This protocol outlines a routine method for using serial block-face scanning electron microscopy (SBF-SEM), a powerful 3D imaging technique. Successful application of SBF-SEM hinges on proper fixation and tissue staining techniques, as well as careful consideration of imaging settings. This protocol contains practical considerations for the entirety of this process.
Serial block-face scanning electron microscopy (SBF-SEM) allows for the collection of hundreds to thousands of serially-registered ultrastructural images, offering an unprecedented three-dimensional view of tissue microanatomy. While SBF-SEM has seen an exponential increase in use in recent years, technical aspects such as proper tissue preparation and imaging parameters are paramount for the success of this imaging modality. This imaging system benefits from the automated nature of the device, allowing one to leave the microscope unattended during the imaging process, with the automated collection of hundreds of images possible in a single day. However, without appropriate tissue preparation cellular ultrastructure can be altered in such a way that incorrect or misleading conclusions might be drawn. Additionally, images are generated by scanning the block-face of a resin-embedded biological sample and this often presents challenges and considerations that must be addressed. The accumulation of electrons within the block during imaging, known as “tissue charging,” can lead to a loss of contrast and an inability to appreciate cellular structure. Moreover, while increasing electron beam intensity/voltage or decreasing beam-scanning speed can increase image resolution, this can also have the unfortunate side effect of damaging the resin block and distorting subsequent images in the imaging series. Here we present a routine protocol for the preparation of biological tissue samples that preserves cellular ultrastructure and diminishes tissue charging. We also provide imaging considerations for the rapid acquisition of high-quality serial-images with minimal damage to the tissue block.
Serial block face scanning electron microscopy (SBF-SEM) was first described by Leighton in 1981 where he fashioned a scanning electron microscope augmented with an in-built microtome which could cut and image thin sections of tissue embedded in resin. Unfortunately, technical limitations restricted its use to conductive samples, as non-conductive samples such as biological tissue accumulated unacceptable levels of charging (electron buildup within the tissue sample)1. While coating the block-face between cuts with evaporated carbon reduced tissue charging, this greatly increased imaging acquisition time and image storage remained a problem as computer technology at the time was insufficient to manage the large file sizes created by the device. This methodology was revisited by Denk and Horstmann in 2004 using a SBF-SEM equipped with a variable pressure chamber2. This allowed for the introduction of water vapor to the imaging chamber which reduces charging within the sample, making imaging of non-conductive samples viable albeit with a loss of image resolution. Further improvements in tissue preparation and imaging methods now allow for imaging using high vacuum, and SBF-SEM imaging no longer relies on water vapor to dissipate charging3,4,5,6,7,8,9. While SBF-SEM has seen an exponential increase in use in recent years, technical aspects such as proper tissue preparation and imaging parameters are paramount for the success of this imaging modality.
SBF-SEM allows for the automated collection of thousands of serially-registered electron microscopy images, with planar resolution as small as 3-5 nm10,11. Tissue, impregnated with heavy metals and embedded in resin, is placed within a scanning electron microscope (SEM) containing an ultramicrotome fitted with a diamond knife. A flat surface is cut with the diamond knife, the knife is retracted, and the surface of the block is scanned in a raster pattern with an electron beam to create an image of tissue ultrastructure. The block is then raised a specified amount (e.g., 100 nm) in the z-axis, known as a "z-step," and a new surface is cut before the process is repeated. In this way a 3-dimensional (3D) block of images is produced as the tissue is cut away. This imaging system further benefits from the automated nature of the device, allowing one to leave the microscope unattended during the imaging process, with the automated collection of hundreds of images possible in a single day.
While SBF-SEM imaging primarily uses backscattered electrons to form an image of the block-face, secondary electrons are generated during the imaging process12. Secondary electrons can accumulate, alongside backscattered and primary-beam electrons that do not escape the block, and produce "tissue charging," which can lead to a localized electrostatic field at the block-face. This electron accumulation can distort the image or cause electrons to be ejected from the block and contribute to the signal collected by the backscatter detector, decreasing the signal-to-noise ratio13. While the level of tissue charging can be decreased by reducing the electron beam voltage or intensity, or reducing beam dwell time, this results in a diminished signal-to-noise ratio14. When an electron beam of lower voltage or intensity is used, or the beam is only allowed to dwell within each pixel space for a shorter period of time, less backscattered electrons are ejected from the tissue and captured by the electron detector resulting in a weaker signal. Denk and Horstmann dealt with this problem by introducing water vapor into the chamber, thereby reducing charge in the chamber and on the block face at the cost of image resolution. With a chamber pressure of 10-100 Pa, a portion of the electron beam is scattered contributing to image noise and a loss of resolution, however this also produces ions in the specimen chamber which neutralizes charge within the sample block2. More recent methods for neutralizing charge within the sample block use focal gas injection of nitrogen over the block-face during imaging, or introducing negative voltage to the SBF-SEM stage to decrease probe-beam-lading energy and increase signal collected6,7,15. Rather than introducing stage bias, chamber pressure or localized nitrogen injection to decrease charge buildup on the block surface, it is also possible to increase the conductivity of the resin by introducing carbon to the resin mix allowing for more aggressive imaging settings16. The following general protocol is an adaptation of the Deerinck et al. protocol published in 2010 and covers modifications to tissue preparation and imaging methodologies we found useful for minimizing tissue charging while maintaining high resolution image acquisition3,17,18,19. While the previously mentioned protocol focused on tissue processing and heavy metal impregnation, this protocol provides insight into the imaging, data analysis, and reconstruction workflow inherent to SBF-SEM studies. In our laboratory, this protocol has been successfully and reproducibly applied to a wide variety of tissues including cornea and anterior segment structures, eyelid, lacrimal and harderian gland, retina and optic nerve, heart, lung and airway, kidney, liver, cremaster muscle, and cerebral cortex/medulla, and in a variety of species including mouse, rat, rabbit, guinea pig, fish, monolayer and stratified cell cultures, pig, non-human primate, as well as human20,21,22,23. While small changes may be worthwhile for specific tissues and applications, this general protocol has proven highly reproducible and useful in the context of our core imaging facility.
