New developments in electron microscopy for serial image acquisition of neuronal profiles

Abstract

Recent developments in electron microscopy largely automate the continuous acquisition of serial electron micrographs (EMGs), previously achieved by laborious manual serial ultrathin sectioning using an ultramicrotome and ultrastructural image capture process with transmission electron microscopy. The new systems cut thin sections and capture serial EMGs automatically, allowing for acquisition of large data sets in a reasonably short time. The new methods are focused ion beam/scanning electron microscopy, ultramicrotome/serial block-face scanning electron microscopy, automated tape-collection ultramicrotome/scanning electron microscopy and transmission electron microscope camera array. In this review, their positive and negative aspects are discussed.

3D reconstruction, SBEM, ATUM, FIB-SEM, neuron

Introduction

Three-dimensional (3D) reconstruction from serial electron micrographs (EMGs) has many advantages for neuronal profile analysis, synaptic density analysis, organelle distribution analysis and morphometric analysis of cross-sectional area, surface area, length, synapse junction area, etc. The 3D reconstruction method was introduced in neuroscience using real objects reconstructed in three dimensions, using 3.2-mm thick wooden sheets of a neuronal profile obtained from the cortical stellate cell of 80-nm thick ultrathin serial EMGs observed under transmission electron microscopy (TEM) [1]. Reconstruction of the stellate cell dendrites and soma allowed the reliable quantitation of the spatial synapse distribution on the cell. However, given the incredibly laborious process used, it could analyze only a small portion of the stellate cell. Nevertheless, this method provided new insights into cortical microcircuit analysis. In 1999, dendrites of three subtypes of hippocampal interneurons were manually reconstructed, and excitatory and inhibitory synapse distributions were compared among different dendritic segments [2]. Later, a computer-assisted 3D reconstruction system was developed to reduce the time and improve the quality of the reconstructed image [3]. The system includes custom-made software and hardware, extending the Visilog program (NOESIS). Using this system, we analyzed two striatal interneuron subtypes: fast spiking (FS) and low threshold spiking (LTS) cells. The striatal FS cell was found to have synaptic junction area, well correlated with the target dendritic circumference; however, the synaptic junction area of the LTS cell was always small and independent of the target dendrite size [4]. Fiala [5] developed the free downloadable reconstruction software ‘Reconstruct’, which could be run in a conventional Windows OS PC, http://synapses.clm.utexas.edu/tools/index.stm. ‘Reconstruct’ provided an easy access to the 3D reconstruction technique and allowed morphological measurement for fine neuronal profiles including cross-sectional area and circumference of the dendritic segment and synaptic junction area that cannot be measured using light microscopy (LM) because of the limitation of resolution [6–8], which is ∼0.5 µm and larger than the range of neural fine profiles such as spines, synapses and microtubules. Furthermore, synapse density can be analyzed easily. However, the 3D reconstruction process was still very laborious and demanded highly technical skills, such as thin sectioning with an ultramicrotome, collection of the serially cut sections, capturing and manual alignment of the serial EMGs, and segmentation of neuronal profile of interest [9]. Recently, new electron microscopy (EM) systems were developed to improve the time-consuming EMG capturing processes, such as focused ion beam/scanning electron microscope (FIB/SEM), diamond knife serial block-face scanning electron microscopy (SBEM), automated tape-collection ultramicrotome (ATUM) and transmission electron microscope camera array (TEMCA). These new EM systems are designed to capture serial EMGs automatically.

In this review, the techniques and concerns of the new EM systems for automated serial imaging are introduced briefly with current best knowledge. These techniques are developing, and new methods, machines or protocols are almost likely to be introduced in the near future.

