The NIH/NIGMS
Center for Integrative Biomedical Computing

Serial-Section Electron Microscopy for Analysis of Ganglion Connectivity

Professor Robert Marc

Biomedical imaging (i.e. image acquisition) is progressing at an extraordinary pace, and biologists, engineers, and clinical researchers are finding that the bottleneck in producing the next generation of scientific achievements is not the acquisition of images, but the analysis of those images. Serial section microscopy is one of the most compelling examples of this gap between image acquisition and analysis. The answers to many important biological questions depend on a better understanding of cellular ultrastructure, which forms the interface between biochemistry and anatomy. Detailed, data-driven descriptions of microscopic structures are especially important in neurobiology. Dr. Robert Marc, Mary H. Boesche Professor of Ophthalmology, conducts research toward reconstructing the complete connectivity patterns of ganglion cells in the mammalian retina.

The complete retinal reconstruction effort is important for four topical areas of neurobiology. First, classification of retinal neurons is incomplete despite a century of effort. Second, complete reconstruction provides a "ground truth" to assess whether a given connectivity creates neuronal feature selection, modifies simple spatiotemporal features, or merely represents buffering of signals. Third, while robust tools have been developed for modeling, virtually all models rely on inferred or partial connectivity patterns. Providing a complete network for every GC class uniquely defines modeling parameters. Further, the detailed attributes of some GCs (e.g. uniformity detectors, local-edge detectors, orientation detectors) are poorly known because they are rare or very small. Complete circuitry allows detailed prediction of properties for physiological testing. Finally, reconstruction analysis is currently a key tool in determining the mechanisms of CNS plasticity, but is hampered by the need to use small sample volumes and sample numbers. The goal of this collaboration is to provide tools for processing very large volumetric, serial-section datasets that contain complex, interconnecting structures.