Neural Coding of Distinct Statistical Properties of Reward Information in Humans

Open Access
Publisher Website
Google Scholar
Abstract
Brain processing of reward information is essential for complex functions such as learning and motivation. Recent primate electrophysiological studies using concepts from information, economic and learning theories indicate that the midbrain may code two statistical parameters of reward information: a transient reward error prediction signal that varies linearly with reward probability and a sustained signal that varies highly non-linearly with reward probability and that is highest with maximal reward uncertainty (reward probability = 0.5). Here, using event-related functional magnetic resonance imaging, we disentangled these two signals in humans using a novel paradigm that systematically varied monetary reward probability, magnitude and expected reward value. The midbrain was activated both transiently with the error prediction signal and in a sustained fashion with reward uncertainty. Moreover, distinct activity dynamics were observed in post-synaptic midbrain projection sites: the prefrontal cortex responded to the transient error prediction signal while the ventral striatum covaried with the sustained reward uncertainty signal. These data suggest that the prefrontal cortex may generate the reward prediction while the ventral striatum may be involved in motivational processes that are useful when an organism needs to obtain more information about its environment. Our results indicate that distinct functional brain networks code different aspects of the statistical properties of reward information in humans.