All animals were handled according to the guidelines described in the Association for Research in Vision and Ophthalmology Statement for the Use of Animals in Vision and Ophthalmic Research and the University of Houston College of Optometry animal handling guidelines. All animal procedures were approved by the institutions in which they were handled: Mouse, rat, rabbit, guinea pig, and non-human primate procedures were approved by the University of Houston Animal Care and Use Committee, zebrafish procedures were approved by the DePauw University Animal Care and Use Committee, and pig procedures were approved by the Baylor College of Medicine Animal Care and Use Committee. All human tissue was handled in accordance with the Declaration of Helsinki regarding research on human tissue and appropriate institutional review board approval was obtained.
1. Tissue Processing
2. Block Preparation
NOTE: The method will depend on how the sample is oriented in the block and how the sectioning is to take place. However, the most common tissue orientation finds the tissue centered in the tip of the resin block, perpendicular to the long end of the resin block.
3. SEM Settings for Imaging the Block Face
NOTE: The imaging settings that follow were produced on the device used by the authors, which is listed in the Table of Materials provided. While this device is capable of variable pressure imaging, best results were captured under high vacuum.
Mouse Cornea
This protocol has been applied extensively to the mouse cornea. Using SBF-SEM imaging a network of elastin-free microfibril bundles (EFMBs) were shown to be present within the adult mouse cornea. It was previously believed that this network was only present during embryonic and early postnatal development. SBF-SEM revealed an extensive EFMB network throughout the cornea, with individual fibers found to be 100-200 nm in diameter when measured in cross-section. It was also found that this EFMB network was organized in distinct layers, with fibers closely associated with keratocytes, even lying within shallow invaginations on the keratocyte surface (Figure 5). The discovery of EFMB fibers in the adult cornea led to immunogold-labeling transmission electron microscopy (TEM), fluorescence and confocal studies to further understand the nature of this network23.
Further application of this protocol led to the discovery of a previously unknown population of central corneal nerves that fuse with basal epithelial cells at the stromal-epithelial border (Figure 6). Previously, it was believed that all nerves interacting with the epithelium at this border penetrated into the corneal epithelium and ramified producing the subbasal and epithelial plexi. In this study, ~45% of central nerves interacting with the basal epithelium underwent cell-cell fusion rather than simple penetration. Using stereological methods applied to SBF-SEM data sets, it was possible to show this novel nerve population had a surface-to-volume ratio roughly half that of penetrating nerves, consistent with their "swollen" appearance (Nerve Fusion – 3.32±0.25, Nerve Penetration – 1.39±0.14, p ≤ 0.05). 3D reconstructions of penetrating and fusing nerve bundles and their mitochondria were created, highlighting the lack of mitochondria in fused portions of the nerve bundles. The discovery of neuronal-epithelial cell fusion using SBF-SEM led to fluorescence studies verifying membrane continuity between the two fused cells21.
The central cornea is an avascular tissue, and as such the peripheral limbal vasculature is of particular importance to the overall health of the cornea. The cell-cell relationships and ultrastructure of this region is complex; however, the ability to appreciate these cell-cell interactions and ultrastructural connections has been limited in fluorescence and single section TEM studies. For this reason an SBF-SEM image stack containing limbal vasculature, nerve bundles, and associated cells was manually segmented for 3D reconstruction (Figure 7). In this image the close association between vascular endothelial junctions and an overlaying pericyte, the individual granules of a perivascular mast cell, the nucleus and leading edge of a neutrophil crawling along the outer surface of the blood vessel wall, as well as a passing nerve bundle can be seen.
Taken together, this body of work demonstrates the capability of this protocol to produce high quality 3D electron microscopy data sets in tissues rich in extracellular matrix and epithelium, as well as vasculature and associated cells.
Higher Order Primate Retina – Nerve Plexus and Vascular Network
The retinal nerve fiber layer (RNFL) of higher order primates contains and depends on an extensive vascular network. Often, diseases of the retina involve changes in both parameters of the retinal nerve fiber layer as well as the vasculature found within it. Understanding the relationship between the RNFL and its vascular network in healthy, non-pathological tissue is the first step to understanding any changes that may occur as a result of disease. In order to better understand this relationship, the SBF-SEM protocol was applied to normal higher order primate retina and the reconstruction of the vascular network was performed and volumetric data extracted from this reconstruction (Figure 8). This 4,642,307 µm3 region of the RNFL contained a vascular bed 1.207×10-4 µL in volume, making up 2.6% of the total volume of the RNFL. This work demonstrates the capability of this protocol to produce high quality 3D electron microscopy data sets in dense neurological tissue.
