Separable neural mechanisms contribute to feedback processing in a rule-learning task

https://doi.org/10.1016/j.neuropsychologia.2007.08.009Get rights and content

Abstract

To adjust performance appropriately to environmental demands, it is important to monitor ongoing action and process performance feedback for possible errors. In this study, we used fMRI to test whether medial prefrontal cortex (PFC)/anterior cingulate cortex (ACC) and dorsolateral (DL) PFC have different roles in feedback processing. Twenty adults completed a rule-switch task in which rules had to be inferred on the basis of positive and negative feedback and the rules could change unexpectedly. Negative feedback resulted in increased activation in medial PFC/ACC and DLPFC relative to positive feedback, but the regions were differentially active depending on the type of negative feedback. Whereas medial PFC/ACC was most active following unexpected feedback indicating that prior performance was no longer correct, DLPFC was most active following negative feedback that was informative for correct behavior on the next trial. The current findings show that inconsistent results about the role of prefrontal cortex regions in feedback processing are most likely associated with the informative value of the performance feedback. The results are consistent with the hypothesis that medial PFC/ACC is important for signaling expectation violation whereas DLPFC is important for goal-directed actions.

Introduction

An important function of adaptive control is using feedback to monitor our own performance for possible errors and make rapid adjustments. This type of performance monitoring has been well documented in the classic Wisconsin card sorting task (WCST) (Demakis, 2003, Miltner, 1963). In this task, participants must sort cards according to different possible sorting rules, related to color, shape or number. Feedback after each sorting trial indicates whether performance was correct or incorrect. Once the participant has discovered the correct rule, the rule changes unexpectedly and the participant needs to use positive and negative feedback cues to find the new correct sorting rule.

An important feature of the WCST is the processing of negative feedback (Bayless et al., 2006; Monchi, Petrides, Petre, Worsley, & Dagher, 2001), which can have several meanings (Barcelo & Knight, 2002). First, negative feedback that indicates that previous performance is no longer correct can be referred to as first-warning negative feedback. This scenario entails a violation of expectation due to the unexpected nature of the changes in the rule that has to be used (Monchi et al., 2001). Second, negative feedback that can be used to test hypotheses about the new sorting rule can be referred to as efficient negative feedback. In this case, the feedback indicates that possibilities for future actions are constrained, and that the participant can use goal-directed behavior to apply the correct rule on the next trial (Walton, Devlin, & Rushworth, 2004). Third, negative feedback can indicate that a performance error was committed. This feedback signals an internal violation of expectations and also requires the participant to exert goal-directed behavior on the next trial (Holroyd & Coles, 2002).

Neuroimaging studies may allow us to examine the possible dissociability of underlying mechanisms for these different types of negative performance feedback. To date, most neuroimaging studies have focused on the differences in neural activity associated with negative relative to positive feedback (Holroyd et al., 2004; van Veen, Holroyd, Cohen, Stenger, & Carter, 2004), but the results are inconclusive (Nieuwenhuis, Slagter, Alting von Geusau, Heslenfeld, & Holroyd, 2005). These studies have not attempted to dissociate between different types of feedback processing, and a review suggests that the performance feedback types summarized above may indeed be neurally separable.

Event-related potential (ERP) and source localization studies have suggested that the medial prefrontal cortex (PFC) is important for processing errors and negative feedback (Holroyd & Coles, 2002; Miltner, 1963; Nieuwenhuis, Holroyd, Mol, & Coles, 2004). The feedback-related negativity (FRN) is a negative deflection over (fronto-) central scalp locations originating in or near medial PFC. It peaks 250–300 ms after negative performance feedback and is thought to correlate with the evaluative function of the feedback (Gehring & Willoughby, 2002; Holroyd & Coles, 2002; Miltner, Braun, & Coles, 1997; Nieuwenhuis et al., 2004). Studies using functional magnetic resonance imaging (fMRI) have confirmed that medial PFC, mostly the anterior cingulate cortex (ACC), is active following errors and error feedback (Holroyd et al., 2004, Mars et al., 2005; Ullspeger & von Cramon, 2003). The finding that the medial PFC/ACC is important for processing errors and negative feedback suggests that this region provides the first indication of outcomes worse than anticipated (Brown and Braver, 2005; Holroyd et al., 2004, Kerns et al., 2004; Rushworth, Walton, Kennerley, & Bannerman, 2004). However, other neuroimaging studies have not consistently shown that medial PFC/ACC differentiates between negative and positive feedback (Nieuwenhuis et al., 2005, van Veen et al., 2004).

