The “cytoarchitectural” complexity of the mammalian brain is staggering. Depending upon the organism, the brain may consist of billions of cells, that are organized into a large number of distinctive regions with specific functional properties. The tissue in each brain region is built up of a large number of cell types, notably neurons, microglia, astrocytes, and oligodendrocytes, supported by an extensive network of blood vessels. Each of these cell types can be divided into several distinct subtypes. Each cell can, in turn, be in a variety of states of activation and health. A long list of molecular markers of cell health and activation states have been discovered. Cells do not act alone, and are often organized into localized multi-cellular units, notably the vast cortical cell layers, and the “niches” of neural stem cells. These regions are richly interconnected by axonal fibers, and studying brain connectivity is a science in its own right, and referred to as the “connectome”. Overall, despite major initiatives by groups around the world, much remains unknown about the brain’s structure, and a comprehensive mapping of an entire brain has not yet been achieved. What remains abundantly clear though, is the importance of high-performance computing to brain mapping.
The focus of our research is on the application of high-performance computing to quantify the cytoarchitectural alterations inflicted by pathophysiological conditions like traumatic brain injury, ischemic stroke, and experimental drug treatments that attempt to treat these conditions. It is vital to detect and quantify these alterations in a sensitive manner, and in a spatially comprehensive manner since alterations can be spread across brain regions that are distant from the primary injury site. The consequence of missing cellular alterations is high, since missed changes in critical brain regions may eventually manifest as confounding clinical conditions, and possibly dangerous side effects (e.g., violent/suicidal tendencies). In this regard, current histological imaging methods fail to capture the structural and molecular complexity of brain tissue. Importantly, visual examination of these images cannot provide objective and quantitative data, and simply cannot reveal subtle, but nevertheless important cell-state changes that do not affect the cell morphology.
To overcome the limitations of current methods, we are collaboratively developing a comprehensive quantitative open-source histology system named FARSIGHT (www.farsight-toolkit.org) by integrating multiplex immunolabeling, fluorescence imaging using robotic computer-controlled microscopes, and automated image interpretation technologies. Our collaborators at the Texas Medical Center in Houston (Drs. Pramod Dash and John Redell) perform the animal studies designed to test experimental drugs to treat brain disorders. Our collaborator Dr. Dragan Maric at the National Institutes of Health in Bethesda, Maryland is developing a comprehensive brain tissue imaging system capable of capturing a detailed molecular signature of each cell (10 – 30 markers) to capture cell type, functional states, and activity markers of interest. Using automated image stitching algorithms, this system can provide seamless coverage of extended brain regions comprising multiple whole-brain slices (i.e., 8-10 m thick sections of rat brain ranging from 5-40 mm long and 5-20 mm wide that are serially cut along the coronal or sagittal planes of interest) with sub-cellular resolution (300 nm/pixel). This system produces massive images (about 1 terabyte per brain slice), and the task of interpreting them is large. The FARSIGHT image analysis system performs automated segmentation and computational sensing of molecular and morphological alterations in each cell in the extended images, overcoming the limitations of visual analysis.
All of the computations are performed on a cluster of Dell servers making extensive use of C++, the Insight Toolkit Library (ITK), and OpenMP. Multivariate statistical tools allow us to compare normal and perturbed brains to detect, quantitate and create a comprehensive profile description of cytoarchitectural alterations. The results are visualized using interactive tools written using the open source Visualization Toolkit (VTK) running on desktop systems equipped with nVIDIA graphical accelerators.
Figure 1. Comprehensive imaging of a rat brain slice affected by ischemic stroke on the right hemisphere at 300nm/pixel resolution and 10 molecular markers reveals tissue alterations in unprecedented detail.
Figure 2. Computational parcellation of brain tissue into brain regions as defined by the Paxinos brain atlas.
Figure 3. Comprehensive color-coded map of cytoarchitectural alterations for each brain region.