Volumetric Reconstruction of a Human Retinal Pigment Epithelial Cell Reveals Specialized Membranes and Polarized Distribution of Organelles

Abstract

Purpose: Despite the centrality of the retinal pigment epithelium (RPE) in vision and retinopathy our picture of RPE morphology is incomplete. With a volumetric reconstruction of human RPE ultrastructure, we aim to characterize major membranous features including apical processes and their interactions with photoreceptor outer segments, basolateral infoldings, and the distribution of intracellular organelles.

Methods: A parafoveal retinal sample was acquired from a 21-year-old male organ donor. With serial block-face scanning electron microscopy, a tissue volume from the inner-outer segment junction to basal RPE was captured. Surface membranes and complete internal ultrastructure of an individual RPE cell were achieved with a combination of manual and automated segmentation methods.

Results: In one RPE cell, apical processes constitute 69% of the total cell surface area, through a dense network of over 3000 terminal branches. Single processes contact several photoreceptors. Basolateral infoldings facing the choriocapillaris resemble elongated filopodia and comprise 22% of the cell surface area. Membranous tubules and sacs of endoplasmic reticulum represent 20% of the cell body volume. A dense basal layer of mitochondria extends apically to partly overlap electron-dense pigment granules. Pores in the nuclear envelope form a distinct pattern of rows aligned with chromatin.

Conclusions: Specialized membranes at the apical and basal side of the RPE cell body involved in intercellular uptake and transport represent over 90% of the total surface area. Together with the polarized distribution of organelles within the cell body, these findings are relevant for retinal clinical imaging, therapeutic approaches, and disease pathomechanisms.

The retinal pigment epithelium (RPE) plays a prominent role in the biology of vision and the pathophysiology of age-related macular degeneration (AMD), the major blinding disease of aging.¹ The RPE is a promising treatment target, including replacement therapy, in AMD and other retinal diseases.^2,3 RPE cell biology, physiology, and biochemistry has been intensively studied.⁴ Notably, RPE cells are metabolic gatekeepers to the photoreceptors and form the outer blood-retina barrier.^5–7 Apical RPE participates in outer segment renewal, and basal RPE interacts with the circulation. Relative to other mammalian cells, RPE has unique properties of non-spherical melanosomes, reverse apical-basal distribution of Na⁺ K⁺ ATPase, and expression of a glucose transporter that is prominent at blood-neural barriers.^8–10 

Although major functions and morphological features of RPE are recognized, several outstanding questions would benefit from new information at the ultrastructural level, as foundational studies on human retina are now over 40 years old. In part, knowledge gaps are due to scarcity of tissue that is adequately preserved for subcellular detail while also still attached to photoreceptors. In previous studies, we analyzed the arrangement of pigmented, autofluorescent, and reflective organelles important for clinical imaging by reconstructing image stacks produced by serial block face scanning electron microscopy (SBFSEM).^10,11 In combination with other comprehensive visualization techniques,¹² these data provided new insight into the separate origins, lifecycles, and spatial distributions of lipofuscin and melanolipofuscin (herein called electron-dense organelles).¹³ Similarly, the anatomy of mouse RPE has been investigated using manual segmentation with some automatation.¹⁴ Meleppat et al. correlated in vivo imaging with ex vivo electron microscopy of the same region and found that RPE in the abca4^−/− mouse had fewer melanosomes and more lipofuscin than wildtype mice.¹⁵ 

We herein extend our prior work on human RPE by combining SBFSEM with deep learning-based automated segmentation, with the goal of visualizing an entire native cell including specialized plasma membranes and additional organelles. A high-resolution 3D view of RPE is desired to visualize how the RPE interacts with photoreceptors. Early studies emphasized a role for apical process in guiding shed disks to the cell body.^16,17 The extension of apical processes far beyond the outer segment tips, as revealed by subsequent studies, suggested additional roles in metabolic exchange.^18,19 Other signature RPE activities are highly coordinated across specific regions along the apical to basal axis.³ For example, through a complex signaling pathway, a light-induced increase in calcium initiated at the apical surface ultimately increases fluid absorption at the basal surface.³ Thus, the distributions of mitochondria and endoplasmic reticulum, each with defined roles in intracellular calcium regulation,²⁰ are of interest. 

