Tissue fixation and processing

All rat experiments were conducted in accordance with UK Home Office regulations.

Rat Kidney: Intracardiac perfusion of a glutaraldehyde fixative containing lanthanum and dysprosium to stain the glycosaminoglycans (Lanthanum Dysprosium GlycosAminoGlycan adhesion method (LaDy GAGa)) was carried out as previously described [17]. A male Sprague Dawley rat was injected with a lethal dose of 0.7 ml/kg sodium pentobarbitone (Lethobarb, Ayrton Saunders Ltd UK), the thorax promptly opened and the left side of the heart perfused at 100 mmHg pressure with a flush solution of HEPES buffered mammalian Ringer containing 0.5% LaNO₃.6H₂O, 0.5% DyCl₃.6H₂O, (room temperature, pH 7.3). Heparin anticoagulant was not used because of potential cross-reaction with the lanthanides in solution. Flushing with a 3 ml bolus was followed by 100 ml of 2.5% glutaraldehyde, 2% sucrose in the same solution as above (room temperature, pH 7.3). Tissues started to stiffen within 30 seconds. Samples 1 mm in diameter were cut from each kidney and stored at 4ºC for 2 days to several weeks in buffered glutaraldehyde. Tissues were washed with HEPES buffer and transferred to 0.1 M sodium cacodylate buffer (pH 7.3), postfixed in 1% OsO₄ in 0.1 M cacodylate buffer (pH 7.3) for 1 h and rinsed in 0.1 M sodium cacodylate buffer (pH 7.3). After Osmication, tissues were washed in distilled water, Fixed/stained with 2-3% aqueous uranyl acetate [~12 hrs, 4ºC], dehydrated in a graded series of ethanol, infiltrated with Araldite resin mixtures with propylene oxide and Araldite resin alone and then embedded in Araldite resin (Agar Scientific, UK).

A Human tissue block was prepared from an unused transplant kidney by perfusion fixation. However, additional aqueous heavy metal steps were applied using an adapted West et al[18] protocol (See Additional file 1). Dehydration and resin infiltration used the rat protocol. Research involving human subjects was performed with appropriate consent, and after review by UK research ethics committee (study approval 07/H0102/45) in accordance with UK legislation and the Helsinki Declaration.

Tissue block preparation

SBF-SEM

Tissues blocks with glomeruli of interest were demarcated and each trimmed to approximately a 500 μm × 500 μm block face with less than 2 mm depth using a razor blade. The blocks were mounted using cyanoacrylate adhesive (Permabond) onto a 3 mm head diameter aluminium mounting pin. The pin head and block were sputter-coated with gold ready for SEM.

The trimmed mounted block was attached to the chuck and the resin embedded glomeruli were sectioned using a Gatan 3view ultramicrotome within an FEI Quanta 250 scanning electron microscope (SBF-SEM). This technique, based on the work of Denk and Horstman [19] allows a backscattered electron image of the block face to be collected after each diamond knife cut (Figure 2A).

Once set up inside the microscope an automatic routine imaged the surface, advanced the specimen and sectioned the block face removing a 150 nm thick resin section (50 to 200 nm thickness is possible for resin embedded material). The knife was moved away from the block and the surface re-imaged. The sequence of advance, cut, clear (the block), image was repeated until all the images were acquired. Multiple images were stacked together to build up a three-dimensional reconstruction of the original sample. The data stack was imaged using backscattered electron detection at 2.5 kV, on a 2048 × 2048 array with 25 μs dwell time at a chamber pressure of 0.56 mbar in a water vapour atmosphere to earth and reduce sample charging. Each cut was a 150 nm slice. The detected depth is expected to be under 10 nm where there is stain. Five hundred images were taken giving an overall volume of 84 μm × 84 μm × 75 μm. This relates to a 41 nm × 41 nm × 150 nm voxel size.

Analysis

Image analysis was performed with ImageJ (NIH) [24] but additionally a combination of IMOD [23], Amira (Visage Imaging) and Fiji [25] were also used. Tom-TEM reconstructions and general large file data stack manipulations of FIB-SEM and SBF-SEM were performed with IMOD. Amira was used for segmentation. Fiji was used for image registration, 3D visualisation and other analysis.

