Abstract

We report a 3D analysis of the neuronal circuits of human cerebral cortex. Neuronal circuits, which are essential for brain functions, are built up by neurons as a 3D network, so tracing the 3D neuronal network of human cerebral cortex is the first step to understanding the mechanism of human brain functions. The cortical microstructures were visualized by X-ray microtomographic imaging of adult frontal cortex tissue stained with metal impregnation. Skeletonized wire models were built by tracing the 3D distribution of X-ray absorption coefficients. The obtained neuronal models were composed of 240 pyramidal neurons and 131 interneurons. Capillary vessel structures along with blood cells in the capillary lumen were also visualized and traced to build capillary network models. Possible neuronal circuits were analytically resolved from the skeletonized wire models. The operating mechanism of the resolved circuits is discussed on the basis of neurotransmission in the circuits. The results also indicate that X-ray microtomography is a potential method of visualizing the neuronal circuits of the brain.

Introduction

Cerebral neuronal circuits are essential for brain functions. Somata and neuronal processes of cortical neurons build up circuits as 3D networks. Therefore, tracing the 3D neuronal networks of human cerebral cortex is the first step to understanding the mechanism of human brain functions.

The primary method for visualizing 3D structures of soft tissues, including cerebral tissue, is confocal light microscopy. Transgenic strategies for visualizing the neuronal network by genetically labeling neurons with multiple fluorescence dyes have been reported as the Brainbow technique (Livet et al. 2007), though transgenic methods cannot be applied to human samples. Automated serial sectioning along with light- and electron microscopic analyses has been proposed as a method for imaging neuronal circuits (Denk and Horstmann 2004; Micheva and Smith 2007; Knott et al. 2008). However, hundreds to thousands of sections should be mechanically prepared to reconstruct a 3D structure of a functionally relevant volume of biological tissue.

Computed tomography (CT) is a noninvasive technique for visualizing 3D structures. Tomographic slices are reconstructed from a series of projection images acquired by a rotational scan, which is a very straightforward process compared with serial sectioning. The application of X-ray microscopic approaches to CT analysis (Bonse and Busch 1996; Salomé et al. 1999; Uesugi et al. 1999) has led to the visualization of the 3D microstructures of biological samples (Bonse et al. 1994; Salomé et al. 1999; Shimizu et al. 2000). We have recently reported 3D microtomographic imaging of human cerebral cortex stained with high–atomic number (high-Z) probes (Mizutani, Takeuchi, Uesugi, Ohyama et al. 2008). The obtained structure indicated that the neuronal networks and capillary vessel architectures were visualized as the 3D distribution of X-ray absorption coefficients that describe electron densities. The capillary microstructures were used as a template to fabricate a resin model of capillary blood vessels (Mizutani, Takeuchi, Uesugi, Takekoshi et al. 2008).

Neuronal processes visualized in the microtomographic image can be traced to build neuronal network models. In the study reported here, human frontal cortex stained with the rapid Golgi impregnation was subjected to microtomographic visualization and network tracing. The obtained models were analyzed to resolve the neuronal circuits of the frontal cortex responsible for human brain functions.

Materials and Methods

Preparation of Tissue Sample

Human samples were obtained with informed consent, using protocols approved by the Clinical Research Review Board of Tokai University Hospital. Normal brain tissue (middle-aged male) from the middle part of the inferior frontal gyrus of the right hemisphere was dissected at autopsy and fixed with 10% formaldehyde for 7 days. Anatomical analysis found no abnormality in the brain tissue. The cortex tissue sample was further dissected into a 10-mm block and subjected to Golgi impregnation, as reported previously (Mizutani, Takeuchi, Uesugi, Ohyama et al. 2008) by the following procedure. The tissue was washed for 5 min in a solution containing 2.5% potassium dichromate and 4% formaldehyde and then incubated at 25 °C for 7 days in a solution containing 2.5% potassium dichromate. After being blotted with filter paper, the tissue was further incubated at 25 °C for 48 h in 0.75% silver nitrate. Residual silver nitrate was washed away with water. These potassium dichromate and silver nitrate treatment steps were repeated 3 times.

The stained sample was sequentially immersed in ethanol for 24 h and n-butylglycidyl ether for 16 h at room temperature. The dehydrated tissue was then incubated in Petropoxy 154 epoxy resin (Burnham Petrographics, Rathdrum, ID) at 4 °C for 24 h. The epoxy resin was degassed in vacuum before use. The tissue was again placed into a new resin solution and incubated for an additional 24 h at 4 °C. A prismatic sample with dimensions of 0.30 × 0.30 × 2.45 mm was prepared from the stained tissue block using razor blades and transferred into a borosilicate glass capillary (W. Müller Glas, Berlin, Germany) filled with epoxy resin. The outer diameter of the capillary was 0.7 mm. The sample capillary was then incubated at 90 °C for 16 h to cure the epoxy resin.

