Physiology & Behavior 77 (2002) 477–482

The anterior cingulate as a conflict monitor: fMRI and ERP studies

Vincent van Veena,b, Cameron S. Cartera,b,c,d,*

aDepartment of Psychology, University of Pittsburgh, Pittsburgh, PA 15213, USA bCenter for the Neural Basis of Cognition, Pittsburgh, PA 15213, USA cDepartment of Psychiatry, University of Pittsburgh, Pittsburgh, PA 15213, USA dDepartment of Neuroscience, University of Pittsburgh, Pittsburgh, PA 15213, USA Received 31 July 2002; accepted 22 September 2002

Abstract

We propose that the anterior cingulate cortex (ACC) contributes to cognition by detecting the presence of conflict during information processing, and to alert systems involved in top-down control to resolve this conflict. Here, we review several functional magnetic resonance imaging (fMRI) and event-related potential (ERP) studies that have used simple response interference tasks, and propose that ACC activity is activated prior to the response during correct conflict trials and reflected in the frontocentral N2, and immediately following error trials and reflected in the error-related negativity (ERN). Furthermore, we suggest that certain disturbances in cognition and behavior in common mental disorders such as schizophrenia and obsessive-compulsive disorder (OCD) can be understood as resulting from alteration in performance monitoring functions associated with this region of the brain. D 2002 Elsevier Science Inc. All rights reserved.

Keywords: Cingulate; ; Conflict; Interference; Event-related potentials; Functional magnetic resonance imaging; N2; Error-related negativity; Schizophrenia; Obsessive-compulsive disorder

1. Introduction current review will focus on discussing evidence support- ing the conflict theory. In particular, we focus on data The anterior cingulate cortex (ACC), located on the obtained using two methodologies from modern cognitive medial surface of the frontal lobes, has been the subject of neuroscience that we have used to test predictions of this increasing interest during the previous decade. It is usually theory: functional magnetic resonance imaging (fMRI) and thought of as playing an important role in attentional and ERPs. fMRI has excellent spatial resolution but relatively motivational processes [1]; however, its exact role in poor temporal resolution, while ERPs have excellent cognition has been the subject of debate. We have recently temporal resolution but relatively poor temporal resolution. proposed a comprehensive theory of human ACC func- The two methodologies are thus mutually complementary tioning that we believe accounts for most, if not all, of the in developing a spatio-temporal map of human brain neuroimaging and event-related potential (ERP) data avail- activity during cognition. able about this region of the brain [2,3]. According to this view, the ACC is activated in response to conflict occur- ring between incompatible streams of information process- 2. fMRI studies of ACC functioning ing. Following conflict detection, regions of the lateral prefrontal cortices and other areas associated with atten- An earlier, popular view of ACC functioning main- tional control are engaged to resolve the conflict. The tained that it was involved in exerting and selection for action [4]. According to this view, the ACC guides the selection and processing of stimuli that need to be acted upon or are otherwise relevant to the * Corresponding author. Western Psychiatric Institute and Clinic, 3811 O’Hara Street, Pittsburgh, PA 15213, USA. Tel.: +1-412-624-5007; fax: +1- currently active task or goal representation. In contrast, 412-624-0223. we maintain that ACC activity is often observed when E-mail address: [email protected] (C.S. Carter). control is engaged because a strong engagement of

