Three-Dimensional Visualization of the Podocyte Actin Network Using Integrated Membrane Extraction, Electron Microscopy, and Machine Learning

Significance Statement

Podocytes have interdigitated foot processes with intricate three-dimensional structures that are crucial for glomerular filtration. Electron microscopy shows podocytes’ complex morphology, but direct visualization of their cytoskeleton and definitive identification of the proteins that comprise the cytoskeletal structures have remained elusive. The authors describe a novel technique that reveals the three-dimensional organization of the podocyte cytoskeleton, finding that actin cables inside foot processes are connected directly to slit diaphragms, to form a continuous mesh-like sheet covering the glomerular basement membrane. Their findings also reveal these actin cables to be part of an extensive, contiguous actin network surrounding the major processes and the podocyte cell body. Applying this technique may help elucidate the mechanobiologic mechanisms regulating podocyte architecture and reveal the ultrastructural changes in the actin network on podocyte injury.

Visual Abstract

Abstract

Background

Actin stress fibers are abundant in cultured cells, but little is known about them in vivo. In podocytes, much evidence suggests that mechanobiologic mechanisms underlie podocyte shape and adhesion in health and in injury, with structural changes to actin stress fibers potentially responsible for pathologic changes to cell morphology. However, this hypothesis is difficult to rigorously test in vivo due to challenges with visualization. A technology to image the actin cytoskeleton at high resolution is needed to better understand the role of structures such as actin stress fibers in podocytes.

Methods

We developed the first visualization technique capable of resolving the three-dimensional cytoskeletal network in mouse podocytes in detail, while definitively identifying the proteins that comprise this network. This technique integrates membrane extraction, focused ion-beam scanning electron microscopy, and machine learning image segmentation.

Results

Using isolated mouse glomeruli from healthy animals, we observed actin cables and intermediate filaments linking the interdigitated podocyte foot processes to newly described contractile actin structures, located at the periphery of the podocyte cell body. Actin cables within foot processes formed a continuous, mesh-like, electron-dense sheet that incorporated the slit diaphragms.

Conclusions

Our new technique revealed, for the first time, the detailed three-dimensional organization of actin networks in healthy podocytes. In addition to being consistent with the gel compression hypothesis, which posits that foot processes connected by slit diaphragms act together to counterbalance the hydrodynamic forces across the glomerular filtration barrier, our data provide insight into how podocytes respond to mechanical cues from their surrounding environment.

ESKD is a major public health burden that can arise when the podocytes responsible for blood filtration lose their function.¹ Although this loss of function is associated with podocyte foot process effacement and detachment, the mechanobiological factors underlying podocyte shape regulation and adhesion have not been fully characterized.² A challenge is that these cells are best studied in the natural environment of the glomerular filtration barrier (GFB), which comprises an intricate epithelial podocyte layer lying atop the glomerular basement membrane (GBM) with the glomerular endothelium on the opposite side of the GBM. In this environment, podocytes develop tentacle-like foot processes that interdigitate with those of neighboring podocytes and form slit diaphragms that stretch in stacked zipper-like arrays of 40 nm–long protein assemblies over the intervening gaps.³ However, due to these length scales being too small for conventional microscopy, and to the need to image podocytes within the context of a much larger, curved physiologic structure,^4,5 the three-dimensional (3D) actin structures that give rise to the shape and function of podocytes have never been quantified.

Given the dominant role of the actin cytoskeleton in determining cell shape,^6–10 understanding actin organization is especially critical for understanding how podocyte shape is regulated in both health and injury. The actin cytoskeleton is particularly responsive to mechanical stress, with both its structure and its adhesions to the matrix (via receptors) varying strongly with mechanical loading.^11–14 The mechanical environment of podocytes is thus critical, with stresses arising from blood pressure, cytoskeletal contraction, and fluid flow through slit diaphragms.¹⁵ These factors further motivate the need to study the podocyte cytoskeleton in the context of an intact glomerulus.

Many studies have focused on overcoming the challenges of scale and geometry to characterize podocytes. Transmission electron microscopy (TEM) and field emission-scanning electron microscopy (SEM) have revealed broadening of podocyte foot processes as they efface^16–18 and changes to endothelial cells associated with injury in specific glomerular diseases.¹⁹ Two recently developed techniques, block-face SEM^20,21 and focused ion beam (FIB)–SEM,^22–25 have uncovered key aspects of glomerular and podocyte ultrastructure in 3D, including ridge-like promenades,²² the irregular GBM of Alport syndrome,²⁰ podocyte-parietal cell interconnectivity,²⁶ and changes to podocytes during foot process effacement.²⁴ Especially important are recent FIB-SEM studies; however, these have yet to succeed in revealing the details and identities of the protein fibers that comprise the cytoskeletal network.^24,27

From the perspective of mechanobiology, actin is a critical mediator of cell shape and function, and a critical mediator of cell-matrix interactions and mechanosensing.^28–31 The challenge of imaging actin structures in vivo has been a factor limiting the understanding of many cell types, leading to underlying uncertainties about the role of structures, such as stress fibers, in the physiology of mature, healthy cells.^32,33 There is thus a pressing need for a broadly applicable technology to image the actin cytoskeleton at high resolution. We therefore developed such a technique, and applied it to characterize the podocyte’s 3D actin network in unprecedented detail.

Materials and Methods

Animal Models

C57BL/6J mice (The Jackson Laboratory) were used for isolation of wild-type healthy glomeruli. Mice were anesthetized using a ketamine/xylazine cocktail. Glomeruli were isolated by perfusing mice with magnetic Dynabeads as previously described,³⁴ but without collagenase digestion.

Glomerular Isolation and Membrane Extraction

Isolated glomeruli were subjected to membrane extraction as previously described.³⁵ All rinses utilized a super-magnet to pull down the pellet and prevent loss of glomeruli during washes. The extraction buffers (KHMgE buffers) contained 10 µM phalloidin, 10 µM Taxol, 2 mM ATP, 1 mM phenylmethylsulfonyl fluoride, 10 µM leupeptin in 0.1 M KCl, 30 mM HEPES at pH 7.2, 10 mM MgCI₂, and 2 mM EGTA at pH 7.1. Extraction solutions contained 0.01% saponin and 0.5% Triton X-100. Before the extraction, samples were preincubated with the KHMgE solution for 1.5 minutes. They were then detergent extracted for 1.5 minutes in the respective extraction solution, before fixing the sample for 10 minutes in 4% PFA in KHMgE buffer, with 0.01% Saponin, 0.5% Triton X-100 before fixing it further in 2% PFA/2% glutaraldehyde in PBS at room temperature (RT).

