How ecological features of a virtual environment shape neuronal activity in the mouse visual processing system

Abstract

Animals must continuously parse rich sensory streams and translate them into behaviour that is adaptive to the current context. This thesis examines how ecologically meaningful visual information and self-initiated behaviour are represented in the mouse brain during interaction with a head-fixed virtual reality (VR) environment. I developed a VR corridor paradigm that preserves experimental control while enabling naturalistic behaviours such as locomotion, exploration, approach to a reward port, and spontaneous licking, so that sensory processing and behavioural dynamics could be quantified alongside neuronal population activity.

In Chapter 1, I present an overview of the literature relating to the visual processing system, and how VR can be used to study these processes within a context more ecological than found in traditional lab experiments. Chapter 2 articulates the aims of the project, namely the development of a paradigm utilizing VR which enables investigation of behaviour and visual processes of ecological relevance.

Chapter 3 documents iterative optimization of three training protocols. Pilot data obtained during development of the protocols confirmed robust engagement with the reward zone across protocols but revealed limited use of visual trial cues for discrimination, motivating a shift in emphasis from decision performance to sensory processing and neural representation.

Chapter 4 details the final experimental design and recording procedures. Mice controlled position on a treadmill while navigating a VR corridor divided into stimulus and reward areas (S+ vs. S- trials defined by wall gratings). Behavioural variables (position, velocity, licking) and extracellular neuronal activity were recorded simultaneously, primarily in V1 with additional subcortical sites. I describe data acquisition, spike sorting, and two complementary activity representations: occupancy-adjusted activity for position-locked analyses and conventional firing-rate metrics for modelling.

Chapter 5 shows that mice adopted ecologically interpretable strategies: licking concentrated in the reward zone over sessions, with idiosyncratic differences across animals; locomotion exhibited a robust bimodal velocity distribution reflecting alternation between stationary and running states; and time investment differed across task zones and individuals. Signal Detection Theory analysis indicated decision performance hovered near chance, consistent with weak reliance on the visual trial cue for guiding reward-seeking behaviour.

Chapter 6 demonstrates that individual neurons exhibited distinct activation patterns related to spatial position and task area, and that population activity displayed coordinated structure across cortical depth. Occupancy-adjusted maps revealed neurons and populations with preferential firing at specific positions or zones, providing a substrate for spatial encoding during active exploration.

Chapter 7 evaluates population representations using dimensionality-reduction and supervised approaches. Unsupervised methods (PCA, t-SNE, UMAP) produced embeddings largely dominated by temporal proximity, with limited structure aligned to task variables. CEBRA-Time captured smooth temporal organization and enabled accurate decoding of time, whereas CEBRA-Behaviour captured dynamics related to locomotion, but otherwise provided limited separation for task labels beyond time. Regularized regression (Lasso, Ridge) predicted task/behavioural variables at near-chance levels when temporal correlations were controlled via trial-wise splits, suggesting weak decodability of these variables from recorded visual populations under the present conditions.

Chapter 8 provides a general discussion, arguing that VR provides a powerful compromise between ecological relevance and experimental control for studying sensory processing. Show more