FIB/SEM

FIB/SEM was developed to examine the fine structure of semiconductors and other items in material science research [10]. It was introduced in the field of neurobiology in 2008 to identify synaptic contacts [11]. In this method, a gallium ion beam digs into the block surface of the tissue and SEM captures the EMG of the freshly emerged block surface. The angle of the FIB and SEM beam is 54° (Carl Zeiss Microscopy GmbH, Oberkochen, Germany) or 52° (FEI, Eindhoven, The Netherlands) (Fig. 1a). This spatial double beam arrangement allows one to obtain the serial images from any part of the block surface conveniently. The electron microscopic image in these systems is reduced to 80% in one direction during imaging the slanting block surface (Fig. 1b), and it is then rescaled by calculation using trigonometric functions to the original image size (Fig. 1d). In contrast, the Hitachi FIB/SEM arranges the two beams at 90° (MI-4000L; Hitachi, Japan) (Figs. 1c and 2e), and the SEM can thus capture the block surface image without any tilt (Fig. 1d). However, this beam arrangement allows the image capturing only at the margin of the block surface, so the tissue block must be trimmed to place the region of interest at the margin of the block. The correlation of light microscopic and electron microscopic observation can be done easily with this system, because the light microscopic and electron microscopic imaging planes are identical (Figs. 1c and 2) [12]. On the other hand, the identification of small profiles such as dendrites and axons identified under LM is quite difficult in the FIB/SEM with 54° cross-beam SEM, because the imaging electron microscopic plane is perpendicular to the light microscopic imaging plane and the electron microscopic image does not coincide with the light microscopic image (Fig. 1a) [13].

Fig. 1.

FIB/SEM cross-beam angle configuration. (a) FIB mills the tissue block (blue) perpendicularly to the block surface (green). Freshly formed block surface (pink) is imaged by SEM at 54°/52° with FIB/SEM of 54°/52° cross-beam angle. The block surface imaging plane (pink) is perpendicular to the light microscopic imaging plane (green). Each beam line is shown with broken lines. (b) The original captured image by this FIB and SEM arrangement in (a) is shrunk to ∼80% in one direction. (c) FIB mills the tissue block perpendicularly to the light microscopic imaging plane of the block surface (green). Freshly formed block surface (pink) is imaged by SEM in the right direction. The light microscopic imaging plane (green) is parallel to the block surface (pink). Each beam line is shown with broken lines. (d) Captured image by FIB and SEM arrangement in (c) has no tortuosity. Images in (b) and (d) are makeshift EMGs with image modification for easy understanding.

Fig. 2.

Correlation study of LMG and EMG using FIB/SEM. eYFP labeled dendrite (red) with synaptic markers (yellow and green) in superficial layer of mouse in vivo visual cortex. (b) The light micrograph of DAB-stained dendritic segments shown in (a). (c) The light micrograph of trimmed block for serial EMGs imaging. Area in the yellow rectangle is similar to image area in (b). (d) The electron micrograph in lower magnification showing the same block in (c). Crossed lines and marker dots were carved on the block surface with FIB as landmarks to identify ROI. (e) Hitachi FIB/SEM MI-4000L. (f) The surface block SEM image showing the ROI area underlying the stained dendrite shown in (b). The dendritic segments were captured in seven square images shown in rectangles. (g) Dendritic segments (green) reconstructed from serial EMGs, shown in (a–c).