Adherence of Glomeruli to Glass

In preparation for EM, glass coverslips were cut into 3 mm × 3 mm squares, cleaned in chromo- sulfuric acid solution and washed extensively with dH2O. Coverslips were then incubated in 5 mg/ml LMW poly-l-lysine in 0.1 M KCl for 30 minutes, and then washed 3 × 5 minutes in 0.1 M KCl. Coverslips were placed on a parafilm-coated super-magnet, the residual KCl solution wicked off the top of the slip, and a 10 µl drop of suspended glomeruli in KHMgE buffer (70 mM KCl, 30 mM HEPES 5 mM MgCl, 3 mM EGTA pH 7.0) was deposited onto the slip, and allowed to adhere 5 minutes at RT. After adhesion, coverslips were transferred into 2% glutaraldehyde in KHMgE buffer for 30 minutes, then either washed 3 × 10 minutes 0.15 M Na Cacodylate with 2mM CaCl₂, pH 7.4, and processed for SEM or washed 3 × 5 minutes in dH2O and processed by Quick-Freeze Deep-etch EM (QF-DEEM). A super-magnet was utilized to help secure adhered glomeruli to the coverslip surface for solution transfers hereafter.

SEM Preparation

Isolated glomeruli were subjected to membrane extraction as previously described.³⁵ Samples were stained with 1% osmium tetroxide in 0.15 M Na cacodylate with 2 mM CaCl₂, pH 7.4 for 1 h at RT in the dark. After 3×10-minute rinses in dH2O, samples were dehydrated in a graded ethanol series (50%, 70%, 90%, 100%, and 100%) for 10 minutes each and then critical point dried (Leica CPD 300). Glass slips were adhered to a conductive carbon adhesive tab on aluminum stubs. Sample stubs were sputter coated with 6-nm iridium, (Leica ACE 600) and imaged by SEM (Zeiss Merlin equipped with a Gemini II electron column) at an accelerating voltage of 3 keV with a beam current of 200 nA with secondary electrons detected by an Everhart-Thornley (SE2) detector.

QF-DEEM

Quick-freezing was accomplished by previously reported protocol with minor modifications.^36–41 Briefly, coverslips containing extracted glomeruli were rinsed in dH2O and frozen by abrupt application of the sample against a liquid helium–cooled copper block with a Cryopress freezing machine. Frozen samples were transferred to a liquid nitrogen-cooled Balzers 400 vacuum evaporator, fractured at −100°C, etched for 20 minutes at −80°C, and rotary replicated with approximately 2 nm platinum deposited from a 20° angle above the horizontal, followed by an immediate, approximately 10 nm stabilization film of pure carbon deposited from an 85° angle. Replicas were floated on a dish of concentrated hydrofluoric acid, and transferred with a glass rod through three rinses of dH2O, one rinse of household bleach, then three additional rinses of dH2O. All solutions containing a loopful of Photo-flo. Replicas were picked up on Luxel grids (LUXEL), and photographed on a JEOL 1400 microscope with attached digital camera (AMT, AMT Imaging in Woburn, MA).

Airyscan Sample Preparation and Imaging

Isolated mouse glomeruli were membrane extracted in the presence of Alexa 564–phalloidin (Invitrogen) as actin stabilizer. After the membrane extraction, glomeruli were fixed and mounted on a poly-L-lysine-coated slide with the assistance of a super magnet. Samples were either imaged directly or further stained with mouse-anti-nestin (DSHB, #Rat-401, 1:20) and chicken anti-vimentin antibody (Novus, #NB300–223, 1:500). The secondary antibodies used for this study were: Alexa 488–labeled anti-mouse IgG1 and Cy5-labeled anti-chicken antibodies (Jackson ImmunoResearch). Samples were imaged using the Airyscan 880 (Zeiss) and volume rendered using Fiji and Amira software. This experiment was repeated three times.

Ultrathin-Cryo Sectioning and Immunogold Sample Preparation

Membrane-extracted glomeruli were placed on filter paper and immersed in 2.3 M sucrose to gradually remove the remaining water. The filter papers holding glomeruli were then transferred onto specimen pins before being frozen and stored in liquid nitrogen. The use of filter paper helped to move numerous glomeruli as a bulge, while also allowing them to sit at the tip of the specimen pin. Ultrathin cryosections were cut at 240-nm-thickness with a Leica EM-FC6 cryo-ultramicrotome equipped with a diamond knife. Sections were collected on carbon-coated coverslips that had been newly glow-discharged at 2×10⁻² mbar.

For immunogold labeling, ultrathin cryosections were rinsed briefly in PBS to remove the sucrose, incubated in a quenching solution containing 50 mM lysine, 50 mM glycine, and 50 mM NH4Cl-PBS for 30 minutes, followed by a quick rinse in PBS, and blocking in a 1% bovine serum albumin-PBS solution for another 30 minutes, all performed at RT. The sections were then incubated at 4°C overnight with anti–gamma actin (clone 2A3; MABT824, Sigma-Aldrich) and anti-vimentin (NB300–223, Novus Biologicals) separately at 1:200 dilution and 1:500 dilution, respectively. Detection of the primary antibody binding sites was achieved with an incubation in 18 nm colloidal gold goat anti-mouse (115–215–146, Jackson ImmunoResearch) and 12 nm colloidal gold donkey anti-chicken (703–205–155, Jackson ImmunoResearch), both diluted at 1:15, for 30 minutes at RT. After 3×5-minute washes in PBS, the sections were postfixed with 0.5% glutaraldehyde–PBS for 30 minutes.

London Resin White Embedded Samples and Post-Embedding Immunogold Labeling

Membrane-extracted glomeruli containing magnetic beads were collected at the bottom of a 0.5-ml microfuge tube containing 0.5 ml of PBS with the aid of a super-magnet, which was utilized in all further collection steps to avoid centrifuging. Conventional London Resin (LR) White embedding for immunocytochemistry was performed as described: (https://www.emsdiasum.com/microscopy/technical/datasheet/14380_LR_white.aspx), with the omission of tannic acid/phosphotungstic acid staining. After overnight curing at 60°C in an airtight chamber, samples were sectioned at 60 nm thickness, and then transferred by a hair tool to a small drop of dH2O atop a 20×40 mm² carbon-coated and glow-discharged coverslip. The coverslip was then placed atop a 60°C heat plate until the water drop evaporated, and sections adhered tightly to the coverslip. The 5×5 mm² chips containing sections were cut from the coverslip with a diamond scribe and processed for actin and vimentin immunogold labeling as described for the ultrathin cryosections. After secondary gold labeling, sections were washed 3×5 minutes with PBS, followed by 3×5 minutes dH2O, water wicked away, and glass chips air dried, respectively. Sections on glass were then floated off onto a dish of hydrofluoric acid, and transferred to a petri dish of dH2O containing a loopful of Photo-Flo 200 solution. Sections were picked up on slot grids and post-stained with 4% uranyl acetate.

