The advent of transmission electron microscopy (TEM) in the 1950s represented a fundamental step in the study of the nervous system since it allows the visualization of axons, dendrites and glial processes, as well as synapses and subcellular organelles. Furthermore, it is an essential method to study synaptic circuits and the synaptic organization in the brain. The application of this technique soon led to the realization that the numbers and types of synapses vary depending on the brain region, as well as during the course of normal life and under certain pathological or experimental circumstances. Ever since the arrival of TEM, one of the main goals in neuroscience has been to define simple and accurate methods to study the synaptic organization and to estimate the magnitude of these changes. If three-dimensional data from brain tissue are required, reconstruction from serial sections is the gold standard technique. TEM is a well-established method for this purpose. However, serial section TEM is a time-consuming and technically demanding task, since ribbons of consecutive sections must be obtained with an ultramicrotome equipped with a diamond knife. One of the main limitations of conventional TEM is that long series of ultrathin sections are very difficult to obtain, often making it impossible to reconstruct large volumes of tissue. With the development of automated techniques of ultramicrotomy, this limitation has been overcome, and long series of sections can now be obtained.
FIB-SEM is one such automated technique. It combines a Focused Ion Beam (FIB) and a Scanning Electron Microscope (SEM). The FIB column directs a gallium (Ga+) ion beam towards the specimen. Low currents of Ga+ can be used to visualize the surface of the sample, since they detach secondary electrons and ions from it. However, the FIB is mainly used to mill the surface of the sample, since moderate and high currents of Ga+ can remove material from the sample surface. Since the FIB can be focused and controlled on a nanometer scale, the selected region of interest can be milled very precisely, removing a thin layer of material of a specified thickness. Once the region of interest has been milled, the SEM column is used to acquire a backscattered electron image from the surface. The milling/imaging cycle is then repeated automatically, so we can obtain long series of images without the intervention of the user. Since the images are separated by a known distance, they can be stacked with dedicated software to reconstruct a three-dimensional volume of tissue (Figure 1).
The preparation of samples for FIB-SEM is similar to the preparation for conventional TEM and, in general, any protocol that can be performed in TEM can be adapted for FIB-SEM. This technique has several advantages over conventional TEM, leaving aside the fact that serial sectioning is performed automatically. First, serial images are acquired from a surface that is progressively milled, without any mechanical interaction with the specimen, so the process is more reliable and stable than sectioning performed with a diamond knife. Artifacts such as wrinkles, holes and deformations, which are common in conventional ultramicrotomy, are avoided. Moreover, the milling thickness can be much smaller than the thickness of ultrathin sections. For example, a milling thickness of 20 nm, which would be impossible to obtain with a diamond knife, can be routinely achieved with FIB-SEM; in fact, even smaller thicknesses are possible and they can be adjusted as needed. Finally, although part of the sample is destroyed while milling, the window that you need to mill to acquire a stack of serial images is only a few tens of microns wide, so multiple stacks of serial sections can be obtained from the same or adjacent regions. In this way, the variability within a given region, or between regions, can be analyzed statistically
In our laboratory, we have applied FIB-SEM imaging to the study of the ultrastructure of the cerebral cortex in rodents, including the densities, distribution and sizes of synapses, the organization of mitochondria and the distribution of multivesicular bodies. In humans, it has been applied to the study of the synaptic organization in normal and pathological brain tissue. We have also used FIB-SEM in combination with different labeling techniques, such as genetic labeling, tract tracing and intracellular injections.
Once a long series of images has been acquired, we need to align the sections and stack them to reconstruct the volume of tissue in 3D. For image visualization, segmentation in 3D and analysis, we use EspINA software (http://cajalbbp.es/espina), which we specifically designed and developed for 3D electron microscopy. We usually begin with the identification and classification of synapses. The identification of synapses is carried out based on the presence of the pre- and postsynaptic densities, and on the accumulation of synaptic vesicles in the presynaptic terminal. The segmentation of synapses relies on the fact that the pre- and postsynaptic densities are very electron-dense. A grey-level threshold is chosen and the segmentation algorithm will select the pixels of the synaptic junction that are darker than the chosen threshold. The resulting segmentation is a 3D object that comprises the pre- and postsynaptic densities, since these are the two darkest regions of the synaptic junction. We next classify synapses on the basis of morphological criteria. Synapses with a prominent postsynaptic density are tagged as “asymmetric” and synapses with a thin postsynaptic density are tagged as “symmetric”. Once all synapses in the stack have been segmented, we calculate the density of synapses by using an unbiased three-dimensional counting frame. The size of synapses is measured as the area of the synaptic apposition surface (SAS), which is equivalent to the area of the postsynaptic density that faces the synaptic cleft.