Zebrafish and Giant Danio Heart – Striated Muscle and Developing Vasculature
Both the zebrafish and the giant danio are important models for heart development and regeneration. Historically, the zebrafish heart is considered to consist of two anatomically distinct myocardial segments functioning together in support of the physiological demands of the zebrafish. However, the interface between these two ventricular layers was not well understood. Using this protocol, a previously unrecognized junctional region was discovered consisting of a thin sheet of fibroblasts. It was found that openings within this sheet allowed two separate myocardial segments to come into contact and form complex adhesions junctions including desmosomes and fascia adherens22.
This protocol has been utilized in further work examining the vascular network of the developing giant danio heart (Figure 9). This method allows for the 3D appreciation of the developing cardiac myocyte network and its relationship with developing microvasculature. Taken together, this work demonstrates the capability of this protocol to produce high quality 3D electron microscopy data sets in muscle and highly vascularized tissues.
Image Settings, Charging, and Resolution
While appropriate fixation and heavy metal staining is necessary for quality SBF-SEM imaging, equally important is the use of conductive resin and proper imaging settings for the questions being addressed. In this protocol, the use of carbon black is employed in order to increase the conductivity of the sample block and provide a conduit to the mounting pin for the clearance of secondary electrons from the block-face. This has proven effective in combating tissue charging which often degrades image quality in tissues not prepared with carbon black16. In addition, the silver paint and gold sputtering applied to the block provides a dissipation pathway for electron buildup. Some devices allow for the addition of a focal charge compensator, which reduces charging by applying a puff of nitrogen over the block-face, however we have had similar success with the use of carbon black and the application of silver paint and gold sputtering to the block15. Lack of sample conductivity can lead to electron buildup visible as tissue charging (Figure 1), as well as discharges that are visible as abrupt image shifting and warping which dramatically diminish image quality (Figure 10B & F). The use of carbon black allows for imaging under high-vacuum and the use of image settings that result in high signal-to-noise ratio and improved image resolution. One such setting that leads to improved image quality is pixel dwell time. The SBF-SEM imaging process involves the raster scanning of an electron beam across the sample surface to generate backscattered electrons which the microscope detector can collect and interpret as signal. The length of time this beam is allowed to dwell within the space of each pixel leads to a more accurate pixel value being assigned to each pixel location (Figure 3A & B)2. There is a balance that must be struck between increased signal-to-noise, resolution and damage dealt to the block-face however. The beam effectively irradiates the block-face with high energy electrons which can break down and soften resin resulting in image degradation and cutting complications (Figure 10)28. The thinner the z-resolution required, the more difficult it becomes to maintain high-resolution imaging. We generally use z-steps of 100-200 nm, however z-step sizes of 25-50 nm have been reported5,29,30,31. With z-steps of this size, the break-down and softening of resin due to beam damage can lead to either compression of the resin causing the knife to miss a cut or cut the block-face but with "chatter" where the knife skips across the surface of the block creating ripples and bands13. While small z-steps are an attractive prospect, it is important to keep the specific research question in mind when choosing an appropriate z-step. Over-sampling can lead to substantial data-storage considerations as well as an increase in time required to produce 3D reconstructions.
Tissue Fixation and Staining
Prior to heavy metal incubation, tissues must be fixed in glutaraldehyde. While we highly recommend microwave fixation under vacuum for the preservation of tissue ultrastructure27, if a laboratory grade microwave is not available a commercial inverter microwave with variable wattage can be substituted32,33,34,35. If this is done, extra care should be employed to ensure that tissue distortion does not occur. Improper tissue fixation can result in altered tissue morphology as can be seen in Figure 10E. This protocol, like most modern SBF-SEM staining protocols, has been adapted from the staining procedure outlined by Deerinck in 201017, based on the osmium-thiocarbohydrazide-osmium stains created by Willingham and Rutherford in 198436. The heavy metals utilized in this protocol add contrast to the cellular structures within a tissue sample (Figure 1). The initial osmium incubation occurs with reduced osmium which binds to C=C bonds in unsaturated fats leading to membrane and lipid staining37,38. Osmium is reduced by potassium ferrocyanide, which assists in the staining of saturated lipids and also acts to stabilize phospholipids39,40. Thiocarbohydrazide is subsequently added as a mordent that binds to the osmium from the first incubation, acting as a bridge on which further osmium is bound at a later stage in the protocol41. Uranyl acetate, which is a uranium salt, is an effective contrasting agent for lipids, nucleic acids and proteins, while lead citrate enhances contrast of proteins and glycogens. The varying affinities of these agents for cellular components further enhances the overall contrast within tissues over and above the osmium incubations42.
Imaging the Block-Face
Figure 11, Figure 12, Figure 13 illustrate the combined effects of voltage, pixel dwell time and beam intensity. Conventional practice suggests a combination of low voltage, short dwell time and low beam intensity are necessary for optimal imaging and preventing beam damage to the sample block. Contrary to these settings, Figure 11, Figure 12, Figure 13 illustrate that higher voltages (e.g., 7 kV), longer dwell times (32 µs/px) and higher beam intensities (setting 6 in our case) can produce superior image quality over conventional settings.