Platinum Replica TEM Imaging

Ultrathin-cyro sections stained with actin and vimentin were prepared for platinum replication after a previously published protocol with minor modifications.^42,43 Cryosections adhered to carbon-coated glass were fixed in 2% glutaraldehyde for 1 hour, and then stored in PBS at 4°C. Before replication, samples were rinsed in 3×5 minutes exchanges of dH2O, 3×5 minutes exchanges of 100% ethanol, followed by 2×1 minute exchanges of hexamethyldisilizane. Cover slips were then placed on a piece of filter paper to dry. Dried samples were rotary shadowed with 4 nm platinum and 4 nm carbon in a Leica Ace900 vacuum evaporator.

FIB-SEM Sample Preparation

FIB-SEM sample preparation and imaging were performed as described.⁴⁴ Samples were fixed in 2.5% glutaraldehyde and 2% paraformaldehyde in 0.15 M cacodylate buffer containing 2 mM CaCl₂, pH 7.4, overnight at 4°C. The samples were then stained as described.44 In brief, samples were rinsed in cacodylate buffer 3×10 minutes each, subjected to a secondary fixation for 1 hour in 2% osmium tetroxide/1.5% potassium ferrocyanide in cacodylate buffer for 1 hour, rinsed in ultrapure water 3×10 minutes each, and stained in an aqueous solution of 1% thiocarbohydrazide for 1 hour. Samples were then again stained in aqueous 2% osmium tetroxide for 1 hour, rinsed in ultrapure water 3×10 minutes each, and stained overnight in 1% uranyl acetate at 4°C. Samples were then washed in ultrapure water 3×10 minutes each, and en bloc stained for 30 minutes with 20 mM lead aspartate at 60°C. After staining was complete, samples were briefly washed in ultrapure water, dehydrated in a graded acetone series (50%, 70%, 90%, 100% ×2) for 10 minutes in each step, infiltrated with microwave assistance (Pelco BioWave Pro, Redding, CA) into Durcupan resin, and cured in an oven at 60°C for 48 hours.

FIB-SEM Imaging

After resin curing, samples were exposed with a razor blade, and 70-nm-thick sections were prepared on silicon wafer chips. These chips were then adhered to SEM pins with carbon adhesive tabs, and large areas were imaged at high resolution in a FE-SEM (Zeiss Merlin, Oberkochen, Germany) at 5 KeV and 3 nA using the ATLAS (Fibics, Ottowa, Canada) scan engine to tile resin sample surface and identify regions of interest (ROI) for further analysis. Once ROI were identified, resin blocks were mounted onto SEM pins with silver epoxy and sputter coated with 6 nm of iridium (Leica ACE 600, Vienna, Austria) at a 45° angle with rotation on a planetary stage to ensure the entire block was coated.

Two ROIs showing different glomeruli were chosen for FIB-SEM. Sample blocks were loaded into the FIB-SEM (Zeiss Crossbeam 540, Oberkochen, Germany), and the ROI were located by secondary electron imaging at 5 KeV and 900 pA. Once a ROI was found, the sample was prepared using the ATLAS (Fibics, Ottawa, Canada) 3D nanotomography routine.⁴⁴ In short, a platinum pad was deposited on a 20 µm×20 µm ROI at 30 KeV and 1.5 nA. Three vertical lines for focus and stigmation and two angled lines for Z-tracking were milled into the platinum pad at 300 pA, then filled with carbon at 50 pA to fill the tracking/alignment marks, followed by an additional deposition of a protective platinum pad at 1.5 nA. A rough trench 40-µm-wide and 25-µm-deep was then milled at 30 nA and polished at 7 nA. Once polished, face detection, focusing, and Z-tracking were all performed on the fiducial marks. Serial block-face imaging was performed at 2.5 KeV and 900 pA using the SE2 and ESB detector with a grid voltage of 1050 V. The block was milled at a current of 300 pA, with 10 nm slices and images were acquired at a resolution of 10 nm/pixel with a dwell of 4 µs and a line average of 5 for a total Z-depth of about 15 µm. The stack of acquired images was aligned using Atlas 5 (Fibics, Ottawa, Canada).

Post-Imaging Analysis and Manual Segmentation

Image processing and data analysis were performed using a Data Workstation with Dual Intel Xeon E5–2650 2.3Ghz 10-Core Hyper-threaded Processors and 256 GB DDR4–2133MHz RAM in the Washington University Center for Cellular Imaging. For manual segmentation, selected ROI were cropped from the stack using Fiji/ImageJ and segmented using Amira software (v9.3). Using an interactive tablet system (INTUOS Pen Tablet medium), the GBM, nuclei, foot processes, and podocyte cell body were selected manually on each slice, then assembled into a 3D object.

Machine Learning Model and Supervised Segmentation

To segment the GBM and podocyte actin cytoskeleton from FIB-SEM image stacks, Fiji/ImageJ (version 1.53C) was utilized with the Waikato Environment for Knowledge Analysis (Weka) trainable segmentation plugin (version 3.2.34).^45,46 Before training the model, features were selected. For edge detection features, the Sobel filter, Hessian, and Difference of Gaussian options were selected. For texture description features, the Structure and Neighbors options were selected. For noise removal and edges detection features, Gaussian blur and Membrane projections were utilized.⁴⁶

To train the machine learning models, we generated independent trained classifiers for each dataset. For each classifier, five elements were classified and assigned to the selected region on the image stack via the graphic user panel. These five elements represented “actin,” “GBM,” “background,” “microtubules,” and “gaps,” where “gaps” indicated the peripheral fading boundary between the podocyte actin cytoskeleton and the “background” in the image. The development and training of machine learning models for each of the four conditions were on the basis of an ROI within FIB-SEM image stacks. The training ROI was 251 slices at 587×218 pixels. The classifiers were trained on the basis of the ROI and the selected features, one for each of the image stacks. The classifiers were then applied to the ROI to generating probability maps.