Microtomographic Analysis

The microtomographic analyses were performed at the BL20XU beamline (Suzuki et al. 2004) of the synchrotron radiation facility SPring-8. The sample capillary was mounted on the goniometer head of the microtomograph using a brass fitting specially designed for the glass capillary sample. Transmission radiographs were recorded with a charge-coupled-device-based X-ray imaging detector (AA50 and C4880-41S; Hamamatsu Photonics, Hamamatsu, Japan) using monochromatic radiation of 12.000 keV. The field of view and effective pixel size of the detector were 1.00 × 0.65 mm and 0.50 × 0.50 μm, respectively. A total of 1800 images were acquired with a rotation step of 0.10° and exposure time of 300 ms. It took 35 min to acquire each dataset. The data acquisition conditions are summarized in Table 1. Since the viewing field of this microtomograph is limited to 0.65 mm in height, datasets of the sample with an approximate height of 2.4 mm were taken in 4 batches. Each dataset was recorded by displacing the sample 0.600 mm along the longitudinal axis.

Table 1

Data acquisition conditions

Beamline BL20XU 
X-ray energy (keV) 12.000 
Pixel size (μm)a 0.50 × 0.50 
Detector size (pixels)a 2000 × 1312 
Detector viewing field (μm)a 1000 × 656 
Rotation/frame (degrees) 0.10 
Exposure/frame (ms) 300 
Frame/dataset 1800 
Dataset acquisition time (min) 35 
Beamline BL20XU 
X-ray energy (keV) 12.000 
Pixel size (μm)a 0.50 × 0.50 
Detector size (pixels)a 2000 × 1312 
Detector viewing field (μm)a 1000 × 656 
Rotation/frame (degrees) 0.10 
Exposure/frame (ms) 300 
Frame/dataset 1800 
Dataset acquisition time (min) 35 
a

Width × height.

The obtained radiographs for 4 datasets were subjected to reconstruction calculation using the program RecView (available from http://www.el.u-tokai.ac.jp/ryuta/) accelerated with CUDA parallel computing processors. Convolution–back-projection calculation of a 2000 × 2000 pixel tomogram from 1800 projection images took about 2.3 s using a PC equipped with a Quadro FX3700 board (nVIDIA, San Francisco, CA) as a CUDA computing environment.

To determine the precise positional relationship of the 4 reconstructed structures, we superposed each end of the 3D structures by minimizing the root-mean-square difference in voxel density. The whole image was built up by stacking these datasets. Finally, voxels corresponding to the glass wall and marginal resin were removed. These steps were performed by using the program suite SLICE (Nakano et al. 2008; available from http://www-bl20.spring8.or.jp/slice/). The data size of the obtained image stack composed of 4921 tomograms of 620 × 740 pixels reached 2.2 GB.

Model Building

The obtained tomograms were divided into 10 horizontal sections composed of 500 tomograms. Each section was sequentially subjected to model building by using the program MCTrace (available from http://www.el.u-tokai.ac.jp/ryuta/). Skeletonized wire models were built by placing and connecting nodes in the electron density maps. The model-building procedures were similar to those reported for crystallographic analyses of macromolecular structures (Jones and Kjeldgaard 1997), while neuronal processes were automatically traced by using a 3D Sobel filter (Al-Kofahi et al. 2002).

First, soma nodes were placed at the centers of dense bodies by inspecting coarse maps. Neuronal processes were traced from the soma node by tracking the density distribution down to the linear absorption coefficient of 7.0 cm1. This corresponds to 1.08 times the standard deviation of the linear absorption coefficients in Layer IV. The neuronal tracing was performed automatically using the Sobel filter, while the tracing process was monitored continuously via a graphical user interface to allow intervention in the process. Electron densities around the working models were also explored to place starting nodes to build further models. Finally, electron density maps of the entire section were examined to find empty densities to be modeled.

The cell types of the working models were determined from the morphology of somata and dendrites. Some large dendritic spines were observed as densities projecting from dendritic shafts. Such spine densities were manually modeled if they existed within 3 μm of the dendritic shaft and if they showed a solid shape greater than approximately 1 μm3. The obtained models were reexamined and modified to best fit the electron density maps, giving the final models. The model statistics are summarized in Table 2. Cartesian coordinate files of the models are available from http://www.el.u-tokai.ac.jp/ryuta/.

