Neural Coding of Distinct Statistical Properties of Reward Information in Humans

TLDR

Reward processing in the brain underlies learning and motivation, and primate studies suggest the midbrain encodes two statistical parameters of reward: a transient error‑prediction signal that scales linearly with reward probability and a sustained, highly nonlinear signal peaking at maximal uncertainty (probability = 0.5). Using event‑related fMRI, we disentangled these two signals in humans by systematically varying monetary reward probability, magnitude, and expected value in a novel paradigm. The midbrain showed both transient activation for the error‑prediction signal and sustained activation for reward uncertainty, while the prefrontal cortex tracked the transient signal and the ventral striatum tracked the sustained signal. These results demonstrate that distinct functional brain networks encode different statistical properties of reward information, implying that the prefrontal cortex generates predictions and the ventral striatum supports motivational processes that facilitate information gathering.

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.