Herein, we provide the first 3D picture and quantification of RPE surface membrane specializations in attached retina, including apical processes and basolateral infoldings.^{16,17,21–25} We determined the apportionment of subretinal space, including the interphotoreceptor matrix, a gel-like substance that maintains structural integrity and facilitates nutrient and retinoid exchange.²⁶ We clearly illustrate a dense network of endoplasmic reticulum and a thick basal cushion of mitochondria, in relation to the plasma membrane. Finally, we report for the first time a distinctive spatial pattern of pore complexes on the nuclear envelope of RPE cells.^27,28 

Methods

Tissue Recovery and Preparation

A whole globe of a 21-year-old white male organ donor and vehicular accident victim was recovered within 20 minutes of the termination of life support during organ recovery. The enucleated eye was transected at the limbus, drained of vitreous, placed in warm oxygenated culture medium (Ames’ Medium; Sigma-Aldrich, St. Louis, MO, USA), and maintained in this medium at 37°C. The retina with RPE–choroid intact was dissected from the sclera and isolated from the globe by a cut at the optic nerve head and immersion fixed in 4% glutaraldehyde in 0.1-M phosphate buffer. Under a dissecting microscope, the optic disc, foveal depression, foveal center, arcuate bundles, and major meridians were visually localized. Small parafoveal pieces (approximately 1–2 mm in length) were cut and rinsed in cacodylate buffer (0.1 M, pH 7.4) and incubated for 1 hour in a 1.5% potassium ferrocyanide and 2% osmium tetroxide (OsO₄) solution in 0.1-M cacodylate buffer. After washing, the tissue was placed in a freshly made thiocarbohydrazide solution (0.1 g in 10 mL double-distilled H₂O heated to 60°C for 1 hour) for 20 minutes at room temperature (RT). After another rinse at RT, the tissue was incubated in 2% OsO₄ for 30 minutes at RT. The samples were rinsed again and stained en bloc in 1% uranyl acetate overnight at 40°C; they were then washed and stained with Walton's lead aspartate for 30 minutes. After a final wash, retinal pieces were dehydrated in a graded alcohol series, placed in propylene oxide at RT for 10 minutes, and then embedded in a water-soluble epoxy resin (Durcupan; Sigma-Aldrich). To define a region of interest for serial block-face sectioning, semi-thin vertical sections (0.5–1 µm thick) were stained with toluidine blue. 

Sample Processing and Serial Imaging

A region of interest was identified approximately 2 mm from the foveal center, the block was trimmed, gold-coated by standard methods to prevent charging during sectioning, and mounted in a ThermoFisher Volumescope (ThermoFisher Scientific, Waltham, MA, USA). The block face was imaged in four 40 µm × 40 µm tiles (2 horizontal, 2 vertical, and 10% overlap between tiles), including 1 whole and 2 partial RPE cells, and vertically spanning from the inner segment/outer segment junction to below Bruch's membrane. Scanning was performed with a 5-nm x-y resolution, giving a single 40 µm tile dimension of approximately 8000 × 8000 pixels. Sections were cut at 50 nm thickness. The resulting TIFF images were contrast normalized, aligned within and across layers into a volume using TrakEM2 software (plug-in for ImageJ FIJI; National Institutes of Health, Bethesda, MD, USA).²⁹ SBFSEM produces little or no distortion due to sectioning, because the undisturbed block face is imaged. This contrasts with sections that are cut and mounted on a grid for transmission electron microscopy. For this reason, we utilized the TrakEM2 “Rigid” alignment option both within a layer and across layers. This method permitted translation and rotation of image tiles within a layer and layers relative to each other; however, size scaling and shear (affine adjustments) were not allowed. No additional regularization was performed. Once cross-sectional images were aligned, a downsampled volumetric cube measuring 30 µm × 30 µm × 40 µm containing 578 slices was generated within Dragonfly. EM image data are available upon reasonable request. 

