Illuminating Mouse Renal Proximal Tubule Architecture through High-Resolution Volume EM and Machine Learning Analysis

Visual Abstract

Abstract

Key Points

• High-resolution 3D imaging reveals new features of proximal tubule ultrastructure that suggested mechanisms for regulating kidney function.

• Our studies illuminate novel connections between membranes of the endoplasmic reticulum, plasma membrane, and apical endocytic compartments.

• The endoplasmic reticulum in proximal tubule cells has subdomains characterized by proteins involved in distinct biochemical functions.

Background

Kidney epithelial cells perform complex vectorial fluid and solute transport at high volumes and rapid rates. Their structural organization both reflects and enables these sophisticated physiologic functions. However, our understanding of the nanoscale spatial organization and intracellular ultrastructure that underlies these crucial cellular functions remains limited.

Methods

To address this knowledge gap, we generated and reconstructed an extensive electron microscopic dataset of mouse renal proximal tubule epithelial cells at isotropic resolutions down to 4 nm. We used artificial intelligence–based segmentation tools to identify, trace, and measure all major subcellular components. We complemented this analysis with immunofluorescence microscopy to connect subcellular architecture to biochemical function.

Results

Our ultrastructural analysis revealed complex organization of membrane-bound compartments in proximal tubule cells. The apical endocytic system featured deep invaginations connected to an anastomosing meshwork of dense apical tubules, rather than discrete structures. The endoplasmic reticulum (ER) displayed distinct structural domains: fenestrated sheets in the basolateral region and smaller, disconnected clusters in the subapical region. We identified, quantified, and visualized membrane contact sites between ER, plasma membrane, mitochondria, and apical endocytic compartments. Immunofluorescence microscopy demonstrated distinct localization patterns for ER resident proteins at mitochondrial and plasma membrane interfaces.

Conclusions

This study provides novel insights into proximal tubule cell organization, revealing specialized compartmentalization and unexpected connections between membrane-bound organelles. We identified previously uncharacterized structures, including mitochondria–plasma membrane bridges and an interconnected endocytic meshwork, suggesting mechanisms for efficient energy distribution, cargo processing, and structural support. Morphologic differences between 4 and 8 nm datasets indicate subsegment-specific specializations within the proximal tubule. This comprehensive open-source dataset provides a foundation for understanding how subcellular architecture supports specialized epithelial function in health and disease.

Introduction

Our understanding of the cellular structures that create the scaffold for kidney function has been limited by the lack of a rich portrait of the three-dimensional (3D) design of renal epithelial cells. The proximal tubule is characterized by densely packed microvilli forming the brush border to maximize apical surface area, an elaborate endocytic apparatus for protein reabsorption, extensive basolateral membrane infoldings for solute transport, and abundant mitochondria supporting its high metabolic demands. These cells are further specialized along the proximal tubule length, with S1 and S2 segments showing distinct organizational patterns. The complexity of the subcellular architecture of renal tubule epithelial cells has been visualized by scanning electron microscopy (SEM)¹ and is suggested by elegant experiments performed almost 40 years ago, in which en bloc osmium staining was used to reveal the distribution of the darkly stained endoplasmic reticulum (ER) in each cell type of the rat renal tubule.^2–5 These studies showed that in the proximal tubule, the S1 and S2 segments contain ER composed of both fenestrated saccules and tubular structures. The fenestrated saccules span the entire cell height and are particularly well-developed within basolateral membrane infoldings, while tubular extensions reach toward the apical brush border. Notably, these studies highlighted specialized features such as close contacts between fenestrated saccules and lateral plasma membrane, especially near tight junctions, and extensive interactions with mitochondria.

Until recently, high-resolution images of complex tissues such as the kidney could encompass only small volumes. Technologic advances in focused ion beam SEM (FIB-SEM) have enabled 3D imaging of large biologic specimens, including kidney tubules and glomeruli, with remarkable precision, achieving isotropic voxel resolution as fine as 4 nm.^6–9 When combined with artificial intelligence–based segmentation tools, this technology allowed for precise identification, tracing, and measurement of subcellular components and their interactions.^7–11 Enhanced FIB-SEM allowed relationships between structure and function in the cells of complex tissues to be explored and was especially valuable in studying membrane contact sites (MCSs). These include ER–plasma membrane contacts¹² and ER–mitochondria contacts,¹³ which facilitate lipid and calcium transport. Studies in hepatocytes have demonstrated that these contacts are both functionally significant and dynamically regulated.^14,15 The organization and functional properties of such contacts have just begun to be elucidated in renal epithelial cells.^16–18

We have performed enhanced FIB-SEM analysis at 4 and 8 nm resolution of the mouse proximal tubule. To analyze the resultant datasets, we have created neural network models that enable 11 cellular structures to be distinguished, traced, and quantitated, including the basolateral and apical plasma membranes, the ER, mitochondria, endosomes, lysosomes, the Golgi complex, and dense apical tubules (DATs). To define further the functional properties of the subcellular structures that the FIB-SEM analysis reveals, we determined the localizations of compartment-specific markers through immunofluorescence microscopy. These studies demonstrate that the highly developed endomembranous networks of renal proximal tubule epithelial cells exhibited spatially defined, biochemically distinct domains that appeared to subserve different physiologic roles. These studies thus provide new and previously inaccessible insight into the relationships between renal proximal tubule epithelial cell structure and function.

Methods

FIB-SEM Sample Preparation

All experimental procedures adhered to the guidance set forth by the Institutional Animal Care and Use Committee of Janelia Research Campus, under protocol number 21-0206. Fixative-perfused kidneys from an 8-week-old C57BL/6 male mouse were used for the study. Detailed methods for sample acquisition, preparation, and imaging are provided in the Supplemental Material. After embedding, two samples (mouse kidney [MK]-8 nm and MK-4 nm) were trimmed to a small block containing region of interest, as previously described.¹⁹ Trimmed samples were coated 10 nm gold, followed by 100 nm carbon. A graphical description of dataset preparation is shown in Figure 1A.

Figure 1.