The best lateral resolution of SEM image is about a few nanometers. The finest FIB diameter can be up to a few nanometers, so the serial EMGs can be captured in a few nm z-step. The high image resolution including depth is a great advantage in acquisition of fine cellular structures, such as mitochondria, microtubules and synaptic contacts. Microtubules form the cytoskeleton of dendrites and axons of neurons and have a diameter of ∼25 nm. With serial EMGs of 5 nm pixel size in 10 nm z-step, the microtubules in dendritic segments were identified and traced beyond the next section easily [8]. We showed that the cross-sectional area and the number of microtubules were well correlated in dendrites. The high z-resolution allows a fine isometric voxel image, in which the image resolution in x, y and z is equivalent, and this allows for excellent analysis of synapse structure. The synaptic contact is defined by the occurrence of a cleft structure at the junction, presynaptic small vesicles and a postsynaptic density (PSD). Those indispensable structures for identification of the synaptic contact are in the range of 20–50 nm in size and require at least 5–10 nm pixel resolution for imaging. Therefore, images for the fine isometric voxel for analysis of these neuronal structures should be 5–10 nm or less. It is difficult to identify synaptic contacts if the cleft plane is situated parallel to the cutting plane in TEM [7]. In contrast, the synapse can be identified in any plane including the diagonal plane in the isometric/semi-isometric voxel image with 5 or 10 nm resolution or less obtained by the FIB/SEM (Fig. 3) [14]. The downsides of FIB/SEM are the slower cutting time, and smaller image size than the other new SEM systems. Single FIB mills the tissue block surface in a series of parallel lines, and this feature requires some seconds for the block surface milling, which is dependent on the FIB current and acceleration voltage as well as milling area size. It requires only a couple of seconds to mill one slice of small block surface area such as 10 µm by 10 µm, but larger area milling requires more time. Dwell time is usually ∼1–10 µs/pixel depending upon an SEM detector and the image quality. For instance, the surface cutting and image-capturing cycle for the serial images of 4k × 4k image size can be 4–5 min, and 300 serial images can be obtained in 24 h with FIB/SEM ‘Auriga’ or ‘Cross Beam’ (Carl Zeiss Microscopy GmbH). The image size can be larger by tiling the images, although it takes more time. Recently, we reconstructed two long pyramidal cell dendritic segments of mouse visual cortex with FIB/SEM (Hitachi MI-4000L, Tokyo, Japan). After in vivo imaging of eYFP expressed in pyramidal cell tuft dendrites in the cortical superficial layer through a cranial window using a 2-photon microscope, the mouse was perfused. Cortical sections of 50 µm thickness were cut tangentially to the cortical surface and the neuronal profiles were 3,3-diaminobenzidine tetrahydrochloride (DAB) stained with immunohistochemistry using antiserum against eGFP [6,15]. The dendritic segments were imaged serially using the ‘Multi Cut & See’ function of the Hitachi MI-4000L, which captures the image of the long dendrites efficiently using a mosaic, which comprised seven imaging frames (Fig. 2). Each frame was composed of 1183–1641 serial images of 2k by 2k image size, 5 nm pixel size and 12 nm z-step. The milled block surface was about 100 µm depth by 75 µm width from the side of the block. The milling time per slice was 26 s. The imaging time per slice was about 280 s (10 µs/pixel, 40 s/image) in the presence of the In-lens detector. Time for a round was about 5 min and 30 s and the total acquiring time was about 7 days.

Fig. 3.

Orthogonal views of a synaptic contact in a semi-isometric stack image using the FIB/SEM. (a) Synaptic contact with cleft plane parallel to the cutting imaging plane (arrow). (b and c) Orthogonal xz (b) and yz (c) views of the synapse in (a) (arrow) at the vertical and horizontal white line, respectively. It clearly shows PSD in the postsynaptic spine and small vesicle aggregation at the presynaptic membrane.

SBEM

Leighton introduced block surface imaging by SEM coupled with an in-chamber microtome to sequentially remove slices [16]. Denk and Horstmann developed this idea further and demonstrated its use for 3D reconstruction in neurobiology [17]. In this method, a diamond knife cuts the entire block surface and the fresh surface image is captured with the SEM (Fig. 4d and e). The positive aspects of SBEM are faster cutting and imaging times, and larger image size than FIB/SEM. The diamond knife cuts the entire block surface in a few seconds and the dwell time per pixel is about half or a few microseconds. Practically, the serial image acquisition of 4k square images can be less than half a minute and >3000 serial images can be captured in 24 h. The diamond knife cuts the entire block surface, so large images can be captured. Practically, the image size depends on the SEM performance. Therefore, SBEM can obtain fairly large image sizes, up to 32 k × 32k pixels. For instance, the 32k × 32k pixel image with a 5-nm pixel size has a 160-µm field-of-view. The drawback is that using <3 nm pixel size and longer dwell time is difficult because of electron charging by the scanning beam, resulting in the low image quality. For this reason, SBEM is better suited than FIB/SEM for acquisition of lower magnification and larger-sized serial images that require low electron dose imaging conditions and for analysis of larger structures such as mitochondria [18], retinal cells [19] and mouse whole brain reconstruction [20].