Automated and Manual Segmentation

The plasma membranes of the cell body, its apical processes, and basolateral infoldings were manually delineated, reconstructed by three annotators (authors M.L., D.K., and A.S.) in the aligned EM cross-sections with computer assistance using the TrakEM2 plug-in of the Fiji framework and a pen-display (Cintiq 22HDT, Wacom, Kazo, Japan).³⁰ Apical processes were delineated by placing a node in their center on each slice and marking the start and end points. The downsampled volume of this cell was rendered and imported into Dragonfly (Release version 2022.2; Object Research Systems, Montreal, Quebec, Canada) for further analysis. 

To initiate automatic segmentation of individual organelles, cytoplasm, and extracellular space, a small number of single 2D EM images were manually segmented using Dragonfly (out of 578 images, 5 slices were annotated in their entirety and 5 slices were annotated covering approximately 50% of their area). Slices labeled in this way were then used as ground truth for training a convolutional neural network (CNN) within Dragonfly's deep learning toolkit. Because the U-Net deep learning model architecture is well suited for medical image tasks,^31,32 all segmentations were performed with models of this architecture. Training was considered complete when the model failed to improve over multiple training iterations, as measured by performance metrics, such as Categorical Cross Entropy and Validation Loss available through Dragonfly, as well as by visual inspection. The model was then used to segment the entire volumetric EM cube, from which the organelles of the index cell were separated using the above-mentioned manual segmentation as a cropping mask. 

To perform a volumetric analysis of the subretinal space, we trained a separate model to segment apical processes, photoreceptor outer segments, and interphotoreceptor matrix. A 13.5 µm³ cube of this segmentation, spanning from the junction of the photoreceptor inner and outer segments to the RPE actin belt, was then extracted for analysis. This segmentation was used to determine the relative volumes of each component. Separate segmentation models were also trained in the manner described above for the basal membrane surface, nucleus, and endoplasmic reticulum. 

To measure total cell surface area and volume, a 3D mesh generated from manual slice-by-slice contouring of the binary images representing the plasma membrane of the entire cell was imported into Dragonfly. To minimize artifacts introduced by high anisotropy of the sample, we manually set the dimensions of a structured grid of 5 nm × 5 nm × 5 nm to create a high-resolution region of interest (ROI), into which the mesh was imported. To assess the contribution of apical processes to total cell surface and volume, we cropped this ROI to include only the surface apical to the actin belt, a level at which the cell area, viewed en face, measured 433 µm². The actin belt was chosen for reasons of time and repeatability, as it is easily identified in EM cross sections, and cell surface area between apical processes was considered negligible. 

Basolateral infoldings of the index cell were less well aligned than other parts of the cell. Because of this issue, we did not fully trace basolateral infoldings in the same way as apical processes during manual segmentation in Fiji, but only on slices in which they were seen as still attached to the cell body. Therefore, the assessment of their contribution to cell volume is likely an underestimate. For this reason, we estimated their surface area from manual segmentation of a 7.5 × 7.5 µm sampling region of a well-aligned neighboring cell within the same tissue block and used this to calculate the infoldings’ contribution to the index cell's total volume and surface area. The surface area (µm² of surface area per mm² of cell area) was calculated using the area of the index cell at the basal aspect just before basolateral infoldings appear, which measured 319.7 µm². 

The remaining ROI, apically cut at the height of the actin belt, and basally directly before infoldings appear basally, was considered the cell body without any membrane specializations for measurements. Once the individual ROIs were created, their volume and surface area, using Lindblad's method,³³ were computed in Dragonfly's “Scalar Generator” tool. We report total area, computed in a piece-wise manner as described, as a sum of areas of apical processes + cell body + basal infoldings. Percentages of these components are referenced to the total area. 

Results

Reconstruction Overview

Each individual cross-section in this rapidly preserved eye shows a full attachment of the neurosensory retina to the RPE with exquisite preservation of cellular membranes and ultrastructure (Figs. 1A, 1B, 1C). A rod:cone ratio of 3.3:1 (see below) indicates a parafoveal location.^34,35 Figures 1D to 1I show an overview of a single complete RPE cell resulting from a hybrid manual-automated 3D reconstruction approach. Plasma membranes of the cell body including apical processes and basolateral infoldings in their entirety (see Fig. 1E) are shown with internal structures including actin belt (see Fig. 1D), melanosomes (see Fig. 1G), and electron-dense organelles (see Fig. 1H), mitochondria (see Fig. 1I), nucleus (see Fig. 1G), and endoplasmic reticulum (see Fig. 1F). 