Kidney sample preparation, imaging, and organelle segmentation for 3D visualization. (A) Schematic of tissue harvest, preparation, imaging, and deep learning training for visualization. See Methods for details. (B) Overview of the proximal tubule section MK-8 nm imaged in 8 nm resolution in the x, y, and z planes; scale bar: 1 mm. (C) Overview of the proximal tubule section MK-4 nm imaged in 4 nm resolution in the x, y, and z planes; scale bar: 5 µm. (D) A single plane of MK-4 nm dataset with ER annotations (in magenta) that represent the ground-truth for training of the model for the ER; scale bar: 200 µm. (E) The same image in (D) with the ground-truth in magenta and the model predictions of the ER in cyan; scale bar: 200 µm. (F) Representative plane of the MK-4 nm dataset with full tissue segmentation. Apical membrane is yellow, plasma membrane is white, ER is green, DATs are pink, endosomes are light purple, peroxisomes are orange, mitochondria are red, lysosomes are black, Golgi are dark blue, nuclei are dark purple, and basement membrane is light blue; scale bar: 2 µm. (G) Representative plane of the MK-8 nm dataset with full tissue segmentation. Apical membrane is yellow, plasma membrane is white, ER is green, DATs are pink, endosomes are light purple, peroxisomes are orange, mitochondria are red, lysosomes are black, Golgi are light blue, nuclei are dark purple, and basement membrane is dark green; scale bar: 5 µm. CT, computed tomography; DAT, dense apical tubule; ER, endoplasmic reticulum; FIB-SEM, focused ion beam scanning electron microscopy; MK, mouse kidney.

FIB-SEM 3D Large Volume Imaging

Two FIB-SEM prepared samples, MK-8 nm and MK-4 nm, were imaged by sequential ablation–imaging cycles using two customized Zeiss FIB-SEM systems as previously described.⁷ Parameters are described in details in the Supplemental Material. MK-8 nm was imaged with x-y pixel resolution set at 8 nm. After each image acquisition, a focused Ga⁺ beam ablated away 8 nm of the surface. The ablation–imaging cycle continued about once every half minute for 7 days to complete the FIB-SEM imaging of MK-8 nm at 8 nm x-y pixel resolution. The acquired image stack was postprocessed as described in Methods and consists of a final isotropic volume of 90×60×177 µm³ at 8 nm voxels, which can be viewed in any arbitrary orientations (Figure 1B and Supplemental Movie 1: https://doi.org/10.6084/m9.figshare.30105073). MK-4 nm was imaged with x-y pixel resolution was set at 4 nm, and the milling thickness was also set at 4 nm. The ablation–imaging cycle continued about once every 3 minutes for 26 days and resulted in an image stack of 50×50×50 µm³ at 4 nm voxels throughout the entire volume, which can be viewed in any arbitrary orientations (Figure 1C and Supplemental Movie 2: https://doi.org/10.6084/m9.figshare.30105073). MK-4 nm was cropped to 29.2×24.8×8 μm³ to focus analysis on the proximal tubule. The MK-8 nm dataset remained uncropped.

Image Analysis

To train custom semantic deep neural networks (DNNs), we used the ZEISS Arivis Cloud deep learning toolkit, which uses a 2D U-Net architecture²⁰ with EfficientNet encoder and PixelShuffle decoder to train custom semantic DNNs. Training data were selected to represent conditions throughout the image volumes, uploaded (as tagged image file formats) to cloud storage, and registered to separate directories for each organelle. Ground-truth annotations for DNN training were made manually based on established morphologic features in the literature. Organelles exhibiting greater morphologic diversity required larger/broader ground-truth images and manual annotations.

Instead of training all classes into a single multiclass model, we trained a separate model for each organelle, with a unique background class and organelle class (Figure 1D), enabling us to refine each model iteratively and use each model separately (or in combination) in analysis pipelines. Using the models as individual building-block operations facilitated the addition of preprocessing and postprocessing operations such as object math, object masking, and morphology when needed.

For the actual network trainings (to find effective data representations through customized convolutional neural networks), annotations and image tiles ran through the standard Arivis Cloud training protocol. The protocol splits ground-truth annotations into two subsets: training data—to train the model, and validation data—to monitor the accuracy of predictions through each training epoch. Training minimized loss between ground-truth and validations, increasing accuracy and improving intersection over union of pixel predictions. Training automatically ended when these metrics plateaued over many iterations of the full dataset (epochs; Supplemental Figure 1). After each training series, we inspected the predictions from each model on numerous images to ensure we achieved high-quality segmentation consistent with the ground-truth (Figure 1E) and the literature. Training yielded models capable of detecting 11 whole cell features. These included whole tissue, ER, mitochondria, plasma membrane, lysosome, endosome, DAT, nuclei, apical membrane/microvilli, peroxisome, Golgi, and basement membrane. The final model for the ER is a combination of two trainings, focused separately on the basolateral ER and apical ER, based on morphologic differences discussed in the results. Organelle segmentation can be seen in representative planes in both the MK-4 nm (Figure 1F) and MK-8 nm (Figure 1G) datasets. For reconstruction of the plasma membrane, a DNN was trained on orthogonal views of the raw data volume and predictions were averaged from multiple directions and processed to improve automatic segmentations of individual cell bodies.

To enhance the segmentation quality for certain classes, we applied simple semiautomated refinements using processing operations in ArivisPro software (ZEISS). These refinements included smoothing, size filtering, masking, watershed, and agglomeration. For other classes, biologic priors and finished classes were used to improve raw predictions. In addition, we derived additional object classes such as contact sites, whole tubule objects, and cell bodies within the software.

Immunofluorescence Staining

Standard methods were used for immunofluorescence. Details are provided in the Supplemental Material. The primary antibodies used are listed in Supplemental Table 1.

Results

Cell Morphology

Isolation of individual cells based on the trained plasma membrane boundaries revealed that proximal tubule cells in the MK-4 nm and MK-8 nm datasets were highly interdigitated with one another, with extensive shared surface area (Figure 2, A–C, Supplemental Figure 2, A–C, and Supplemental Movies 3 and 4: https://doi.org/10.6084/m9.figshare.30105073), similar to previous reports.^21,22 The 3D rendering further showed that individual cells were extremely large, ranging from 1593 to 2567 µm³ and averaging 1980±49 µm³, and spanned across large segments of the tubule diameter. There was an average 5.29-fold increase in surface area observed over that of a simple cuboidal epithelial cell with the same dimensions (mean 5073±726.07 µm² gained). MK-4 nm contained 10 partial cells, while MK-8 nm included 25 whole cells and 19 partial cells.

