Using fMRI to decompose the neural processes underlying the Wisconsin Card Sorting Test

Abstract

The specific role of particular cerebral regions with regard to executive functions remains elusive. We conducted a functional magnetic resonance imaging (fMRI) study to segregate different network components underlying the Wisconsin Card Sorting Test (WCST), a test widely applied clinically to assess executive abilities.

Three different test variants of the WCST, differing in task complexity (A > B > C), were contrasted with a high-level baseline condition (HLB). Cognitive subcomponents were extracted in a serial subtraction approach (A–C, A–B, B–C). Imaging data were further subjected to a correlational analysis with individual behavioral parameters.

Contrasting A with the HLB revealed the entire neural network underlying WCST performance, including frontoparietal regions and the striatum. Further analysis showed that, within this network, right ventrolateral prefrontal cortex related to simple working memory operations, while right dorsolateral prefrontal cortex related to more complex/manipulative working memory operations. The rostral anterior cingulate cortex (ACC) and the temporoparietal junction bilaterally represented an attentional network for error detection. In contrast, activation of the caudal ACC and the right dorsolateral prefrontal cortex was associated with increased attentional control in the context of increasing demands of working memory and cognitive control. Non-frontal activations were found to be related to (uninstructed relative to instructed) set-shifting (cerebellum) and working memory representations (superior parietal cortex, retrosplenium).

The data provide neural correlates for the different cognitive components involved in the WCST. They support a central role of the right dorsolateral prefrontal cortex in executive working memory operations and cognitive control functions but also suggest a functional dissociation of the rostral and caudal ACC in the implementation of attentional control.

Introduction

The ability of a subject to effect contextually appropriate behavior involves a range of executive functions. The term ‘executive functions’ refers to a number of higher-order cognitive processes including planning, initiation, hypothesis generation, cognitive flexibility, decision making, regulation, judgment, feedback utilization and self-perception.

A clinically widely used test to assess executive functions is the Wisconsin Card Sorting Test (WCST) (Grant and Berg, 1948). Although the WCST is sensitive to frontal lobe dysfunction, it has proved difficult to determine the specific contributions of different cerebral regions to the task because of its complexity. A number of previous imaging studies used the WCST, aiming to delineate the neural representations of selected task components such as set-shifting, working memory, or inhibitory control processes (Monchi et al., 2001, Konishi et al., 1999, Konishi et al., 2003). However, integrating the findings of these studies in order to elucidate the specific contributions of particular brain regions to specific cognitive components has been problematic due to differences in experimental design. We therefore aimed at integrating and extending the findings of previous studies by fractionating the different cognitive components underlying the WCST within the same group of subjects. We thereby intended to translate the clinically well-established WCST into a functional neuroimaging paradigm, which can be used to delineate the neural correlates of the different task components in each subject. Determining the ‘neural signature’ of specific cognitive components in healthy subjects may allow dysexecutive syndromes (e.g. Parkinson's disease) to be characterized on a pathophysiological basis. Such simple efficient experimental designs could then be used in future studies of patient populations to establish functional imaging as an additional diagnostic and prognostic tool. With this ultimate aim in mind, we implemented a computerized version of the WCST with 3 different test variants (A, B, C) in a simple but efficient blocked-design, which allowed us to segregate different cognitive components of the WCST within the same individual.

A cognitive gradient in task demands was introduced across conditions (A > B > C), while the experimental setting remained otherwise unchanged. Task A was very similar to the original WCST, in which subjects have to match a stimulus card to one of four reference cards. The four reference cards show different geometric figures; they also differ in the number of objects displayed and their color (see Fig. 1). The stimulus card can be matched to one of the four reference cards by either color, shape, or number of objects. The subject is not informed about the sorting dimension but rather deduces the sorting rule in part by trial and error, relying upon the feedback provided by the investigator. If the correct sorting criterion has been found and if four correct answers have been given consecutively, the sorting dimension is changed by the examiner. The subject then has to adapt to the change and has to search again for the new correct sorting criterion. In task B, subjects are explicitly instructed about the dimensional changes, which occur once every four trials. The (instructed) dimension thus only has to be maintained/retrieved during the following four trials, until the next dimensional change occurs. In task C, the task is further simplified by providing the respective sorting dimension prior to each trial, thereby reducing working memory demands to a minimum. In the HLB, each stimulus card is identical to one of the four reference cards. The HLB thus includes no set-shifting operations and basically controls for perceptual, motor and identity matching and response selection components inherent to the WCST.

Accordingly, contrasting C with the HLB delineates the basic cognitive network engaged in the matching task (i.e. matching a non-identical card according to a criterion, relative to identifying identical cards as involved in the HLB) plus the neural processes underlying (instructed) set-shifting relative to no set-shifting. Contrasting B with the HLB additionally reveals simple working memory operations. Contrasting A with the HLB shows the whole neural network underlying WCST performance. Additionally, contrasting A with B depicted the neural correlates of error detection/feedback utilization, complex working memory operations and (uninstructed relative to instructed) set-shifting. Contrasting B with C extracts neural representations of simple working memory operations. Based upon previous data, we hypothesized that A–B would reveal increased neural activity in the dorsolateral prefrontal cortex (DLPFC), while B–C would show activity in the ventral prefrontal cortex, reflecting known functional dissociations within the prefrontal cortex regarding simple and complex working memory operations (Petrides, 2000). We furthermore expected the ACC to be increasingly engaged the higher the task demands are (A > B > C), reflecting processes of response selection (Barch et al., 2000), attentional control (Posner and Petersen, 1990) and error detection (Gehring et al., 1993).