Connectomes are characteristic network graphs with complete adjacency matrices. Their mapping remains a grand challenge in neuroscience. We have used deep connectomics mapping (mapping to statistical saturation) via automated transmission electron microscope (ATEM) imaging at 5,000 images/day, automated image volume assembly, comprehensive molecular tagging to classify cells, Viking annotation to trace networks, cell visualization with 3D renderers, and network analysis with database browsers and connectivity viewers. We have built the first 2 nm-resolution retinal connectome (RC1) and are exploring it with teams of expert annotators. This resolution allows unambiguous identification of synapses and gap junctions. We have discovered many new retinal features, including novel networks for type A(II) amacrine cells, the critical fan-out element in mammalian rod-cone networks. We have also defined the three fundamental signal processing roles for retinal amacrine cells. The over 30 different amacrine cell classes of the mammalian retina engage in three distinct modes of inhibition: (1) nested feedback/feedforward signal processing; (2) ON-OFF channel crossover motifs; and (3) long-range scotopic-photopic control loops. The next stage in connectomics is populating adjacency matrices with synaptic and coupling weights to generate realistic models.
展开▼