Figure 1.

 

Deep learning assisted volumetric reconstruction of a parafoveal human RPE cell. (A) Low magnification vertical EM shows the parafoveal region of a 21-year-old male donor. (B) Magnified view (inset in A) of the outer retina comprising of the photoreceptor-RPE complex. (C) Annotated vertical EM section (approximate location shown in B) show sub-cellular organelles and membranes of a single RPE cell, illustrated in panels D to I. (D) The 3D renderings generated using manual and deep-learning assisted segmentation illustrates the dense packing of organelles enclosed by the RPE plasma membrane (gray). (E) Membrane specializations formed by apical processes and basal infoldings (shown in respective insets). (F) The extra-organellar space is occupied with cytoplasm and endoplasmic reticulum (purple). A ring-like actin belt (green arrowhead) is closely related to a ring of tight junctions with neighboring RPE cells at a slightly apical plane that is not shown here. (G) Spindle-shaped melanosomes (orange) tend to localize in the apical portion of the RPE cell maintaining the trajectory of individual apical processes. The RPE cell nuclei is shown in gray. (H) Electron dense organelles (melanolipofuscin and lipofuscin granules, yellow) occupy the middle portion of the cell body. (I) Mitochondria (magenta) are densely packed near the basal potion of the cell body (closer to choroidal circulation) and less dense toward the apical portion of the cell. Scale bars: panel A = 100 µm; panel B = 50 µm; and panels C and D = 5 µm.

Ultrastructure at different levels of the index cell, as seen in en face sections reconstructed from vertical slices, is highlighted in Figure 2. In this view, the cell is seen as heptagonal, as are approximately 20% of normal human parafoveal RPE.³⁶ A dense meshwork of apical processes and photoreceptor outer segments in various planes localize to the subretinal space (Fig. 2B). Strands of actin frequently appear as small electron dense dots inside apical processes, stretching through the length of processes. Closer to the RPE cell body, apical processes also contain electron-dense organelles (Figs. 2C, 2D). In the top fifth of the cell body, where the cell is widest, junctional complexes (Fig. 2E) are seen as an electron-dense region between neighboring RPE cells associated with a cytoplasmic band of actin filaments. Moving basally, the nucleus and electron-dense organelles first appear (Figs. 2F, 2G) followed by a layer of mitochondria near the basal surface (Fig. 2H). The basal surface has convoluted infoldings of the plasma membrane and substantial basolateral clefts between cells (Fig. 2I). Organelle volume percentages are listed in the Table. 

Figure 2.

 

Cross-sectional views through different levels of the RPE cell. (A) The 3D EM Cube showing approximate location of panels B to I, utilizing reconstructed orthogonal views generated from original EM tiles. (B) Near IS/OS junction, mostly rod and cone outer segments surrounded by some apical processes (C) Slightly lower, some melanosomes can be found inside the numerous apical processes wrapping around outer segments. (D) Directly above the apical border of the cell, a dense mesh of apical processes with pigment granules inside them surround end tips of outer segments. Due to the slightly convex nature of the apical RPE surface, we can also see electron-dense organelles (melanolipofuscin and lipofuscin), which are positioned above the actin belt but not inside apical processes. (E) Positioned directly on the actin belt, which delineates the outer edge of the cell. Melanolipofuscin and lipofuscin granules begin to appear in this area of the cell along with melanosomes. The area of the cell at this level is approximately 430 µm². (F) The apical pole of the nucleus and growing number of pigment granules. (G) The center of the nucleus, with pigment granules arranged circularly around the cell border. (H) Basal half of the cell, showing dense mitochondria and the basal pole of the nucleus. The area of the cell at this level is approximately 320 µm². (I) Basal infoldings directly internal to the basal lamina (and below that, Bruch's membrane and the choriocapillaris).

Table.