Figure 2.

Organelle size and density in both MK-4 nm and MK-8 nm datasets. (A) Single plane from the MK-4 nm dataset with one cell segmented in pink; scale bar: 2 µm. (B) 3D reconstruction of all ten cells segmented in the MK-4 nm dataset. Each cell is a different color; scale bar: 2 µm. (C) 3D reconstruction of the MK-4 nm cells with dark blue and pink cells removed compared with (B) to better visualize the large cell in white; scale bar: 2 µm. (D–G) In the MK-4 nm dataset, five of the ten cells imaged include both apical and basal regions and were used for quantification. In the MK-8 nm dataset, 25 cells were used as a representation of the dataset for quantification. (D and E) The mean volumes of organelles were calculated using the ArivisPro Software per cell, with mitochondria being the most abundant in the MK-4 nm (D) and MK-8 nm (E) dataset. Nucleus (11% 8 nm), mitochondria (32% 4 nm, 26% 8 nm), ER (9% 4 nm, 6% 8 nm), peroxisomes (3% 4 nm, 1.0% 8 nm), lysosomes (2% 4 nm, 2% 8 nm), DATs (0.6% 4 nm, 0.7% 8 nm), endosomes (0.5% 4 nm, 0.8% 8 nm), and Golgi apparatus (0.5% 4 nm, 0.7% 8 nm). (F and G) The polarized distribution of organelles was determined as a percentage of volume within the tubule of interest in either the apical or basal region of the cell. Cells were segregated into apical and basal regions bounded by the tips of the mitochondria closest to the apical membrane, which is why mitochondria are not represented in the graphs. As expected, DATs are essentially only found on the apical side, while peroxisomes are present mostly within the basal region of the cell in both the MK-4 nm (F) and MK-8 nm (G) datasets. Mt, mitochondria; Perox, peroxisomes.

Using the plasma membrane boundaries, we quantified the densities of organelles in partial cells in MK-4 nm and in all whole cells in MK-8 nm (Supplemental Figure 2, D and E). Mitochondria, ER, peroxisomes, and lysosomes were preferentially represented in the MK-4 nm dataset, whereas DATs, endosomes, and Golgi apparatus occupied higher fractional volumes in the MK-8 nm dataset (Figure 2, D and E). Analysis confirmed the polarization of some of these organelles toward either the apical or basal regions of the cell (Figure 2, F and G).

Representative images of proximal tubule cell organelles were selected and rendered as 3D reconstructions (Figure 3 and Supplemental Figure 3). These renderings revealed new features of cellular compartments previously unappreciated from conventional ultrastructural analyses, as described below.

Figure 3.

Organelle segmentation and structure in MK-4 nm and MK-8 nm datasets. Segmented cellular organelles and compartments were rendered in 3D to reveal their structural complexity. The left image of each panel shows a representative FIB-SEM plane, the middle image shows the same image with the object segmentation overlaid, and the right image shows the representative object in a 3D view. (A) Apical membrane microvilli and membrane invaginations in the MK-4 nm dataset, with an inset highlighting the apical invaginations; scale bar: 500 nm. (B) Apical membrane microvilli and membrane invaginations in the MK-8 nm dataset, with an inset, highlighting the apical invaginations; scale bar: 50 µm. (C) Early endosome (MK-4 nm dataset); scale bar: 100 nm. (D) Apical vacuoles (MK-4 nm dataset). Each colored object represents a separate endosome; scale bar: 200 nm. (E) DATs (MK-4 nm dataset) form a highly connected network (purple object). The apparently individual DATs represented by other colors may also be connected to this object outside of the planes that are shown; scale bar: 200 nm. (F) Autophagosomes (MK-4 nm dataset) were frequently segmented as endosomes based on their luminal density, but in 3D were observed to include regions having the characteristic crescent shape of forming autophagosomes (blue); scale bar: 100 nm. (G) Lysosome (MK-4 nm dataset); scale bar: 200 nm. (H) Nucleus (MK-8 nm dataset); scale bar: 1 µm. (I) Golgi stacks (MK-4 nm dataset) are located surrounding and above the nuclei. Individual Golgi stacks with connecting transitional zones are represented as distinct colored objects; scale bar: 200 nm for EM, 500 nm for 3D reconstruction. (J) Mitochondria (MK-4 nm dataset); scale bar: 200 nm. Note the connection between three apparently distinct mitochondria when viewed in 3D. (K) Basal-lateral ER (MK-4 nm dataset) lines the plasma membrane and surround mitochondria. This is a section of ER (in 50 planes) in the MK-4 nm dataset, where each color represents a unique ER object, with pink ER being segmented as one continuous object; scale bar: 500 nm. (L) Subapical ER (MK-4 nm dataset) is fragmented and lacks the fenestrated structure of basal-lateral ER. Each color represents a discrete ER object; scale bar: 500 nm, 50 planes. The nuclear envelope was not trained as part of the ER, and as it is visually distinct from the rest of the ER, it was not designated as ER in our training. EM, electron microscopy.

Organelle Morphology and Interactions

Apical and Subapical Regions

Our analysis revealed that proximal tubule cells in both MK-4 nm and MK-8 nm sections contained densely packed microvilli with diameters of approximately 100–120 nm and lengths of approximately 3 µm. We observed many microvilli that were forked, and some microvillus tips contained vesicles (Supplemental Figure 3, A and B, and Supplemental Movies 5 and 6: https://doi.org/10.6084/m9.figshare.30105073). Several proximal tubule cells in the MK-8 nm dataset exhibited large apical extrusions containing sparse distributions of organelles, including ER, lysosomes, and mitochondria (Supplemental Figure 3C). These extensions included microtubule cores similar to cilia and in some cases contacted the apical cell surfaces of cells on opposing faces of the tubule lumen (Supplemental Movie 7: https://doi.org/10.6084/m9.figshare.30105073). Importantly, these apical membranous extensions were consistently present across multiple samples and preparations; their morphology is distinct from typical fixation-induced blebs, suggesting that they represent genuine cellular structures rather than artifacts. Similar structures have recently been reported to facilitate the extrusion of organelles and nanoparticles from proximal tubule cells.²³

Apical endocytosis in proximal tubule cells occurs through clathrin-dependent invaginations that form at the base of microvilli.²⁴ Ruthenium red staining of fixed kidney blocks before sectioning for electron microscopy showed that these invaginations can extend deep into the apical cytoplasm.²⁵ Consistent with this, the MK-4 nm dataset revealed numerous deep invaginations in the apical plasma membrane extending into the subapical region (Figure 3A and Supplemental Movie 8: https://doi.org/10.6084/m9.figshare.30105073). These invaginations displayed notably different morphologies between datasets, appearing narrower in MK-4 nm sections and larger in MK-8 nm sections (Figure 3B and Supplemental Movie 9: https://doi.org/10.6084/m9.figshare.30105073). In both cases, the invaginations branched into distinct tubules and formed contacts with various organelles.

