Files

Abstract

Understanding how a premature circuit adapts to environmental stimuli and matures could greatly improve our understanding of learning rules in a neural system. The development of synapses in the mammalian visual cortex during the critical period plays a significant role in the proper functioning of the visual system. Despite extensive studies with electrophysiology and light microscopy, there has not been a detailed investigation with connectomics. The rapid development of electron microscopy and computing technology in the last few years gave rise to ever-increasing data size and made previously costly connectomics reconstruction feasible. However, tremendous computational challenges remain in the field. Better computing resource accessibility and a higher level of automation performance are required for the field to expand into mainstream neuroscience. Also, skepticism in connectomics remains on whether/what biological insights it can offer at the current scale(100 um cubes). This thesis discusses three projects on different yet interconnected aspects of connectomics, addressing these concerns. The first project describes a software pipeline for end-to-end EM connectomics reconstruction, integrating multiple advanced automation tools into High-performance computing(HPC) environment, utilizing the previously untapped power of supercomputers at national labs for connectomics for the first time. The second project discusses a novel computational framework for neuronal subcompartment morphology classification and how it can perform neurite type classification at a massive scale with state-of-the-art accuracy. It is also demonstrated to detect and correct merge errors from segmentation, addressing a significant bottleneck in improving EM reconstruction accuracy. A comparative study is conducted in the third project to reconstruct mouse primary visual cortex layer 4 samples at age P14, the beginning of the critical period, and P105, in adulthood. Automatic saturated segmentation and synapse prediction are run with HPC, and focused proofreading of dendritic synapses is conducted. It is found that in the adult visual cortex, synapses increase substantially in both density and size, challenging a pruning centric hypothesis of circuit development. Also, a sharp increase of perisomatic shaft synapse density is observed, supporting the significance of modulatory inhibitory inputs, and the lack thereof, in determining the maturity of neurons in the visual cortex. Also, a significant increase in mitochondria coverage and its correlation with synapse density is reported, suggesting a vital role of mitochondria in the maturation of circuits.

Details

Actions

PDF

from
to
Export
Download Full History