Subspace analysis identifies low-dimensional interactions between cortical and subcortical brain regions in social reward computation.

Animals living in social environments evaluate rewards given to themselves and others. We previously showed that single-unit activities in cortical (the medial prefrontal cortex, MPFC) and subcortical (the dopaminergic midbrain nuclei, DA, and the lateral hypothalamus, LH) regions reflect social reward computation. Extending the single-neuron analyses, this study employs matrix and tensor decomposition methods to characterize population activity within single regions and interactions between pairs of regions. First, we determined the dimensionality of population activity and corresponding components in a single brain region. The dimensions of MPFC and LH were comparable, indicating similarities in the population activities of the two regions. In contrast, the dimensions of DA were considerably smaller, indicating that the activities were idiosyncratic. Further, "subspaces" shared between MPFC and DA, between MPFC and LH, and between LH and DA were identified. We found that a few components in MPFC and LH explained a large portion of population activities in DA, indicating that the neural computation of social rewards resides in a small subspace. Our findings demonstrate that a limited number of neural components within cortico-subcortical circuits regulate the social monitoring of rewards to oneself and others.