Final Report Summary - FEEDBACKMOUSEVISION (The functional role of feedback signals to mouse primary visual cortex)
The aim of this project was to understand the role of early sensory cortex when we learn the significance of sensory stimuli. To this end, we monitored responses in the visual cortex of mice, while they learned visually guided tasks.
We developed a novel visually-guided task in a virtual reality environment, in which mice were running through a virtual corridor and had to learn to discriminate two visual patterns on the corridor walls. Mice learnt this task very quickly (within a week), perhaps because they actively interacted with their visual environment, as compared to most previous studies in head-fixed primates and rodents. After mice had learned this task, we additionally trained them to either perform a visual discrimination or an odor discrimination in the same visual environment, which allowed us to also study how neural responses to the same stimuli depended on the task that the mouse was performing.
We localized the different visual areas with intrinsic signal imaging. Bilateral optogenetic silencing in PV-ChR2 mice showed that V1 was required for the visual discrimination task. Increasing light intensity reduced performance in the visual discrimination task but not in the odour discrimination task.
We built a custom high-speed 2-photon microscope setup and used repeated in vivo two-photon imaging of genetic calcium indicators to monitor the activity of the same layer 2/3 cells in V1 while animals learned the task. This allowed us to precisely determine how single cells alter their processing of visual stimuli as these become behaviorally relevant to an animal.
We found a strong and progressive increase in the ability of neural populations to discriminate the task-relevant visual stimuli. Day-to-day improvements in behavioural performance were associated with increasingly distinguishable neural representations of task-relevant stimuli. This was the result of both an increase in the stabilization of existing neurons and the recruitment of new neurons with stimulus-selective responses.
We further show that underlying this learning-related increase in stimulus selectivity was a permanent component that was task-independent, as well as another component that was only apparent when the animals engaged in visual discrimination. This indicates that top-down feedback signals contribute to enhanced processing of behaviorally relevant stimuli. These signals acted globally on the visual circuit because they increased the selectivity of neurons encoding both the rewarded and non-rewarded stimuli.
In contrast, we discovered two types of non-sensory signals related to the task structure which emerged during learning. One type reflected the animal's expectation about the appearance of the visual stimuli, and the other type reflected the animal’s behavioral choice. Interestingly, these signals developed to specifically influence only a subpopulation of neurons whose firing predicted the reward.
Our results demonstrate that the earliest stage of cortical visual processing exhibits a remarkable flexibility to tailor its processing to task requirements and the learned behavioral relevance of sensory stimuli. Mice have become a crucial model for studying the circuit mechanisms of perception, behavior and learning. We believe our novel naturalistic behavioral paradigms will open up new possibilities for further elucidating the mechanisms that underlie these functions, and help bridge the gap between studies in the rodent and human brain.
We developed a novel visually-guided task in a virtual reality environment, in which mice were running through a virtual corridor and had to learn to discriminate two visual patterns on the corridor walls. Mice learnt this task very quickly (within a week), perhaps because they actively interacted with their visual environment, as compared to most previous studies in head-fixed primates and rodents. After mice had learned this task, we additionally trained them to either perform a visual discrimination or an odor discrimination in the same visual environment, which allowed us to also study how neural responses to the same stimuli depended on the task that the mouse was performing.
We localized the different visual areas with intrinsic signal imaging. Bilateral optogenetic silencing in PV-ChR2 mice showed that V1 was required for the visual discrimination task. Increasing light intensity reduced performance in the visual discrimination task but not in the odour discrimination task.
We built a custom high-speed 2-photon microscope setup and used repeated in vivo two-photon imaging of genetic calcium indicators to monitor the activity of the same layer 2/3 cells in V1 while animals learned the task. This allowed us to precisely determine how single cells alter their processing of visual stimuli as these become behaviorally relevant to an animal.
We found a strong and progressive increase in the ability of neural populations to discriminate the task-relevant visual stimuli. Day-to-day improvements in behavioural performance were associated with increasingly distinguishable neural representations of task-relevant stimuli. This was the result of both an increase in the stabilization of existing neurons and the recruitment of new neurons with stimulus-selective responses.
We further show that underlying this learning-related increase in stimulus selectivity was a permanent component that was task-independent, as well as another component that was only apparent when the animals engaged in visual discrimination. This indicates that top-down feedback signals contribute to enhanced processing of behaviorally relevant stimuli. These signals acted globally on the visual circuit because they increased the selectivity of neurons encoding both the rewarded and non-rewarded stimuli.
In contrast, we discovered two types of non-sensory signals related to the task structure which emerged during learning. One type reflected the animal's expectation about the appearance of the visual stimuli, and the other type reflected the animal’s behavioral choice. Interestingly, these signals developed to specifically influence only a subpopulation of neurons whose firing predicted the reward.
Our results demonstrate that the earliest stage of cortical visual processing exhibits a remarkable flexibility to tailor its processing to task requirements and the learned behavioral relevance of sensory stimuli. Mice have become a crucial model for studying the circuit mechanisms of perception, behavior and learning. We believe our novel naturalistic behavioral paradigms will open up new possibilities for further elucidating the mechanisms that underlie these functions, and help bridge the gap between studies in the rodent and human brain.