The current model for proximal tubule apical endocytosis posits that invaginated membranes bud from the apical surface and fuse with growing apical early endosomes (AEEs) that are believed to mature into larger apical vacuoles.^26,27 Although discrete AEEs were readily identifiable in the MK-4 nm dataset (Figure 3, C and D), the MK-8 nm dataset showed that these structures predominantly tethered to apical invaginations (Supplemental Movie 9: https://doi.org/10.6084/m9.figshare.30105073).

Recycling of internalized membrane receptors and other proteins is thought to occur through tubules with a characteristic luminal density, termed DATs, that emanate from AEEs and apical vacuoles. Segmentation showed that DATs formed an extensive connected meshwork across the subapical region, rather than existing as individual tubules (Figure 3E). We frequently observed direct luminal contiguity between this meshwork and both AEE-like structures and apical invaginations (Supplemental Figure 3, D and E).

The subapical region contained highly irregular, moderately sized to large apical vacuoles beneath the DAT meshwork. These vacuoles exhibited characteristic granular texture and at times contained intracellular vesicles or membrane fragments from various organelles (Supplemental Figure 3, F and G). We documented instances of apical vacuoles appearing to engulf dense lysosomes (Supplemental Figure 3H) and form autophagosome-like crescent extensions (Figure 3F and Supplemental Movie 10: https://doi.org/10.6084/m9.figshare.30105073). ER and endosomal membranes have previously been demonstrated to contribute to autophagosome formation.²⁸ Supporting this observation, we identified cases of ER appearing to fuse with vacuolar structures (Supplemental Figure 3I).

Finally, degradation of organelles and internalized ligands begins within the acidic interior of multivesicular bodies and autophagosomes, eventually forming hybrid fusions with lysosomes that become increasingly electron-dense as they age.^29–31 We observed electron-dense lysosomes, likely representing terminal compartments, in both subapical and basal cellular regions (Figure 3G). Some of these structures appeared to be undergoing apical extrusion (Supplemental Figure 3J). We also identified electron-dense objects resembling lipid droplets, although their similarity to lysosomes made reliable differentiation challenging through our current modeling approach.

Middle and Basal Cell Regions

In proximal tubule cells, the nucleus visually dominated the subcellular space below the apical region. Although appearing predominantly spherical in single planes, 3D analysis revealed that many nuclei possessed distinctive structural features, including crescent-shaped invaginations and lobular extensions (Figure 3H).

The Golgi complex exhibited a characteristic organization of stacked membrane cisternae, with individual stacks connected by transitional vesicular regions.³² Consistent with typical polarized epithelial cell architecture, these Golgi stacks clustered primarily around the apical aspect of the nucleus. Some stacks extended further into the cytoplasmic extensions that project from the nuclear region (Figure 3I).

The midregion of proximal tubule cells contained numerous peroxisomes, identifiable as spherical compartments with uniform luminal density (Supplemental Figure 3K). We observed and quantitatively determined the presence of a larger number of volume-adjusted peroxisomes in the cells captured in our MK-4 nm dataset as compared with the MK-8 nm dataset (Figure 2, D and E).

Mitochondria dominated the basal and midregions of proximal tubule cells. 3D reconstruction revealed that these organelles formed a densely packed, interdigitated network characterized by remarkable structural diversity, consistent with previous findings.³³ Individual mitochondria exhibited considerable variation in both size and morphology, with many displaying extensive branching patterns (Figure 3J). The density and complexity of this mitochondrial network likely reflects the intense energy demands to which proximal tubule cells are subjected.^34–36

ER Physiology and Architecture

The ER exhibits remarkable structural diversity across cell types, with its architecture responsive to both cell cycle progression and metabolic state.^14,37 Our 3D analysis revealed distinct structural domains of the ER with characteristic subcellular distributions. In the basolateral region, the ER formed extensive fenestrated sheets that frequently established close contacts with both the plasma membrane and mitochondria (Figures 3K and 4, A–D). Although we observed occasional connections between these fenestrated sheets and the nonfenestrated outer nuclear membrane (Supplemental Figure 3L and Supplemental Movie 11: https://doi.org/10.6084/m9.figshare.30105073), our segmentation model—trained specifically on basolateral ER—did not recognize the nuclear envelope as ER, reflecting its distinct properties. Notably, we discovered an unexpected abundance of ER membrane in the subapical region, although this ER population differed markedly from its basolateral counterpart, appearing as smaller, disconnected clusters lacking fenestration (Figure 3L).

Figure 4.