Table 2

Model statistics

Image size (μm)a 310 × 370 × 2460.5 
Number of observed image voxels 2 257 754 800 
Number of model nodes 173 265 
Total number of structural constituents 2356 
    Pyramidal neurons 240 
    Interneurons 131 
    Orphan processes 1933 
    Blood capillaries 30 
    Other cells 22 
Cumulative length of structural constituents (mm) 372.3 
    Pyramidal neuron (mm) 148.4 
    Interneuron (mm) 74.6 
    Orphan process (mm) 128.5 
    Blood capillary (mm) 12.5 
    Other cells (mm) 8.3 
Image size (μm)a 310 × 370 × 2460.5 
Number of observed image voxels 2 257 754 800 
Number of model nodes 173 265 
Total number of structural constituents 2356 
    Pyramidal neurons 240 
    Interneurons 131 
    Orphan processes 1933 
    Blood capillaries 30 
    Other cells 22 
Cumulative length of structural constituents (mm) 372.3 
    Pyramidal neuron (mm) 148.4 
    Interneuron (mm) 74.6 
    Orphan process (mm) 128.5 
    Blood capillary (mm) 12.5 
    Other cells (mm) 8.3 
a

Width × depth × longitudinal height.

Tissue Photomicrographs

Brain tissue adjacent to the sample for the microtomographic analysis was embedded in paraffin for sectioning. Thin sections 10 μm thick were stained using the Kluver–Barrera method. Photomicrographs of the obtained sections were taken using an Eclipse 80i microscope with a Plan Fluor ×10 objective lens and DXM1200F camera head (Nikon, Tokyo, Japan). Digital images of 1280 × 1024 pixels were recorded to cover the gray matter region of the section. The paraffin embedding caused tissue shrinkage as it has been reported previously (Abeles 1991; Quester and Schröder 1997). The shrinkage was corrected by simply magnifying the photomicrographs by taking account of the original tissue dimensions.

Results

Microtomographic Analysis

Electron density maps and a skeletonized wire model of pyramidal neuron 3013 in Layer III of the gray matter are shown in Figure 1A. A 3D rendering of electron densities around Layer IV is shown in Supplementary Movie 1. The voxel width of the acquired images was set to 0.5 μm, giving a spatial resolution of 1.0 μm, which corresponds to the X-ray optics resolution (Uesugi et al. 2001). Although this spatial resolution is typically coarser than light-microscopy resolutions, we put priority on achieving a viewing field that can accommodate the entire sample. The obtained images demonstrated that this spatial resolution was sufficient in practice to visualize the 3D structure of somata and neuronal processes of the human cerebral cortex. The 3D coordinates of the soma position were placed at the center of the cell soma. Neuronal processes were traced in the 3D electron density maps, giving the wire models of neurons. Capillary network models were built along the cylindrical axis of the lumen (Fig. 1B).

Figure 1.

Stereo drawings of electron density maps superposed on wire models. Maps are drawn in gray and the models in black. (A) Structure of pyramidal neuron 3013. Electron density maps in a 1.0-μm grid are contoured at 40.2 cm1. The closed circle indicates the soma coordinate. (B) Structure of blood capillary vessel 15. Electron density maps in a 0.5-μm grid are contoured at 26.9 cm1.

Figure 1.

Stereo drawings of electron density maps superposed on wire models. Maps are drawn in gray and the models in black. (A) Structure of pyramidal neuron 3013. Electron density maps in a 1.0-μm grid are contoured at 40.2 cm1. The closed circle indicates the soma coordinate. (B) Structure of blood capillary vessel 15. Electron density maps in a 0.5-μm grid are contoured at 26.9 cm1.

Neural cell models were classified on the basis of their morphology. Cells with apical dendrites, including fusiform neurons, were categorized as pyramidal neurons. Cells without apparent apical dendrites were categorized as interneurons. Neural cells with thin, numerous, and intricate processes observed adjacent to capillary vessels were categorized as possible glial cells. Besides neuronal processes emanating from somata, process densities that cannot be linked to any somata were built as orphan process models. Blood cells were visualized in the capillary vessels, as shown in Figure 1B.

3D Model of Cortex Structure

The 3D structure of the obtained models is shown in Figure 2 and Supplementary Movie 2. The models are composed of 240 pyramidal neurons, 131 interneurons, 9 glial cells, 13 uncategorized cells, 1933 orphan processes, and 30 capillary traces. The soma distribution determined from the microtomographic image corresponds to the structure observed in a photomicrograph of an adjacent thin section stained using the Kluver–Barrera method (Fig. 2). In Layer III, large pyramidal neurons were observed in both the photomicrograph and the microtomographic image. Models of these pyramidal neurons drawn in red can be found in the right panel of Figure 2. In Layer II, interneurons built as green wire models in the microtomographic image were also observed in the photomicrograph. Most of the neurons in Layer IV observed in the 2D photomicrograph were granular interneurons, while several pyramidal neurons in addition to the interneurons were identified in the 3D image. In Layers IV and V, intense dendrite networks of neurons in these layers were observed in the microtomographic image. The cortical layers were defined from these structures observed in the photomicrograph and in the microtomographic image. The Layer II/III border was located 445 μm deep from the pia (2000 μm for the z-coordinate of the model), the III/IV border at 1220 μm (z = 1225 μm), the IV/V border at 1595 μm (z = 850 μm), and the V/VI border at 2050 μm (z = 395 μm).

