Reward expectation signals in the superior colliculus during associative learning

Predicting future rewards allows neutral sensory cues to gain motivational importance and facilitate goal-directed actions that optimize reward acquisition. These predictions can be modified when outcomes deviate from expectations, resulting in an error in prediction. Dopamine neurons in the midbrain serve as a representative neural marker for these prediction error signals and transmit them to various brain regions.

These acquired reward predictions ultimately impact neural processes that enable animals to choose, prepare, and execute actions towards anticipated rewards. Expectations about rewards have been detected in different sensory and motor circuits, where they can regulate sensory processing and decision-making about actions. This leads to the inquiry about how sensorimotor circuits combine information about reward expectations with sensory and motor factors that drive reward-seeking behavior.

The superior colliculus (SC), a midbrain area known for its role in orienting and selecting actions, is substantially affected by reward context. Expected rewards influence visual and premotor activity, the value associated with objects changes based on learning, and a recent reward alters visual processing in the mouse SC. The SC also serves as an input source for midbrain dopamine neurons and can transmit or acquire conditioned sensory responses that impact dopaminergic activity.

However, SC projections can also trigger orienting movements without relying on reinforcement, which suggests that activities related to cues might represent salience or motor preparation instead of reward prediction. In this study, researchers utilized an associative-learning task in which auditory cues predicted varying reward probabilities, alongside continuous tracking of intermediate SC population activity throughout the learning process.

They aimed to determine if the SC acquires signals related to reward probability, whether these signals can be distinguished from anticipatory licking, and whether outcome responses display the characteristics of reward-prediction errors. The findings revealed that cue-evoked SC activity gradually distinguished reward probabilities over the course of learning, and cue-associated reward probability accounted for neural activity beyond anticipatory licking.

Therefore, the intermediate SC carries a learned reward-expectation signal that is intertwined with, but not interchangeable with, motor preparation. Conversely, outcome responses did not reveal a uniform population-level indication of reward-prediction errors. In summary, these results identify the SC as a location where predicted rewards are integrated with preparatory actions during associative learning.