0031-9384/02/$ – see front matter D 2002 Elsevier Science Inc. All rights reserved. PII: S0031-9384(02)00930-7 478 V. van Veen, C.S. Carter / Physiology & Behavior 77 (2002) 477–482 control is needed in situations in which conflict is high When considering interference tasks, an important issue [3], conflict and control being confounded in most stud- to address is to determine at what level of processing ies. conflict takes place [10]. In the same vein, we might ask Two recent fMRI studies from our lab have contrasted what levels of processing are monitored by the ACC for the the control view with the conflict theory. One study [5] did presence of conflict. In the incongruent condition of tasks so by analyzing between-trial reaction time (RT) adjust- such as the Eriksen or Stroop tasks, one dimension of the ments and their relationship to ACC activity during a stimulus is associated with the correct response and another version of the Eriksen flanker interference paradigm [6]. dimension of the stimulus leads to an incorrect response. In this paradigm, participants respond to a centrally pre- Thus, it has been argued that incongruent stimuli in the sented target while trying to ignore simultaneously pre- Eriksen or Stroop interference tasks not only elicit response sented flanker stimuli, responding to the identity of the conflict, but also conflict at the level of stimulus encoding or target with a left or right button press. It is usually found target detection, and a comparison between congruent and that RTs and error rates are smallest when these flankers incongruent stimuli would confound these two sources of are identical to the target (congruent), and highest when interference [11]. they are mapped onto the opposite response hand (incon- A first attempt to start addressing the issue regarding gruent), indicating that conflict, and thus the need for what kinds of conflict the ACC responds to was done using control, are high during such trials. In this paradigm, a version of the Eriksen flanker task [12]. In addition to a control is dynamically adjusted depending on the trial congruent condition (CO), in which flankers and target are type, as reflected in between trial-RT adjustments; follow- identical, and a response-incongruent condition (RI), in ing high-conflict trials, control is more strongly engaged which flankers and target are mapped onto opposite [7]. Thus, when an incongruent trial is preceded by an response fingers, we used a condition in which flankers incongruent trial (iI trial), RT is shorter than when it is are different than the target, but mapped onto the same hand preceded by a congruent trial (cI). This phenomenon as the target, simply by associating two letters with each provided a way to dissociate between conflict and control: response finger. This latter condition presumably reflects during iI trials, control is highly engaged and conflict is conflict at the level of stimulus encoding or target detection relatively low; during cI trials, control is lower and conflict without response conflict (stimulus-incongruent, SI), and is highest. The selection-for-action hypothesis of ACC RTs in this condition have been found to be greater than RTs functioning would thus have predicted stronger ACC to CO stimuli, but smaller than RTs to RI stimuli [6]. Thus, activation during iI trials. However, fMRI showed that comparing CO to SI trials presumably isolates stimulus the ACC was most strongly engaged in response to the cI conflict, whereas comparing SI to RI trials isolates response trials, supporting the conflict theory [5]. conflict. While RTs indeed showed both stimulus and A second experiment designed to critically distinguish response conflict (CO < SI < RI), comparing brain activity between conflict and control-based explanations of ACC elicited by these three trial types revealed that the ACC did functioning manipulated participants’ expectancy about not distinguish between the CO and SI stimuli [12]. ACC the frequency of high-conflict versus low-conflict trials activity was enhanced only to RI stimuli, suggesting that in [8]. It has been shown that, when most trials in a block this kind of interference task, the ACC only monitors for the are congruent, RT interference is very large. Since most presence of response conflict. The activation found in the trials are low-conflict in such a condition, control does Eriksen flanker study is in the same location as that in our not need to be strongly engaged, thereby making incon- earlier study using a different version of this paradigm [5]. gruent trials very high in conflict. In contrast, when most Specifically, the ACC activation seen in these and many trials are incongruent, RTs interference is much smaller, other interference tasks is located in the rostral cingulate presumably reflecting the fact that control is more motor zone [13]. From its strong connectivity to other parts strongly engaged [9]. As in the previous study, this study of the motor system, it is not surprising that this area is most provided the opportunity to contrast conflict and control. strongly activated to response conflict during these kinds of Instead of the Eriksen flanker interference task, it used a tasks (Fig. 1). version of the Stroop color-word naming interference Preliminary data from our lab, however, suggest that paradigm. In this paradigm, participants have to name dependingonthetask,ACCactivationmightnotbe the color in which a word is written; the usual finding is limited to response conflict, but might respond to other that RTs and error rates are smallest when color and word sources of conflict as well. This would be consistent with are identical, and highest when the color and the word are previous findings that suggest that ACC activity can be incongruent (e.g., the word ‘‘red’’ written in green letters, elicited in situations without motor requirements: ACC the correct response being ‘‘green’’) [10]. Results, again, activity has been observed in response to conflict induced supported the conflict theory; fMRI showed that ACC by unexpected error feedback in the Wisconsin Card activity to incongruent trials was highest during blocks Sorting Task [14], to conflict occurring at a conceptual when these were relatively infrequent, rather than frequent level by having participants read stories that do not form [8]. an integrated narrative [15], or to conflict induced when V. van Veen, C.S. Carter / Physiology & Behavior 77 (2002) 477–482 479