Lateral prefrontal cortex (lateral PFC) has also been implicated in processing negative performance feedback. For example, set-shifting studies that have used paradigms closely matching the WCST have consistently reported activation in dorsolateral (DL) PFC following switch cues (Dove, Pollmann, Schubert, Wiggins, & Von Cramon, 2000; Konishi, Hayashi, Kikyo, Takahashi, & Miyashita, 2002; Lie, Specht, Marshall, & Fink, 2006). In addition, Dias, Robbins, and Roberts (1997) reported that damage to lateral PFC in marmosets results in impaired performance on extradimensional switch trials.

Lateral PFC, including mid-DLPFC, is generally thought to be involved in monitoring of task sets in working memory (Kerns et al., 2004, Monchi et al., 2001). A feedback-learning study by Monchi et al. (2001) indicated that ventrolateral PFC is more active following negative feedback compared to positive feedback. In contrast, Nieuwenhuis et al. (2005) found in an fMRI study and ERP dipole source analyses that the right superior frontal gyrus was activated more strongly by positive than by negative feedback. Thus, regions within lateral PFC can be sensitive to both negative and positive feedback. It is possible that regions within lateral PFC are sensitive to the informative value of the feedback, rather than the valence of the feedback. This hypothesis finds support in studies showing that lateral PFC damage in humans impairs corrective behavior. Thus, the lateral PFC possibly interacts with medial PFC/ACC in monitoring behavior (Gehring & Knight, 2000).

Besides medial PFC/ACC and DLPFC, lateral orbitofrontal cortex (lat-OFC, area BA 47) (O’Doherty, Critchley, Deichmann, & Dolan, 2003) and the caudate nucleus (Monchi et al., 2001; Tricomi, Delgato, McCandliss, McClelland, & Fiez, 2006) have also been implicated in the processing of feedback in set-shifting tasks such as the WCST. Whereas lat-OFC has been implicated in more emotionally salient feedback processing (Dias et al., 1997, O’Doherty et al., 2003), the caudate has been found to be active following both positive and negative feedback (Tricomi et al., 2006). The latter result has been interpreted to mean that the caudate is involved in the general process of learning, but it remains unclear whether this region is sensitive to the informative value of the feedback.

The current study examined the role of the medial PFC/ACC and mid-DLPFC using a rule-learning task based on the classic Wisconsin card sorting task (WCST). Participants were presented with a stimulus display containing four horizontally spaced locations. A stimulus was randomly placed in front of one of the four locations. Participants were instructed to apply one of three stimulus–response mapping rules that had been practiced prior to scanning. After applying a rule, participants received positive or negative performance feedback (Crone, Ridderinkhof, Worm, Somsen, & van der Molen, 2004). The location task has two advantages over the classic WCST. First, all rules were based on the same dimension (location), eliminating the demands for extradimensional switches. Second, in the classic WCST some stimuli can be sorted according to multiple rules (e.g. color and shape), which can lead to ambiguous trials. This possibility was eliminated in the present design, because each location rule was specifically mapped to one response (see also Walton et al., 2004). During scanning, participants were not told which rule to apply and were instructed to use the positive or negative feedback in order to deduce the correct rule. After two, three or four correct consecutive rule applications the rule changed without warning, and participants were instructed to use positive and negative feedback to find the correct rule again.

We differentiated between three types of negative feedback, each of different informative value for the participant when performing the task. The first type of negative feedback (negFB) is the first-warning negFB, which indicates a rule-switch. An efficient negFB indicates that the rule chosen when searching for the correct rule is not correct. Therefore, the participant knows with certainty what the correct rule should be on the next trial. An error is only scored as efficient negative feedback when it is preceded by a first-warning negFB and followed by a correctly applied rule. The third type of negative feedback is the error-related negFB, which indicates that the participant failed to use the correct rule in a set of correct trials.