Mitochondria maintain distinct contact sites with both ER and plasma membrane. (A) Using ArivisPro, mitochondria and ER were isolated on a per-cell basis using the plasma membrane objects. A 3D rendering of mitochondria (cyan) and ER (red) are within one cell in a clipped volume; scale bar: 1 µm. (B) The mitochondria and ER objects in (A) were extracted and subjected to 180° vertical axis rotation; scale bar: 1 µm. (C) Organelle segmentation of a representative section of ER (red), mitochondria (cyan), and plasma membrane (white) shown in 3D of the first 20 planes including part of the segmentation in (A), with an higher magnification inset corresponding to the area in the yellow box. Scale bar: 2 µm (left), 500 nm (inset). (D) Representative FIB-SEM plane of the MK-4 nm dataset showing segmentation of ER (red), mitochondria (cyan), and plasma membrane (white), revealing close association among all three structures. Scale bar: 500 nm. (E) High magnification of mitochondria and ER MCSs (red arrow heads) and mitochondria and plasma membrane contact sites (white arrow heads); scale bar: 200 nm. (F) Box plot displaying the distribution of mitochondria interactions with the ER (left) and with the plasma membrane (right), each dot representing a mitochondrion. (G) Scatter plot illustrating the inverse relationship between the percentage of mitochondrial surface area interacting with the plasma membrane and that interacting with the ER. The red line represents a linear regression fit, with the surrounding gray shaded area indicating the 95% confidence interval of the model (Cor=−0.43, R²=0.183, P = 5.84×10⁻¹¹). (H) 3D rendering of mitochondria (magenta) and ER (blue) showing mitochondria–ER contact (yellow) and mitochondria–plasma membrane contact (red); scale bar: 1 µm. (I) The same rendering of mitochondria and contact sites without ER objects; scale bar: 1 µm. MCS, membrane contact site; Mito, mitochondria; PM, plasma membrane.

These observations of close organellar apposition prompted us to examine MCSs more broadly. MCSs are characterized by close membrane apposition between organelles (typically within 40 nm) often featuring visible physical connections.^38,39 This proximity provides a structural prerequisite for direct communication and material exchange between the organelles.^38,40 Analysis of both MK-4 nm and MK-8 nm datasets revealed diverse heterotypic contacts, including interactions between plasma membrane–ER (Supplemental Figure 4A), mitochondria–lysosome (Supplemental Figure 4B), and endosome contacts with various organelles (ER, lysosomes, and mitochondria; Supplemental Figure 4, C and D). We also identified instances of homotypic contacts, such as those between adjacent mitochondria (Supplemental Figure 4E).

ER–Plasma Membrane–Mitochondria MCSs

Our reconstruction of the mitochondria and ER highlighted the dense arrangement of tubular and fenestrated ER surrounding mitochondria in proximal tubule epithelial cells (Figure 4, A and B), as well as extensive sheets of fenestrated ER residing in close apposition to the plasma membrane (Figure 4, C and D). Leveraging high z-axis resolution, we quantified and generated 3D reconstructions of MCSs. MCSs were defined as regions where the distance between the two organelles was 40 nm or less (Figure 4E).

Within a subset of our MK-4 nm that included 679 mitochondria, we found n=65,562 counts of mitochondria–ER contact sites and n=13,529 counts of mitochondria–plasma MCSs. Nearly all of the mitochondria in the sample, 677 (99%), exhibited interactions with the ER, with a median contact size of 1.5 nm². In addition, 216 mitochondria (32%) interacted with the plasma membrane, with a median interaction size of 0.72 nm². Quantification revealed that, on average, 17% of a mitochondrion surface area was within interaction proximity of the ER, while approximately 1% was within interaction proximity of the plasma membrane (Figure 4F).

We found that mitochondria located farther from the basement membrane exhibited a slightly positive association with increased interactions with the ER (correlation=0.35). In addition, the extent of mitochondrial interaction with the plasma membrane was inversely correlated with its interaction with the ER (correlation=−0.43; Figure 4G). This pattern was likely attributable to spatial constraints and proximity to basolateral infoldings. However, the total number of interactions manifested by an individual mitochondrion was not associated with its distance from the basement membrane.

ER–Apical Membrane–Endosome–DAT MCSs

Given the unexpectedly high abundance of ER in the subapical region, we investigated whether the apical ER was closely associated with the apical endocytic apparatus. We documented large patches of close apposition between the ER and endocytic vesicles (Figure 5 and Supplemental Movie 12: https://doi.org/10.6084/m9.figshare.30105073). Using our established criterion of ≤40 nm separation for MCSs, we observed instances where the ER appeared to bisect transitions between apical invaginations and DATs (Figure 5, A–F). Furthermore, we found ER wrapping around apical membrane invaginations and positioned between apical endosomes and DATs (Figure 5, G and H).

Figure 5.

Apical ER is closely associated with endocytic vesicles. (A) Representative FIB-SEM plane of subapical region in the MK-4 nm dataset; scale bar: 200 nm. (B) Same FIB-SEM plane with organelle segmentation of the apical membrane (blue), DATs (orange), endosomes (magenta), and ER (white). ER contact sites are visualized with the apical membrane (yellow), DATs (green), and endosomes (dark purple). The organelle segmentation is for a representative section for 28 planes; scale bar: 200 nm. (C) Organelle segmentation visualized in 3D; scale bar: 200 nm. (D) Same segmentation zoomed in, with an inset to further show ER segmenting DATs and apical membrane; scale bar: 200 nm, zoom in: 100 nm. (E) Same inset in (D) angled slightly to show the ER wrapping around the organelles; scale bar: 100 nm. (F) 3D visualization of the ER with the interactions with the apical membrane and DATs only; scale bar: 100 nm. (G) 3D visualization of ER wrapping around a section of apical membrane with interactions visualized in yellow; scale bar: 100 nm. (H) 3D visualization of ER wrapping around an endosome (magenta) with interactions visualized in dark purple; scale bar: 200 nm. (I and J) ER interactions were calculated using ArivisPro technology, where the ER objects in the apical portion of the cells are within 40 nm of apical membrane, DATs, and endosomes. Percent interactions of each organelle with ER are shown in table format for both the MK-4 nm (I) and MK-8 nm (J) datasets.

We quantified this close apposition using our MCS analysis with the 40 nm threshold. Analysis of the two datasets revealed distinct patterns of apical ER object interactions. In the MK-4 nm dataset (30,235 ER objects), 50% of apical ER objects interact with apical membrane, 48% with DATs, and 30% with endosomes. The corresponding surface area proportions were 40% (±25%) for apical membrane, 26% (±20%) for DATs, and 34% (±24%) for endosomes (Figure 5I and Supplemental Figure 5, A–C). The larger MK-8 nm dataset (319,862 ER objects) demonstrated higher interaction frequencies with apical membrane (65%) and DATs (64%), but lower endosome interactions (12%). Surface area proportions were 45% (±25%) for apical membrane, 19% (±16%) for DATs, and 22% (±19%) for endosomes (Figure 5J and Supplemental Figure 5, D and E). Notably, the MK-8 nm dataset exhibited more frequent ER–apical membrane contacts involving larger surface areas than MK-4 nm, while MK-4 nm showed more extensive ER–endosome interactions (Supplemental Figure 5, A, C, D, and F). These differences between interaction distributions could indicate that the datasets represent different proximal tubule subsegments.