detection that an error was made, based on a comparison between the actual movement and the intended movement; a mismatch would give rise to the ERN [23]. However, a recent fMRI study using a ‘‘guilty knowledge’’ task observed strong ACC activation during ‘‘lies’’ [24]. This is inconsistent with a comparison-based account of error detection, because there is no mismatch between the inten- ded and the actual response; during lies, these are identical. We interpret the existence of the ERN as consistent with a conflict-based explanation of ACC functioning [2].Itis known that following error commission, stimulus processing continues, leading to fast activation of the correct response [25–27]. The moment at which the correct and incorrect responses are both maximally activated, thus, the moment of Fig. 1. Area of the anterior cingulate cortex engaged by response conflict greatest conflict, occurs therefore immediately following the but not conflict at the stimulus level during the Eriksen flanker task during error, which is of course the moment that the ERN peaks. functional MRI (from [Ref. 12]). Evidence suggests that the timing of events during correct incongruent trials is somewhat similar. After trial participants are instructed to inhibit sexual arousal while onset, the incorrect response is prepared based on partial watching erotic films [16]. stimulus analysis; however, before its activation reaches response threshold, the correct response manages to over- ride the incorrect response activation leading to a correct 3. ERP correlates of ACC activity response [26,27]. Thus, during correct incongruent trials, ACC activation is expected prior to the response, whereas Another popular theory of ACC functioning was based during error trials the ACC is activated immediately fol- on research using ERPs. In the presence of response lowing the response (Fig. 2). We hypothesized that ACC conflict, people are known to sometimes make fast, impuls- activity during correct conflict trials is reflected in the ive errors based on partial, incomplete analysis of the frontocentral N2 component of the ERP [2]. stimulus. Such impulsive errors are known as ‘‘slips’’ We used the same version of the Eriksen task described [17]. Research has shown that within 50–150 ms following earlier [12] using high-density EEG. Importantly, we found the commission of a slip, a large, negative deflection is that the ERN and N2 had identical scalp distributions, and observed in the ERP, with a maximum at frontal or central dipole source localization suggested the same caudal ACC sites [18,19]. Subsequent dipole modeling of this so-called generator for both peaks, strongly suggesting that these two error-related negativity (ERN) suggested an ACC generator components reflect the same process [2]. As we collected [2,20]. Later fMRI studies confirmed that the ACC showed enough trials to analyze error rates, we found that accuracy increased activity during error trials [21,22]. The discovery was reduced in response to RI trials, but was similar to CO of the ERN led to the hypothesis that it might reflect the and SI trials, showing that only the RI trials elicited

Fig. 2. Conflict-related (left, stimulus-locked) and error-related activity (right, response-locked) ERP waveforms observed as subjects performed the Eriksen flanker task (from Ref. [2]). CO: congruent condition in which targets and flankers are the same, SI: stimulus-incongruent condition in which targets and flankers are different but map to the same response hand, RI: response-incongruent condition in which targets and flankers are different and map to responses on opposite hands, ERN: error-related negativity. 480 V. van Veen, C.S. Carter / Physiology & Behavior 77 (2002) 477–482 increased response conflict. Since fMRI data had shown that Interestingly, fMRI research had previously indicated that the ACC is activated only to response conflict in this task, the rostral ACC was engaged during error trials [22]. Our we expected the N2 too to be increased only to RI stimuli, dipole analysis of the response-locked waveforms suggested but similar in amplitude to CO and SI stimuli. This was that the rostral ACC was activated following the ERN, and indeed what we found [2]. These data provide evidence that reflected in the ‘‘error positivity,’’ the component that the ERN and N2 reflect conflict detection by the ACC. follows the ERN [2]. Both the error positivity [32] and the Indeed, in a variety of studies using different tasks, includ- rostral ACC [1] have been associated with affective pro- ing Go/NoGo, target search and other conflict tasks, the cesses, and our dipole model is thus consistent with previous frontal-central N2 is largest to those conditions that involved findings and theories, linking the two lines of research. the most conflict (as reflected in increased error rates and/or One prediction that follows from the hypothesis that both RTs) [2,27–31]. For instance, much like in the fMRI study the ERN and the frontocentral N2 reflect conflict detection described earlier [5], the N2 has been observed to be larger by the ACC is that both should have the same cognitive to conflict trials that followed low-conflict trials than to properties. Interestingly, it has been shown that both the conflict trials that followed conflict trials [29]. Thus, we ERN and the N2 can be elicited unconsciously. During an suggest that the ACC activation observed during correct anti-saccade task, in which participants have to make a conflict trials in fMRI studies is the same as measured by saccade to the opposite side of a peripheral abrupt onset cue, the frontocentral N2 in ERP studies (Fig. 3). participants are often unaware of their errors. The ERN,

