Nucleus accumbens dopamine release reflects Bayesian inference during instrumental learning.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Behavioral Neuroscience

Background:

• Dopamine release in the nucleus accumbens is theorized to represent reward prediction error, a key concept in reinforcement learning.
• Understanding how the brain maps sensory information to reward predictions is crucial but remains incomplete, especially with ambiguous environmental cues.

Purpose of the Study:

• To investigate whether dopamine signaling in mice reflects probabilistic belief updating during an instrumental learning task.
• To test the hypothesis that dopamine release is conditional on the brain's evolving beliefs about the environment's hidden state.

Main Methods:

• Mice performed a two-armed bandit task with a stochastically switching hidden state.
• Behavioral choices were recorded, alongside dopamine release in the nucleus accumbens using dLight imaging.
• Cognitive models, including Bayesian and non-Bayesian approaches, were compared against behavioral and neural data.

Main Results:

• Behavioral data analysis favored computational models that employed Bayesian updating of probabilistic beliefs.
• These Bayesian models also provided a superior quantitative fit to the measured mesolimbic dopamine (dLight) signals compared to non-Bayesian models.
• The findings indicate that probabilistic belief computation is integral to instrumental task performance and is mirrored in dopamine signaling.

Conclusions:

• Mesolimbic dopamine signaling in mice reflects the computation of probabilistic beliefs about environmental states.
• This study provides evidence for a biological basis of reinforcement learning, where dopamine signals are linked to Bayesian inference.
• The results underscore the role of probabilistic belief updating in instrumental decision-making and its neural correlates.