Predictive signals in visual cortex: How experience and expectations shape visual processing

Our visual perception is not simply a continuous picture of the world. It is heavily influenced by context, prior experience and most importantly: predictions. We constantly predict the movements, actions and sensory information of everything around us and quickly detect any deviation from these predictions. In this thesis, I studied how visual processing is affected by experience and predictions and what happens in our brains when these predictions are violated. In Chapter 2, we developed a novel calcium imaging analysis toolbox that allowed us to extract high-quality neuronal signals from our calcium imaging datasets. Although not directly related to visual predictions, the work in this chapter was an important step to support the analytical work necessary for the other experimental chapters in this thesis, which both use calcium imaging as a main technique. In Chapter 3, we studied the role of chandelier cells in the primary visual cortex (V1) of mice in relationship to visuomotor predictive processing. We found that chandelier cells have unique connectivity with other neurons in the local network and that they are strongly activated by surprising visual stimuli, such as when these deviate from what is expected based on movement. Furthermore, we discovered that after repeated exposure to visual stimuli in a virtual tunnel, chandelier cells showed plasticity leading to reduced responses to visual input. We also found that in contrast to popular believe, chandelier cells in V1 have relatively little inhibitory control over the activity of nearby pyramidal cells. Finally, in Chapter 4 we describe the role of visual experience on predictive processing of visual information. We investigated how neurons in V1 responded to visual input and to the context of this information by showing mice several different natural scenes. By sometimes partially occluding the images, we were able to measure how neurons in V1 responded when contextual information, but no direct visual input, was presented. We discovered that when we showed partially occluded images, neurons in V1 responded to the occluded area with stimulus-specific activity: we could infer which image was shown based on their responses. These signals were strongly influenced by the extent to which the mouse was familiar with the images that we showed. We also found that separate populations of neurons in V1 responded to the occluded and not occluded pictures. Finally, we found that pictures that had not been shown during training of the mice, making them unexpected, induced stronger responses in V1. Taken together, the research described in this thesis provides a better understanding of the role of experience and predictions on visual processing.