Figure 2.

Entire tissue structure. The brain surface is to the top. Cortical layers are indicated with labels. Scale bar: 50 μm. Left, photomicrograph of a section adjacent to the sample for microtomographic analysis. The 10-μm section was stained using the Kluver–Barrera method. Center, volume-rendered figure of 3D electron density distribution. Densities were rendered from linear absorption coefficients of 0–149.6 cm1 by the maximum projection method using the program VG Studio MAX (Volume Graphics, Heidelberg, Germany). Right, the entire model structure determined from the electron density maps. Structures of pyramidal neurons are drawn in red, interneurons in green, glial cells in blue, blood capillary vessels in purple, and orphan processes and uncategorized cells in gray. Closed circles indicate soma coordinates.

Figure 2.

Entire tissue structure. The brain surface is to the top. Cortical layers are indicated with labels. Scale bar: 50 μm. Left, photomicrograph of a section adjacent to the sample for microtomographic analysis. The 10-μm section was stained using the Kluver–Barrera method. Center, volume-rendered figure of 3D electron density distribution. Densities were rendered from linear absorption coefficients of 0–149.6 cm1 by the maximum projection method using the program VG Studio MAX (Volume Graphics, Heidelberg, Germany). Right, the entire model structure determined from the electron density maps. Structures of pyramidal neurons are drawn in red, interneurons in green, glial cells in blue, blood capillary vessels in purple, and orphan processes and uncategorized cells in gray. Closed circles indicate soma coordinates.

Since the neurons were stained with the Golgi impregnation, a small proportion of their total number was visualized. The total density of neurons in the gray matter of human cerebral cortex has recently been reported to be 1.95 × 104 neuron per mm3 (Azevedo et al. 2009). The obtained models consist of 371 neurons in a gray matter volume of 0.30 × 0.30 × 2.45 mm3, corresponding to 1.68 × 103 neuron per mm3. Therefore, 8.6% of neurons were stained and visualized by the microtomographic analysis along with the Golgi impregnation.

Most of the orphan processes that could not be linked to somata were localized in Layers IV and V. Dendrites of neurons in Layers IV and V were densely distributed in these layers. Therefore, the orphan processes in Layers IV and V should be dendritic arbors of Layer IV/V neurons, of which somata were not observed in the visualized volume.

Dendrite smoothness was examined for 33 interneurons, of which dendrites were traced for longer than 0.5 mm. Dendritic spines with large head volumes were observed in the electron density maps even at the spatial resolution of 1 μm. Electron density maps and a density plot of large spines built as short branches on dendritic shafts are shown in Figure 3. The large-spine density of 27 pyramidal neurons in Layers IV, V, and VI showed a broad distribution in the range from 0.018 to 0.077 per μm, while the large-spine density of interneurons exhibited a peak at 0 per μm in addition to a broad distribution. Thus, 19 interneurons with spine densities of less than 0.01 per μm were assigned as smooth neurons.

Figure 3.

Dendritic spines with large head volumes observed in electron density maps. (A) Stereo drawing of electron density maps of dendritic spines of interneuron 1019 superposed on the wire model. Maps are drawn in gray and the model in black. Maps with a 0.5-μm grid are contoured at 8.3 cm1. (B) Density of large dendritic spines. Spine densities of 33 interneurons are indicated with open bars and those of 27 pyramidal neurons with hatched bars.

Figure 3.

Dendritic spines with large head volumes observed in electron density maps. (A) Stereo drawing of electron density maps of dendritic spines of interneuron 1019 superposed on the wire model. Maps are drawn in gray and the model in black. Maps with a 0.5-μm grid are contoured at 8.3 cm1. (B) Density of large dendritic spines. Spine densities of 33 interneurons are indicated with open bars and those of 27 pyramidal neurons with hatched bars.

Microcircuit Analysis

In order to analyze circuits from the neuronal models, we should distinguish the axon of each neuron from dendrites. We found that neuronal processes with low densities could be discerned by plotting the electron density trajectory along the wire models (Fig. 4). Using this method, we examined neuronal processes that emanated directly from the soma of interneurons or that emanated from the soma base of pyramidal neurons; among these processes, we defined 22 of those having lower densities as axons. It is known that the myelinated part of the axon is not fully visualized by Golgi impregnation (Abeles 1991). Since the microtomographic analysis was performed using adult tissue in this study, it is probable that axons were observed as low-density processes and difficult to trace along a long trajectory.

Figure 4.