Based on prior studies indicating that medial PFC/ACC is active when there is high conflict (Mars et al., 2005) or a large number of response alternatives (Walton et al., 2004), we predicted increased activation in this region relative to DLPFC following a first-warning negFB. In contrast, we expected that DLPFC would be more active than medial PFC/ACC following an efficient negFB, because there are fewer response alternatives and the demand for goal-directed behavior is higher (Miller & Cohen, 2001). Both medial PFC/ACC and DLPFC were expected to be active following error-related negFB. In addition to medial PFC/ACC and DLPFC, we also tested the role of lat-OFC (Monchi et al., 2001; O’Doherty et al., 2003) and the caudate nucleus (Monchi et al., 2001, Tricomi et al., 2006) in feedback processing.

Section snippets

Participants

Twenty adults between 18 and 25 years of age (M = 20.6, S.D. = 1.92, 8 male) were recruited from Leiden University. The participants, all of whom received course credit or a fixed payment, were healthy right-handed volunteers with no history of neurological or psychiatric problems. Informed consent was obtained and the study was approved by the Internal Review Board at the Leiden University Medical Centre. Two subtests of the Wechsler adult intelligence scale (WAIS-III; Wechsler, 2000),

Behavioral data

Out of 300 trials there were an average of 58.9 (S.D. = 5.29) first-warning error trials, for which the negative feedback signaled a sorting rule change. On approximately half of the rule shifts, participants first made an efficient error in order to find the correct sorting rule (M = 26.5, S.D. = 5.29). Finally, on approximately half of the rule shifts participants committed a performance error when applying the correct rule (M = 20.64, S.D. = 14.53). The difference between the number of efficient

Discussion

The aim of this study was to investigate the neural correlates of processing performance feedback with different informative values. Medial PFC/ACC, DLPFC, and lat-OFC were fractionated in order to test their differential sensitivity to the informative value of negative and positive feedback. In prior studies, it had been shown that these regions play a role in feedback processing (Konishi et al., 2002, Lie et al., 2006, Monchi et al., 2001), but their contributions to different types of

Acknowledgements

The third and last authors (SARBR and EAC) were supported by NWO-VENI/VIDI grants. We thank Santani Teng and Michiel Westenberg for helpful comments on an earlier version of the manuscript, and Wouter Teeuwisse for technical assistance.

References (42)

  • C.A. Cocosco et al.

    Brain web: Online interface to a 3D MRI simulated brain database

    NeuroImage

    (1997)
  • E.A. Crone et al.

    Developmental Science

    (2004)
  • E.A. Crone et al.

    Neurocognitive development of the ability to manipulate information in working memory

  • G.J. Demakis

    A meta-analytic review of the sensitivity of the Wisconsin card sorting test to frontal and lateralized frontal brain damage

    Neuropsychology

    (2003)
  • N.U.F. Dosenbach et al.

    Distinct brain networks for adaptive and stable task control in humans

    Proceedings of the National Academy of Science of the United States of America

    (2007)
  • A. Diamond

    Normal development of prefrontal cortex from birth to young adulthood: Cognitive functions, anatomy, and biochemistry

  • R. Dias et al.

    Dissociable forms of inhibitory control within prefrontal cortex with an analogue of the Wisconsin card sorting test: Restriction to novel situations and independence from “on-line” processing

    Journal of Neuroscience

    (1997)
  • S. Durston et al.

    A shift from diffuse to focal cortical activity with development: The author's reply

    Developmental Science

    (2006)
  • R. Elliott et al.

    Dissociable neural responses in human reward systems

    Journal of Neuroscience

    (2000)
  • W.J. Gehring et al.

    Prefrontal–cingulate interactions in action monitoring

    Nature Neuroscience

    (2000)
  • W.J. Gehring et al.

    The medial frontal cortex and the rapid processing of monetary gains and losses

    Science

    (2002)
  • Cited by (0)

    View full text