Because results revealed the central actin cable was connected to the slit diaphragm, additional training was required to segment one from the other. For that we added one more “connection” class for the slit diaphragm structure, followed by machine training on the basis of that plus the other five classes listed above, using the procedures described. The trained models are available on request from the authors.

All probability maps were projected into 3D models using Amira software (Thermo Fisher Scientific, version 2019.4), with each probability map having one channel for each class. The final channels used for 3D visualization were the “GBM” and “actin” channels, resulting in low 3D visualizations of the structures of the GBM and podocyte actin network. Additional details are provided in the Supplemental Material and Supplemental Methods.

Statistical Analysis

For the microfilament diameter measurements, multiple high magnification QF-DEEF images were used. A two-tailed t test was used to compare two groups. P=0.05 was considered significant. Values are reported as mean±SD. All raw data are provided in the Supplemental Material.

Results

Membrane-Extraction Technique and SEM Imaging of Podocytes

To visualize the podocyte cytoskeleton, we adapted a method for cytoskeleton stabilization and membrane extraction.³⁵ During the membrane-extraction procedure (Figure 1) we exposed isolated glomeruli at RT to the membrane-extraction solution containing two types of nonionic detergents (i.e., Triton and saponin) and microfilament-stabilizing agents that prevent depolymerization of the actin network (phalloidin, 10 µM) and microtubules (Taxol, 10 µM). To verify this approach, we used SEM to compare intact (Figure 2, A and B) and membrane-extracted healthy glomeruli (Figure 2, C and D). We observed a clear lack of membranes in the extracted samples, but substantial preservation of cytoskeletal networks within the major processes (Figure 2D, arrows) and foot processes (Figure 2D, arrowheads; Figure 2, E and F, arrows), and the slit diaphragms (Figure 2, D and F, asterisks) and GBM (Figure 2, E and F). Furthermore, around the podocyte nucleus, we observed microfilaments covering the areas where the major processes and the foot processes branch (Figure 1G, arrows).

Figure 1.

Schematic diagrams show the experimental workflow for the membrane-extraction technique. (1) After anesthesia, mice were perfused through the heart with 4-µm magnetic beads for 3 minutes, after which the kidneys were harvested. (2) (A) Kidneys were minced and (B) passed through a 100-µm cell strainer. (C) Glomeruli were collected by pelleting them in a tube with the assistance of a magnet. (3) Glomeruli were incubated with the membrane-extraction solution for (A) 1.5 minutes before (B) fixing them for 10 minutes in PFA containing membrane-extraction solution. (4) Membrane-extracted glomeruli were used for the appropriate imaging modality.

Figure 2.

SEM of intact and membrane-extracted podocytes show the exposed actin network of the interdigitating foot processes (FPs) after cell membrane removal. (A and B) Low and high magnification SEM images of an intact healthy glomerulus show four podocytes with major processes (MPs) and interdigitating FPs. (C and D) Low and high magnification SEM images of a membrane-extracted healthy glomerulus show a capillary loop with the cytoskeletal filaments of MPs and FPs. Note the podocyte nucleus in the bottom right corner in (C). The zoomed-in image in (D) shows the MPs (arrows), the FPs (arrowheads), and the slit diaphragms in between (asterisks). Note the porous nature of the slit diaphragms. (E and F) Low and high magnification SEM images of a membrane-extracted and fractured healthy capillary show the endothelial cell cytoskeleton (EC-Cytoskeleton, arrows in E), the GBM and FPs (indicated in F) with the slit diaphragms in between (asterisks in F). (G) Low magnification SEM of a membrane-extracted healthy podocyte shows the cell body, the MPs and the FPs. Note the microfilaments surrounding the cell body (arrows). (H and I) QF-DEEM and TEM micrographs of a membrane-extracted healthy capillary wall show the different types of microfilaments in the podocyte FPs and MPs. The microfilaments in the MPs appear as long filaments running above the FPs (arrows in H and I), whereas the FPs contain bundled microfilaments in the QF-DEEM (arrowheads in H) that appear as electron-dense structures on the TEM images (arrowheads in I). (J) Low magnification QF-DEEM image of a membrane-extracted glomerulus shows an en face view of interdigitating FPs (arrows). Note the slit diaphragm areas (asterisks). (K) High magnification image of the boxed area in (J) shows details of the microfilament bundles in the FPs (cyan box, short arrow) and the loose thicker microfilaments above (yellow box, long arrow). Inserts show high magnification views of the microfilaments. (L) Quantification of the thicknesses of the individual filaments in the FPs (represented by the cyan box in [K]) and the long filaments nearby (represented by the yellow box in [K]) are consistent with the first being actin and the latter being intermediate filaments. t test: ****P<0.001.

To gain more insights about the observed structures, we imaged the membrane-extracted glomeruli using two different types of EM approaches, the QF-DEEM (Figure 2H) and TEM (Figure 2I). These techniques yielded similar results, with excellent preservation of cytoskeletal structures in foot processes and across the GBM. Interestingly, both approaches showed the major processes are filled with microfilaments (Figure 2, H and I, arrows) that are different to the foot processes (Figure 2, H and I, arrowheads). Furthermore, en face images of the foot processes as imaged by QF-DEEM platinum replicas confirmed the presence of approximately 7-nm-diameter filaments arranged as bundles in the foot processes (Figure 2, J and K, cyan box area, see insert) consisting of uniform globular subunits with polar orientation, which are characteristic features of assembled actin.^37,47 The thicker filaments with a diameter of approximately 10 nm are presumably intermediate filaments, on the basis of size, structural stability, and frequent proximity to the nucleus (Figure 2, J and K, yellow box area, see insert). Considering the variable deposition of the 2-nm platinum layer evaporated onto the samples resulting from the angle of evaporation and the 3D topography of the specimens, variable increases in filament diameter are expected. This variability is consistent with diameter measurements of approximately 7 nm for actin and approximately 10 nm for intermediate filaments (Figure 2L, Supplemental Table 1). From these observations, we conclude that the observed filaments are likely to be actin and intermediate filaments, respectively.⁴⁸

Immunostaining of Filaments in Membrane-Extracted Glomeruli

To confirm the protein composition of the microfilaments observed by SEM, QF-DEEM, and TEM (Figure 2), we used fluorescence labeling for super-resolution imaging and immunogold staining for EM to attempt to identify both actin and intermediate filaments.