Density trajectory along neuronal processes of pyramidal neuron 4021. Linear absorption coefficients along 8 neuronal processes are, respectively, plotted against trajectory length from the center coordinate of the soma. The process trajectory indicated with closed triangles showed lower densities compared with those observed for other processes. The neuronal process corresponding to this trajectory was defined as an axon.

Figure 4.

Density trajectory along neuronal processes of pyramidal neuron 4021. Linear absorption coefficients along 8 neuronal processes are, respectively, plotted against trajectory length from the center coordinate of the soma. The process trajectory indicated with closed triangles showed lower densities compared with those observed for other processes. The neuronal process corresponding to this trajectory was defined as an axon.

The neuronal circuits were analytically resolved from the obtained neuronal models. Geometric analysis of the models indicated that all possible neuronal contacts were made by axons and dendrites. Axodendritic contacts are mostly mediated by dendritic spines. Quantitative analysis of spine morphologies of human pyramidal neurons (Benavides-Piccione et al. 2002) revealed that the average major axis of the heads of dendritic spines was 1.08 μm. It also indicated that most of the spine necks were shorter than 2 μm. From these estimations, we treated contacts between half-maximum surfaces in electron density maps of axons and dendritic shafts within 3 μm as possible synaptic connections. A pair of pyramidal neurons interacting through the possible synaptic connections is shown in Figure 5. Although smooth neurons form synapses at dendritic shafts, it is still possible that very few spines on the dendritic shafts are involved in synaptic connections. Therefore, this threshold was also applied to smooth neurons, and contact distances are stated in the circuit diagram. The obtained circuits are shown in Figure 6.

Figure 5.

Stereo drawing of a pair of pyramidal neurons 1020 (green) and 1023 (red). The brain surface is to the top. Closed circles indicate soma coordinates. Axons are indicated with arrows. Possible synaptic connections are indicated with black dots. Scale bar: 50 μm.

Figure 5.

Stereo drawing of a pair of pyramidal neurons 1020 (green) and 1023 (red). The brain surface is to the top. Closed circles indicate soma coordinates. Axons are indicated with arrows. Possible synaptic connections are indicated with black dots. Scale bar: 50 μm.

Figure 6.

Neuronal circuits analyzed from the models. Pyramidal neurons involved in the neuronal circuits are indicated with triangles with cell numbers. Smooth interneurons are indicated with shaded crosses. Spiny interneurons are indicated with open crosses. Interneurons of undetermined smoothness are indicated with broken-line crosses. Orphan processes having no linkage to cell soma are indicated with circles. A glial cell is indicated with a hexagon. Dendrites are shown with lines and axons with arrows. Possible synaptic connections are shown with closed diamonds with their contacting distances. Zero distances indicate close contacts without any interspace. Cortical layer borders are indicated with broken lines. Numbers on the left show the approximate z-coordinate in the wire models. (A) Separate circuits in Layers II–VI. (B) Interlayer circuits observed in Layers IV and V.

Figure 6.

Neuronal circuits analyzed from the models. Pyramidal neurons involved in the neuronal circuits are indicated with triangles with cell numbers. Smooth interneurons are indicated with shaded crosses. Spiny interneurons are indicated with open crosses. Interneurons of undetermined smoothness are indicated with broken-line crosses. Orphan processes having no linkage to cell soma are indicated with circles. A glial cell is indicated with a hexagon. Dendrites are shown with lines and axons with arrows. Possible synaptic connections are shown with closed diamonds with their contacting distances. Zero distances indicate close contacts without any interspace. Cortical layer borders are indicated with broken lines. Numbers on the left show the approximate z-coordinate in the wire models. (A) Separate circuits in Layers II–VI. (B) Interlayer circuits observed in Layers IV and V.

Discussion

Neuronal Circuits

Pyramidal neurons 19 and 23 (Fig. 6A) receive excitatory inputs from each other. This local circuit is analogous to the flip-flop circuit known as an electronic temporary memory. The same type of circuit is also observed for the local circuit formed by pyramidal neurons 1017, 1020, and 1023 (Fig. 6B). These neuronal circuits can operate as oscillators that store the input stimulus for a certain period. The stored stimulus should be sent to other neurons in the circuits for further processing of the input information. The analytical approach based on the 3D structure of the neuronal networks allows these discussions of the operating mechanism of the neuronal circuits of the human frontal cortex.

The major types of neuronal interaction proposed for the cat visual cortex (Douglas and Martin 1991) can be found in the resolved circuits. For example, smooth neuron 1010 projects its output to pyramidal neurons 1006 and 1023 and to spiny interneuron 2142 in Layer V. However, the neuronal interactions involving Layer II/III neurons were not observed in the resolved circuits. It has been reported that the Golgi-impregnated image should be regarded as underestimating the true figures of dendritic and axonal branches (Brown and Fyffe 1981; Abeles 1991). The visualization of neuronal processes of Layer II/III neurons was especially incomplete in the sample used in this study, resulting in a smaller contribution from the Layer II/III neurons. Although the staining procedures should be further investigated to visualize the complete neuronal circuits, the analytical approach can delineate not the average but discrete neuronal circuits.