 

Morphometry of a Reconstructed Human RPE Cell

 

Apical Processes and Their Interaction With Photoreceptor Outer Segments

We manually tracked single apical processes from their origin at the cell body surface to endings at individual photoreceptor outer segments. We found that hundreds of apical processes extend from a single RPE cell body into the subretinal space, varying greatly in length. These commonly branch, often multiple times, forming extended neural-like processes within the subretinal space, unlike the uniformly stout tufts microvilli seen on airway brush cells.³⁷ In total, we identified slightly over 800 apical processes originating at the index cell's apical surface. Because 85% of them branched at least once along their length, they terminated in over 3000 end-tips within the subretinal space. Average apical process length and diameter (n = 100, ± standard deviation) was 8.77 ± 6.1 µm and 0.245 ± 1.5 µm, respectively. Processes contained electron-dense thin microfilaments resembling smooth endoplasmic reticulum. However, it is important to note that the diameter of a single apical process can vary along its length, as it expands, contracts, and distorts to form dense clusters around outer segments (Figs. 3D, 3F, 3G). Figure 3A shows a 3D visualization of apical processes emerging from the cell body. 

Figure 3.

 

Reconstruction of apical processes achieved through combination of automated and manual segmentation. (A) Rendering of two photoreceptor outer segments (red) and multiple apical processes (green) surrounding them. (B) Single branching apical process contacting two separate rod outer segments. (C) EM section of apical processes within subretinal space not directly contacting outer segments. (D) Elongated apical process with multiple contact protrusions along an outer segment. (E) Several apical processes surround a cone outer segment. White arrowheads in panels (D) and (E) indicate contact sites between apical processes and outer segments. (F) Cone outer segment end tip showing two sheet like apical processes (stars) which continue to contact the outer segment body for several µm. This particular cone was, however, not contacted by the index cell. (G) Another cone outer segment end tip directly above the RPE cell body, surrounded by apical processes. Note the spindle shaped melanosome directly below. (H) Apical processes below cone outer segment tip, as seen looking toward the apical surface from above, the outer segment was semitransparent. Individual processes separated in shades of green. (I) Apical processes around the outer segment tips of cone and rod (semitransparent red). Abbreviations: RPE = retinal pigment epithelium; OS = outer segments; R = rod outer segment; C = cone outer segment. All scale bars = 1 µm.

In EM cross-sections, multiple contact sites between membranes of apical processes and outer segments can be identified (see Figs. 3, arrowheads in 3D, 3E). Apical processes may run parallel to outer segments or encircle them (see Fig. 3F), appearing tubular-shaped or elongated in EM cross-sections. In the cross-sectional view apical processes and outer segments appear bent unlike their vertical orientation in vivo. This distortion is due to compaction during tissue preparation, an underappreciated artifact in many histological studies of outer retina. Multiple apical processes together can form a stockade around outer segments that almost fully separates them from the surrounding interphotoreceptor matrix (Figs. 3H, 3I). Individual apical processes form contact sites with outer segments at several points along their length and may branch to contact multiple outer segments or re-establish a connection with the same outer segment. We also observed outer segments in close contact with apical processes stemming from neighboring RPE cells, however, 55% of apical processes of the index cell contacted only outer segments directly internal to them. We observed that apical processes have heterogeneous branching patterns and can extend in all directions. Figure 3B shows one process branching to contact 2 photoreceptors. In the index cell, 15% of apical processes did not branch, 27% had 1 branch, 16% had 2 branches, and 42% had multiple branches (up to 19). Apical processes from the index cell contacted 39 photoreceptor outer segments originating from 30 rods and 9 cones (3.3 rod:cone ratio). 

Composition of Subretinal Space

The subretinal space is a tightly regulated tissue compartment between the external limiting membrane and the RPE junctional complexes. Here, the extracellular space between RPE and outer segments is filled with interphotoreceptor matrix, a moderately electron dense material in our specimen. Figure 4 shows a 3D view of this space, including a surface created by apical processes perforated by circular gaps that accommodate outer segments and the surrounding interphotoreceptor matrix. Our analysis revealed that the interphotoreceptor matrix occupies 38% of the subretinal space, with outer segments and apical processes occupying 28% and 34%, respectively. This apportionment changes along an axis stretching from the apical surface of the RPE cell to the junction of inner and outer segments (Fig. 4F), where only the matrix is present. The matrix diminishes to near 0% near the RPE cell body, where apical processes are numerous. 

