Reconstruction, simulation, and analysis of large-scale neuronal connectivity of the mouse brain

Abstract:

Recent advances in spatial transcriptomics and connectomics are mapping out neurons in the mouse brain with increasing precision. However, computational models have yet to fully leverage these data. This thesis presents multiple projects that bridge this gap by integrating multimodal data to analyze, model, and simulate large-scale neuronal dynamics in the mouse brain.

First, we introduce a network-generating algorithm that reconstructs microscopic brain connectivity by combining the molecular and spatial profile of neurons with mesoscopic projection data. A GPU-powered simulation platform is then constructed to simulate a computational model, incorporating connectivity developed in conjunction with a detailed neuronal electrophysiology model.

Secondly, we validated the model by simulating the spatiotemporal dynamics of the mouse cortex in response to sensory stimuli, successfully reproducing realistic patterns of cortical activity. Building on this, we then examined the emergence of macroscopic traveling waves in the mouse cortex. The model successfully generated realistic traveling waves, and we developed computational techniques to quantify their properties in the 3-D computational domain. We discovered that realistic connectivity promotes higher and more flexible macroscopic traveling waves compared to artificially constructed local or uniform networks. The level of macroscopic waves also depends non-monotonically on the coupling strength between neurons in the network and is most prominent during slow oscillations in the delta frequency band (0.5–4 Hz), which is an electrophysiological hallmark of sleep. Extending this work, we also modeled how brain-derived neurotrophic factor (BDNF) induces synaptic changes in the cortical network that modulate local sleep activity.

Lastly, with our experimental collaborators, we developed an experimental-computational pipeline to map and analyze the most active neurons and networks in the mouse brain at different times of the day. Using activity-dependent genetic labeling, light-sheet imaging, machine learning segmentation, and spatial transcriptomic matching, we find distinct region- and layer-specific activation patterns of neurons in the mouse brain, especially increased activity in layer 5 of the cortex during wakefulness. We also observe a dynamic shift in excitatory/inhibitory balance across the brain. We then defined a new computational method that introduces the \textit{active connectivity}, which shows a reorganization of network hubs from subcortical to cortical regions during the wake period.

Together, these studies contribute to the computational infrastructure that interfaces with multiscale, multimodal data for next-level analysis, modeling, and simulation of the neuronal dynamics in the mouse brain.