Hierarchical error representation in medial prefrontal cortex.

The medial prefrontal cortex (mPFC) is reliably activated by both performance and prediction errors. Error signals have typically been treated as a scalar, and it is unknown to what extent multiple error signals may co-exist within mPFC. Previous studies have shown that lateral frontal cortex (LFC) is arranged in a hierarchy of abstraction, such that more abstract concepts and rules are represented in more anterior cortical regions. Given the close interaction between lateral and medial prefrontal cortex, we explored the hypothesis that mPFC would be organized along a similar rostro-caudal gradient of abstraction, such that more abstract prediction errors are represented further anterior and more concrete errors further posterior. We show that multiple prediction error signals can be found in mPFC, and furthermore, these are arranged in a rostro-caudal gradient of abstraction which parallels that found in LFC. We used a task that requires a three-level hierarchy of rules to be followed, in which the rules changed without warning at each level of the hierarchy. Task feedback indicated which level of the rule hierarchy changed and led to corresponding prediction error signals in mPFC. Moreover, each identified region of mPFC was preferentially functionally connected to correspondingly anterior regions of LFC. These results suggest the presence of a parallel structure between lateral and medial prefrontal cortex, with the medial regions monitoring and evaluating performance based on rules maintained in the corresponding lateral regions.