Neurons and the synaptic basis of the fMRI signal associated with cognitive flexibility Anja Stemme, a, T Gustavo Deco, b Astrid Busch, a and Werner X. Schneider a a Department of Psychology, Ludwig-Maximilian University Munich, Leopoldstr. 13, D-80802 Munich, Germany b ICREA Research Professor, Computational Neuroscience, Universitat Pompeu Fabra, Passeig de Circumvallacio ´ 8, 08003 Barcelona, Spain Received 23 July 2004; revised 20 January 2005; accepted 21 January 2005 Available online 31 March 2005 The Wisconsin Card Sorting Test (WCST) is well known to test cognitive flexibility in terms of set-shifting capabilities. Many fMRI studies with behaving monkeys as well as human subjects have shown transient neural activity in the Prefrontal Cortex (PFC), as indicated by an increase in the fMRI signal, following a rule change in the WCST or when using a WCST-like paradigm. We present a computational model, covering a limited number of PFC neurons and using precise biophysical descriptions, which is able to simulate WCS-like tests. Further, the detailed neuronal representation of the model allows us to calculate the resulting fMRI signal. Thus, we are able to analyze the adequacy of the model and its structure by comparing the calculated fMRI signal with the experimental data which in turn provides promising insights into the neural base of the increase in the fMRI signal. D 2005 Elsevier Inc. All rights reserved. Keywords: Event-related fMRI; Cognitive set-shifting; Wisconsin Card Sorting Test; Attention; Working memory; PFC Introduction The prefrontal cortex (PFC) plays an important role in the cognitive control of behavior and the Wisconsin Card Sorting Test (WCST) is one of the major tests to detect respective PFC dysfunction (Konishi et al., 1999; Miller, 2000a,b). The main aspect tested by the WCST is the ability of subjects to switch from a previously relevant context to a new context. This so-called set- shifting capability was first shown to be impaired in patients with frontal lobe damages by Milner (1963) and seems to be affected also in patients suffering from Schizophrenia or Parkinson’s disease (see, for example, Everett et al., 2001; Owen et al., 1993). Event-related fMRI studies with behaving monkeys as well as healthy humans have shown an increase in the fMRI signal following a rule change in WCS-like tests in the posterior part of the interior frontal sulci (Konishi et al., 1999; Nakahara et al., 2002). Konishi et al. (1999) introduced a modification of the WCST, which the authors refer to as the dinstructionT condition as opposed to the doriginalT condition. In the instruction variant of the WCST, the subjects were explicitly informed about the new rule they had to choose thus removing the trial-and-error load. Konishi et al. observed that the increase of the fMRI signal is comparatively lower in the instruction condition compared to the original condition. In Schizophrenics, on the other hand, a substantial lack of activation was found in studies using the original WCST (see Mitchell et al., 2001; Volz et al., 1997). However, the reason for the increase in the fMRI signal in healthy subjects or for the inability of the various patient groups to shift attention to a new context still remains unclear. Several computational models based on neural network algorithms have been developed to simulate Wisconsin Card Sorting tests and especially the impaired set-shifting capabilities (also referred to as dperseverative errorsT). Dehaene and Changeux (1991) developed a very interesting neural network architecture to model subject behavior in a WCST. The model comprised several neuronal clusters consisting of simple inhibitory and excitatory units with weighted connections. However, while successfully predicting the existence of drule-coding clustersT (kind of which were later detected by White and Wise, 1999) and simulating the perseverative errors of patients with frontal lobe damages their model exhibits some limitations. First, it relies on error-based modification of weights during the simulations to establish the switch to a new valid rule. Apart from the question whether this kind of short-term depression is biological plausible, the general absence of a detailed biophysical description does not permit to calculate a potential resulting fMRI signal. This argument holds especially for the weight-modification-based rule-switching proc- ess: it is not possible to translate the weight-modification process into neuronal activity which would be necessary to estimate a resulting fMRI signal. Furthermore, the PFC lesions were simulated by modifying connection strengths rather than the neural units themselves which is considered as well to be rather critical. Thus, although Dehaene and Changeux (1991) provided a 1053-8119/$ - see front matter D 2005 Elsevier Inc. All rights reserved. doi:10.1016/j.neuroimage.2005.01.044 T Corresponding author. E-mail addresses: stemme@psy.uni-muenchen.de (A. Stemme)8 Gustavo.Deco@upf.edu (G. Deco). Available online on ScienceDirect (www.sciencedirect.com). www.elsevier.com/locate/ynimg NeuroImage 26 (2005) 454 – 470