Anatomy of an error: ERP and fMRI

Abstract

Successful inhibition of pre-potent responses involves conflict; failed inhibition involves both conflict and errors, complicating the study of errors. Event-related potential (ERP) and functional magnetic resonance imaging (fMRI) data were collected while ten subjects (26–55 years old) performed a Go/NoGo task, with pre-potent responses (88% Go) and inhibition of responses (12% NoGo). We measured error-related negativity (ERN) to false alarms (FA), correct-related negativity (CRN) to hits, NoGo N2 to correct rejections (CR) and Go N2 to hits. ERP difference scores were derived (ERN-CRN; NoGoN2-GoN2) for correlating with fMRI contrasts (FA–hits; CR–hits). Age effects were removed from ERN and N2 difference scores, and conflict effects, reflected in N2 difference scores, were removed from ERN. The resulting ERN correlated with fMRI activations in both caudal and rostral anterior cingulate cortex (ACC), while N2 correlated with fMRI activations in caudal ACC and in executive control regions including dorsolateral prefrontal cortex. Thus, error and conflict monitoring may be dissociable, subserved by both overlapping and distinct ACC regions.

Introduction

About 10 years ago, two laboratories reported a unique brain response to errors, the error-related negativity (ERN) (Gehring et al., 1993) and the negativity associated with errors (Ne) (Falkenstein et al., 1991). This negative component of the event-related brain potential (ERP) starts at the onset of the overt error response, peaks ≈100 ms thereafter, and is maximal at frontocentral midline scalp sites (Gehring et al., 1993, Kopp et al., 1996, Kopp and Rist, 1999). Although it is generated when the subject knows the correct response but fails to execute it (Dehaene et al., 1994), it is dissociable from error awareness (Nieuwenhuis et al., 2001). It is independent of corrective motor responses, occurring after errors of commission in Go/NoGo tasks even though no corrective actions are possible (Falkenstein et al., 1995, Scheffers et al., 1996, Falkenstein et al., 2000), and also occurring in response to external error feedback (Miltner et al., 1997, Badgaiyan and Posner, 1998, Luu et al., 2000). The ERN was interpreted as a reflection of error detection or attempted error inhibition, resulting from a comparison of the actual and intended responses (Gehring et al., 1993, Scheffers et al., 1996, Scheffers and Coles, 2000). That there might be such an error detection system was proposed by Rabbitt (1966) to explain our attempts to correct our errors “in flight” before they are complete.

Although the initial view of the ERN was that it reflected an error-monitoring system in the brain, an alternative theoretical account of the ERN that has gained prominence in recent years is that it reflects the detection of response conflict, rather than errors per se (Carter et al., 1998, Botvinick et al., 1999, Cohen et al., 2000, Botvinick et al., 2001). Carter et al. (1998) first reported functional magnetic resonance image (fMRI) activation in the anterior cingulate cortex (ACC) associated with errors of commission in the AX-continuous performance task (AX-CPT) and showed that this same ACC particle was activated to correct trials that were high in conflict. They argued that errors were simply a special case of response conflict in which the incorrect response channel prevailed in its competition with the correct response channel. The conflict hypothesis of the ERN, formalized by Botvinick et al. (2001), postulated that the ERN is emitted by the ACC as part of a conflict monitoring system that detects high degrees of response competition and recruits greater top-down control from the dorsolateral prefrontal cortex (DLPFC) to improve task performance and reduce conflict. In spite of conflict on correct trials, the ERN does not occur on these trials, possibly because response conflict is resolved before a correct response is emitted (Nieuwenhuis et al., 2003). An ERP reflection of conflict on correct trials is best seen in the stimulus-locked ERPs prior to response execution, perhaps as an N2 component—the second major negative component of the ERP, peaking between 200 and 400 ms. A study by van Veen and Carter (2002a) provided support for this hypothesis by showing conflict modulation of the N2 component of the stimulus-locked ERP.

The N2 was first linked to response conflict when it was shown to be associated with successful response inhibition in Go/NoGo tasks (for review see Falkenstein et al., 1999). Indeed, it has been suggested that the ERN might be a delayed NoGo N2; that is, ERN ‘may reflect a late and hence unsuccessful attempt to inhibit the false alarm’ (Falkenstein et al., 1999, p. 271). However, Falkenstein et al. compared ERPs associated with successful (N2) and unsuccessful (ERN) response inhibitions in visual and auditory Go/NoGo tasks and found that N2 and ERN can be distinguished based on scalp topography, sensory modality and group performance differences. Nevertheless, using dipole source localization analyses, van Veen and Carter (2002a) could not distinguish N2 and ERN elicited during a choice reaction time (RT) task, locating the source of both in the same area of the ACC.

The fact that an enhanced N2 is elicited by correct responses to high response-competition trials in choice RT tasks (van Veen and Carter, 2002a) and correctly withheld responses to NoGo trials in Go/NoGo tasks (Falkenstein et al., 1999) is consistent with the view that the stimulus-locked N2 reflects conflict in both tasks. Although Go/NoGo tasks involve the monitoring and inhibition of a single response, whereas choice RT tasks involve monitoring competition between two or more responses, Go/NoGo tasks can be considered to evoke conflict between competing responses in the sense that participants must choose to go or not to go based on the stimulus presented. Competition between emitting a response or withholding it generates conflict on NoGo trials, particularly when highly probable Go trials create a pre-potent tendency to respond. Indeed, participants must inhibit the pre-potent Go response on NoGo trials just as they must inhibit the pre-potent tendency to read the color words in a Stroop task or respond in accord with incongruent flankers in the Eriksen flankers task. Thus, both the Go/NoGo task and choice RT tasks, such as the Stroop or flankers tasks, involve conflict and require some degree of pre-potent response inhibition. A further argument supporting the induction of conflict on the NoGo trials is the fact that NoGo P300 is later than the Go P300 (e.g., Pfefferbaum and Ford, 1988), consistent with other examples of a time cost associated with high conflict trials in choice RT tasks.