Fig. 3. Scalp topography and dipole source solutions of the N2 and ERN in the Eriksen flanker task. Note identical topographies and source locations in the caudal region of the ACC (from Ref. [2]). V. van Veen, C.S. Carter / Physiology & Behavior 77 (2002) 477–482 481 however, is still present following such unconscious errors These two lines of research suggest a role for the ACC in [33]. Likewise, targets preceded by subliminally presented both the lack of control in schizophrenia and the excess of primes elicited enhanced RTs and error rates, and an control in OCD. We believe that the approach outlined in increased N2, when the prime was incongruent with the the studies described above is also likely to be a useful one target [34]. These findings thus suggest that conflict mon- in the investigation of neural mechanisms underlying a itoring is a process that can occur outside of awareness. In number of other neuropsychiatric disorders, including neuroimaging studies, too, ACC activity has sometimes ADHD and substance abuse disorders, in which disturban- been found to be engaged by unconsciously occurring ces in cognitive control and internal monitoring are prom- conflict [35]. Although the hypothesis that the (caudal) inent features. ACC monitoring system is an implicit one still has to be addressed more thoroughly, the findings so far are in sharp contrast to earlier views that regarded the ACC as a structure References contributing in an important way to consciousness [36]. [1] Bush G, Luu P, Posner MI. Cognitive and emotional influences in anterior cingulate cortex. Trends Cogn Sci 2000;4:215–22. 4. Implications for neuropsychiatric disorders [2] van Veen V, Carter CS. The timing of action-monitoring processes in the anterior cingulate cortex. J Cogn Neurosci 2002;14:593–602. Disturbances in the function of the ACC have been [3] Botvinick MM, Braver TS, Barch DM, Carter CS, Cohen JD. Conflict monitoring and cognitive control. Psychol Rev 2001;108:624–52. reported in a number of psychiatric disorders, including [4] Posner MI, Dehaene S. Attentional networks. Trends Neurosci 1994; schizophrenia and obsessive-compulsive disorder (OCD). In 17:75–9. schizophrenia, cognitive control is impaired, and the level of [5] Botvinick MM, Nystrom LE, Fissell K, Carter CS, Cohen JD. Conflict cognitive dysfunction in the illness is correlated with monitoring versus selection-for-action in anterior cingulate cortex. behavioral disorganization and with social and occupational Nature 1999;402:179–81. [6] Eriksen BA, Eriksen CW. Effects of noise letters upon the identifica- disability [37]. The conflict monitoring theory of schizo- tion of a target letter in a nonsearch task. Percept Psychophys 1974; phrenia would predict that if ACC dysfunction is contrib- 16:143–9. uting to these deficits, decreases in both error-related and [7] Gratton G, Coles MGH, Donchin E. Optimizing the use of informa- conflict-related activity would be observed in the ACC, and tion: strategic control of activation of responses. J Exp Psychol Gen these would be associated with decreases in trial-to-trial 1992;121:480–506. [8] Carter CS, MacDonald III AM, Botvinick MM, Ross LL, Stenger VA, adjustments in performance. Many PET and fMRI studies Noll D, et al. Parsing executive processes: strategic vs. evaluative have shown decreased ACC activity in schizophrenia