Fig. 4.

Correlation study of LMG and EMG with SBEM. (a) A 50-nm thick section of cortex processed with conventional OsO₄ staining. (b) A 50-µm thick cortical section processed with rOTO and lead staining. The section is completely dark, and light penetrated through only perpendicularly oriented blood vessels. (c) SBEM: 3View (Gatan, Pleasanton, USA)/Sigma (Carl Zeiss Microscopy GmbH). (d) Ultramicrotome in the SBEM chamber. (e) Tissue block showing fresh block surface (arrow). Silver paste covers the base. (f) DAB-stained pyramidal cell dendrite that was injected with lucifer yellow; c, capillary. (g) EMGs of the DAB-stained dendrite shown in (f) captured by the SBEM; c, capillary. (h) 3D reconstructed dendrite from serial EMGs. The dendritic segment in the rectangle is the correlated 3D image of the dendritic segment shown in (f) and (g).

The electron dose, which is defined by the equation below, should be controlled under 20e⁻/nm², for good quality image acquisition and to allow serial smooth sectioning for brain tissue embedded in Durcupan ACM Fluka (Sigma-Aldrich, St. Louis, USA) using 3View (Gatan, Pleasanton, USA)/Merline (Carl Zeiss Microscopy GmbH). The optimal dose may be varied in different imaging conditions for tissue preparation processes, embedding plastics, machine types and so on; however, the range of the optimal dose for SBEM should be around 20 e⁻/nm², which is smaller than that available with other new EM systems.

$$\begin{array}{c}\text{electron}\text{dose}({\text{e}}^{-}/\mathrm{n}{\text{m}}^2)=\text{beam}\text{current}(\mathrm{amperes})\times (6.24151\times \text{1}{\text{0}}^{-18}(\mathrm{coulombs}/\mathrm{electron}))\\ \times \text{pixel}\text{dwell time (s)/pixel siz}{\text{e}}^2(\mathrm{nm})\end{array}$$

The image quality gets better in image acquisition in higher dosing, because the SEM detector receives sufficient signals from the block surface. However, the high-dose imaging process induces extra polymerization of the plastic in the scanned portion of the block surface and this may change the surface shape. This surface shape change is a serious problem for cutting the section evenly [21]. We experienced serious troubles because of this shape change, which include uneven cutting, every other sectioning, wavy partial cutting, cracking or dim membrane (Fig. 5, Supplementary data online, video). An extraordinary overdose causes too much charge on the block surface, and it becomes very difficult to capture a good electron microscopic image for saturation. The charging issue may be reduced by in-chamber coating of palladium [22] or perhaps using a conductive plastic, which has not yet been developed. So far, we can get good serial imaging in 50 µm z-step sectioning with a dose of 20 e⁻/nm², for instance, a 50-pA beam current, 0.5-µs dwell time and 2.8 nm pixel size. The electric charging issue is not so critical in other new EM systems. The other downside is that LMG–EMG correlation is very difficult even in 50 µm thick tissue sections because of the coal-black color of ferrocyanide (reduced)-OsO₄-thiocarbohydrazide-OsO₄ (rOTO) [20,23] following the lead-staining protocol (Fig. 4b) [24]. DAB-stained neurons could not be seen in sections after the strong metal staining. This makes the correlation of LMG and EMG very difficult. However, we succeeded to capture EMGs of DAB, a stained dendrite that was identified by LM, and make the 3D reconstructed image (Fig. 4f–h), because we chose small blood vessels perpendicularly penetrating the tissue section (Fig. 4b) as landmarks to identify the rough location of the stained dendrites. Therefore, the LMG–EMG correlation is possible with careful preparation.

