journal of Comparative Neurology, 320, 145–160. Evidence from an antisaccade task. Psychophysiology, 38, Sheehan, D., Lecubrier, Y., Harnett-Sheehan, K., Janavs, J., 752–760. Weiller, E., Bonara, L. I., Keskiner, A., Schinka, J., Knapp, E., Nieuwenhuis, S., Ridderinkhof, K., Talsma, D., Coles, M. G. H., Sheehan, M. F., & Dunbar, J. C. (1997). Reliability and validity Holroyd, C. B., Kok, A., & Van Der Molen, M. W. (2002). A of the Mini International Neuropsychiatric Interview computational account of altered error processing in older (M.I.N.I.): According to the SCID-P. European Psychiatry, age: Dopamine and the error-related negativity. Cognitive, 12, 232–241. Affective, and Behavioral Neuroscience, 2, 19–36. Spencer, T., Willens, T., Biederman, J., Faraone, S. V., Ablon, Oades, R. D. (1982). Search strategies on a hole-board are J. S., & Lapey, K. (1995). A double-blind crossover impaired in rats with ventral tegmental damage: Animal comparison of methylphenidate and placebo in adults model for tests of thought disorder. Biological Psychiatry, with childhood-onset attention-deficit hyperactivity 17, 243–258. disorder. Archives of General Psychiatry, 52, 434–433. Phillipson, O. T. (1979). Afferent projections to the ventral Stemmer, B., Segalowitz, S. J., Witzke, W., Lacher, S., & tegmental area of Tsai and interfascicular nucleus: A Scho¨nle, P. W. (2000). Do patients with damage to the horseradish peroxidase study in the rat. Journal of anterior cingulate show an error-related negativity (ERN)? Comparative Neurology, 187, 117–144. Psychophysiology, 37, 95. Rammsayer, T. H. (1999). Neuropharmacological evidence for Stemmer, B., Witzke, W., & Scho¨nle, P. W. (2001). Losing the different timing mechanisms in humans. Quarterly Journal error related negativity in the EEG of human subjects: An of Experimental Psychology, 52, 273–286. indicator of willed action. Neuroscience Letters, 308, Richelson, E. (1989). Antidepressants pharmacology and 60–62. clinical use. In Treatments of psychiatric disorders: A task Suri, R. E., & Schultz, W. (1999). A neural network model with force report of the American Psychiatric Association dopamine-like reinforcement signal that learns a spatial (p. 1776). Washington, DC: American Psychiatric Association. delayed response task. Neuroscience, 91, Ridderinkhof, K. R., de Vlugt, Y., Bramlage, A., Spaan, M., Elton, 871–890. M., Snel, J., & Band, G. P. H. (2002). Alcohol consumption Sutton, R. S., & Barto, A. G. (1998). Introduction. In impairs detection of performance errors in mediofrontal Reinforcement learning (pp. 1–23). Cambridge, MA: cortex. Science, 298, 2209–2211. MIT Press.

Zirnheld et al. 1111 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/0898929041502779 by guest on 26 September 2021 Taga, C., Sugimoto, Y., Nishiga, M., Fujii, Y., & Kamei, C. Semprex-D and diphenhydramine on learning in young (2001). Effects of vasopressin on histamine H1 receptor adults with seasonal allergic rhinitis. Annals of Allergy, antagonist-induced spatial memory deficits in rats. Asthma and Immunology, 76, 247–252. European Journal of Pharmacology, 423, 167–170. Wallesch, C. W., Karnath, H. O., Papagno, C., Zimmermann, P., Takasawa, N., Takino, R., & Yamazaki, K. (1990). Event-related Deuschl, G., & Lucking, C. H. (1990). Parkinson’s disease potentials derived from feedback tones during motor patient’s behaviour in a covered maze learning task. learning. Japanese Journal of Physiological Psychology and Neuropsychologia, 28, 839–849. Psychophysiology, 8, 95–101. Weiler, J. M., Bloomfield, J. R., Woodworth, G. G., Grant, Tucker, D. M., Hartry-Speiser, A., McDougal, L., Luu, P., & A. R., Layton, T. A., Brown, T. L., McKenzie, D. R., Baker, deGrandpre, D. (1999). Mood and spatial memory: Emotion T. W., & Watson, G. S. (2000). Effects of fexofenadine, and right hemisphere contribution to spatial cognition. diphenhydramine, and alcohol on driving performance. Biological Psychology, 50, 103–125. A randomized, placebo-controlled trial in the Iowa driving Vidal, F., Hasbroucq, T., Grapperon, J., & Bonnet, M. (2000). simulator. Annals of Internal Medicine, 132, 354–363. Is the ‘‘error negativity’’ specific to errors? Biological Williams, J. H., Wellman, N. A., Geaney, D. P., Feldon, J., Psychology, 51, 109–128. Cowen, P. J., & Rollins, J. N. (1997). Haloperidol enhances Vurrman, E. F. P. M., van Vegel, L. M. A, Sanders, R. L., latent inhibition in visual tasks in healthy people. Muntjewerff, N. D., & O’Hanlon, J. F. (1996). Effects of Psychopharmacology, 133, 262–268.

1112 Journal of Cognitive Neuroscience Volume 16, Number 6 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/0898929041502779 by guest on 26 September 2021