Compartmentalization and Spatial Distributions of Biochemical Functions of the ER

To examine whether the distinctive ER spatial morphology and distribution correlates with variations in ER function, we performed immunofluorescence microscopy on 0.8-µm sections of mouse kidney tissue using antibodies against proteins representing these functions.³⁷ This included the sarco/endoplasmic reticulum calcium ATPase (SERCA) calcium ATPase and calreticulin, which facilitate calcium accumulation in the ER lumen; protein disulfide isomerase (PDI) and the specific amino acid sequence (Lys-Asp-Glu-Leu) that regulates retrieval of proteins into the ER retention sequence shared by GR78, GRP94, and PDI, components of the ER protein folding and quality control machinery; and vesicle-associated membrane protein-associated protein A (VAP-A), which mediates lipid transfer at ER contact sites with other organelles. In addition, we examined Climp63, a structural protein that bridges opposing ER membrane surfaces to form extended sheets.

Our results revealed distinct localization patterns for these ER-resident proteins (Figure 6). The SERCA calcium ATPase and calreticulin exhibited broad distribution within the central regions of the cell, suggesting its presence in ER sheets surrounding mitochondria, which were similarly positioned (Figure 6, A and B). Costaining with an antibody directed against the mitochondrial marker mitofilin demonstrates that the SERCA calcium ATPase and calreticulin were indeed associated with regions of the ER that surround mitochondria (Supplemental Figure 6, A–F). PDI and specific amino acid sequence (Lys-Asp-Glu-Leu) that regulates retrieval of proteins into the ER sequence-containing proteins were concentrated in subapical and supranuclear regions, indicating that protein biosynthesis and folding occur primarily in the nonfenestrated, less-organized ER structures in this area (Figure 6, C and D). By contrast, VAP-A localized predominantly to regions of the ER contacting the lateral plasma membrane and basolateral plasma membrane infoldings (Figure 6E and Supplemental Figure 6, G–I). Climp63, in turn, was exclusively localized to ER regions closely opposed to the lateral plasma membrane, reinforcing its role in forming extended flat ER sheets (Figure 6F and Supplemental Figure 6, J–L).

Figure 6.

Functional compartmentalization of the ER markers in the wild-type mouse kidney. Representative immunofluorescence images depicting the localization of ER proteins (red) in proximal tubule cells, identified by co-staining with the lectin LTL (yellow). The SERCA protein (A), which mediates ER calcium uptake, and calreticulin (B), which participates in ER calcium handling, are associated with elements of the ER that occupy central regions of the cell that are richly endowed with mitochondria. The PDI (C) and KDEL sequence-containing proteins (D), which participate in the maturation and quality control of newly synthesized proteins inserted into or across the ER membrane, are localized around the nuclei and in subapical regions. Both VAP-A (E) and CLIMP-63 (F) are detected in portions of the ER that reside in close apposition to the lateral plasma membrane and to basolateral infoldings. All images were obtained from semithin frozen sections (0.8 µm). Scale bar: 10 µm. A ROI identified by a white square is shown as a 3× digital enlargement below each image (A–F). Scale bar: 10 µm. GRP94, glucose-regulated protein 94; KDEL, specific amino acid sequence (Lys-Asp-Glu-Leu) that regulates retrieval of proteins into the ER; LTL, lotus tetragonolobus lectin (Fluorescein); PDI, protein disulfide isomerase; ROI, region of interest; SERCA, sarco/endoplasmic reticulum calcium ATPase; VAP-A, vesicle-associated membrane protein-associated protein A.

These findings strongly support the conclusion that, despite appearing to form a single continuous network (Figure 3, K and L), the ER in proximal tubule cells consists of functionally distinct subdomains that are spatially specialized for specific roles. To further investigate its architectural organization, we refined our segmentation model to trace ER lumens throughout the dataset. Although the ER surface appeared continuous in our reconstructions (Figure 4, A and B), lumen analysis revealed subdivision into numerous discrete compartments, with some sharing common luminal spaces while others remained isolated (Supplemental Figure 7).

Discussion

Our detailed 3D ultrastructural analysis of proximal tubule cells revealed novel insights into their subcellular organization and suggests new models and testable hypotheses for how these cells manage their extraordinary transport and metabolic demands. The findings highlight specialized compartmentalization and unexpected connections between membrane-bound organelles that likely support proximal tubule cell function.

Our ultrastructural analysis revealed several fundamental insights about proximal tubule endocytic organization: the complex architecture of apical invaginations, a continuous rather than discrete recycling compartment, extensive involvement of ER with cargo processing structures, and a revised understanding of degradative compartments. These findings necessitate updates to the classical model of proximal tubule cell endocytic trafficking.

The entry point for endocytosis—clathrin-coated invaginations between microvilli—has long been recognized. However, our 3D analysis revealed unprecedented details about these structures, showing remarkably tortuous invaginations that extended deep into the subapical region and formed intimate associations with downstream compartments. Most strikingly, we observed that DATs, rather than existing as discrete structures emerging from AEEs, formed an interconnected meshwork with direct connections to both apical invaginations and endosomes. This anastomosing network of DAT tubules was previously observed using SEM and FIB-SEM approaches,^9,41 although connections to apical invaginations were not documented.⁴² The apparent delivery of recycling molecules to internalization sites may represent a necessary adaptation of the endocytic pathway to the spatial constraints of proximal tubule apical membrane structure. Alternatively, these connections between the DAT network and apical invaginations may highlight a previously unappreciated mechanism required to maintain rapid membrane dynamics.