To visualize actin, we first used Alexa Fluor 536–labeled phalloidin as an actin stabilizer during membrane extraction, and processed the glomeruli for Airyscan 3D super-resolution imaging using glass-bottomed dishes.⁴⁹ Z-stack imaging of the membrane-extracted glomeruli (Figure 3, A–F, Supplemental Video 1) showed intense phalloidin labeling at in the capillary wall areas (Figure 3, Ca, blue arrows) and in the mesangial cells. At the glomerular capillary walls, super-resolution imaging distinguished between two types of phalloidin staining, an intense exterior and a faint interior staining (Figure 3, A–F, Supplemental Video 1), which suggests the former belongs to the podocyte foot processes layer and the latter to the endothelial layer of the capillary walls (Figure 3, A–F, white arrowheads).

Figure 3.

Immunolabeling of actin structures in membrane-extracted glomeruli. (A–C and D–F) Z-stack Airyscan super-resolution imaging of a whole membrane-extracted glomerulus using Alexa546-Phallodiin as a stabilizer in the extraction procedure. Superficial (A–C) and deeper (D–F) optical sections of the membrane-extracted glomerulus show phalloidin staining at the glomerular capillary wall areas (A–F, cyan arrows, Cap) and at the periphery of the podocyte cell bodies. Note the phalloidin-positive structures imaged en face in two podocyte cell bodies (green arrows, Pod). (B and E) DIC images show the podocyte cell bodies and the capillary loop areas. (G–H) 3D reconstruction of the Z-stack images in (A–F) shows one podocyte cell body (boxed in G). (H and I) Two 3D views of the reconstructed podocyte cell body (Supplemental Video 1) show the actin cables surrounding the podocyte cell body. (J–M) TEM images of LR-White ultrathin sections of a membrane-extracted glomerulus labeled with an antibody against γ-actin. (J) An en face image of podocyte MP and interdigitating FPs shows actin immunogold labeling only in the FPs (arrows). (K) TEM micrograph shows a capillary wall with the FPs atop the GBM labeled with actin immunogold. Although the microfilaments in the MPs are not labeled, this micrograph demonstrates the electron dense areas in the extracted FPs are actin positive. Note the labeling in the endothelial fenestrae area as well (arrowheads). (L and M) Two high-magnification TEM images show the actin staining in the central actin cables (CA) in the FPs and in the peripheral actin cables (PA) that reach to the slit diaphragm areas. (N–Q) Micrographs showing HMDS-Platinum replica electron microscopy (HMDS-EM) images of membrane-extracted glomeruli labeled with anti–γ-actin immunogold nanoparticles (yellow enhancement). (N) En face HMDS-EM image shows membrane-extracted podocyte FPs labeled with actin immunogold. Note that the slit diaphragms (SDs) bridging adjacent FPs and the GBM areas are both negative for actin immunoGold labeling. (O–Q) Low- and high-magnification HMDS-EM images show membrane-extracted FPs labeled with actin immunogold nanoparticles. (O) Although no staining of GBM, SDs, or intermediate filaments (IFs) was observed, FPs were labeled for actin. (P and Q) High magnification images of the boxed areas in red and yellow in (O) show actin immunogold nanoparticles labeling the actin filaments in the FPs. SDs (asterisks).

Interestingly, Z-stack imaging (Figure 3, A–C) and 3D reconstruction (Figure 3, G–I; Supplemental Video 1) showed the podocyte cell bodies (pod, green arrows) are surrounded with phalloidin staining that is linked to the actin staining at the capillary wall areas, which can clearly be observed in the deeper optical sections (compare Figure 3, A–C with Figure 3, D–F). These data confirm the actin network in the membrane-extracted glomeruli is indeed intact. Our data also suggest the actin network in the foot processes is connected to the periphery of the podocyte cell bodies.

To directly identify the actin filaments at the EM level, we applied immunogold staining using both LR-White and ultrathin-cryosections of membrane-extracted glomeruli. LR-White sections stained with an antibody against γ-actin shows clear staining in the electron dense areas in the foot process, with very little staining in the surrounding major processes and the GBM areas (Figure 3, J and K, arrows). Zoomed-in images show the anti-actin stained both the central actin cables at the tip of the foot processes, and the sides, and often could be observed near the slit diaphragm areas (Figure 3, L and M, arrows), suggesting the presence of actin cables connecting both (i.e., the central actin cables and the slit diaphragms). Consistent with the phalloidin labeling (Figure 3, A–F), we observed some actin staining in the endothelial cell fenestrae (Figure 3, J and K, arrowheads).

Similarly, we stained ultrathin cryosections of the membrane-extracted glomeruli with γ-actin and generated platinum replicas, which allows us to observe the type of microfilaments that are stained with the immunogold nanoparticles. As expected, only the microfilaments in foot processes and major processes were labeled (Figure 3, N and O; note the lack of staining of GBM and intermediate filaments in the major processes). We often observed immunogold staining at the microfilaments connecting the foot processes to the slit diaphragm (Figure 3, P and Q). Collectively, these data show an uncharacterized actin network that surrounds that podocyte cell bodies, which links podocytes to the actin network in the foot processes. It also shows the podocyte foot processes are connected to the slit diaphragms via side actin cables, which in turn are connected to the central actin cables in the podocyte foot processes.

Because healthy mouse podocytes express the intermediate filament proteins nestin and vimentin,^50,51 we used antibodies against these two proteins followed by Airyscan imaging. Z-stack imaging showed that both nestin and vimentin are present in the podocyte cell bodies and in the major processes, and they colocalize in most of the observed areas (Figure 4, A–D; Supplemental Figure 1).

Figure 4.

Immunolabeling of IFs in membrane-extracted glomeruli. (A–D) Airyscan images show membrane-extracted glomeruli stained with phalloidin (red), anti-nestin (green), and anti-vimentin (cyan). (A–B) and (C–D) are two different optical sections of the same glomerulus showing partial colocalization of vimentin and nestin in the podocyte cell bodies (Pod) and MPs (arrows) but not in FPs of the capillary walls (Cap). (E and F) TEM images of a LR-White ultrathin section of a membrane-extracted glomerulus labeled with an antibody to vimentin. Low (E) and high (F) magnification TEM images show that the microfilaments in MPs but not in FPs are labeled with anti-vimentin immunogold nanoparticles. (G and H) Low and high magnification HMDS-EM images show a membrane-extracted podocyte labeled with anti-vimentin immunogold nanoparticles (yellow enhanced particles). These images demonstrate that only the microfilaments in MP contain vimentin, and they are therefore IFs.