SBF-SEM allows for the collection of serial electron microscopy images which can be collected as a 3D data set comprised of voxels. While this is an incredibly powerful use of SBF-SEM, this method also allows for the rapid and repeatable imaging of rare biological events or cells. Image acquisition using SBF-SEM can be monitored for rare events, and imaging paused in order to collect higher magnification/resolution images of these events. Furthermore, the block can be removed from the microscope chamber and the block-face sectioned for transmission electron microscopy (TEM) imaging. In this way large datasets of rare events can be collected using SBF-SEM as well as appreciated at the angstrom scale using TEM.
Figure 1: SBF-SEM and TEM comparisons at various steps in the protocol. This protocol contains multiple steps in which sample tissue is stained with heavy metals. This affects not only tissue contrast and appreciation of cellular structures and organelles, but also the levels of charging that occurs when the tissue is imaged. This figure contains three distinct views of prepared tissue: a low magnification view (A, D & G), a high magnification view (B, E & H), and a TEM comparison of prepared mouse cornea (C, F & I). It can be noted that higher magnification images can result in increased tissue charging, as the electron beam is concentrated in a smaller region of tissue. The top row (A-C) is a representative sample from tissue processed through the completion of step 1.8, and has been impregnated with potassium ferrocyanide, osmium tetroxide, and thiocarbohydrazide. The arrows in the first two columns show the epithelial-stromal interface as a reference point. Note the low level of contrast in comparison to the bottom two rows, as well as the increased levels of tissue charging. The sample in the middle row (D-F) was processed through the completion of step 1.10 and benefits from an additional osmium tetroxide step, and is visibly more contrasted than the sample in the top row. While cellular structures are discernible, charging is still present. The sample in the bottom row (G-I) benefits from the full staining protocol and has minimal tissue charging. TEM imaging reveals tissue contrast levels imparted by the heavy metals present at each step (right column): organelles in the corneal endothelium (*) are more contrasted and apparent as tissue processing continues through the protocol. Additionally, stromal collagen and fibrillin details become more visible (arrowhead) as the protocol is completed. Panel A, D & G scale bar = 50 µm. Panel B, E & H scale bar = 10 µm. Panel C, F & I scale bar = 1 µm. Please click here to view a larger version of this figure.
Figure 2: Schematic of embedded tissue block, specimen pin, and final preparation. (A) Tissue should be placed in a known orientation at the very tip of the resin mold and the upper third of the mold filled with carbon black saturated resin. The region of the mold furthest from the tissue should remain clear so that the experiment label can be clearly seen. (B) Specimen pin surface should be scratched to produce a grid pattern, this allows for a greater area of contact for the cyanoacrylate glue to harden between the prepared specimen block and pin. (C) The carbon black saturated resin should make a wide area of contact with the specimen pin head, however the region that is cut by the diamond knife should be no greater than 1×1 mm. It is good practice to taper the block towards the tip. This minimizes cutting forces on the diamond knife and by having a wider base, the block is more resistant to separating from the pin during sectioning. Please click here to view a larger version of this figure.
Figure 3: Comparison of image capture settings. (A & B) Panels A and B compare image quality and resolution as a function of pixel dwell time. Panel A was created using a 32 µs/pixel dwell time at 4 kV and suffers from a diminished signal to noise ratio as is apparent in the "grainy" appearance of the enlarged inset. Panel B was created using a 100 µs/pixel dwell time at 4 kV. Increasing the pixel dwell time increases the signal to noise ratio and reveals an increased level of cellular detail, however increased pixel dwell time has the potential to lead to tissue charging and/or heat build-up which softens the block and introduces cutting artefacts (chatter) when sectioning. Panels C and D compare images captured under identical exposure conditions but with two different beam kV values. Tissue in these panels was impregnated with gold-toned nanogold particles to make differences in beam-penetration depths more apparent. Panel C was captured at 9 kV while panel D was captured at 21 kV. Increased kV has the advantage of increased contrast (D), however details are lost as result of gathering electrons from a greater depth of tissue (C). As a result of sampling a larger cross section, larger numbers of immunogold particles specific for GAP 43 are visible while non-specific labeling remains the same resulting in an increased signal-to-noise ratio. Panel A & B scale bar = 2 µm. Panel C & D scale bar = 1 µm. Please click here to view a larger version of this figure.
Figure 4: Beam intensity, kV and spot size. (A) Upon contacting the tissue sample, the electron beam (light blue) yields a teardrop-shaped interaction volume, from which varying forms of energy are produced from the interaction between beam electrons and the tissue sample. The teardrop shape is a function of tissue density and heavy metal staining along with beam energy, and the tilt angle of the electron beam43. While x-rays, auger electrons, and tertiary electrons are produced during SBF-SEM imaging, the primary concern is with backscattered (dark blue) and secondary (green) electrons13. The image produced with SBF-SEM imaging is produced by collecting backscattered electrons. These electrons originate from elastic interactions between the beam and the sample, and the signal collected is highly dependent on the atomic number of atoms interacted with – hence the need for heavy metal staining44. Secondary electrons originate from inelastic interactions between the beam and the sample and detection of their signal is highly dependent on surface orientation. Because the block-face is flat in SBF-SEM, secondary electrons do not contribute meaningfully to the signal collected13. In fact, secondary electron accumulation on the surface of the block can be a major source of charging and has a deleterious effect on image quality2. (B) This graph shows the relationship between beam intensity, beam kV, and spot size. The spot size is the spatial resolution of the beam, and determines the resolution limit of the images being produced. Lowering kV increases the spot size, but also decreases the imaging depth allowing for finer appreciation of detail. This has the effect of decreasing the detectable signal as well. Increasing beam intensity offers an initial improvement on spot size and signal detection, but rapidly increases levels of tissue charging. Ultimately, the beam intensity and kV values chosen are sample dependent and best determined empirically in relation to the scientific question being asked. Please click here to view a larger version of this figure.