Model building is the main rate-limiting step in analyzing the neuronal circuits of brain tissue. It took approximately 460 person-hours to build the entire models composed of 372.3 mm of structural constituents. The performance of model building was 1.2 h/mm of processes. This human working load is comparable to that reported for building a 3D model from Brainbow images (Helmstaedter et al. 2008). This suggests that our model-building procedures, which are similar to those used in macromolecular crystallographic studies, can be an alternative method of obtaining 3D models of neuronal structures. However, visualizing the complete neuronal network of the same tissue volume would take at least 670 working days (5350 h) since only 8.6% of neurons were visualized, as described above.

The sparseness of the Golgi-stained image has allowed light microscopic analyses of neural cells. However, this structural sparseness is not essential for microtomographic visualization. We have reported a microtomographic study of neural tissue in which most of the neurons were stained with aurate dyes (Mizutani et al. 2007). The complete neuronal circuits should further be revealed from such comprehensive images since the Golgi-stained neurons would not be representative of the entire neuronal networks.

Microtomographic Analysis

Hard X-rays with photon energy of typically greater than 10 keV can pass through soft tissues composed of light elements. The resultant transmission image obtained from hard X-rays gives little contrast. The contrast can be enhanced by using soft X-rays with lower energy. However, soft X-rays are absorbed even by water, so block tissues cannot be analyzed using soft X-rays. The application of phase contrast techniques for observing interferometric images has been reported (Snigirev et al. 1995; Momose et al. 1996; Wilkins et al. 1996). The phase contrast image of soft tissue gives the distribution of native electron densities, while the electron density itself has little relationship with biological functions or cellular organization. In clinical diagnosis, luminal structures of a living body are visualized by using X-ray contrast media. These contrast media contain high-Z elements that absorb hard X-rays efficiently. Therefore, X-ray visualization of microstructures of soft tissues should be performed by contrasting each biological constituent with high-Z element probes, which corresponds to fluorescent labels in light microscopy. The distributions of multiple high-Z probes can be individually visualized by using the X-ray absorption edge of each probe element (Mizutani, Takeuchi, Akamatsu et al., 2008).

Any soft tissues stained with high-Z element probes, including metals and metal conjugates, can be analyzed using X-ray imaging. We have recently achieved microtomographic visualization of Drosophila nerve tissue by using antibody–gold conjugates (Mizutani et al. 2009). Although it has been reported that the permeability of antibodies can be improved by heat treatment (Evers and Uylings 1997), immunocytochemical staining is rather difficult for block tissues. Therefore, efficient immunocytochemical method for staining block tissues should be developed to visualize the 3D distribution of immunological markers. Such a staining method could also facilitate light microscopic observation of 3D structures. Although transgenic strategies cannot be applied to human samples, specific expression of electron-rich proteins such as ferritin could be another method of visualizing biological structures by using X-ray imaging.

We have reported a radiographic visualization of neural tissue at 160-nm resolution (Mizutani et al. 2007). While light microscopic imaging beyond the conventional diffraction limit (∼200 nm) has been demonstrated (Betzig et al. 2006; Donnert et al. 2006), the recent application of zone-plate X-ray optics to the microtomographic analysis (Uesugi et al. 2006; Mokso and Cloetens 2007) achieved nanometer-scale resolution. Since the wavelength of hard X-rays is several thousand times shorter than that of the visible light used in light microscopy, the diffraction-limited resolution of X-ray microscopy is much higher than that of light microscopy. In addition, the transmissive and less refractile nature of X-rays with respect to biological tissue enables 3D radiographic analysis without any clearing procedure such as those developed for light microscopy (Alanentalo et al. 2007; Dodt et al. 2007). Therefore, X-ray microtomography is a potential method of visualizing the neuronal circuits of the brain, like X-ray crystallography in molecular biology.

Supplementary Material

Supplementary material can be found at: http://www.cercor.oxfordjournals.org/.

Funding

Subsidy for education and research provided by School of Engineering, Tokai University (FY2008).

We thank Noboru Kawabe (Teaching and Research Support Center, Tokai University School of Medicine) for sectioning tissue. The synchrotron radiation experiments were performed at SPring-8 with the approvals of the Japan Synchrotron Radiation Research Institute (proposal nos. 2006B1716, 2007A1844, 2007B1102, and 2009A1113). Conflict of Interest: None declared.