Because these proteins colocalize, we decided to use vimentin for immunogold EM. Similar to the actin staining (Figure 3), we used anti-vimentin antibody and immunogold to label ultrathin sections of membrane-extracted glomeruli for both LR-White and platinum replicas for EM. Consistent with the Airyscan images, we observed vimentin staining only in the podocyte cell bodies and the major processes but not in the foot processes, in both LR-White (Figure 4, E and F) and hexamethyldisilizane-EM platinum replicas (Figure 4, G and H).

Imaging of Membrane-Extracted Podocytes by FIB-SEM

Because SEM only allows visualization of surface topology, we sought a method to view the whole depth of the GFB. We chose FIB-SEM, which generates a large image stack in the Z-direction (Supplemental Figure 2), thus providing a view of the structures located deep in the tissue.²⁴ Imaging membrane-extracted glomeruli with FIB-SEM showed excellent preservation of glomerular structures, including basement membranes (GBM and Bowman’s capsule) and nuclei (Figure 5, A and B; Supplemental Videos 2 and 3).

Figure 5.

FIB-SEM imaging of a healthy membrane-extracted glomerulus. (A) FIB-SEM overview image shows a healthy glomerulus with intact nuclei and GBM. Note the lack of cell membranes and the electron dense actin cables in FPs. (B) Zoomed-in area of the yellow box in (A) shows a capillary wall with intact GBM and FPs. Note the continuity between the FPs and the adjacent SDs (arrows). (B’ and B’’) Two high magnification images taken from the boxed areas in (B) demonstrate the connection between the FP CA and the SD areas (asterisks) via the side actin cables (SA, orange arrows). (C–E) 3D segmentation of 150 frames of the capillary loop in the white boxed area in (A). (C) Top, (D) side, and (E) bottom views of the segmented area demonstrate intact GBM and show the interdigitated FPs.

Detailed analysis of the FIB-SEM stacks of various capillary wall areas revealed that the central actin cables in the foot processes (Figure 5, B, B’, and B’’, central actin cables) are connected with the neighboring foot processes (from hereon called side actin cables) via slit diaphragms (Figure 5, B’ and B’’, asterisks), forming a continuous structure (Figure 5B, Supplemental Video 4). Using Amira software, 3D visualization (Supplemental Figure 2) of the various podocyte structures showed the actin cables in foot processes interdigitated with one another in a similar fashion to the intact foot processes (Figure 5, C–E).

Manual segmentation of the different structures in the glomeruli (Figure 6, A and B; Supplemental Figures 2 and 3, boxed area) allowed for a more accurate 3D visualization of the GBM areas (Figure 6, C–G) and the actin structures within the foot processes (Figure 6, H–K). Manual segmentation of the actin cables within two neighboring foot processes in different colors (Figure 6H) revealed the actin cables interdigitate similar to the intact foot processes (Figure 6, I–K), suggesting a high level of structural preservation during the membrane-extraction procedure. Collectively, this 3D visualization proves that membrane extraction faithfully preserves the structures observed by TEM and SEM, and actin cables in foot processes and nephrin-based slit diaphragms form one continuous structure.

Figure 6.

FIB-SEM 3D visualization of a membrane-extracted healthy glomerulus by manual segmentation reveals the detailed architecture of the capillary wall. (A and B) One ROI for segmentation of the actin network (boxed area in Supplemental Figure 3) with the two GBM segments colored red in (B). (C and D) Top (C) and side (D) views of the 3D visualization of the GBM in (B) after manual segmentation (approximately 150 frames). (E–G) Three different views of the segmented GBM areas superimposed with an orthogonal image from the FIB-SEM image stack. (H) Manual segmentation of both the podocyte cell body and the actin cables of neighboring podocytes atop the GBM (highlighted in green and blue). The nucleus is yellow and the electron dense patches at the periphery of the cell body are pink. (I) and (J) Two different 3D visualizations of manually segmented GBM, FPs, and cell body. (K) 3D visualization of manually segmented FPs with an orthogonal image from the FIB-SEM image stack. This 3D image demonstrates that the actin cables in the adjacent FPs are interdigitated.

Actin Cables Within the Podocyte Cell Body

One interesting observation in the FIB-SEM stacks of membrane-extracted glomeruli is the presence of peripheral, electron-dense patches around podocyte cell bodies (Figure 7, A and B, red arrows, boxed area in Supplemental Figure 4, Supplemental Video 2) and major processes (Figure 7G, red arrows, Supplemental Video 3). Surprisingly, when we segmented those patches in the FIB-SEM stacks and rendered them in 3D, they were revealed as thick cables, running in parallel to one another along the longitudinal axes of the cell body (Figure 7, B–F), and around the major processes (Figure 7, G–K, and Supplemental Video 5).

Figure 7.

Manual segmentation of peripheral actin cables surrounding a podocyte cell body and MPs. (A) Overview FIB-SEM single-frame image of a membrane-extracted glomerulus shows two podocyte cell bodies with electron dense patches (red arrows) at the periphery of the cell bodies. (B) One frame of the ROI (boxed in Supplemental Figure 4) shows an example of a peripheral actin patch highlighted for manual segmentation. The peripheral electron-dense patches surrounding the cell bodies is highlighted in magenta, and the GBM is highlighted in white. Two adjacent FPs are highlighted as well, in green and cyan. (C) 3D reconstruction of the segmented patches in (B) shows that they are parts of thick parallel actin cables along the longitudinal axis of the podocyte. (D, E, and F) 3D renderings of the podocyte’s peripheral actin cables and the adjacent GBM show the position of these longitudinal cables relative to the GBM. (G) A FIB-SEM slice of a membrane-extracted podocyte shows a podocyte with two MPs extending to form FPs that attach to two different capillary loops. Similar to the cell bodies, this image demonstrates that the MPs are also surrounded by peripheral electron-dense areas (red arrows). (H–K) Overview 3D visualization (H and I) and zoomed-in views (J and K) of the segmented areas in (G) show thick, branched actin cables around the MPs (red arrows). For orientation, white arrows indicate the openings of the capillaries. Scale: tickmarks in the x- and y-axes are in nm.