Figure 5: Elastin-free microfibril bundle network in the mouse cornea. 3D reconstruction of microfibrils (white) closely associated with keratocytes (yellow, orange & green) within the corneal stroma. The microfibrils can be seen adjacent to, and in some cases within shallow grooves in, corneal keratocytes (arrows) (A). This network of elastin-free microfibrils are organized in distinct layers within the corneal stroma (B). Scale bar = 2 µm. The image block reconstructed is 45×45 µm in the x & y axis, and 30 µm in the z axis with voxel a resolution of 22x22x100 nm. Please click here to view a larger version of this figure.
Figure 6: Reconstruction of corneal nerves passing through basal lamina at the stromal-epithelial border. 3D reconstruction of a penetrating nerve (purple) as it passes through the basal lamina (green). This nerve can be seen to bifurcate prior to penetration. After penetrating into the epithelium, both nerve branches underwent ramification. Mitochondria (yellow) are visible in the stromal and epithelial portions of the nerve bundle. Scale bar = 2.5 µm. The image block reconstructed is 25×25 µm in the x & y axis, and 14 µm in the z axis with a voxel resolution of 12x12x100 nm. Please click here to view a larger version of this figure.
Figure 7: Limbal vasculature and associated cells in the peripheral mouse cornea. A single image (A) from a 3D image block (B) can be seen through which a vessel, nerve bundle, and associated cells travel. Panel C shows a reconstructed vessel (red) with an associated pericyte (gray) wrapped around it covering the endothelial cell junctions. A nerve bundle (blue) bifurcates in close proximity to this vessel as it travels through the tissue. A neutrophil (yellow) can be seen parallel to the long axis of the vessel, with its polymorphic nucleus visible within its cell body and the trailing uropod visible as a protrusion towards the right of the image. A mast cell (magenta) is visible on the underside of the vessel. Panel D isolates this mast cell, where its granules (green) can be more easily seen overlaying the nucleus (purple) within the cell. Panel E highlights the cellular structures overlaid on the cellular reconstructions, with endothelial nuclei denoted in blue, and adherent microparticles visible in the vessel lumen (orange). Arrows show cell-cell borders between endothelial cells, which can be further seen as raised ridges extending along the cells on the luminal side of the vessel. Panel A scale bar = 2 µm. The image block used to reconstruct these cells is 30×30 µm in the x & y axis, and 42.5 µm in the z axis with a voxel resolution of 14.6×14.6×100 nm. Please click here to view a larger version of this figure.
Figure 8: Reconstructed vascular network of the non-human primate retinal nerve fiber layer. (A) A 200×200 µm SBF-SEM image of the primate retina taken at 8192×8192 px. The location sampled is ~500 microns from the inferior temporal rim margin of a healthy eye with no pathology. The image series reconstructed in panels C & D were captured at 2048×2048 px, with imaging paused so that regions of interested could be imaged at 8192×8192 px. Panel B is the inlayed region of panel A, taken directly from the original image. Note the large number of axons and their mitochondria. (C) Orthoslice section through a 200x200x200 µm tissue volume of a control eye inferior temporal nerve fiber layer, with vasculature segmented. (D) Z-projection of the nerve fiber layer vasculature. This series illustrates the resolution possible in a large field using this methodology. Panel A scale bar = 20 µm. Panel B scale bar = 2 µm. Image series voxel resolution is 97.6×97.6×500 nm. Region of interest pixel resolution is 24.4×24.4 nm. Please click here to view a larger version of this figure.
Figure 9: Segmentation and 3D volume rendering of vessels in the giant danio (Devario malabaricus) compact heart. (A) Two-dimensional micrograph in an image stack, showing the profile of a central venular-size vessel (arrow) and an endothelial nucleus (arrowhead), with surrounding cardiac myocytes rich in mitochondria and well organized sarcomeres (*). (B) Two-dimensional micrograph of the image stack with a capillary-size vessel (arrow). (C) Biorthogonal projections of the micrograph stack showing the capillary in panel B projected through one orthogonal slice. (D) 3D rendering of segmented endothelial cells lining the reconstructed vessel. Illustrated in green, red, blue, and purple are four separate endothelial cells; the endothelial cell depicted in blue can be seen in cross section in panel B (arrow), while the endothelial cells depicted in red (arrow) and green (arrowhead) are seen in cross section in panel A. Panels A & B scale bar = 2 µm. The image block reconstructed is 30×30 µm in the x & y axis, and 16 µm in the z axis with a voxel resolution of 14.6×14.6×100 nm. Please click here to view a larger version of this figure.