Figure 4.

 

Deep learning-based reconstruction of the Subretinal space. (A) The 3D volume EM cube combined with automated segmentation of apical processes (green), showing the gaps formed by rod and cone outer segments. (B) Rod and cone outer segments. (C) Combination of both apical processes and outer segments. (D) Automated 3D segmentation of subretinal space, showing gaps in the interphotoreceptor matrix left by outer segments and apical processes. (E) Single section EM image, showing automated segmentation (left half) and unmodified (right half). (F) Percentual distribution of subretinal space components in a 13.5 nm³ measurement area spanning from the actin belt to the IS/OS junction. Abbreviations: RPE = retinal pigment epithelium; IS = inner segments. All scale bars = 10 µm.

Spatial Distribution of Electron-Dense Organelles, Mitochondria, and Endoplasmic Reticulum

Almost all electron-dense organelles in apical processes were spindle-shaped, and were considered melanosomes. Similar to Steinberg and Wood's initial findings, we observed 72 such organelles within slim apical processes (Fig. 5A). Within the cell body, however, we could not readily differentiate among the three types of electron-dense pigment granules (lipofuscin, melanolipofuscin, and melanosomes) due to the osmium post-staining protocol. Therefore, we refer to a combined population of electron-dense organelles in Figure 5C to contrast with mitochondria. 

Figure 5.

 

Polarization of pigment granules and mitochondria. (A) Single RPE cell rendered in 3D. Plasma membranes (forest green) with internal mitochondria (magenta), pigment granules (yellow), and melanosomes (orange) and actin belt (neon green). The SBFSEM block it was segmented from shown on the lower half. Scale bar = 10 µm. (B) Cutaway of full RPE cell segmentation embedded in the SBFSEM block. Scale bar = 10 µm. (C) Volume analysis graph showing prominence of electron-dense (ED) organelles (731 µm³), mitochondria (261 µm³), actin belt, and nucleus (379 µm³) along a basal to apical axis. Dashed lines indicate approximate location of apical cell border, actin belt, and RPE basal lamina. (D) The ED organelles and mitochondria distribution in the apical, middle, and basal portions of the cell. The red dashed line in the first column denotes the viewing plane in images in that row. All scale bars = 10 µm. Abbreviations: ED organelles = electron-dense organelles; RPE = retinal pigment epithelium.

In total, the cell body contained 714 electron-dense organelles with a total surface area and volume of 5032 µm² and 731 µm³_, respectively. These strongly localized to the apical half of the cell body (Fig. 5) with far fewer in the basal half. A characteristic circular arrangement of these organelles was best visualized in en face views primarily at the level of the nucleus (see Fig. 2G). Mitochondria are ovoid with electron-dense internal membranes creating shelf-like cristae that group in the cell's basal half, many closely opposed to the basolateral infoldings. However, a few mitochondria were also observed in the apical half (Fig. 5D). In total, we counted 475 mitochondria in the cytoplasm with a surface area and volume of 3463 µm² and 261 µm³, respectively. Figure 5C plots the area occupied by these organelles along an apical-to-basal axis, in which the actin belt represents 0. In addition to three previously described stacked organelle cushions (electron-dense organelles, electron-dense organelles mixed with mitochondria, and mitochondria), this graph shows a plane between the actin belt and the apical surface with few mitochondria and no electron-dense organelles. 

Membranous tubules and sacs in the cytoplasm of the RPE cell body typical of endoplasmic reticulum occupied a fifth of the total cell body volume, with a total surface area and volume of 5321 µm² and 988 µm³, respectively. This network extends throughout the entire cell body, enters the lower parts of some apical processes, and tends to avoid the basolateral infoldings (Fig. 6). We did not distinguish between smooth and rough endoplasmic reticulum, or Golgi apparatus in these reconstructions. 

Figure 6.

 

Automated segmentation of the endoplasmic reticulum. (A) Vertical cross section showing endoplasmic reticulum (ER), that of the index cell highlighted (purple), the actin belt is shown in green. The yellow box designates location of close-ups in B and C, showing cytoplasm with and without automated ER segmentation. (D) RPE cell cut in half showing the abundant ER. (E) Horizontal cut of the same segmentation at the height of the actin belt, gaps formed by pigment granules and mitochondria.

