Parametric modulation of error-related ERP components by the magnitude of visuo-motor mismatch

Abstract

Errors generate typical brain responses, characterized by two successive event-related potentials (ERP) following incorrect action: the error-related negativity (ERN) and the positivity error (Pe). However, it is unclear whether these error-related responses are sensitive to the magnitude of the error, or instead show all-or-none effects. We studied error-monitoring with ERPs while healthy adult participants performed ballistic pointing movements towards a visual target with or without optical prisms, in alternating runs. This allowed us to record variable pointing errors, ranging from slight to large deviations relative to the visual target. Behavioural results demonstrated a classic effect of prisms on pointing (i.e. initial shifts away from targets, with rapidly improving performance), as well as robust prismatic after-effects (i.e. deviations in the opposite direction when removing the prisms after successful adaptation). Critically, the amplitude of both ERN and Pe were strongly influenced by the magnitude of errors. Error-related ERPs were observed for large deviations, but their amplitudes decreased monotonically when pointing accuracy increased, revealing a parametric modulation of monitoring systems as a function of the severity of errors. These results indicate that early error detection mechanisms do not represent failures in an all-or-none manner, but rather encode the degree of mismatch between the actual and expected motor outcome, providing a flexible cognitive control process that can discriminate between different degrees of mismatch between intentions and outcomes.

Introduction

The rapid detection of errors is crucial for adaptive and flexible behaviours. Accordingly, the recording of event-related brain potentials (ERPs) has revealed the existence of rapid error monitoring systems in the human brain, centered on the anterior cingulate cortex (Bush et al., 2000, Dehaene et al., 1994). Following erroneous action, two ERP specific components have been consistently identified in many studies, across various tasks and stimuli (Falkenstein, Hoormann, Christ, & Hohnsbein, 2000). First, the error-related negativity (ERN) peaks 0–100 ms after the occurrence of an incorrect response (either a false alarm or a discrimination error), with a maximum amplitude over fronto-central leads (see Gehring, Coles, Meyer, & Donchin, 1990). Next, the positivity (Pe) peaks 150–200 ms after the incorrect response, with a maximum amplitude over central leads (Falkenstein, Hohnsbein, Hoormann, & Blanke, 1991). Whereas the ERN is thought to reflect an early automatic detection of errors (i.e. the rapid appraisal of a mismatch between the actual and intended motor response), the Pe is assumed to reflect higher-order behavioural or motivational processes associated with the subsequent adjustment of performance (Nieuwenhuis et al., 2001, Scheffers et al., 1996).

Typically, the ERN and Pe components are recorded while participants perform interference tasks such as Stroop, Flanker, or go/nogo tasks, and occasionally make unwanted response errors. In the two former cases, discrimination errors may arise, whereas false alarms are frequently produced in the latter case; but in all these instances, comparing ERPs to incorrect vs. correct responses classically reveals prominent ERN and Pe components. However, because of this dichotomous distinction between correct and incorrect responses in most of the error monitoring studies, it is unclear if these ERP components may also code for the degree or “severity” of mismatch between the actual and intended motor response. Errors can be either large or small, and thus require different degrees of behavioural adjustment. Presumably, early error-detection systems, as reflected by the ERN and Pe ERP components, might be sensible to this factor, enabling flexible cognitive control and learning mechanisms. However, a simple comparison between correct vs. incorrect responses does not allow to test this prediction.

According to the dominant error detection theory of Falkenstein et al. (2000), the greater the amount of mismatch between the executed and intended correct response, the larger the probability to detect this error, and by extension, the larger the ERN amplitude, since this component is directly related to the detectability of errors (Maier, Steinhauser, & Hubner, 2008). However, it has been shown that undetected (i.e. unconscious) errors can also elicit a sizeable ERN component in the absence of any Pe component (Endrass et al., 2007, Nieuwenhuis et al., 2001, Scheffers and Coles, 2000), suggesting that the early ERN may reflect an automatic (perhaps subcortical) “all or nothing” error signal associated with alerting and learning mechanisms. Alternatively, a residual ERN to unconscious errors might indicate that some detection of mismatch may still operate on sensorimotor representations unavailable to conscious awareness. By contrast, the Pe was ascribed to a conscious (cortical) remedial action process, pointing to a functional dissociation between the ERN and Pe components (Nieuwenhuis et al., 2001). Thus, until now, ERP results have remained inconclusive regarding the link between the ERN and the degree of error detectability (Maier et al., 2008).

It is noteworthy that recent models of error monitoring have postulated that the ERN might reflect reinforcement learning signals mediated by dopaminergic pathways (Holroyd and Coles, 2002, Nieuwenhuis et al., 2004). In both humans and primates, there is evidence that the phasic firing rate of midbrain dopamine neurons encode the difference between the predicted and experienced reward of an event, consistent with a prediction error model (Schultz, Tremblay, & Hollerman, 2000). Moreover, this dopaminergic signal shows a linear response according to the extent to which expectations are violated (see Abler, Walter, Erk, Kammerer, & Spitzer, 2006 for human fMRI evidence; Fiorillo, Tobler, & Schultz, 2003), such that larger deviations from expectations lead to larger firing bursts of dopaminergic neurons. On the other hand, because errors correspond to events that turn out to be worse than anticipated, they are thought to induce a phasic suppression of dopamine, which can in turn produce a transient release of activity in the dorsal ACC activity, and thus generate the ERN/Ne (Holroyd & Coles, 2002). Based on these data (Abler et al., 2006, Fiorillo et al., 2003), one may therefore predict a linear relationship between the magnitude of errors and the amplitude of the ERN/Ne component (being under dopaminergic influences), although this hypothesis has received no empirical validation so far.

In the current study, we addressed this question using a novel method. We recorded response-related ERPs while healthy participants were asked to perform a simple visuo-motor pointing task in which variable errors could be systematically induced by prism goggles. Prism goggles create a compelling deviation of the visual field that does not correspond anymore to the motor space (Redding and Wallace, 1996, Rossetti et al., 1993). As a result, pointing movements towards a seen target become inaccurate and deviate away from the actual target location. To counteract this optical displacement induced by prisms, adaptive visuo-motor processes take place so as to progressively correct the pointing movements (see Redding, Rossetti, & Wallace, 2005 for a review). This so-called prismatic adaptation effect can fully restore pointing accuracy after a few trials (∼15 on average). But once adaptation has occurred, the active nature of this adjustment will lead to a deviation in the opposite direction when prisms are removed (the so-called aftereffect). These effects suggest rapid plasticity mechanisms and cerebral reorganisation in response to the visuo-motor mismatch. Interestingly, a recent fMRI study of prism adaptation in healthy participants (Luaute et al., 2009) reported that the magnitude of error produced by prismatic shift was correlated with activation of the ACC, consistent with the notion that this region plays crucial role in error detection (Kerns et al., 2004).

Here, to determine whether such recruitment of dorsal ACC during prism adaptation may correspond to typical error monitoring systems observed in ERPs, we recorded ERP in healthy participants who performed a similar pointing task with ballistic movements towards a visual target, with or without prisms in alternating runs. This enabled us to collect many pointing errors with variable magnitudes of deviation, and hence determine whether the amplitude of the ERN and Pe components may vary with the severity of deviations. We tested the prediction that the ERN (and Pe to a lesser extent) might reflect error detection mechanisms under the influence of reinforcement learning signals (perhaps related to dopaminergic midbrain structures; see Holroyd & Coles, 2002), whose degree of recruitment should be modulated by the amount of mismatch between the actual and intended motor response. In other words, the larger the deviation, the larger the amplitude of the error-related ERPs should be.