Report

Micro-Optical Sectioning Tomography to Obtain a High-Resolution Atlas of the Mouse Brain

See allHide authors and affiliations

Science  03 Dec 2010:
Vol. 330, Issue 6009, pp. 1404-1408
DOI: 10.1126/science.1191776

Better Brain Maps

A high-resolution atlas of the complete neuronal connectivity in a whole brain should fundamentally advance our understanding of the organization and function of animal nervous systems. Now, A. Li. et al. (p. 1404, published online 4 November) describe an automated system, micro-optical sectioning tomography, that allowed three-dimensional mapping of the morphology and spatial location of neurons and traces of neurites in a whole, intact mouse brain.

Abstract

The neuroanatomical architecture is considered to be the basis for understanding brain function and dysfunction. However, existing imaging tools have limitations for brainwide mapping of neural circuits at a mesoscale level. We developed a micro-optical sectioning tomography (MOST) system that can provide micrometer-scale tomography of a centimeter-sized whole mouse brain. Using MOST, we obtained a three-dimensional structural data set of a Golgi-stained whole mouse brain at the neurite level. The morphology and spatial locations of neurons and traces of neurites could be clearly distinguished. We found that neighboring Purkinje cells stick to each other.

One of the most important aims in neuroscience is to obtain an interconnection diagram of the whole brain. In mammals, individual neurons are considered the basic units of the brain, but the complex functions of the brain depend more on the fine anatomical architecture of a very large number of neurons and their connections (1). Although modern neuroscience has made great progress in brain studies at both the system and cellular levels, our empirical knowledge of neuroanatomical connectivity remains inadequate, limiting the progress of brain studies (2). Thus, it is necessary to gain new insights into the morphology, localization, and interconnectivity of neural circuits throughout the whole brain at an appropriate resolution. Individual synapses are the finest functional element in circuits, but it is currently not technologically feasible to study the brainwide connectivity of complex vertebrate organisms (e.g., mice) at the synaptic level. In contrast, mesoscale techniques are currently more feasible and applicable for understanding specific neural functions (2).

Light microscopy has remained a key tool for neuroscientists to observe cellular properties at the mesoscopic level (3). A whole mouse brain is centimeter-sized, however, and is beyond the field of view of modern light microscopy techniques such as confocal and multiphoton microscopy (4, 5). Before microscopic imaging, histological sections are usually prepared for observation of the internal microstructure of large specimens (68), but it is difficult to perform serial ultrathin sectioning on large specimens by means of a traditional microtome. In the past decade, several new techniques have been developed to explore the neural circuits of the whole brain (914). A common characteristic of the techniques is the use of various kinds of automated sectioning or sectioning-like approaches to increase the detection depth of light microscopes. Some methods start by building section libraries automatically and then imaging these libraries with light or electron microscopy (12, 13). Another method is to perform imaging and sectioning simultaneously without collecting sections, which is a faster, more automated approach (14). Although these techniques have been successfully applied in specific fields, they have not provided a brainwide architecture atlas at the neurite level (3).

We developed an automatic micro-optical sectioning tomography (MOST) instrument to determine the neuronal connectivity of the whole mouse brain at the mesoscopic level. The MOST system is composed of a microtome, light microscope, and image recorder, and it performs imaging and sectioning simultaneously (Fig. 1A and fig. S1). While working, the microtome slices the specimen into ribbons about 450 μm in width. Once separated from the specimen block, the ribbons are imaged immediately. The light microscope is a reflected bright-field microscope in which the illumination beam is perpendicular to the rake face of the knife and coincides with the imaging beam. As the ribbons are being imaged in motion, a time-delay integration line-scan charge-coupled device (CCD) is used to improve the sensitivity of image recording. All of these design features, especially the optical design, can effectively reduce the system complexity to ensure automation stability and high resolution during time-consuming data acquisition without interruption.

