A reinforcement learning algorithm shapes maternal care in mice

Abstract

The neural substrates for processing classical rewards such as food or drugs of abuse are well-understood. In contrast, the mechanisms by which organisms perceive social contact as rewarding and subsequently modify their interactions are unclear. Here we tracked the gradual emergence of a repetitive and highly-stereotyped parental behavior and show that trial-by-trial performance correlates with the history of midbrain dopamine (DA) neuron activity. We used a novel behavior paradigm to manipulate the subjects expectation of imminent pup contact and show that DA signals conform to reward prediction error, a fundamental component of reinforcement learning (RL). Finally, closed-loop optogenetic inactivation of DA neurons at the onset of pup contact dramatically slowed emergence of parental care. We conclude that this prosocial behavior is shaped by an RL mechanism in which social contact itself is the primary reward. One-Sentence SummaryMaternal interactions with offspring are shaped by a dopaminergic reinforcement learning mechanism.