Figure 10: Imaging complications and artefacts. (A) The wavy and distorted nature of this image is the result of imaging using a pixel dwell time that is too long. This heats the resin block, leaving the block face soft and rubbery which results in a distorted image upon cutting. (B) This image contains a host of artefacts. The asterisk indicates a wavy distortion caused by prior imaging at a higher magnification and similar to panel A, concentrating the beam on a smaller region with a longer pixel dwell time has softened the resin in this region of interest. While the higher magnification image collected was free of artefacts, this can lead to a subsequent series of images where the sample underlying the region of interest appears distorted. This panel also illustrates the issue of debris accumulation on the block face (arrow) during imaging, also denoted by the arrow in panel E. If this becomes a persistent imaging problem, it will be necessary to break the vacuum, open the chamber and blow away debris accumulated on the diamond knife and around the sample. Small discharges of electrons from the block-face can lead to the rapid contrast changes and lines denoted by the white arrowhead. (C) This image illustrates knife scratches on the block face. This can occur due to a damaged knife, or debris accumulation on the edge of the knife. (D) The artefact denoted (arrow) is a result of the electron beam focused on (without sectioning) the block face for an extended period of time with the sample still in the imaging chamber. (E) Improper fixation of tissue can lead to separation of cellular structures and connective tissue (*). (F) If a large amount of charging occurs in your tissue or resin block, subsequent accumulation and discharge can occur which leads to the image "skipping" as is seen in this image. Note the distortion of the tissue in the image at these skipping points (arrows). Panel A scale bar = 1 µm. Panel B scale bar = 2 µm. Panel C scale bar = 5 µm. Panel D scale bar = 2 µm. Panel E scale bar = 25 um. Panel F scale bar = 50 um. Please click here to view a larger version of this figure.
Figure 11: Imaging tissue at 3 kV using various pixel dwell times and beam intensities. All images were collecting using a 3 kV beam, beam intensity is on a device-specific scale ranging from 1 to 20. The field imaged is of the vascular lumen containing white and red blood cells. At this low kV it is difficult to appreciate cellular detail. Increasing the pixel dwell time had little effect. Increasing beam intensity to 6 improved image contrast. Please click here to view a larger version of this figure.
Figure 12: Imaging tissue at 7 kV using various pixel dwell times and beam intensities. All images were collected using a 7 kV beam, beam intensity is on a device-specific scale ranging from 1 to 20. The field imaged is of the vascular lumen containing white and red blood cells. At 7 kV, increasing beam intensity and pixel dwell time contributed to higher quality imaging. Please click here to view a larger version of this figure.
Figure 13: Imaging tissue at 12 kV using various pixel dwell times and beam intensities. All images were collected using a 12 kV beam, beam intensity is on a device-specific scale ranging from 1 to 20. The field imaged is of the vascular lumen containing white and red blood cells. At 12 kV, imaging is optimized by adjusting pixel dwell time and beam intensity. Charging is reduced/absent at shorter pixel dwell times while cellular detail and image contrast are best with a longer pixel dwell time and higher beam intensity. Please click here to view a larger version of this figure.
The purpose of this methods paper is to highlight the tissue preparation and imaging methodology that has allowed our lab to reliably capture high-resolution serial electron microscopy images, and to point out critical steps that lead to this outcome as well as potential pitfalls that can occur when conducting SBF-SEM imaging. Success using this protocol requires proper fixation of tissue, impregnation of heavy metals into the sample, modifications of the embedding resin to reduce charging, as well as an understanding of the microscope and imaging settings used to collect images. The maxim, "quality in, quality out" is an appropriate axiom for SBF-SEM imaging. As the goal of SBF-SEM often is the appreciation or quantification of ultrastructural detail, extra care must be given to fixation strategy in order to ensure that tissue distortion does not occur. If tissue becomes distorted at any point in the preparation of samples (i.e., undergoes swelling, shrinking, or disruption of cellular morphology), then tissue reconstruction and quantization will not yield accurate data. Furthermore, the use of incorrect imaging settings can lead to loss of data that cannot be recaptured as SBF-SEM imaging is a destructive process. Additionally, care must be used when loading a tissue sample as the delicate diamond knife can be damaged by hasty or incorrect sample preparation. This can result in chips or breaks in the knife, which can leave visible scratch marks in images (Figure 10C). The diamond knife can also be damaged by calcified structures, hard granules, or accidentally embedded glass (e.g., from reagent ampules).
