演題詳細
Poster
Learning-driven enhancement of top-down control in the primary visual cortex
- P1-173
- Hiroshi Makino:1 Takaki Komiyama:1
- 1:University of California, San Diego, USA
Perception is a constructive process where incoming sensory information is interpreted in light of past experiences and knowledge, so called top-down modulations. This suggests that the activity of sensory areas in the brain is determined not just by feedforward sensory inputs but also by feedback inputs from higher brain areas. Although theoretical studies suggest a dynamic shift in the balance between the two pathways through learning, circuit mechanisms underlying this process is poorly understood. Here using two-photon calcium imaging in head-fixed mice, we examined the activity of L2/3 excitatory neurons in the primary visual cortex and two major inputs to them during passive sensory experience and reinforcement learning over days. Both passive experience and reinforcement learning caused a decrease in the activity of L4 excitatory neurons, the major source of bottom-up sensory inputs to L2/3, and a concurrent increase in top-down inputs from the posterior cingulate cortex (PCC) arriving at L1. Furthermore, in contrast to naive and passive experience conditions in which L2/3 excitatory neurons respond in a manner faithful to stimulus presentation, learning changed the temporal pattern of L2/3 responses such that the activity peaks at the time of the associated event. This learning-specific temporal pattern was present in the PCC inputs but not in L4 activity, suggesting an enhanced influence of top-down inputs exerted on the activity of L2/3 neurons. Compared to passive experience, the activity of somatostatin-expressing inhibitory neurons (SOM-INs), a type of inhibitory neurons that mainly inhibit distal dendrites in L1, was much weaker in learning, and optogenetic enhancement of SOM-IN activity after learning was sufficient to revert the temporal pattern of L2/3 responses to what is observed in naive and passive conditions. These results reveal a dynamic shift in the balance of bottom-up and top-down inputs in primary visual cortex during reinforcement learning and uncover a role for SOM-INs in controlling this process.