Hierarchical Distribution of Reward Representation in the Cortical and Hippocampal Regions

Dopaminergic inputs to various brain regions, such as the striatum, orbitofrontal cortex, and amygdala, play a critical role in processing reward acquisition information. While reward-related activity is also observed more broadly in motor, parietal, and hippocampal regions, the functional significance and potential hierarchy of reward-related representation across these latter areas remain unclear. We investigated this by quantifying neural predictive power using machine learning. Specifically, neural activity was examined in six brain areas-the primary and secondary motor cortices (M1 and M2), posterior parietal cortex (PPC), dorsal and ventral CA1 (dCA1 and vCA1), and lateral entorhinal cortex (LEC)-in male rats performing a self-initiated left-right choice task. Machine learning models classified rewarded versus nonrewarded trials based on neuronal firing properties significantly above chance for all regions. Crucially, classification revealed a clear performance gradient, forming a functional hierarchy: models using hippocampal data (dCA1 and vCA1) performed best, followed by LEC and PPC, with M1 and M2 performing lowest. Furthermore, SHapley Additive exPlanations (SHAP) analysis revealed a qualitative transformation in coding strategies along this hierarchy: while neocortical regions relied on subtle, distributed high-order statistics, the hippocampus utilized precise, categorical representations. At this apex, distinct strategies emerged: dCA1 primarily utilized temporally precise post-reward spike distributions with transient increase of response, while vCA1 integrated both spike timing and firing rate changes with suppressive response. These findings provide quantitative evidence for a functionally hierarchical and qualitative evolution of reward-related representation, highlighting distinct roles of dCA1 and vCA1 in encoding reward-related events to potentially guide future behavior.