Abstract
Mitochondrial networks are defined as a continuous matrix lumen, but the morphological feature of neuronal mitochondrial networks is not clear due to the lack of suitable analysis techniques. The aim of the present study is to develop a framework to capture and analyze the neuronal mitochondrial networks by using 4-step process composed of 2D and 3D observation, primary and secondary virtual reality (VR) analysis, with the help of artificial intelligence (AI)-powered Aivia segmentation an classifiers. In order to fulfill this purpose, we first generated the PCs-Mito-GFP mice, in which green fluorescence protein (GFP) could be expressed on the outer mitochondrial membrane specifically on the cerebellar Purkinje cells (PCs), thus all mitochondria in the giant neuronal soma, complex dendritic arborization trees and long projection axons of Purkinje cells could be easily detected under a laser scanning confocal microscope. The 4-step process resolved the complicated neuronal mitochondrial networks into discrete neuronal mitochondrial meshes. Second, we measured the two parameters of the neuronal mitochondrial meshes, and the results showed that the surface area (μm2) of mitochondrial meshes was the biggest in dendritic trees (45.30 ± 53.21), the smallest in granular-like axons (3.99 ± 1.82), and moderate in soma (27.81 ± 22.22) and silk-like axons (17.50 ± 15.19). These values showed statistically different among different subcellular locations. The volume (μm3) of mitochondrial meshes was the biggest in dendritic trees (9.97 ± 12.34), the smallest in granular-like axons (0.43 ± 0.25), and moderate in soma (6.26 ± 6.46) and silk-like axons (3.52 ± 4.29). These values showed significantly different among different subcellular locations. Finally, we found both the surface area and the volume of mitochondrial meshes in dendritic trees and soma within the Purkinje cells in PCs-Mito-GFP mice after receiving the training with the simulating long-term pilot flight concentrating increased significantly. The precise reconstruction of neuronal mitochondrial networks is extremely laborious, the present 4-step workflow powered by artificial intelligence and virtual reality reconstruction could successfully address these challenges.