These data are consistent with the Z-stack Airyscan imaging that shows phalloidin staining around the cell body, and the major processes are surrounded with phalloidin-positive actin cables (Figure 3, Supplemental Video 1). Importantly, the observed actin structures are reminiscent of the synaptopodin-positive patches that we previously described using STORM, which disappear after podocyte injury.⁴⁹

Applying a Machine Learning Model for 3D Image Segmentation of Podocytes

Because segmenting FIB-SEM image stacks manually is time consuming and prone to user bias, we applied a machine learning tool that allowed us to segment the GBM and the actin structures surrounding it using the trainable Weka plug-in in Fiji/ImageJ software.⁴⁶ In glomeruli, machine learning and GBM segmentation in the training stack (15 frames) were straightforward, but accurate identification of actin cables in the foot processes required multiple rounds of training (see the Methods for details). Applying this approach to a training stack from a set of images (Figure 8A, Supplemental Methods) generated a trained model for segmentation of the GBM and the actin assembly in the foot processes (Figure 8B) that was then applied to a larger image stack. Applying this to a FIB-SEM stack of 251 images (Supplemental Movie 4) generated an extensive probability map of the two classes of interest (Figure 8C) that subsequently was used to generate a 3D model of the GBM and actin cables in foot processes (Figure 8, D–F; Supplemental Video 6). Moreover, by adding one more class for the slit diaphragm, we trained imaging stacks from the glomerulus (Figure 8G) that allowed segmentation of actin in foot processes from the slit diaphragm (Figure 5B). The probability map generated (Figure 8I) was used to generate a 3D model for the GBM, the central actin cables, and the slit diaphragms (Figure 8, J–L; Supplemental Video 7). Collectively, these data indicate a successful machine learning model to segment podocyte actin structures.

Figure 8.

Machine learning-mediated segmentation of FIB-SEM images of a membrane-extracted healthy mouse glomerulus reveals the architecture of the capillary wall. (A) Inverted single-frame FIB-SEM image of a membrane-extracted healthy glomerulus shows the GBM and the CA of the FPs (arrows). (B) Example of the training process for deep learning segmentation of the image in (A) shows five classes assigned to a selected training stack (15 images). (C) The probability map of FP (green) and GBM (red) channels shows the result of the deep learning classifier. (D) Front view of the 3D visualization of approximately 250 FIB-SEM segmented images showing interdigitating podocyte FPs on top of the GBM. (E and F) Bottom and flipped views of the 3D visualization in (D). GBM, yellow; FPs, silver. (G) Inverted single-frame FIB-SEM image of a membrane-extracted glomerulus shows the GBM, the CA of FPs, and the SDs. (H) Example of the training process for deep learning segmentation of the image in (G) shows six classes assigned to a selected training stack (15 images). (I) The probability map of FP (green), GBM (red), and SD (blue) channels shows the result of the deep learning classifier. (J) Front view of the 3D visualization of a podocyte shows the FP with SDs. (K and L) Bottom and flipped views of the 3D visualization in (J). GBM, yellow; FPs, silver; SDs, light blue. Scale: tickmarks in the x- and y-axes are in nm. (M) Schematic summary diagram shows the layout of the actin cables and IFs within the podocytes. (1) Although IFs (cyan lines) are restricted to the cell body and the MPs, actin cables are connected to the SDs (red lines) via the SA (green and magenta lines) that are connected to the CA within the FPs (block dots) and to the actin cables within the MPs (black lines), which in turn connect to one another and to the actin cables at the periphery of the podocyte cell body (thick magenta lines). (2) Representative single frame image of the FIB-SEM stacks demonstrates the actin cables at the periphery of the cell body (magenta). Other structures are highlighted as well: the podocyte cell nucleus (yellow) and adjacent FPs (green and cyan). (3) Representative single frame image of the FIB-SEM stacks shows the actin assembly within the FPs. SA (green arrows) connect the CA (white arrows) to the adjacent SDs (red arrows).

Discussion

We report a new method to visualize the actin and intermediate filament networks of podocytes in their native environment. This approach has provided interesting and important new insights into the podocyte’s ultrastructure:

• (i)The central actin cables within foot processes are connected to the slit diaphragms. This is consistent with previous observations that predicted this connection biochemically.^52,53 It is interesting that the mild detergents used here did not disrupt the slit diaphragm assembly, in agreement with the finding that nephrin and podocin are solubilized only when the actin cytoskeleton is disassembled.⁵²

The actin cables in foot processes with the slit diaphragms form one continuous, mesh-like, electron-dense sheet that covers the GBM in healthy cells. This network should act as a unit capable of resisting the pulsating forces generated across the capillary wall from both intracapillary blood pressure against the wall and the shear stress of fluid flow through the filtration slits. Our observations are consistent with the recently proposed GBM compression model,⁵⁴ in which foot processes connected by slit diaphragms act together to counterbalance the hydrodynamic forces across the GFB; the resulting compression of the GBM leads to pore sizes that restrict albumin’s passage.

Recent experimental modeling has shown that mutations in slit diaphragm genes such as NPHS2, which encodes podocin (Pod^R231Q/A286V), weakens the GFB, and over time would lead to proteinuria. Importantly, the glomerular capillary walls of those mutant mice were prone to dilation when the mice were exposed to hemodynamic changes that lead to an increased glomerular filtration rate (such as angiotensin II injection).^55,56 This is consistent with our finding that the actin cables in foot processes and the slit diaphragms form one continuous, mesh-like sheet. We predict that any disturbances in this layout would weaken the overall ultrastructure of the podocyte and the slit diaphragms, which eventually would prevent the compression necessary to maintain the proper GBM pore sizes⁵⁴ and lead to proteinuria. In such a mouse model, it would be interesting to use the membrane-extraction technique to directly observe the connections between the actin network in the foot processes and the adjacent slit diaphragms.

• (ii)Podocyte cell bodies and major processes are reinforced by thick actin bundles along their longitudinal axes. On the basis of (1) similarities between actin cables imaged by FIB-SEM (i.e., viewed in two dimensions [2D] as actin patches, Figure 3) and synaptopodin-positive patches in the podocyte cell body and major processes observed by STORM⁴⁹; and (2) the localization of myosin IIA high in the podocytes (only in the cell body and major processes⁴⁹), we speculate that these actin bundles are contractile and generate tension necessary for maturation of podocyte adhesive complexes at the GBM, analogous to tension thresholds for adhesion of other cells.^57–59 On the basis of STORM imaging,⁴⁹ we believe these thick actin cables are important for the proper force distribution inside the podocytes, because they disappear in injured podocytes in various mouse models of nephrotic syndrome, such as Lamb2 KO, Cd2ap KO.⁴⁹ Here, it would be interesting to apply the membrane-extraction technique to these models and models for FSGS such as Inf2 KO and Actn4 KO.