Fig. 1

(A) Schematic representation of the MOST system. The specimen is mounted in a chamber, the motion of which is controlled by a series of mechanical translation stages that can move in three directions (the chamber and stages are not shown). Slicing is performed by moving the specimen along the x axis to generate ribbons, and each ribbon is simultaneously imaged. The illuminating beam passes through the beam splitter, mirror, and objective and irradiates the ribbon. After it passes through the mirror, beam splitter, and tube lens, the imaging beam collected by the objective is then recorded by a line-scan CCD. (B) Schematic representation of slicing. The slicing produces ribbons that glide forward along the knife face. The illuminating and imaging areas are indicated by a circle and a red line, respectively. To expand the detection range, we performed slicing with a lateral and an axial scan (15). (C) An image stack acquired using MOST. The stack is composed of many subimages. The subimages aligned along the x axis were produced from a ribbon. These image sequences reconstitute the entire cross section of the specimen along the y axis. A subimage of the cortex, hippocampus, and corpus callosum is also indicated.

We used MOST to image a whole brain of an adult Kunming (KM) mouse. The brain was from a 5-week-old male with a body weight of 27.57 g and a body length of 173 mm. During specimen preparation (15), staining was performed before embedding; we used a modified Golgi-Cox approach for staining and Spurr resin for embedding. The brain was kept intact throughout specimen preparation. The three anatomical axes of the brain—left-right, dorsal-ventral, and anterior-posterior—corresponded to the three movement directions of the specimen stage, X, Y, and Z. Throughout data acquisition, the section thickness was 1.0 μm, and a water-immersion objective (40×, numerical aperture 0.8) was used for imaging. Uninterrupted data acquisition lasted for about 242 hours, covering 15,380 coronal sections. The total amount of uncompressed volume data exceeded 8 terabytes, and the average acquisition rate was about 0.11 μs per voxel, or 0.001 mm3/s.

For optimal visualization of the highly complicated neuronal architecture, the raw mass data should be preprocessed to solve problems with periodic noise and nonuniform brightness. The preprocessed data can then be reconstructed and the volume rendered in three dimensions with the use of the program Amira 5.2.2 to show the whole anatomical sections, small regions of the brain, and individual neurons (15).

Figure 2 shows a series of coronal images reconstructed along the anterior-posterior axis of the entire mouse brain with a spacing of 1 mm. We could also use the original coronal images to reconstruct virtual sagittal images; Fig. 3A shows the reconstruction of such sagittal images. The selected location of the sagittal plane and the abbreviations used in Fig. 3A conform to those in Franklin and Paxinos’ atlas (16), and almost all major regions of the brain can be seen. Details of the cerebral cortex, hippocampus, and cerebellum in Fig. 3A are shown in Fig. 3, B to D. The original raw images and additional reconstructions of cross sections are shown in fig. S5 and figs. S7 to S10.

Fig. 2

(A) Locations of images shown in (B) to (L), spaced 1 mm apart, from the olfactory bulb to the cerebellum. (B to L) A series of images reconstructed from a stack of 100 coronal sections (total thickness of 0.1 mm). The horizontal stripes, caused by refractive index mismatch between specimen and ambient medium, lead to little information loss. The raw data from MOST are 8-bit grayscale images, and the resulting reconstruction uses a green pseudocolor to enhance their visible effects; dorsal-ventral and left-right axes are indicated at lower left. The dashed box in (H) is described in Fig. 4B.

