Different stimulus types and long-term plasticity in mouse primary visual cortex

Abstract

Evidence points to the existence of long-term, experience-dependent plasticity in primary visual cortex (V1). This is surprising given the early location of V1 in the visual processing hierarchy, but is consistent with known anatomical and functional properties of V1 circuitry. Research has uncovered, with training, the emergence of stimulus- specific potentiation and complex “predictive” activity in V1. The Berry lab has previously explored encoding of novel and familiar visual stimuli in layer 2/3 excitatory neurons, along with the short-term stimulus memory capacity of these cells. Prior experiments used sequences of custom stimuli (Gabor patches) and two-photon calcium imaging, which enables in vivo recording from cortical neurons at single-cell resolution. I investigated whether training with sequences—made from different stimuli types—affects how layer 2/3 cells encode sequences. I performed two-day experiments using naturalistic movies, naturalistic scenes, and static gratings. I focused on "sustained" cells, or neurons responding periodically to part of a sequence while it is presented. Sustained responses may encode both spatial and temporal features of a sequence, but such responses have not been compared across different stimuli. For each experiment, analysis questions included how sustained responses changed in magnitude and temporal dynamics within and across days. Results indicate substantial overall plasticity, between days, for all stimulus types—including amplitude enhancement that can be stronger than Gabors-induced learning effects. During experiments with movies, population activity in response to "blank" frame substitutions—when no stimulus is shown—increasingly recapitulated the activity during uninterrupted movies. Such “filling in” had greater fidelity when blank frames are presented in the same fixed position (instead of random positions) within a movie. The findings are consistent with the basic expectations of a predictive coding model for V1. While thesis results have substantial limitations, they also suggest avenues for new experimentation.