Abstract : Models of executive control and prefrontal cortex give a growing importance to reward signals in decision-making. Dopamine could play a key role in signalling the difference between obtained and expected reward (reward prediction error). We combined psychophysics, functional magnetic resonance imaging (fMRI), and positron emission tomography (PET) with a dopaminergic D2/D3 receptor antagonist (11C-raclopride) in order to study the neural bases of reinforcement learning of a motor sequence in human. fMRI allowed us to track the dynamics of this mental effort, which implied a large bilateral prefrontal, parietal and striatal network that activated suddenly during trial and error learning periods and collapsed during ensuing routinized repetition periods. This collapse could be driven by an elementary deduction process that predated the actual reception of reward (autoevaluation).
Moreover, some regions of that network were particularly engaged in processing statistical parameters of reward (the reward prediction error and the quantity of information).
At the same time, we developed a recent method of dynamical evaluation of dopamine release in vivo using PET. We showed that dopamine release increased bilaterally in ventral striatum and caudate nucleus during periods of trial and error search of motor sequences. In order to validate these observations and to evaluate the sensitivity of that method, we employed a standard PET paradigm (measure of raclopride binding potential).This paradigm also allowed us to measure a correlation between dopamine release in the right ventral striatum and subjects' behavioural values. These results are in line with the hypothesis of a role of the striatal dopamine in reinforcement learning in humans.
Thus, for the first time, the combination of fRMI and of dopaminergic receptors labelling with PET allowed us to shed light on both the dynamics of cerebral activation and the “cognitive neurochemistry” involved in a mental effort and reinforcement learning situation.