While the neural circuits that underlie behavior are of interest to a substantial part of the neuroscience community, there have been very few technical approaches that actually provide this kind of information across all levels at which circuits function, including the level of synaptic connections. We have established an electron microscopy core which is explicitly designed to provide the “wiring diagrams” of neural circuits in an efficient way. Much of our effort over the past 5 years has been to transform serial electron microscopy of large volumes (such as the fish nervous system) from a heroic to a more mundane enterprise. This transformation required innovations in hardware and software to abbreviate all the time-consuming steps in the connectomic pipeline. In particular we: 1) automated ultra-thin sectioning (using a tape-based approach), 2) automated image acquisition (using a custom multibeam serial electron microscope), 3) automated stitching and registration of the image data on high performance computing clusters, 4) automated segmentation of neurons and synapses on a GPU cluster, and 5) semi-automated proofreading and rendering of the neural circuits with custom software. Using this infrastructure we have collected tens of thousands of sections losslessly at 30 nm thickness and acquired images of them at lateral resolutions of 4 x 4 nanometers. This voxel size (480 nm3) provided enough detail for human or machine vision methods to trace out the finest aspects of neural connectivity.
Acquiring these circuits is also relevant if neuronal connectivity can be associated with cells of particular types, hence the significant benefit of doing analysis of cell types that have been defined in the fish atlas associated with our overall project.
Importantly, these circuit diagrams provide ground truth for testing and refining computational theories of brain function, and are therefore of obvious interest to theorists working on questions and constraints of circuit function.