To determine glycocalyx fibre spacings in the reconstructed tomogram an autocorrelation method was used from previous publications [26, 27]. In brief: X image areas (150 nm by 150 nm) containing glycocalyx were chosen in a reconstruction slice. These were extracted at the same coordinates for every slice if the area contained glycocalyx. To these areas a taper to the intensity mean was added followed by a fast Fourier transform. After this the autocorrelation function was performed leaving a correlation map in real space. Radial profiles were then taken in a direction parallel to the membrane. All positive peaks in the profile were found using the second differential where the distance is a predominant spacing. The peaks were only used if they were above a threshold value determined by the second differential value of control (blank resin) areas.

Ultrastructure of the glomerular filtration barrier

The complex nature of the GFB is demonstrated by classical transmission electron microscopy (TEM) in Figure 1A where a low power image of two rat glomeruli are seen in cross section after LaDy GAGa staining. The higher power view of a single capillary (Figure 1B) shows the presence of the podocyte, foot processes, endothelium, fenestrations and peripheral urinary space. At 24,500 times microscope magnification (Figure 1C) the detailed structure of the GFB reveals that a complex glycocalyx overlies the endothelium, with a high contrast layer sitting above a less electron dense layer just above the fenestrated endothelium. The glycocalyx layer revealed by this staining is 200-400 nm thick. These images also reveal a thin layer of contrast lining the podocyte foot process cell membrane, indicative of a glycocalyx on the urinary side, although this is an order of magnitude thinner as revealed by this perfusion fixation and staining technique.

SBF-SEM

The benefit of using an SEM is that several electron imaging modes can be acquired at the same time (Figure 2B/C). Figure 2B shows secondary electron and Figure 2C the backscattered electron intensity image from the same block face. Secondary electrons are low energy electrons detected from the surface influenced by topographical information along with stain density. In the image a small knife mark (asterisked) is visible 1/4 of the way across the image running top to bottom. Some beam damage in the empty resin in the bottom left hand corner is also visible indicating that the beam intensity limit of this patch of resin has been reached. The imaging conditions were therefore carefully chosen to balance image quality with speed of acquisition and volume of beam interaction, defining the resolution limit. The back-scattering electron signal (Figure 2C) is predominantly affected by the atomic number of the atoms in the sample, resulting in greater visualisation of the stained material within the sample. By inverting the signal intensity of the backscattered electron micrographs an image similar to a standard TEM micrograph of a stained section can be formed. This method contrasts with Tom-TEM, whereby the sample is tilted on a stage and the images reconstituted to provide a three dimensional view through the section (Figure 2D).

Figure 3A shows six images from a stack of 500 serial sections (150 nm section thickness) through a rat glomerulus stained by LaDy GAGa (Additional file 2: Movie S3A). These representative images indicate that SBF-SEM could be used to image the depth of a whole glomerulus. In this glomerulus, the lanthanum and dysprosium was perfused into the first afferent vessels in the glomerulus and was chosen as it demonstrated clearly the 3D nature of the vessels. The dark staining of the glycocalyx can be seen in the afferent arteriole and the first few glomerular vessels. Figure 3B shows a rotated 3D-reconstruction of the images (see Additional file 3: Movie S3B). With the glycocalyx highlighted and extracted from the image (Figure 3C and Additional file 4: Movie S3C) the glycocalyx coverage of the initial vessels is seen to be extensive and easily defined in three dimensions.

To determine whether this technique could be used to reconstruct human glomeruli, a perfused human kidney was subjected to SBF-SEM. The resulting images clearly show SPS, IPS and PUS in the kidney (Figure 4). Figure 4A shows image slices from the raw data (Additional file 5: Movie S4A and Additional file 6: Movie S4B). Figure 4B shows image slices through a podocyte cell body. Figure 4C and 4D show the reconstruction of the podocyte cell body with and without capillaries. The SPS has been coloured yellow to aid visualisation (Additional file 7: Movie S4C and Additional file 8: Movie S4D). Nine SPS areas from under six podocytes were tracked for their entirety (for details see Table 1). Any exit to another urinary space was noted, the height of SPS exit regions ranged from less than 41 nm (pixel size) to 2000 nm. Most were approximately the same as the height of the SPS, rather than a specific constriction. Some did have no obvious exits, though areas where an exit less than 40 nm could be likely were observed.

