Secondary motor cortex tracks decision value during the learning of a non-instructed task.

Optimal decision-making depends on interconnected frontal brain regions, enabling animals to adapt decisions based on internal states, experiences, and contexts. The secondary motor cortex (M2) is key in adaptive behaviors in expert rodents, particularly in encoding decision values guiding complex probabilistic tasks. However, its role in deterministic tasks during initial learning remains uncertain. Here, we describe a self-initiated deterministic task requiring mice to use their forepaws to make choices without guiding cues. Our findings reveal that spontaneous decisions follow a "race" model between actions, which uncovers underlying decision values. We use in vivo microscopy and modeling to show that M2 neurons in male mice exhibit persistent activity-encoding decision values that predict action-selection probabilities. Optogenetic inhibition of the M2 reduces the reversal performance and alters the decision value. Additionally, updates in decision values determine the rate at which learning is reversed. These results highlight the use of decision values by the M2 to adapt choice during initial learning without instructive cues.