Fig. 3

(A) A sagittal image reconstructed from a stack of 100 virtual sagittal sections (total thickness of 0.1 mm). These sections were transformed from the original coronal sections. The sagittal image was located in the right hemisphere about 0.4 mm lateral to the middle. Dorsal-ventral and anterior-posterior axes are indicated. Almost all major regions of the brain can be seen in this image: OB, olfactory bulb; Cx, cerebral cortex; Hc, hippocampus; f, fornix; ac, anterior commissure; T, thalamus; Cb, cerebellum; Mb, midbrain; P, pons; Md, medulla; cc, corpus callosum; SC, superior colliculus; IC, inferior colliculus; Ht, hypothalamus; Po, preoptic area; ox, optic chiasm; 4V, 4th ventricle; arabic numerals 2 to 10, nine lobules of the cerebellum; Sk, spinal cord. The three regions inside the dashed rectangles are the positions of (B), (C), and (D), from left to right. (B to D) The cerebral cortex (B), hippocampus (C), and cerebellum (D). In the reconstruction of sagittal images, no dislocation was observed along the dorsal-ventral axis; that is, the coronal sections are inherently aligned along the anterior-posterior axis. Arrow in (D) indicates the image described in Fig. 4D.

MOST also allows the 3D imaging of neurons and neuronal processes (Fig. 4). Among our sampled reconstructions of the neurons, we found that the neighboring Purkinje cells are close to each other (Fig. 4D and movie S4), which is different from the corresponding description in Gray’s Anatomy (17). We also used the program V3D (18) to trace the neurites in the ectorhinal cortex (Fig. 4B and movie S5).

Fig. 4

(A) A large volumetric reconstruction of a partial hippocampus of the mouse brain. The cube volume is 1.4 mm by 1.5 mm by 0.9 mm. The multilayered structure of the hippocampus and a large number of transverse fibers of the corpus callosum are clearly visible. Dorsal-ventral, anterior-posterior, and left-medial axes are indicated. (B) We traced the neurites of 17 neurons in the ectorhinal cortex, indicated by a dashed box in Fig. 2H. (C) Three-dimensional reconstruction of three neighboring pyramidal cells in layer 5 of the ectorhinal cortex as in (B). These three cells are distinguished by different colors; other cells, neurites, and blood vessels that densely cover the three cells are not shown. (D) Three-dimensional reconstruction of a pair of neighboring Purkinje cells in the ninth lobule of the cerebellum. The Purkinje cell in green can also be seen in Fig. 3D, indicated by an arrow.

One of the advantages of MOST is that no additional registration is needed because of the accurate spatial positioning of the images (15). Most mechanical slicing methods suffer from serious deformation, which can be overcome when using MOST. Figure 3 shows the smooth contours of the brain tissue surface, and Fig. 4A shows a fine inner structure of the mouse brain; both demonstrate that no registration is needed for the MOST data set. Even at the finer levels, the complex distribution of neurites is shown (e.g., fig. S7). The images are saved with a unified dimension, and their spatial position information is recorded during data acquisition. If we stack these subimages in three-dimensional space, a digital whole mouse brain with a voxel size of 0.33 μm by 0.33 μm by 1.0 μm can be reconstructed.

The Golgi method stains a limited number of cells at random in their entirety (15). It is likely that individual neurons or typical structures of the whole brain will be recognized in a complex background and that it will be possible to analyze architectural features with limited computing resources. MOST is capable of fluorescence imaging. Combined with new developments in specimen preparation techniques—for example, the multilabeled transgenic animal models (19)—the MOST system will help us to obtain a better connectivity map of the entire brain. Such studies will play an important role in functional studies of neural systems and in understanding and treating various kinds of nervous system diseases.

Supporting Online Material

www.sciencemag.org/cgi/content/full/science.1191776/DC1

Materials and Methods

SOM Text

Figs. S1 to S10

Movies S1 to S5

References

References and Notes

  1. See supporting material on Science Online.
  2. We thank P. Wu, J. B. Wu, and L. Wu for their early work in MOST; L. Fu, T. H. Xu, and W. Zhou for helpful discussions; Z. Dong and Z. Feng for assistance in experiments; J. Zhou for the nanowire sample; and Diatome AG, Switzerland, for the diamond knife. Supported by the National Natural Science Foundation of China (grants 30727002, 60025514, 30700214, and 30925013) and the Program for Changjiang Scholars and Innovative Research Team in University.
View Abstract

Navigate This Article