Fig. 5.

High-dose imaging with SBEM causes uneven cutting with cracking and dim membrane imaging. (a–f) Serial EM images of SBEM with 173 e⁻/nm² dose (1.8 kV, 100 pA beam current, 1 µs dwell time and 1.9 nm pixel size). Spines (arrowhead), mitochondria (asterisk) and other areas of the section are not evenly cut in the sequential block face images. Cracking (arrow) and dim membrane (thin arrow) are observed.

ATUM

ATUM is an ultramicrotome equipped with an automated tape-collecting system developed by Lichtman [25] and now available as ATUMtome (Fig. 6) (Boeckeler Instruments, Tucson, USA). The ultrathin sections are collected on carbon or indium tin oxide (ITO)-coated Kapton tape or a conductive tape, for instance, a thin copper tape (Fig. 6d–f). The tape with the serially cut ultrathin sections is cut into strips and glued onto a 4-inch size wafer whose surface is covered with double-sided, adhesive, conductive tape (Fig. 6d–f). The sections are observed using an SEM equipped with either ‘Array Tomography/ATLAS' (Carl Zeiss Microscopy GmbH) or ‘Wafer mapping’, which is the original software developed by the Lichtman laboratory. These systems capture EMGs of the same neuronal profile in the serial ultrathin sections one by one automatically after manual positioning of the sections. LMG and EMG correlation analysis can be easily done with the ATUM–SEM system. A positive aspect of this system is that the ultrathin sections are preserved on the tape and can be observed repeatedly. In contrast, the FIB/SEM and SBEM methods cut the sections away and re-observation of the sections is thus impossible. Using the ATUM system, the longitudinal distribution of myelin on pyramidal cell axons was well analyzed over a fairly long distance [26].

Fig. 6.

ATUMtome. (a) Ultramicrotome-equipped ATUM. (b) Side view of ATUM. Thin copper tape to collect the ultrathin sections is guided to diamond knife boat by two guide wheels (arrows). (c) Tape guide tip is put in the diamond knife boat directly to collect the ultrathin sections right after the sectioning. (d) The Kapton tape stripes with large ultrathin sections are glued onto a 4-inch diameter conductive wafer. (e) Ribbons of the serial tiny ultrathin sections on the thin copper tape. (f) Low magnification EMGs of the ultrathin section ribbon.

TEMCA

A custom high-throughput transmission electron microscope camera array (TEMCA) was built by Reid to achieve a 10-terabyte EM-imaged volume, with each image section represented by a 120 000 × 80 000 pixel composite image (4 nm/pixel), encompassing ∼450 × 350 × 50 μm including cortical layers 1, 2/3 and upper 4 by tiling about 60 by 45 images with 1153 serial sections (40–45 nm thickness) [27]. The large volume data set was simply obtained with conventional serial ultrathin sections on one whole grid for TEM observation. Four 2k × 2k array 4 MB CCD cameras were aligned to get 4k × 4k image size at the bottom of a 226-mm diameter scintillator where the EMG image was culminated through a custom-built 1.2-m vacuum chamber extension with a 120 kV TEM (JEOL 1200 EX, Tokyo, Japan). The obtained images were stitched and aligned using the home-made software. The analysis indicated that cortical inhibitory interneurons received synaptic inputs from nearby pyramidal cells with a broad range of preferred orientations. The big advantage of TEMCA is that it obtains images with the better quality of TEM compared with the other systems using SEM. The disadvantage is that it requires the laborious manual manipulation of the serial ultrathin sections using a conventional ultramicrotome, although automated serial image acquisition had been done with a custom-made TEM operation system. The imaged volume data set has been opened to the public (http://wholebraincatalog.org and http://ccdb.ucsd.edu/, accession number MP8448), and it has resulted in an interesting article, which used the results from the data set combined with other data sets obtained with the ATUM method [26].