While the majority of SBF-SEM literature to date uses beam acceleration voltages in the range of 1 to 3 kV alongside pixel dwell times closer to 1-5 µs/px (Figure 11)45,46,47,48,49, the current protocol uses acceleration voltages of 7-12 kV and a pixel dwell time of 12 µs/px for serial imaging and 32 µs/px for imaging regions of interest (Figures 12 & 13). These settings, coupled with a slice thickness of 100-200 nm allows for high-quality and high resolution imaging of a wide range of biological tissue. Increased acceleration voltage allows for an increase in contrast, resolution, as well as signal-to-noise ratio. Increased dwell time further increases resolution and signal-to-noise ratio, while increased slice thickness leads to decreased charging on the block surface during sectioning and combats beam-induced damage in subsequent images14. While this imaging method may differ from convention, the images and datasets produced speak for themselves. If we had to speculate on the reason for this success, it is possible that it is a result of our unique combination of high kV values, longer pixel dwell times, and block preparation. Increasing imaging kV results in an increased interaction volume between the electron beam and the sample. This interaction volume is both deeper as well as wider resulting in a theoretical increase in the number of electrons detected that originate from deeper within the sample block, or from a wider cross section of tissue as the spot size teardrop increases in diameter. As SBF-SEM is interested in the surface detail of the block, this results in a theoretical decrease in signal-to-noise ratio. However, the increase in kV also pushes electrons deeper into the sample where they are less likely to escape the block and contribute to the electrons collected by the detector. With the added benefit of an increased signal via longer pixel dwell times and higher beam intensity, it is possible that this imaging method results in a greater increase in signal from the sample surface in relation to noise originating within the interaction volume. Additionally, the increased sample conductivity introduced with carbon black as well as silver and gold coating helps to ameliorate charge buildup which now occurs deeper within the block and further from the block-face. Indeed, Figure 11, Figure 12, Figure 13 show that as kV is increased sample charging begins to diminish as it is potentially pushed deeper into the block. Samples imaged at low magnification can be captured with adequate contrast using the conventional settings, however these images often lack detail upon close inspection. Our data clearly show that when using relatively high magnification where the goal is cellular detail, increasing the conventional settings can produce exceptional results. The 2020 article by P. Goggin, et al provides a useful table outlining the effect of changing imaging settings on final image quality, and is a helpful reference to consult if optimizing the protocol for novel tissues becomes necessary14. The 100-200 nm slice thickness recommended in this protocol has the added benefit of allowing the collection of large SBF-SEM data sets at a rapid rate. While collecting images at 12µs/px for example, imaging through a 100 µm depth at 2048×2048 px requires ~14 hours while sectioning at 100nm/section but would require ~56 hours if sectioned at 25nm/section. While x,y resolution remains unchanged as a result of section thickness, not accounting for the added ability to image using higher kV values and pixel dwell times that come with larger sections, it is important to note that the resolution along the z-axis does suffer. The loss of z-resolution is an important consideration and should be contemplated when deciding how tissue should be oriented in the resin block and in relation to the imaging plane, and has the potential to preclude the study of smaller cell features or interactions (e.g., synaptic invaginations or intracellular features on the scale of tens of nanometers). However, in addition to rapid imaging time, this protocol has additional added benefits in that it rapidly produces ideal datasets for stereological analysis as well as the study of rare biological events or cells. Larger section thickness can also aid in manual 3D reconstruction, as a 100 µm region sectioned at 100 nm/section would require manual segmentation of 1,000 images while the same region sectioned at 25 nm/section would require manual segmentation of 4,000 images.
SBF-SEM has the benefit of generating large datasets in a relatively short period. While data analysis can be performed using quantitative methods such as stereology, which will be discussed below, it can often be informative to create 3D reconstructions via image segmentation. An image stack created using SBF-SEM can be thought of as a collection of voxels, while segmentation is the process of assigning these voxels to user-defined objects thereby creating 3D representations of tissue structures. These reconstructions often provide a heretofore unseen perspective on tissue ultrastructure and cell-cell interaction (Figure 5, Figure 6, Figure 7, Figure 8, Figure 9). Furthermore, once reconstructions have been created it is possible to use data inherent in the reconstructions to extract a wealth of information from segmented tissue. Parameters ranging from surface area, volume, length and distance, as well as angular data are all readily available once a reconstruction has been created50,51. While this can be incredibly useful, especially when paired with videos and images pulled from reconstructed data sets, the time required for manual segmentation is an important consideration when attempting to extrapolate data from SBF-SEM datasets. There are currently a host of both free and purchasable software available for the manual and semi-manual segmentation of SBF-SEM image stacks. One free option for reconstruction software is the image processing package Fiji for ImageJ, an open source image processing program, which contains a segmentation editor plugin that allows for manual segmentation52,53. Additionally, the software Reconstruct offers an alternative free segmentation option54 (Figure 8). While potentially expensive, purchasable options often contain more robust feature sets, such as semi-automated segmentation processes or movie and image creation suites. One such option was used to create the reconstructions found in Figure 5, Figure 6, Figure 7 and Figure 9 (Details available in Table of Materials). Additionally, tools are available for the creation, analysis, and rendering of contrast-based 3D reconstructions using virtual reality with the potential to greatly speed up the reconstruction process20,55. While not always available for all applications, a host of software tools are available for computer assisted manual segmentation which have the potential to greatly decrease the time required for segmentation56,57,58. Regardless of the software used, considerable forethought and an understanding of the question being answered, or gap in knowledge to be filled, by serial reconstructions should precede segmentation, as the process is laborious and time-intensive.
The production of 3D reconstructions comes with its own considerations. With larger data sets processing power can be a limiting factor, and so optimizing the use of system resources can be critical for maintaining a productive workflow and speeding up the reconstruction and rendering process. When rendering a 3D reconstruction, most software converts segmented image stacks into a surface comprised of interconnected triangles. If a reconstruction project is large or intricate, the rendering of these triangles can require a great deal of computing power. While working on a 3D reconstruction, it can be helpful to limit the number of triangles the reconstruction software can use to convert the segmented images into reconstructed surfaces. This can be useful for monitoring the progress of a 3D reconstruction during the segmentation process. Once segmentation is complete, the triangle limit can be removed before rendering images or videos of reconstructions. Alternatively, and if the reconstruction software allows for it, we have found success monitoring the progress of a reconstruction using volume rendering rather than surface generation. Volume rendering, while not as suitable for images or videos meant for publication or presentation, requires far less processing power and as such can be helpful in providing a smooth experience when reconstructing and preparing images and videos of reconstructions. Additionally, it is best practice when manually segmenting an SBF-SEM data set to define every object to be reconstructed with its own unique identifier. If a field of epithelial cells is being reconstructed for example, rather than assigning all epithelial cells to a voxel group entitled "epithelium," each epithelial cell should be assigned its own moniker (i.e., Epi1, Epi2, Epi3, etc.). This affords greater freedom when the reconstruction is complete, as each cell can be either included or excluded from the final rendering, assigned different colors or transparencies, or removed or introduced individually if a video is being produced. Furthermore, this allows metrics such as surface area or volume to be collected from each reconstructed object rather than the object group as a whole.