Nuclear Pore Complex Topology

The nucleus is spherical with a total surface area and volume of 232 µm² and 379 µm³, respectively. In en face views, it rests centrally in the cell body, making up 7% of its volume. In cross-sections, it lies slightly below the middle of apical-basal axis. The nuclear envelope consists of a double membrane that functions as a physical barrier between the nucleoplasm and cytoplasm. Basally, this membrane borders the infoldings, directly contacting them in some areas. Transport in and out of the nucleus occurs almost exclusively through nuclear pore complexes, which are organized protein structures distributed across the envelope.³⁸ 

Nuclear pores showed a distinctive spatial distribution appearing in distinct rows along the nuclear membrane. In some instances, rows extended nearly halfway around the perimeter of the nucleus. In other areas, short rows of only a few pores were interrupted by other rows positioned at a different angle (Fig. 7). This pattern appears to follow the underlying chromatin distribution,³⁹ with nuclear pore complexes appearing almost exclusively directly adjacent to euchromatin (electron-lucent areas). 

Figure 7.

 

Nucleus and nuclear pore complex. (A) Automated segmentation of the nucleus reveals a distinct pattern of spaces left by nuclear pore complexes (NPC), appearing in rows that span across the membrane. (B) The EM image (left) of the nuclear membrane, and the same image processed through automated segmentation (right). The pattern of NPC (yellow) appears tied to the distribution of Heterochromatin (red) and Euchromatin (blue) within the nucleus. (C) In the membrane located directly around the nucleolus (nu), pores appear less frequently, however, it does not appear to have an influence on the patterns. (D) Rows of NPCs visible in EM cross sections. (E) The NPCs seen on the basal pole of the nucleus. (F) Cutaway showing one of the two nucleoli (green). (G) Rows of NPC stretching across the entire side of the nucleus. All scale bars = 1 µm.

The nucleoplasm can be broadly split into electron-dense heterochromatin and electron lucent euchromatin. Nuclear components, such as Cajal bodies and various types of speckles, were not analyzed individually but were included in these two classes for simplicity. We found that the nucleoplasm consisted of 31% heterochromatin and 65% euchromatin. Our model was also able to detect the nucleolus, a large structure essential for ribosome biogenesis and easily identifiable in cross-sectional images; the index cell nucleus had two nucleoli. 

Reconstruction of Basolateral Infoldings

Basolateral infoldings allow the transfer of nutrients and retinoids from circulation and eliminate metabolic byproducts along the same route in reverse. We found that basolateral infoldings resemble elongated filopodia originating from a consistent level of the cell body and extending toward the RPE basal lamina (Figs. 8A, 8B, 8C). Infoldings had a median height of 1.97 µm (interquartile range = 1.54–2.69, range = 0.13–8.45). Three-dimensional renderings (Figs. 8F, 8G) revealed a complex and elaborate labyrinth of ridge-like folds (Fig. 8H). The infoldings occasionally contain endoplasmic reticulum (see Fig. 8C). Elaborate, convoluted whorls extended inward as far as the nuclear envelope (Fig. 8E) near to mitochondria and endoplasmic reticulum. In the basolateral clefts between the lateral surfaces of juxtaposed RPE cells (Fig. 8D), membrane-bounded cytoplasmic projections resembled basolateral infoldings. Some extended toward the RPE basal lamina and contained endoplasmic reticulum. The position of these infoldings in the basolateral clefts (see Fig. 8D) suggested roles in transport among neighboring cells.²¹ The physical depth of infoldings, represented by distance from the RPE cell body to the basal lamina, was greater in the basolateral clefts than those under the RPE (Fig. 8I). 

Figure 8.