Z-step calibration

Accurate thickness measurement of the ultrathin sections for 3D reconstruction from serial EMGs is a critical first step for the morphological measurement of reconstructed profiles, which affects the size in the z-axis [7]. The thickness of ATUM ultrathin section can be measured using a color laser confocal microscope that has sufficient z resolution, 0.5 nm, as conventional ultrathin sections for TEM observation [7]. The cutting step thickness of the block face can be measured using the measurement function of the SEM, which may be easier for the new SEM system users. The step made by block face partial cutting can be a good way to measure the z-step size for the new SEM systems.

Comparison of EMGs among the systems

EMGs of synaptic contacts from the same rat cortical tissue with the same histological conditions using the new EM systems are compared for image quality with a filmed TEM image of the tissue under similar conditions. In general, the image quality in high magnification is better in the filmed TEM image (Fig. 7a), good in the FIB/SEM image (Fig. 7c) and acceptable while using SBEM (Fig. 7e). Low magnification images look OK for all three methods (Fig. 7b, d and f). This illustrates the current technical standards, although it might be different under different image acquisition conditions such as a different detector, machine type, dwell time, current and voltage of the SEM beam and histology protocol (Fig. 7).

Fig. 7.

Image quality of EMGs from the various EM systems. (a) Synaptic contacts in rat cortex (arrow) filmed using conventional TEM serial ultrathin sections with 15k magnification. The film was scanned at 800 dpi and the final resolution was 2.1 nm per pixel. Only the single osmium process was applied in tissue preparation. (b) 45% reduced-size image including the synapse shown in (a) (asterisk). (c) A synaptic contact from cortex (arrow) imaged using FIB/SEM (Auriga) with 36k e⁻/nm² dose (1.5 kV, 1 nA beam current, 21 µs dwell time and 5 nm pixel size with EsB detector). Heavy metal staining with the rOTO/lead protocol was applied. Image size is enlarged to fit with the image pixel size of (a). (d) 45% reduced-size image including the synapse shown in (c) (asterisk). (e) A synaptic contact from cortex (arrow) imaged using SBEM (3View/Merlin) with 17.3 e⁻/nm² dose (2 kV, 50 pA beam current, 0.5 µs dwell time, 3 nm pixel size with InLens detector). Heavy metal staining with rOTO/lead was applied. Image size is enlarged to fit with the image pixel size of (a). (f) 45% reduced-size image including the synapse shown in (e) (asterisk).

In conclusion, all the new EM systems provide the ability to acquire serial EMGs and this results in large volume data sets for neural tissue. However, even with the new EM systems, it takes a few weeks, or even months, to acquire a large neural volume. A new SEM, MultiSEM 505 (Carl Zeiss Microscopy GmbH), with 61 SEM beams working in parallel to achieve 1220 megapixel/s has just been introduced. This could significantly accelerate the working speed.

Funding

This work was supported by Grant-in-Aid for Scientific Research (B)(25290012), Grant-in-Aid for Scientific Research on Innovative Areas “Neural creativity for communication (No. 4103)” (24120718) and “Adaptive circuit shift (No. 3603)” (26112006) from the MEXT of Japan; The Imaging Science Program of National Institutes of Natural Sciences (NINS).

Acknowledgements

We thank Drs Elly Nedivi, Katherine L. Villa and Kalen P. Berry in MIT, Drs Yoh Yamamoto and Xin Man in Hitachi, Sayuri Hatada and Alsayed A. Mohamed in NIPS for providing figures. We thank Drs Steven R. Vincent in UBC, Shawn Mikula and Mrs Sarah Mikula in MPI for their constructive comments.