The ER appeared to play multiple roles in this endocytic system. Previous studies in nonpolarized cells have documented ER contact with endosomes and lysosomes, and our findings suggest that such interactions may support the apical endocytic pathway in polarized epithelial cells.⁴³ We observed ER membranes wrapping around both subapical invaginations and DATs, with apparent fusion events between ER and apical endosomes or vacuoles during autophagosome formation. This organization suggested that the ER may help create microdomains that segregate soluble and membrane components, potentially serving as a physical gatekeeper for receptor recycling at apical invaginations. The ER could also contribute to receptor-ligand dissociation, as the calcium-dependent binding of megalin and cubilin could be modulated by ER-mediated calcium dynamics.⁴⁴ Alternatively, the apical ER structure may facilitate efficient delivery of newly synthesized membrane and secreted proteins to the apical surface, perhaps through a Golgi bypass route.⁴⁵

Our findings also revised the understanding of degradative compartments in proximal tubule cells. Rather than representing simple maturation products of early endosomes, structures previously classified as apical vacuoles appeared to be amphisomes—hybrid organelles formed by fusion of late endosomes with ER-derived autophagosomes—when visualized in 3D.⁴⁶ These structures can either degrade their contents or release them through plasma membrane fusion.⁴⁷ This interpretation is supported by our observation of extruding lysosomes and apical blebs containing organelle remnants, suggesting new mechanisms of communication between proximal tubule cells and downstream nephron segments.

We found no evidence of structures associated with endocytic activity at the basolateral surface, with endocytic and exocytic traffic-related structures apparently confined almost exclusively to the subapical region. This observation challenges previous models proposing significant transcytotic trafficking in healthy proximal tubule cells.⁴⁸ The basolateral release of fluorescently labeled degradation products from apically internalized albumin, previously attributed to transcytosis of intact protein,⁴⁹ may instead result from transient fusion of lysosomes with the basal surface.

Our 3D analysis revealed an elaborate network of MCSs in proximal tubule cells, with the ER forming extensive contacts with both mitochondria and the basolateral plasma membrane. These contacts appear specialized for distinct functions: bioenergetic regulation at ER–mitochondria interfaces and structural support at ER–plasma membrane junctions.

ER–mitochondria contact sites are orchestrated by several protein complexes that facilitate calcium signaling. Inositol 1,4,5-triphosphate receptor, voltage dependent anion channel, and Grp75 form bridges between these organelles,⁵⁰ while mitochondrial fusion GTPases MFN1 and MFN2 maintain the physical tethering.⁵¹ This organization may support the high metabolic demands of proximal tubule cells. Calcium transfer from the ER to mitochondria through IP3 receptors, and the mitochondrial calcium uniporter enhances oxidative phosphorylation and ATP production.⁵²

Although ER–plasma membrane contacts typically regulate store-operated calcium entry through stromal interaction molecule–Orai interactions, our findings suggest a different specialization in proximal tubule cells. Immunolocalization studies indicate that store-operated calcium entry predominantly occurs at the apical membrane,⁵³ while we found SERCA calcium ATPase localized primarily to mitochondria-associated ER rather than plasma membrane–associated ER. This unexpected distribution suggests that the extensive basolateral ER–plasma membrane contacts serve alternative functions.

We propose that these ER–plasma membrane contacts may provide structural support against the mechanical stresses of fluid transport. Similar to their role in cochlear outer hair cells,^54,55 these contacts may reinforce membranes experiencing significant strain. Proximal tubule cells process over 100 L of fluid daily, subjecting their lateral membranes to substantial mechanical stress and morphologic deformation.⁵⁶ Supporting this structural role, we found Climp-63, an ER sheet–stabilizing protein, exclusively localized to basolateral plasma membrane–associated ER. In addition, VAP-A, a protein involved in lipid transfer at MCSs, specifically localized to these regions, suggesting these contacts may also facilitate delivery of newly synthesized lipids for membrane repair and maintenance.

Mitochondria–plasma membrane contacts have been described in a number of mammalian cell types.⁵⁷ A particularly interesting finding in this study was the identification of electron-dense MCSs connecting mitochondria directly to the basolateral plasma membrane (Figure 4E). Although similar fibrillar bridges exist in photoreceptor cells,⁵⁸ mitochondria–plasma membrane contacts have only been characterized biochemically in yeast, where Num1 mediates these interactions.^59,60 The observed mitochondrial morphology—highly branched and tortuous—likely reflects the high energetic demands of proximal tubule cells that rely exclusively on fatty acid oxidation and oxidative phosphorylation.^61–63 It is tempting to hypothesize that, as has long been suggested, the observed bridges may ensure efficient ATP delivery to Na,K-ATPase pumps, which drive most proximal tubule fluid and electrolyte transport and consume substantial energy.⁶⁴ The molecular composition of these structures remains to be determined, but their identification opens new avenues for understanding how mitochondrial positioning supports cellular energetics.

The morphologic differences between the MK-4 nm and MK-8 nm datasets, particularly in endocytic invagination structures, likely reflect subsegment-specific specializations of the proximal tubule. In the MK-4 nm dataset, apical endocytic invaginations appeared narrow with occasional bulbous tips, whereas those in the MK-8 nm dataset were larger and more dilated. These differences are unlikely to be artifacts of perfusion or fixation, as the imaged tissue blocks were derived from adjacent regions of the same perfused kidney. These larger endocytic invaginations suggest that the MK-8 nm dataset may represent an S2 segment whereas the MK-4 nm dataset may represent an S1 segment, consistent with reports that fluid-phase markers are taken up more robustly in S2 segments.^65–67 However, other morphometric features presented conflicting evidence. We observed no differences in lysosome volume between our datasets, contrasting with previous reports of increased lysosomal content in S2 segments,⁶⁸ nor in microvillar length, which has been reported to be longer in S1 segments.⁶⁹ Conversely, mitochondria constituted a greater fraction of total cell volume in the MK-8 nm dataset compared with the MK-4 nm dataset, while peroxisomal volume was lower (Figure 2, D and E). Based on previous morphometric analyses, this pattern would suggest that MK-4 nm represents an S2 segment, whereas MK-8 nm represents an S1 segment.^69,70 Without capturing the connection between these tubular segments and their respective glomeruli, we cannot definitively determine the identity of either segment.

The use of FIB-SEM to investigate cell ultrastructure is expensive and requires extensive expertise to prepare, image, and interpret the data, which may limit both its routine use and the comparison of multiple samples. Although our analysis was conducted on tissue from only one healthy male mouse, these datasets provide valuable reference points for future investigations. The apparent compartmentalization of ER lumens into small contiguous subdomains could arise from limitations in our segmentation tool, which may not have detected small patent connections. However, our immunolocalization data supported the existence of distinct functional ER subdomains. We anticipate that making these datasets publicly available will stimulate further explorations into the differences between proximal tubule subsegments and the mechanisms through which MCSs contribute to the physiologic functions of proximal tubule cells. Future studies should also focus on identifying the molecular components of the mitochondria–plasma membrane bridges and on understanding how the integrated organization of these cellular compartments responds to physiologic and pathologic challenges.