References

Abeles
M
Corticonics
 , 
1991
Cambridge
Cambridge University Press
Alanentalo
T
Asayesh
A
Morrison
H
Lorén
CE
Holmberg
D
Sharpe
J
Ahlgren
U
Tomographic molecular imaging and 3D quantification within adult mouse organs
Nat Methods
 , 
2007
, vol. 
4
 (pg. 
31
-
33
)
Al-Kofahi
KA
Lasek
S
Szarowski
DH
Pace
CJ
Nagy
G
Turner
JN
Roysam
B
Rapid automated three-dimensional tracing of neurons from confocal image stacks
IEEE Trans Inf Technol Biomed
 , 
2002
, vol. 
6
 (pg. 
171
-
187
)
Azevedo
FAC
Carvalho
LRB
Grinberg
LT
Farfel
JM
Ferretti
REL
Leite
REP
Jacob Filho
W
Lent
R
Herculano-Houzel
S
Equal numbers of neuronal and nonneuronal cells make the human brain an isometrically scaled-up primate brain
J Comp Neurol
 , 
2009
, vol. 
513
 (pg. 
532
-
541
)
Benavides-Piccione
R
Ballesteros-Yáñez
I
DeFelipe
J
Yuste
R
Cortical area and species differences in dendritic spine morphology
J Neurocytol
 , 
2002
, vol. 
31
 (pg. 
337
-
346
)
Betzig
E
Patterson
GH
Sougrat
R
Lindwasser
OW
Olenych
S
Bonifacino
JS
Davidson
MW
Lippincott-Schwartz
J
Hess
HF
Imaging intracellular fluorescent proteins at nanometer resolution
Science
 , 
2006
, vol. 
313
 (pg. 
1642
-
1645
)
Bonse
U
Busch
F
X-ray computed microtomography (microCT) using synchrotron radiation (SR)
Prog Biophys Mol Biol
 , 
1996
, vol. 
65
 (pg. 
133
-
169
)
Bonse
U
Busch
F
Günnewig
O
Beckmann
F
Pahl
R
Delling
G
Hahn
M
Graeff
W
3D computed X-ray tomography of human cancellous bone at 8 microns spatial and 10-4 energy resolution
Bone Miner
 , 
1994
, vol. 
25
 (pg. 
25
-
38
)
Brown
AG
Fyffe
RE
Direct observations on the contacts made between Ia afferent fibres and alpha-motoneurones in the cat's lumbosacral spinal cord
J Physiol
 , 
1981
, vol. 
313
 (pg. 
121
-
140
)
Denk
W
Horstmann
H
Serial block-face scanning electron microscopy to reconstruct three-dimensional tissue nanostructure
PLoS Biol
 , 
2004
, vol. 
2
 pg. 
e329
 
Dodt
HU
Leischner
U
Schierloh
A
Jährling
N
Mauch
CP
Deininger
K
Deussing
JM
Eder
M
Zieglgänsberger
W
Becker
K
Ultramicroscopy: three-dimensional visualization of neuronal networks in the whole mouse brain
Nat Methods
 , 
2007
, vol. 
4
 (pg. 
331
-
336
)
Donnert
G
Keller
J
Medda
R
Andrei
MA
Rizzoli
SO
Lührmann
R
Jahn
R
Eggeling
C
Hell
SW
Macromolecular-scale resolution in biological fluorescence microscopy
Proc Natl Acad Sci USA
 , 
2006
, vol. 
103
 (pg. 
11440
-
11445
)
Douglas
RJ
Martin
KA
A functional microcircuit for cat visual cortex
J Physiol
 , 
1991
, vol. 
440
 (pg. 
735
-
769
)
Evers
P
Uylings
HBM
An optimal antigen retrieval method suitable for different antibodies on human brain tissue stored for several years in formaldehyde fixative
J Neurosci Methods
 , 
1997
, vol. 
72
 (pg. 
197
-
207
)
Helmstaedter
M
Briggman
KL
Denk
W
3D structural imaging of the brain with photons and electrons
Curr Opin Neurobiol
 , 
2008
, vol. 
18
 (pg. 
633
-
641
)
Jones
TA
Kjeldgaard
M
Electron-density map interpretation
Methods Enzymol
 , 
1997
, vol. 
277
 (pg. 
173
-
208
)
Knott
G
Marchman
H
Wall
D
Lich
B
Serial section scanning electron microscopy of adult brain tissue using focused ion beam milling
J Neurosci
 , 
2008
, vol. 
28
 (pg. 
2959
-
2964
)
Livet
J
Weissman
TA
Kang
H
Draft
RW
Lu
J
Bennis
RA
Sanes
JR
Lichtman
JW
Transgenic strategies for combinatorial expression of fluorescent proteins in the nervous system
Nature
 , 
2007
, vol. 
450
 (pg. 
56
-
62
)
Micheva
KD
Smith
SJ
Array tomography: a new tool for imaging the molecular architecture and ultrastructure of neural circuits
Neuron
 , 
2007
, vol. 
55
 (pg. 
25
-
36
)
Mizutani
R
Takeuchi
A
Hara
T
Uesugi
K
Suzuki
Y
Computed tomography imaging of the neuronal structure of Drosophila brain
J Synchrotron Radiat
 , 
2007
, vol. 
14
 (pg. 
282
-
287
)
Mizutani
R
Takeuchi
A
Akamatsu
G
Uesugi
K
Suzuki
Y
Element-specific microtomographic imaging of Drosophila brain stained with high-Z probes
J Synchrotron Radiat
 , 
2008
, vol. 
15
 (pg. 
374
-
377
)
Mizutani
R
Takeuchi
A
Uesugi
K
Ohyama
M
Takekoshi
S
Osamura
RY
Suzuki
Y
Three-dimensional microtomographic imaging of human brain cortex
Brain Res
 , 
2008
, vol. 
1199
 (pg. 
53
-
61
)
Mizutani
R
Takeuchi
A
Uesugi
K
Takekoshi
S
Osamura
RY
Suzuki
Y
X-ray microtomographic imaging of three-dimensional structure of soft tissues
Tissue Eng Part C Methods
 , 
2008
, vol. 
14
 (pg. 
359
-
363
)
Mizutani
R
Takeuchi
A
Uesugi
K
Takekoshi
S
Osamura
RY
Suzuki
Y
Westland
TB
Calton
RN
Three-dimensional microstructural analysis of human brain tissue by using synchrotron radiation microtomographs
Handbook on white matter
 , 
2009
New York
Nova Science Publishers
(pg. 
247
-
277
)
Mokso
R
Cloetens
P
Nanoscale zoom tomography with hard x rays using Kirkpatrick-Baez optics
Appl Phys Lett
 , 
2007
, vol. 
90
 pg. 
144104
 