Figure 5 shows a reconstruction of a human IPS (approximate volume 70 μm³). Figure 5A (Additional file 9: Movie S5A) is an area of an image from the sequence and this has had the relevant areas coloured in Figure 5B (Additional file 10: Movie S5B). Reconstruction indicates that this space is complex, interweaving and tortuous, with narrow exits to the peripheral urinary space. This space has been seen in section before but to fully define these areas serial section stacks would have been prohibitively long (TEM) or too low in resolution (light microscopy). The IPS studied within the image stack (illustrated in Figure 5C) showed evidence of passages to larger areas of PUS of only 500-1000 nm wide represented as a tortuous ribbon (Additional file 11: Movie S5C). These routes interweaved with podocytes cell bodies and foot processes on capillary walls. Furthermore there was a substantial bulb-like space, which appears to have a connection of less than 100 nm at one side and only one narrow ribbon-like connection from the other side (‘N’ in Figure 5C).

FIB-SEM

Figure 6 demonstrates reconstruction of the glomerular filtration barrier using FIB-SEM. Figure 6A (Additional file 12: Movie S6A) shows a single image at low power of a filtration region with no overlying SPS. A magnified region of Figure 6A is shown in Figure 6B. The LaDy GAGa staining was clear and the depths of glycocalyx could be estimated. The three dimensional reconstruction using this technique on 295 sections demonstrates both the variability in the glycocalyx surface (Figure 6C) which appears knobbly and irregular and the complexity of the podocyte foot processes on the other side showing complex overlapping structures (see Additional file 13: Move S6B). Sections taken through the GFB from the lumen (Figure 6D) show the potential for FIB-SEM in determining the spatial relationship between GFB components revealing the relative dispositions of glycocalyx, fenestrations, GBM and foot processes.

Tom-TEM

Using Tom-TEM a volume of 2 μm × 2 μm × 300 nm of the GFB of a rat glomerulus perfused with the LaDy GAGa technique was reconstructed. The resolution was 1.8 nm in the xy directions, meaning that cell membranes could be easily distinguished. The reconstruction resulted in a depth (z) resolution of 4.5 nm due to the missing acquisition wedge and 25% section shrinkage.

Figure 7 demonstrates the raw image data and the reconstruction. As the section was rotated in the EM goniometer, differing views of it were imaged (Figure 7A-D, Additional file 14: Movie S7A). Reconstruction at a single slice level across the GFB (Figure 7E-H, Additional file 15: Movie S7B) demonstrates clearly the SPS foot processes, basement membrane, endothelial cell with fenestrations, and glycocalyx. The glycocalyx coverage of the endothelium is complete being 100-200 nm in depth, and the stained fibres could be resolved. There is a 25 nm thick layer of reduced staining between the membrane and denser glycocalyx layer. Auto-correlation to determine the fibre spacing [17, 26, 27] of the densely stained glycocalyx from 947 image areas over 225 slices (each area approximately 150 nm by 150 nm as in Figure 7E-H) showed an expected mean fibre spacing of 20 nm but no lattice could be determined. Reconstruction in the en-face view of the glycocalyx was also possible using this technique (Figure 7I-L) and the variation in local coverage can be observed. The three dimensional reconstruction of the whole glycocalyx layer and GFB is shown in three different views in Figure 7M-O (Additional file 16: Movie S7C).

