Dynamics of striatal action selection and reinforcement learning.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Systems Neuroscience

Background:

• Spiny projection neurons (SPNs) in the dorsal striatum are central to reinforcement learning models in the basal ganglia.
• Existing models face inconsistencies with known SPN synaptic plasticity rules.

Purpose of the Study:

• To resolve the inconsistency between striatal reinforcement learning models and SPN synaptic plasticity.
• To propose a revised model for learning and action selection in the basal ganglia.

Main Methods:

• Analysis of synaptic plasticity rules for direct-pathway (dSPN) and indirect-pathway (iSPN) neurons.
• Computational modeling of striatal function.
• Validation using striatal recordings.

Main Results:

• Indirect-pathway (iSPN) plasticity, as modeled, hinders learning by reinforcing negative outcomes.
• Simultaneous activation of functionally opposing dSPNs and iSPNs reverses this pathological effect.
• Striatal recordings support the model's predictions regarding SPN activity.

Conclusions:

• A fundamental inconsistency in current striatal reinforcement learning models is identified and resolved.
• A novel model allows for multiplexing of learning and action selection signals without interference.
• This framework supports learning algorithms beyond standard temporal difference models.