Although we tried to minimize potential artifacts due to detergent extraction by fixing quickly and washing extensively, we did observe some vacuolization in the major processes and cell bodies. Yet the cytoskeletal network was undisturbed, perhaps because we stabilized the microfilaments using phalloidin and Taxol. Despite the 1-µM Taxol, extensive microtubule networks were not observed, possibly suggesting any attachment between the microtubules and the remaining cytoskeletal elements was relatively weak after fixation. However, the network of intermediate filaments suggests they were stable and well connected to the actin cytoskeleton, as recently suggested.⁶⁰ Although we show a great level of overlap between the two components of the podocyte intermediate filament network, vimentin and nestin, further studies are needed to discern their exact roles in healthy podocytes.

This study is the first of its kind to use a machine learning approach in combination with FIB-SEM. Because manual segmentation of podocytes requires time and introduces bias,^23,24 a machine learning approach is valuable. Among the many approaches available to facilitate segmentation of biologic samples by machine learning,⁶¹ we chose the open-source Weka in Fiji/ImageJ⁴⁶ because of its simplicity, speed, and flexible use of filters. When performing machine learning segmentation, we had to choose whether to create one single classifier that roughly segments all classes of all datasets, or to create separate classifiers that segment each dataset independently.⁶¹ Given the complexity of our datasets, applying one machine learning classifier model to all models resulted in inaccuracies. Our machine learning achieved satisfactory segmentation results by generating classifiers for each dataset. Weka has been previously used to segment the GBM of TEM images⁶² using similar filters (i.e., random forest). In that study, the authors used Weka for segmenting the GBM in 2D images, which can vary significantly between images and lead to a less accurate generation of the TWS classifier that did not fully recognize the GBM in all of the 2D single images. To overcome this problem, the authors used custom-designed classifiers, applied in MATLAB, for segmenting areas that Weka could not segment (specifically, GBM areas that were morphologically different). In contrast, our study generated a high degree of accuracy when segmenting the GBM areas due to the uniformity of the GBM throughout the image stacks. Overall, our approach allowed for supervised segmentation of podocyte structures with high accuracy versus manual segmentation, with the former validated by choosing smaller datasets from each condition and confirming results manually. The machine learning approach provided a fast, open source, and reliable 3D reconstruction of large FIB-SEM image stacks. This method should have applicability to cells in diverse organs in patients where actin cytoskeletal structures are dependent on the cells’ mechanical microenvironment.

Disclosures

G. Genin reports having consultancy agreements with Grantová Agentura eské Republiky (The Czech Science Foundation) 2020, People’s Republic of China, including participation in a Ministry of Education sponsored Changjiang (Yangtze River) Professorship and a State Council–sponsored Thousand Talents Professorship (2015 to present), Research Grants Council of Hong Kong 2021, Swiss National Science Foundation 2021, and the University of Sydney 2020; reports being hired as a Scientific Expert by Agence Nationale de la Recherche (France) 2020; reports having an ownership interest in Caeli Vascular; reports receiving honoraria from the University of Pennsylvania and Xi'an Jiaotong University; and reports being a scientific advisor or member of ACS Biomaterials Science & Engineering, Acta Mechanica Sinica, Bioengineering, Biophysical Journal, Biophysics Reports, Frontiers in Mechanics of Materials, Frontiers in Physics, International Journal of Mechanical Sciences, Mechanics of Soft Materials, and the Royal Society Open Science. J. Miner reports having consultancy agreements with Alpha Insights, AstraZeneca, Bridge Bio, Deerfield Management, GLG Council, Janssen Biotech, Inc., Kurma, Mantra Bio, Retrophin, The Planning Shop, and the National Institutes of Health; reports receiving research funding from Chinook Therapeutics and Reneo Pharmaceuticals; reports receiving honoraria from the Japanese Society of Pharmacology, NephCure Kidney International, University of Kansas Medical Center, and Western Michigan University Medical School; reports having patents and inventions with Angion, Eli Lilly, Genentech, Kerafast, and Maze Therapeutics; reports being a scientific advisor or member of Journal of Clinical Investigation Consulting Editor, Kidney International Editorial Board, Matrix Biology Editorial Board, and the Matrix Biology Plus Editorial Board; and reports other interests/relationships with the Alport Syndrome Foundation (Scientific Advisory Research Network) and the American Society for Matrix Biology (President-Elect). All remaining authors have nothing to disclose.

Funding

This work was supported by American Heart Association grant 17SDG33420069 (to H.Y. Suleiman), National Institutes of Health grants P30DK020579 (to H.Y. Suleiman), and R01DK058366 and R01DK078314 (to J. Miner), NSF Science and Technology Center for Engineering MechanoBiology grant CMMI 1548571), and the Office of the Vice Chancellor for Research at Washington University. The Washington University Center for Cellular Imaging is supported by Washington University School of Medicine, The Children’s Discovery Institute of Washington University and St. Louis Children’s Hospital (CDI- CORE-2015-505 and CDI-CORE-2019-813), the Foundation for Barnes-Jewish Hospital (3770 and 4642), and the Washington University Diabetes Research Center (National Institutes of Health P30 DK020579). Washington University Center for Cellular Imaging microscopes were purchased with support from the Office of Research Infrastructure Programs, a part of the National Institutes of Health Office of the Director under grant OD021629.

Supplemental Material

This article contains supplemental material online at http://jasn.asnjournals.org/lookup/suppl/doi:10.1681/ASN.2021020182/-/DCSupplemental.

Supplemental Methods. Machine learning: Trainable Weka Segmentation (TWS) approach for 3D view of the membrane-extracted podocytes.

Supplemental Figure 1. Separate channels of intermediate filament immunostaining in membrane-extracted glomeruli.

Supplemental Figure 2. The three methods used to visualize the FIB-SEM image stacks of membrane-extracted glomeruli.

Supplemental Figure 3. FIB-SEM overview slice of a membrane-extracted glomerulus shows the ROI corresponding to the image in Figure 6A.

Supplemental Figure 4. FIB-SEM overview slice of a membrane-extracted glomerulus shows the ROI corresponding to the image in Figure 7A.

Supplemental Table 1. Microfilaments’ diameter measurements.

Supplemental Video 1. 3D reconstruction of Airyscan Z-stack of healthy glomerulus labeled with fluorescent phalloidin.

Supplemental Video 2. Overview WT FIB-SEM image-stack.

Supplemental Video 3. WT FIB-SEM image-stack of the major processes.

Supplemental Video 4. WT FIB-SEM image-stack of the actin cables in the foot processes.

Supplemental Video 5. 3D visualization of the actin cables in the major processes.

Supplemental Video 6. 3D segmentation of WT FIB-SEM image-stack showing actin cables in the foot processes.

Supplemental Video 7. 3D segmentation of WT FIB-SEM image-stack showing the slit diaphragms between the foot processes.