functions of the anterior cingulate cortex. Proc Natl Acad Sci USA [38,39]. ERP studies of the N2 and ERN in this illness 2000;97:1944–8. have shown decreases in these indices of ACC functioning, [9] Lindsay DS, Jacoby LL. Stroop process dissociations: the relationship too [40]. Finally, using fMRI, we observed a reduction of between facilitation and interference. J Exp Psychol Hum Percept Perform 1994;20:219–34. error-related activity in the ACC in schizophrenia, associ- [10] MacLeod CM. Half a century of research on the Stroop effect: an ated with decreases in post-error adjustments in perform- integrative review. Psychol Bull 1991;109:163–203. ance in this illness [39]. These data suggest that a disruption [11] Kornblum S, Stevens GT, Whipple A, Requin J. The effects of irrel- of ACC functioning contributes to impaired cognitive con- evant stimuli: 1. The time course of stimulus–stimulus and stimulus– trol in schizophrenia through an impairment in its perform- response consistency effects with Stroop-like stimuli, Simon-like tasks, and their factorial combinations. J Exp Psychol Hum Percept ance monitoring function. Perform 1999;25:688–714. OCD is characterized by obsessive thoughts and repet- [12] van Veen V, Cohen JD, Botvinick MM, Stenger VA, Carter CS. An- itive behaviors accompanied by a persistent sense of doubt terior cingulate cortex, conflict monitoring, and levels of processing. and dread. Functional imaging studies of OCD have often NeuroImage 2001;14:1302–8. reported increased activity in the ACC in this illness [41]. [13] Picard N, Strick PL. Motor areas of the medial wall: a review of their location and functional activation. Cereb Cortex 1996;6:342–53. An increased ERN has been reported in this illness, which [14] Monchi O, Petrides M, Petre V, Worsley K, Dagher A. Wisconsin has been interpreted as an overactive error monitoring Card Sorting revisited: distinct neural circuits participating in different system [42]. However, OCD patients are often observed to stages of the task identified by event-related functional magnetic res- perform in the normal range, but to take a very cautious onance imaging. J Neurosci 2001;21:7733–41. approach to task performance, suggesting that OCD patients [15] Robertson DA, Gernsbacher MA, Guidotti SJ, Robertson RRW, Irwin W, Mock BJ, et al. Functional neuroanatomy of the cognitive process are unusually sensitive to conflicts. Thus, we propose that of mapping during discourse comprehension. Psychol Sci 2000;11: hyperactivity in the ACC should be seen during both 255–60. incorrect trials as well as correct conflict trials, indicative [16] Beauregard M, Le´vesque J, Bourgouin P. Neural correlates of con- of an overactive conflict monitoring system. We tested this scious self-regulation of emotion. J Neurosci 2001;21(RC165):1–6. hypothesis using event-related fMRI, and confirmed that [17] Reason J. Human error. Cambridge, UK: Cambridge Univ. Press; 1990. [18] Falkenstein M, Hohnsbein J, Hoormann J, Blanke L. Effects of cross- both conflict-related and error-related activity were indeed modal divided attention on late ERP components: II. Error processing increased in the caudal ACC of OCD patients, compared to in choice reaction tasks. Electroencephalogr Clin Neurophysiol 1991; healthy controls [41]. 78:447–55. 482 V. van Veen, C.S. Carter / Physiology & Behavior 77 (2002) 477–482

