From Electron Microscopy to Neural Circuit Hypotheses: Bridging the gap Recent advances in experimental technology promise rapid progress in developing a mechanistic understanding of how neural circuit structure, at the synaptic scale, leads to complex sensory, cognitive, and motor behaviors. Volume EM techniques have advanced to the point that neural tissue can be imaged at synaptic resolution in volumes large enough (~ 100 m) to contain a entire small neuron . Innovations in machine learning approaches to analyzing these EM images have now permitted automated segmentation of these into 3D representations of every neuronal process within the volume. However, significant technical challenges remain to improve EM acquisition and analysis methods to the point of achieving the millimeter-scale volumes required to reconstruct neuronal assemblies underlying complex behaviors. Tools are needed to improve the reliability of EM systems for the acquisition of such large datasets. Automated methods to identify the cell-type of segmented neurites and to characterize the firing patterns of neurons prior to EM imaging are also needed, as are software tools to query and visualize the resulting rich 3D datasets?essentially allowing neuroscientists to carry out ?virtual experiments.? Here we propose to develop a set of tools to address each of these problems. These tools will be developed in the context of the songbird, a powerful model organism to study the neural circuits underlying the production and learning of a complex vocal/motor behavior. Neural recordings in singing birds of every known cell type in cortical, thalamic, and basal ganglia circuits, have led to mechanistic circuit-level hypotheses, which in turn make clear and testable predictions about the ultrastructure of neural circuits. The proposed tools would allow neuroscientists to bridge the gap from vEM data to the virtuous circle of hypothesis driven anatomical experimentation, and data-inspired hypotheses. Our proposal describes three Specific Aims directed toward these goals:
Aim 1 : Develop tools to acq uire and to automatically segment millimeter-scale EM data sets Aim 2: Develop tools to densely identify cell-types of segmented neurites using ultrastructural fingerprints Aim 3: Develop and test a platform for virtual experimentation in segmented and annotated EM datasets
The goal of this BRAIN Initiative project is to develop tools and resources that will facilitate the detailed analysis of brain microconnectivity. We will develop new approaches and optimize existing techniques to facilitate this important new approach to understanding the basic circuitry and computations underlying brain function. The connectomic tools we develop will be of interest to, and will be made accessible to, the broader scientific community; such tools, and the detailed annotated structural datasets they enable, will facilitate the study of neurobiological mechanisms of brain disorders, including autism, schizophrenia, and Huntington?s disease.