Additional file 15: Video S7B. Transmission electron tomography of 300nm thick section of glomerular capillary glycocalyx. z sequence of the reconstruction. Scale Bar=150nm. (MOV 14 MB)

SBF-SEM

Once set up the total run time to section and image 500 slices was 14.5 h on the SBF-SEM, without human intervention. Some of the serial images were out of focus and this was assumed to be due to a small sample fleck interfering with the beam. Even though an extensive staining procedure was required to induce enough electron density within the sample, the overall image quality appeared very good and no differences in ultrastructure could be attributable to the SBF-SEM technique. To create and document this amount of data by traditional serial ultrathin sectioning methods is possible though highly expensive in technical man-hours and EM time. The lanthanum and dysprosium stained glycocalyx had good contrast in the rat glomerulus, but the tissue with no extra heavy metal staining did not, therefore extra heavy metal doping was required to observe general tissue. Even without an improved backscatter detector, SBF-SEM achieved good sub-cellular resolution with a >70 μm cube of tissue. Figure 4 also emphasizes the possibilities of SBF-SEM in allowing identification of multiple features, rapid reconstruction and colour coding to reveal how structures and spaces relate to each other. There is no reason this reconstruction could not have been done for a depth of several hundred micrometres to encompass an entire human glomerulus, something totally impractical at this resolution with manual sectioning.

FIB-SEM

The total run time on the FIB-SEM to section and image 300 slices was 9 h.

The FIB-SEM had better backscatter electron detection (primarily due to the detector quality) than the SBF-SEM used. This accounts for the improved contrast between the SBF-SEM and FIB-SEM on the perfusion fixed rat without the extra heavy metal staining procedure.

In FIB-SEM the estimated resolution was in the order of 25 nm (by fast Fourier transform) in both the x and y directions, though due to the tilt this was not the same number of pixels; oversampling has contributed to a 5 fold lower field of view than could be possible. The 10 nm serial removals of material in the z direction changed the micrograph noticeably indicating a 20 nm resolution, between images but beam damage and artefact is likely on thinner cuts. Unlike SBF-SEM where near perfect alignment of z images is intrinsic, the FIB-SEM required lateral alignment between successive images. However, no significant warping or rotation was observed.

The FIB-SEM differs from the SBF-SEM by having a thinner slice removal capability. This makes it more suited to resolving objects an order of magnitude smaller than in the SBF-SEM. There is also an improvement in signal to noise and resolution due to the improved charge dissipation of the smaller cut surface. The datasets shown were milled in from a glass knife-trimmed surface, but it is relatively straightforward to excavate anywhere from an edge cut into a trimmed block surface, though navigation may be cumbersome without a good surface map from a TEM.

The resulting reconstruction gives novel insights into the 3D nature of the glycocalyx over the filtration barrier. It has a knobbly forest-like appearance, and gaps to the membrane. Such gaps and macrostructure, if physiologically accurate would alter the way in which filtration is modelled and understood. More studies with different stains, or specific studies without staining (freezing techniques), would be needed to determine the level of artefact within this protocol and hence the physiological relevance.

Tom-TEM

Almost all new TEMs can perform tomography, and mostly come with software designed to save user time. The dataset volume is not usually cubic, with the z thickness being approximately10-fold smaller than the xy directions (before shrinkage correction). The resolution is the highest of the three 3D modalities, approaching the limit of resin-embedded sections: the glycocalyx was clear and the fibre spacings discernible (Figure 6E-H), but the missing angles in the tilt series were a limiting factor in the z direction. A dual tilt tomogram (the section is rotated 90° about a central z axis and the tilt sequence repeated) results in reduced missing information. In this case the resolution was approximately half that of the x-y resolution before the corrections for shrinkage were made. For many applications the section shrinkage is not an issue but for quantitative measurements this needs to be corrected for. The main issue is that shrinkage is anisotropic across regions of tissue/resin/stained components in the x, y and z direction.

An issue with conventional electron microscopy and the elucidation of ultrastructure whether in normal or diseased tissues is that sampling error can be enormous. Small sections of small parts of tissues are observed and assumptions are made about ultrastructural relationships in the remaining unseen majority of the structure. Reconstruction of tissues has become easier through the advent of digital imaging and better ultramicrotomes, however, the process is still labour intensive. The techniques described here attempt to addresses this problem by greatly extending the effective volume of tissues analysed and correlating images at different resolutions.