Supplementary Material

jasn-37-747-s001.pdf ^{(1.4MB, pdf)}

jasn-37-747-s002.pdf ^{(8.9MB, pdf)}

Disclosures

Disclosure forms, as provided by each author, are available with the online version of the article at http://links.lww.com/JSN/F460.

Author Contributions

Conceptualization: Mariia Burdyniuk, Michael J. Caplan, Emily M. Lackner, Raj D. Pandya, Song Pang, Ora A. Weisz, C. Shan Xu, Christopher Zugates.

Data curation: Emily M. Lackner, Raj D. Pandya, C. Shan Xu, Christopher Zugates.

Formal analysis: Mariia Burdyniuk, Emily M. Lackner, Raj D. Pandya, Andrea Reyna-Neyra, C. Shan Xu, Christopher Zugates.

Funding acquisition: Michael J. Caplan, Ora A. Weisz.

Investigation: Emily M. Lackner, Wei-Ping Li, Raj D. Pandya, Vraj D. Pandya, Song Pang, Andrea Reyna-Neyra, C. Shan Xu, Christopher Zugates.

Methodology: Mariia Burdyniuk, Emily M. Lackner, Wei-Ping Li, Raj D. Pandya, Vraj D. Pandya, Song Pang, Andrea Reyna-Neyra, C. Shan Xu, Christopher Zugates.

Project administration: Michael J. Caplan, Song Pang, Ora A. Weisz.

Resources: Michael J. Caplan, Song Pang, Ora A. Weisz, C. Shan Xu.

Software: Mariia Burdyniuk, Christopher Zugates.

Supervision: Michael J. Caplan, Song Pang, Ora A. Weisz.

Validation: Emily M. Lackner, Raj D. Pandya.

Visualization: Mariia Burdyniuk, Emily M. Lackner, Raj D. Pandya, Vraj D. Pandya, Andrea Reyna-Neyra, Christopher Zugates.

Writing – original draft: Michael J. Caplan, Emily M. Lackner, Raj D. Pandya, Ora A. Weisz.

Writing – review & editing: Mariia Burdyniuk, Michael J. Caplan, Emily M. Lackner, Raj D. Pandya, Song Pang, Andrea Reyna-Neyra, Ora A. Weisz, C. Shan Xu, Christopher Zugates.

Funding

O.A. Weisz: National Institute of Diabetes and Digestive and Kidney Diseases (DK125049 and DK137329). E.M. Lackner: National Institute of General Medical Sciences (5T32GM133353) and National Institute of Diabetes and Digestive and Kidney Diseases (1TL1DK14327). M.J. Caplan: National Institute of Diabetes and Digestive and Kidney Diseases (DK072612). C.S. Xu, W.-P. Li, and S. Pang: Howard Hughes Medical Institute.

Declarative Statements

All animal experiments were conducted in accordance with the NIH Guide for the Care and Use of Laboratory Animals or an equivalent standard that meets or exceeds the ethical and welfare requirements outlined in the NIH Guide. All protocols were approved by the appropriate institutional animal care and use committee.

Data Availability Statements

Data Type: Image Data. Image data of MK-8 nm are openly available hosted on Janelia servers for research use at https://openorganelle.janelia.org. Image data of MK-4 nm are openly available for research use at https://www.ebi.ac.uk/empiar.

Supplemental Material

This article contains the following supplemental material online at http://links.lww.com/JSN/F461, https://doi.org/10.6084/m9.figshare.30105073.

Supplemental Methods

Supplemental Table 1. Antibodies for indirect immunofluorescence in mouse kidney sections.

Supplemental Figure 1. Analysis of the training data of the ER shown in Figure 1.

Supplemental Figure 2. MK-8 nm 3D cell reconstruction.

Supplemental Figure 3. Segmentation of organelles and unique cellular structures in MK-4 nm and MK-8 nm datasets.

Supplemental Figure 4. Heterotypic and homotypic MCSs between organelles.

Supplemental Figure 5. Distribution of ER interactions with endocytic vesicles.

Supplemental Figure 6. ER proteins involved in calcium handling localize to mitochondria-rich regions.

Supplemental Figure 7. ER lumen continuity is more extensive in the basolateral portion of proximal tubule cells.

Supplemental Movie 1. MK-8 nm dataset (https://doi.org/10.6084/m9.figshare.30105073).

Supplemental Movie 2. MK-4 nm dataset (https://doi.org/10.6084/m9.figshare.30105073).

Supplemental Movie 3. MK-4 nm 3D cell reconstruction (https://doi.org/10.6084/m9.figshare.30105073).

Supplemental Movie 4. MK-8 nm 3D cell reconstruction (https://doi.org/10.6084/m9.figshare.30105073).

Supplemental Movie 5. Forked microvilli (MK-8 nm dataset; https://doi.org/10.6084/m9.figshare.30105073).

Supplemental Movie 6. Vesicles within microvillar tips (MK-4 nm dataset; https://doi.org/10.6084/m9.figshare.30105073).

Supplemental Movie 7. Apical membrane extrusions (MK-8 nm dataset; https://doi.org/10.6084/m9.figshare.30105073).

Supplemental Movie 8. Apical invaginations (MK-4 nm dataset; https://doi.org/10.6084/m9.figshare.30105073).

Supplemental Movie 9. Apical invaginations and endosome formation (MK-8 nm dataset; https://doi.org/10.6084/m9.figshare.30105073).

Supplemental Movie 10. Autophagosome (MK-4 nm dataset; https://doi.org/10.6084/m9.figshare.30105073).

Supplemental Movie 11. ER membrane association with the outer nuclear envelope (MK-8 nm dataset; https://doi.org/10.6084/m9.figshare.30105073).

Supplemental Movie 12. Apical ER is closely associated with endocytic vesicles (https://doi.org/10.6084/m9.figshare.30105073).