Distributed and Mixed Information in Monosynaptic Inputs to Dopamine Neurons.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Systems Neuroscience

Background:

• Dopamine neurons are crucial for encoding reward prediction error (RPE), a fundamental aspect of learning and decision-making.
• Existing models of RPE computation face challenges in experimental validation due to difficulties in assessing neuronal activity and connectivity.
• Understanding the neural basis of RPE requires detailed characterization of inputs to dopamine neurons.

Purpose of the Study:

• To investigate the distribution of neural signals related to reward prediction error computation in inputs to dopamine neurons.
• To experimentally test computational models of RPE by examining the activity and connectivity of dopamine neuron inputs.
• To characterize the firing patterns of monosynaptic inputs to dopamine neurons during a classical conditioning task.

Main Methods:

• Established an awake electrophysiological recording system in mice.
• Utilized rabies virus and optogenetic techniques for cell-type identification of neuronal inputs.
• Performed classical conditioning tasks to elicit reward-related neural activity.
• Analyzed firing patterns of monosynaptic inputs to dopamine neurons.

Main Results:

• Key variables for RPE computation, including actual and predicted reward, were found to be distributed across multiple brain areas projecting to dopamine neurons.
• A significant number of input neurons across different brain regions signaled combinations of these RPE-related variables.
• Unexpected redundancy was observed in the neural representation of this seemingly simple computation.

Conclusions:

• Reward prediction error computation is not localized to specific brain areas but is distributed across a brain-wide network.
• Neural computations, even simple ones like RPE, involve distributed processing and exhibit redundancy.
• The study provides a systematic method for examining both activity and connectivity to understand neural computations.