Momose
A
Takeda
T
Itai
Y
Hirano
K
Phase-contrast X-ray computed tomography for observing biological soft tissues
Nat Med
 , 
1996
, vol. 
2
 (pg. 
473
-
475
)
Nakano
T
Tsuchiyama
A
Uesugi
K
Uesugi
M
Shinohara
K
The 3D image processing tools SLICE
 , 
2008
Hyogo (Japan)
Japan Synchrotron Radiation Research Institute
Quester
R
Schröder
R
The shrinkage of the human brain stem during formalin fixation and embedding in paraffin
J Neurosci Methods
 , 
1997
, vol. 
75
 (pg. 
81
-
89
)
Salomé
M
Peyrin
F
Cloetens
P
Odet
C
Laval-Jeantet
AM
Baruchel
J
Spanne
P
A synchrotron radiation microtomography system for the analysis of trabecular bone samples
Med Phys
 , 
1999
, vol. 
26
 (pg. 
2194
-
2204
)
Shimizu
K
Ikezoe
J
Ikura
H
Ebara
H
Nagareda
T
Yagi
N
Umetani
K
Uesugi
K
Okada
K
Sugita
A
, et al.  . 
Synchrotron radiation microtomography of lung specimens
Proc SPIE
 , 
2000
, vol. 
3977
 (pg. 
196
-
204
)
Snigirev
A
Snigireva
I
Kohn
V
Kuznetsov
S
Schelokov
I
On the possibilities of x-ray phase contrast microimaging by coherent high-energy synchrotron radiation
Rev Sci Instrum
 , 
1995
, vol. 
66
 (pg. 
5486
-
5492
)
Suzuki
Y
Uesugi
K
Takimoto
N
Fukui
T
Aoyama
K
Takeuchi
A
Takano
H
Yagi
N
Mochizuki
T
Goto
S
, et al.  . 
Construction and commissioning of a 248 m-long beamline with x-ray undulator light source
AIP Conf Proc
 , 
2004
, vol. 
705
 (pg. 
344
-
347
)
Uesugi
K
Suzuki
Y
Yagi
N
Tsuchiyama
A
Nakano
T
Development of high spatial resolution X-ray CT system at BL47XU in SPring-8
Nucl Instrum Methods
 , 
2001
, vol. 
A467–A468
 (pg. 
853
-
856
)
Uesugi
K
Takeuchi
A
Suzuki
Y
Development of micro-tomography system with Fresnel zone plate optics at SPring-8
Proc SPIE
 , 
2006
, vol. 
6318
 pg. 
63181F
 
Uesugi
K
Tsuchiyama
A
Nakano
T
Suzuki
Y
Yagi
N
Umetani
K
Kohmura
Y
Development of microtomography imaging system for rock and mineral samples
Proc SPIE
 , 
1999
, vol. 
3772
 (pg. 
214
-
221
)
Wilkins
SW
Gureyev
TE
Gao
D
Pogany
A
Stevenson
AW
Phase-contrast imaging using polychromatic hard X-rays
Nature
 , 
1996
, vol. 
384
 (pg. 
335
-
338
)