The pre- and postsynaptic elements that participate in each synapse are identified next, according to simple criteria. For example, any nerve fiber that is presynaptic to at least one synapse is classified as an axon. Axons can be subdivided into excitatory and inhibitory, depending on the type of synapses that they establish (asymmetric or symmetric, respectively). Postsynaptic elements such as dendritic spines, dendritic shafts or cell bodies can also be identified and tagged. In this way, we can calculate, for example, the proportions of synapses that are established by excitatory axons on dendritic spines or dendritic shafts, and we can compare them with synapses established by inhibitory axons.
Furthermore, any other subcellular structure can be studied. For example, mitochondria and multivesicular bodies can be easily identified. Quantification of these structures may be performed after three-dimensional reconstruction, or with the help of stereological methods. Other structures may be difficult to identify and may require specific labeling. For example, thin glial processes can be easily mistaken for thin dendritic branches. In this case, the intracellular injection of astrocytes has proven useful for the unambiguous identification of the glial processes.
EspINA software provides quantitative information of any structure that has been segmented within the stack of sections. This information includes the number, position, and size of each segmented object, as well as the category to which it belongs (for example, asymmetric synapse, inhibitory axon, or dendritic spine). Different objects may be compared within the stack (for example, the number and size of asymmetric and symmetric synapses; or the volume of multivesicular bodies located in dendrites and axons). Finally, given that multiple stacks of images can be acquired from the same or different regions of the brain, the intra- or inter-regional variability can also be statistically assessed.
We have developed a method that makes it possible to obtain high-resolution images of brain regions in the millimeter to nanometer range (https://doi.org/10.3389/fnana.2018.00014. For this purpose, we use large semithin sections, which are commonly employed to examine large areas of tissue, with an optical microscope to locate and trim the regions that will later be studied with the electron microscope. Ideally, the observation of semithin sections would be from mesoscopic to nanoscopic scales directly, instead of using light microscopy and then electron microscopy. Therefore, we developed a method that makes it possible to obtain high-resolution scanning electron microscopy images of large areas of the brain in the millimeter to nanometer range (Figure 3).
Since these images are obtained from the same sections, in practice it is like zooming and panning at different levels of resolution through different brain regions. Fine detail is maintained throughout all the scales of magnification, so—for example—subcellular structures such as synapses, mitochondria and multivesicular bodies can be identified (Figure 4).
Since our method is also compatible with light microscopy, it is feasible to generate hybrid light and electron microscopic maps. There are hundreds of brain regions and subregions and many of them have not yet been explored at the ultrastructural level. The present method will allow relatively rapid inspection of multiple areas. Even the whole hemisphere of small brains like that of the mouse could be included in a semithin section prepared for SEM This is important not only to further examine the normal brain at the ultrastructural level —such as, for example, the border zones between different regions and subregions— but also to examine changes after different experimental or pathological conditions. Indeed, possible ultrastructural changes under different experimental conditions are usually evaluated in only one particular region like, for example, a certain layer of the CA1 field of the hippocampus. This is because the sections prepared for TEM have to be trimmed to a relatively small block, losing a large portion of the material. However, with this method, it is possible to first evaluate at different levels of magnification and then select specific regions or subregions to be further examined using TEM or FIB-SEM. In fact, the present method can be used to locate the exact regions that will later be studied by FIB-SEM, as in Videos 1–3 (white matter of the somatosensory cortex of the mouse imaged with FIB/SEM). Note the different orientations of the fascicles of myelinated fibers.
Video 1: http://cajalbbp.es/storage/WhiteMatterVideos/White matter 1.mp4
Video 2: http://cajalbbp.es/storage/WhiteMatterVideos/White matter 2.mp4
Video 3: http://cajalbbp.es/storage/WhiteMatterVideos/White matter 3.mp4
A link to all DeFelipe’s laboratory 3D electron microscopy studies focusing on the synaptic organization of the cerebral cortex (experimental animals and humans): https://cajalbbp.es/espina/#publications