As mentioned above, fMRI has been used to distinguish errors and response conflict. Although Carter et al. (1998) showed the same region of the ACC was activated by conflict and errors, more recent fMRI studies (Kiehl et al., 2000, Braver et al., 2001, Ullsperger and von Cramon, 2001) have shown distinct areas of the ACC to be activated by conflict and by errors, suggesting potentially dissociable processes. However, the precise locations of these differential ACC activations to conflict and to errors have not been consistent across studies. For example, at least two studies (Kiehl et al., 2000, Braver et al., 2001) have localized conflict to caudal ACC and errors to rostral ACC. In contrast, Ullsperger and von Cramon (2001) showed that while both errors and response competition activated the frontomedian wall during the flankers task, errors preferentially activated the motor (caudal) region of ACC. To examine if the ERN recorded from the same subjects could have been generated in caudal ACC, a single dipole was placed there and found to account for 91% of the variance. Perhaps because this task did not elicit a large N2 on correct trials, no dipole analysis of the N2 was reported. However, van Veen and Carter (2002a) used dipole source analysis to model the generators of both N2 and ERN. They localized them both to caudal ACC, consistent with fMRI activations on conflict and error trials reported by others (Carter et al., 1998, Kiehl et al., 2000, Menon et al., 2001). They were unable to localize the ERN to the rostral ACC, which some fMRI studies have shown to be selectively activated by errors (Kiehl et al., 2000, Braver et al., 2001, Menon et al., 2001), but they did localize a generator of the later positivity associated with errors, or Pe (Falkenstein et al., 1995), to the rostral ACC. van Veen and Carter (2002b) suggested that the contribution of ACC activity to the generation of the ERN could reflect a more general role of the ACC in conflict detection, rather than error detection per se. Despite these results, the controversy surrounding the role of ACC in conflict versus error detection remains unresolved.

Other frontal lobe structures have also been indirectly implicated in the generation of the ERN. While patients with DLPFC lesions have normal amplitude ERNs following errors compared to age-matched controls, they also have “ERNs” on correct trials (Gehring and Knight, 2000). Similarly, abnormally large ERNs on correct trials, sometimes referred to as ‘CRNs’, have been observed in patients with schizophrenia (Kopp and Rist, 1999, Alain et al., 2002, Mathalon et al., 2002), a disorder known to be associated with DLPFC dysfunction (e.g., Weinberger and Berman, 1996, Goldman-Rakic and Selemon, 1997). Thus, DLPFC input appears to be critical for modulating the ERN signal differentially to correct and error responses. Indeed, bilateral activation of the DLPFC might also contribute to the ERN and could be mistakenly modeled as a single midline dipole in ACC. Similarly, bilateral activation of supplementary motor area (BA6) and pre-supplementary motor area (BA8), which are both preferentially activated to errors (Kiehl et al., 2000, Menon et al., 2001), could also be misrepresented by a single midline dipole in a typical equivalent-dipole model.

While it is unlikely that activity in the basal ganglia could contribute directly to the ERN, it may contribute indirectly (Holroyd and Coles, 2002). Indeed Parkinson’s Disease patients, who are known to have dopaminergic deficits in the basal ganglia, have reduced ERN amplitudes (Falkenstein et al., 2001, Holroyd and Coles, 2002), as do the elderly (Nieuwenhuis et al., 2002, Mathalon et al., 2003) in whom dopaminergic tone is also diminished (Arnsten et al., 1995, van Dyck et al., 2001). Dopaminergic neurons in the basal ganglia exhibit a phasic response when important predictions fail (Schultz et al., 1998). Moreover, one basal ganglia structure, the caudate, has been implicated in stopping an error in progress (Rubia et al., 2001). Another possible source indirectly contributing to the ERN is the cerebellum, which has a well-documented role in motor control and ‘in flight’ error adjustments (Schmanhmann, 1997).

While dipole localization studies provide one approach to distinguishing response conflict reflected in N2 from error detection reflected in ERN, dipole models do not provide unique solutions to the inverse problem. The scalp surface topography of an ERP can be equally well accounted for by many alternative dipole models that may vary greatly in both the number of dipoles and their locations. To further examine the distinction between conflict and error processing, we have adopted a regression analysis approach to isolate ERN variance that is independent of the N2, with subsequent correlation of these Go/NoGo ERP components (N2 and N2-adjusted ERN) with fMRI activations collected during a separate Go/NoGo scan session. fMRI activity associated with successful and unsuccessful response inhibition is likely to reflect more than just response conflict and error monitoring. It may include differential activity associated with sensation, perception, response selection, self-evaluation, planning for the next trial and any number of other processes happening in the 6 s it takes for the hemodynamic response to peak and the subsequent 6–12 s it takes to return to its baseline state. Correlation of these fMRI activations with ERP components allows a relatively focused interrogation of those brain activations that specifically covary with the electrophysiological indices of conflict and error processing that emerge and resolve within the first 400 ms following stimulus presentation (N2) or erroneous response execution (ERN).

To focus on those brain regions predominantly engaged by error monitoring, we correlated ERN data with fMRI activations from unsuccessful response inhibitions. Similarly, to focus on brain regions involved with conflict monitoring, we correlated NoGo N2 data with fMRI activations from successful response inhibitions. Because errors are typically associated with conflict, we removed from the ERN the variance it shares with N2 and examined the correlations of this relatively “conflict-free” measure of error processing with fMRI error activations.