[19] Gehring WJ, Goss B, Coles MGH, Meyer DE, Donchin E. A neural junctions of spatial frequency and orientation as a function of stimulus system for error detection and compensation. Psychol Sci 1993;4: parameters and response requirements. Electroencephalogr Clin Neu- 385–90. rophysiol, Evoked Potentials 1993;88:51–63. [20] Dehaene S, Posner MI, Tucker DM. Localization of a neural system [32] Falkenstein M, Hoormann J, Christ S, Hohnsbein J. ERP components for error detection and compensation. Psychol Sci 1994;5:303–5. on reaction errors and their functional significance: a tutorial. Biol [21] Carter CS, Braver TS, Barch DM, Botvinick MM, Noll DC, Cohen Psychol 2000;51:87–107. JD. Anterior cingulate cortex, error detection, and the online monitor- [33] Nieuwenhuis S, Ridderinkhof KR, Blom J, Band GPH, Kok A. Error- ing of performance. Science 1998;280:747–9. related brain potentials are differentially related to awareness of re- [22] Kiehl KA, Liddle PF, Hopfinger JB. Error processing and the rostral sponse errors: evidence from an antisaccade task. Psychophysiology anterior cingulate: an event-related fMRI study. Psychophysiology 2001;38:752–60. 2000;33:282–94. [34] Leuthold H, Kopp B. Mechanisms of by masked stimuli: [23] Scheffers MK, Coles MGH, Bernstein P, Gehring WJ, Donchin E. inferences from event-related brain potentials. Psychol Sci 1998;9: Event-related brain potentials and error-related processing: an analysis 263–9. of incorrect responses to go and no-go stimuli. Psychophysiology [35] Berns GS, Cohen JD, Mintun MA. Brain regions responsive to nov- 1996;33:42–53. elty in the absence of awareness. Science 1997;276:1272–5. [24] Langleben DD, Schroeder L, Maldjian JA, Gur RC, McDonald S, [36] Posner MI. Attention: the mechanisms of consciousness. Proc Natl Ragland JD, et al. Brain activity during simulated deception: an Acad Sci USA 1994;91:7398–403. event-related functional magnetic resonance study. NeuroImage [37] Carter CS, Barch DM. Attention, memory and language disturbances 2002;15:727–32. in schizophrenia: characteristics and implications. In: Andrade C, ed- [25] Rabbitt P, Vyas S. Processing a display even after you make a re- itor. Advances in psychiatry, vol. 3. London/New Delhi: Oxford Univ. sponse to it: how perceptual errors can be corrected. Q J Exp Psychol, Press; 2000. p. 45–72. A Human Exp Psychol 1981;33A:223–39. [38] Carter CS, Mintun M, Nichols T, Cohen JD. Anterior cingulate [26] Gratton G, Coles MG, Sirevaag EJ, Eriksen CW, Donchin E. Pre- and gyrus dysfunction and selective attention deficits in schizophrenia: 15 poststimulus activation of response channels: a psychophysiological [ O]H2O PET study during single-trial Stroop task performance. analysis. J Exp Psychol Hum Percept Perform 1988;14:331–44. Am J Psychiatry 1997;154:1670–5. [27] Kopp B, Rist F, Mattler U. N200 in the flanker task as a neurobeha- [39] Carter CS, MacDonald III AW, Ross LL, Stenger VA. Anterior cingu- vioral tool for investigating executive control. Psychophysiology late cortex activity and impaired self-monitoring of performance in 1996;33:282–94. patients with schizophrenia: an event-related fMRI study. Am J Psy- [28] Smid HGOM, Jakob A, Heinze H-J. An event-related brain potential chiatry 2001;158:1423–8. study of visual selective attention to conjunctions of color and shape. [40] Kopp B, Rist F. An event-related brain potential substrate of disturbed Psychophysiology 1999;36:264–79. response monitoring in paranoid schizophrenia patients. J Abnorm [29] Wijers AA, Mulder G, Okita T, Mulder LJ, Scheffers MK. Attention to Psychology 1999;108:337–46. color: an analysis of selection, controlled search, and motor activation, [41] Ursu S, Jones M, Shear MK, Stenger VA, Carter CS. Overactive using event-related potentials. Psychophysiology 1989;26:89–109. action monitoring in obsessive-compulsive disorder: evidence from [30] Kopp B, Mattler U, Goertz R, Rist F. N2, P3 and the lateralized functional MRI. Psychol Sci [in press]. readiness potential in a nogo task involving selective response pri- [42] Gehring WJ, Himle J, Nisenson LG. Action-monitoring dysfunction ming. Electroencephalogr Clin Neurophysiol 1996;99:19–27. in obsessive-compulsive disorder. Psychol Sci 2000;11:1–6. [31] Kenemans JL, Kok A, Smulders FT. Event-related potentials to con-