Dopamine says do that again

Abstract

Repeated actions may be reinforced through action prediction error — the difference between an executed action and its prediction given a particular state — but experimental evidence for such a movement-based, value-free teaching signal has been lacking. Greenstreet, Martinez Vergara, Johansson et al. trained mice on an auditory discrimination task and found that dopamine in the tail of the striatum (TS) is needed for learning, and that TS dopamine levels correlate with movement, but not reward, on the choice task. The movement-related dopamine signal in TS decreased over the course of learning, increased when an unfamiliar stimulus was used, and reinforced stimulus–action associations on the choice task; these findings are consistent with dopamine encoding a value-free action prediction error. A dual-controller model incorporating both value-based (reward prediction error via ventral striatum) and value-free (action prediction error via TS) systems learned the task faster than a value-based controller alone and implicated a role for TS during the later stages of learning. These findings suggest that movement-related dopamine signals in TS, and perhaps in other parts of the dorsal striatum, encode a value-free teaching signal that reinforces state–action associations, promoting the repetition of actions. Action prediction error may thus serve as the basis for the implementation of habitual behavior.

Original reference: Nature https://doi.org/10.1038/s41586-025-09008-9 (2025)