Another incredibly powerful tool for extracting quantitative data from SBF-SEM image stacks is the practice of stereology. Stereology takes advantage the inherent mathematical relationships between three-dimensional objects and their two-dimensional representations (i.e., electron micrographs). SBF-SEM data sets are ideal for the application of stereology, as this method for extracting 3D information from large datasets is considerably less time- and labor-intensive when compared to segmented reconstruction. Stereology generally consists of applying geometric grids to random, uniformly sampled images and has been used extensively over the past 50 years in order to produce accurate and unbiased estimates of cell/organelle number, length, surface area, and volume21,59,60,61,62,63. While 3D reconstructions can be impressive and provide a novel perspective on biological tissues, it is often quicker, more accurate, reproducible, and conducive for large sample sizes to use a stereological approach to data extraction. While there are many papers discussing the practical application of stereology64,65,66, a number of textbooks provide useful, in-depth overviews of the methodology as well as provide a number of stereological grids which can be applied to the study of tissue ultrastructure67,68,69.
SBF-SEM is a powerful imaging method that allows for the three-dimensional appreciation of tissue ultrastructure. While the ability to create 3D datasets with SEM resolution puts previously unanswerable questions within our reach, proper tissue preparation and an understanding of SBF-SEM imaging is paramount for the success of studies that utilize this microscopy method. It is our hope that the application of this protocol to future studies will lead to greater and greater insight into the biological mysteries that surround us, and continue to push us further into the frontiers of human knowledge.
The authors have nothing to disclose.
We would like to thank Dr. Sam Hanlon, Evelyn Brown, and Margaret Gondo for their excellent technical assistance. This research was supported in part by National Institutes of Health (NIH) R01 EY-018239 and P30 EY007551 (National Eye Institute), in part by the Lion's Foundation for Sight, and in part by NIH 1R15 HD084262-01 (National Institute of Child Health & Human Development).
1/16 x 3/8 Aluminum Rivets | Industrial Rivet & Fastener Co. | 6N37RFLAP/1100 | Used as specimen pins. |
2.5mm Flathead Screwdriver | Wiha Quality Tools | 27225 | |
Acetone | Electron Microscopy Sciences | RT 10000 | Used to dilute silver paint. |
Aspartic Acid | Sigma-Aldrich | A8949 | |
Calcium Chloride | FisherScientific | C79-500 | |
Conductive Silver Paint | Ted Pella | 16062 | |
Denton Desk-II Vacuum Sputtering Device equipped with standard gold foil target | Denton Vacuum | N/A | This is the gold-sputtering device used by the authors, alternates are acceptable. |
Double-edged Razors | Fisher Scientific | 50-949-411 | |
Embed 812 | Electron Microscopy Sciences | 14120 | |
Gatan 3View2 mounted in a Tescan Mira3 Field emission SEM | Gatan & Tescan | N/A | This is the SBF-SEM device used by the authors, alternates are acceptable. |
Glass Shell Vials, 0.5 DRAM (1.8 ml) | Electron Microscopy Sciences | 72630-05 | |
Gluteraldehyde | Electron Microscopy Sciences | 16320 | |
Gorilla Super Glue – Impact Tough | NA | NA | Refered to as cyanoacrylate glue in text. |
Ketjen Black | HM Royal | EC-600JD | Refered to as carbon black in text. |
KOH | FisherScientific | 18-605-593 | |
Lead Nitrate | Fisher Scientific | L62-100 | |
Microwave | Pelco | BioWave Pro | This is the microwave used by the authors, alternates are acceptable. |
Osmium Tetroxide | Sigma-Aldrich | 201030 | |
Potassium Ferrocyanide | Sigma-Aldrich | P9387 | |
Silicone Embedding Mold | Ted Pella | 10504 | |
Sodium Cacodylate Trihydrate | Electron Microscopy Sciences | 12300 | |
Samco Transfer Pipette | ThermoFisher Scientific | 202 | Used to make specimen pin storage tubes. |
Swiss Pattern Needle Files | Electron Microscopy Sciences | 62115 | |
Thiocarbohydrazide | Sigma-Aldrich | 223220 | |
Uranyl Acetate | Polysciences, Inc. | 21447-25 | |
Reconstruction Software | |||
Amira Software | Thermo Scientific | N/A | Used to create the reconstructions found in figures 5-7 and 9. |
Fiji (Fiji is Just ImageJ) | ImageJ.net | N/A | TrakEM2 can be added to Fiji to asist in manual segmentation. |
Microscopy Image Browser (MIB) | University of Helsinki, Institute of Biotechnology | N/A | |
Reconstuct Software | Neural Systems Lab | N/A | |
SuRVoS Workbench | Diamond Light Source & The University of Nottingham | N/A | |
SyGlass | IstoVisio, Inc. | N/A | Allows for reconstruction in virtual reality and histogram-based reconstruction methods. |