Extending Unbiased Stereology of Brain Ultrastructure to Three-dimensional Volumes

Design and Measurements: Volumes are reconstructed by defining transformations that align the entire area of adjacent sections. Whole-field alignment requires rotation, translation, skew, scaling, and second-order nonlinear deformations. Such transformations are implemented by a linear combination of bivariate polynomials. Computer software for generating transformations based on user input is described. Stereological techniques for assessing structural distributions in reconstructed volumes are the unbiased bricking, disector, unbiased ratio, and per-length counting techniques. A new general method, the fractional counter, is also described. This unbiased technique relies on the counting of fractions of objects contained in a test volume. A volume of brain tissue from stratum radiatum of hippocampal area CA1 is reconstructed and analyzed for synaptic density to demonstrate and compare the techniques. Results and Conclusion: Reconstruction makes practicable volume-oriented analysis of ultrastructure using such techniques as the unbiased bricking and fractional counter methods. These analysis methods are less sensitive to the section-to-section variations in counts and section thickness, factors that contribute to the inaccuracy of other stereological methods. In addition, volume reconstruction facilitates visualization and modeling of structures and analysis of three-dimensional relationships such as synaptic connectivity. ■ J Am Med Inform Assoc. 2001;8:1–16. Information technology has greatly aided visualization and analysis of the nervous system. Computerized scanning-based imaging modalities such as magnetic resonance imaging provide three-dimensional views of the brain down to about 1 mm of resolution. This level of resolution allows regional analyses of brain function but is not sufficient for investigation of intercellular communication at the level of individual neurons. Scanning light microscopy (e.g., confocal microscopy) provides three-dimensional views of individual neurons down to about 1 μm of resolution. Even this higher resolution is not sufficient for visualizing and investigating properties of the tiny connections, called synapses, that occur between neurons. Understanding brain function requires a detailed analysis of connectivity in the brain’s neuropil, that region of the gray matter where axons and dendrites from neurons form a dense plexus of synaptic connections. Electron microscopy (EM) is needed for this level of analysis. However, EM requires us to solve the problem of reconstructing a volume from a series of images (see, for example, Figure 10). Two methods of obtaining a three-dimensional representation of ultrastructure are EM tomography1,2 and ultrathin serial section EM.3,4 This paper describes a three-dimensional reconstruction system based on the latter approach, in which parallel sections are cut and imaged separately. The position of each section in the electron microscope introduces rotational and translational offsets. In addition, each section may be exposed to independent amounts of scaling and nonlinear deformation due to cutting, folding, specimen tilt, temperature changes, and optical distortions in the imaging system.5 Thus, the image of each section needs to be realigned with the images of adjacent sections. Early systems for section realignment were often special-purpose optomechanical devices for re-imaging the sections.4,6,7 In addition to being expensive, these systems corrected only translational and rotational offsets. As computers became more powerful, reconstruction systems began to utilize computer-based registration of images.8–10 These systems still did not fully correct for nonlinear deformations.3 In the next section of this paper (Volume Reconstruction), we describe an inexpensive system for serial section alignment that corrects for nonlinear deformations, can be applied easily to large images, and requires no special-purpose hardware. Stereology is the study of three-dimensional distributions of objects from sections. Looking at a sample of parts, stereologists aim to estimate the whole. Having a reconstructed volume of neuropil allows us to go beyond the conventional types of stereological analysis that are performed on ultrastructural distributions. The conventional techniques are almost exclusively based on single or paired sections.11–14 The one technique that utilizes volumes, the unbiased bricking method, was previously applied to volumes obtained from scanning light microscopes15 but not to large ultrastructural volumes, since such volumes were not easy to generate. In the Volume Analysis section we describe a number of techniques that are uniquely applicable to volumes of neuropil. We introduce a new, unbiased stereological technique, the fractional counter, and show how it can be applied to the reconstructed volumes as well. In the last section of this paper (Applications), we demonstrate volume reconstruction of neuropil from stratum radiatum of hippocampal area CA1. We compare the use of the unbiased bricking, disector, and fractional counter techniques in determining synapse density in this volume. In addition, we show how dendrite reconstructions within the sample volume can be used to build surface models for visualization and to obtain an unbiased estimate of the number of spines per unit length of dendrite. Volume Reconstruction The volume reconstruction system is embodied in a Windows (2000, 95, 98, and NT) application called serial EM (sEM) Align. The software was developed with the funding of the Human Brain Project and is available online at http://synapses.bu.edu./. Production and Imaging of Serial Electron Microscopy Sections Tissue is prepared for serial sectioning and electron microscopy either by intravascular perfusion of fixatives or by microwave-enhanced processing, as detailed elsewhere.16–18 Microwave-enhanced processing speeds up the penetration of fixatives during immersion and leads to improved tissue preservation for electron microscopy.19 After processing and embedding, series of about 100 sections are cut at a thickness of 40 to 60 nm on the basis of the observed color of sections. Later, the actual section thickness is estimated, as described below. Ribbons of sections are mounted on SynapTek pioloform-coated grids (Ted Pella, Inc., Redding, California) and loaded in a rotating stage to facilitate FIALA, HARRIS, Ultrastructure in Three Dimensions 2