 

Basolateral infoldings of the RPE. (A) The 3D rendering of the plasma membrane showing infoldings projecting from the basal aspect of RPE cell body (red). (B) Annotations (red) show the polarity of membrane specializations comprising of apical processes (ap) and infoldings in the basal (bi) and basolateral (bc) aspects of RPE. (C) Ultrastructure of basal infoldings show mitochondria (m) and electron-dense organelles (o) sequestered in the basal portion of the cell body. (D) Infoldings originating from the lateral aspects of two adjacent RPE cells occupying the basolateral cleft (blue arrowheads). (E) Elaborate whorl-like membranous complex extending towards the nuclear envelope (n) surrounded by mitochondria (m). Renderings of basal infoldings showing the 3D surface morphology oblique (F) and horizontal (H) views. (I) Horizontal section generated from software projections shows the labyrinth of basal infoldings. (J) Heatmaps show the depth of infoldings (distance from the RPE cell body to the basal lamina). Region of interest (G) corresponds to Panels F, H, I, and J. Solid white lines denote cell boundaries and dotted circles denote the location of nuclei in multiple RPE cells. Note that the infoldings have a greater depth at the basolateral clefts (D indicates region of basolateral cleft shown in Panel D) and whorls (E) compared infoldings under the RPE (C). All scale bars = 2 µm.

Volume and Surface Area

Total cell volume was 5262 µm³, to which apical processes, as measured from above the actin belt, contributed 1982 µm³, or 38%. Regarding surface area, the entire cell body showed a total of 13,114 µm², and apical processes above the actin belt covered an area of 10,885 µm² comprising 69% of the total surface area. Basolateral infoldings also increased the effective surface area for exchange. In a 7.5 × 7.5 µm sampling region, the surface area was approximately 10 times greater (602.3 µm²) than a flat surface of the same area (56.3 µm²). Extrapolating to the cell area at this level (319.7 µm²), this amounts to 3421 µm² and 22% of the total surface area. Considering apical processes and basolateral infoldings together, approximately 92% of the cell surface area is comprised of elaborate membrane specializations. The ratio of surface area in apical processes to basal infoldings is 70% / 22% = 3.2. 

Discussion

Our volumetric reconstruction of electron microscopy images provides a relatively complete view of a human RPE cell. Manual segmentations revealed a tremendous elaboration of surface area. The interface between RPE and outer segments is surprisingly complex, and although non-synaptic, might be viewed as the first microconnectome in vision. Automated segmentation allowed fast and accurate analysis of organelle distribution, a volumetric analysis of subretinal space, and display of a novel pattern of nuclear pores. 

Morphology of the Apical Processes

Our data indicate that the tip of each photoreceptor is surrounded by multiple apical processes (see Figs. 3H, 3I) and not by unitary cup-like “contact cylinders” as often depicted schematically. Further, the apical processes did not differ depending on rod or cone association. Of the nine cones internal to the reference RPE cell, none were contacted by sheath-like processes, as described for cat retina.^40,37 Previous studies of apical processes in relation to outer segments have been fragmentary due to the challenge of adequately preserving such delicate tissues and technical limitations in capturing electron microscopic images. Classic ultrastructural studies analyzed non-foveal areas of surgically excised human eyes (4-69 years),¹⁷ and the foveal cones of macaque monkey.¹⁶ These studies derived 3D structure from single ultrathin sections that might contain only a few apical processes and from scanning electron micrographs of the RPE apical surface after retina removal. Disc shedding is believed to occur close to the RPE cell body,⁴¹ a process now described as trogocytosis, in which apical processes envelope and translocate shed disks to the cell body for ingestion.³⁷ However, our data, like others, show that apical processes extend much further along outer segments, in some cases, almost reaching the inner segments. This suggests that apical processes have functions beyond dismantling outer segment tips, such as retinoid recycling and fluid regulation. 

We found that a given apical process contacts photoreceptor outer segments at multiple sites (see Fig. 3). Some electron dense contacts between apical processes and outer segments were suggestive of junctional complexes¹⁸ (see Fig. 3E). Unlike simple microvilli, such as those found on enterocytes, apical processes vary in length and shape and can branch several times (see Fig. 3B). Indeed, apical processes originating from a single RPE cell body contacted 39 photoreceptors in our extrafoveal sample. Given the precision with which these samples are localized with respect to the steep gradients of photoreceptor density in human central retina, these values suggest that each RPE cell may contact not only outer segments directly internal to it, as commonly assumed, but also a ring of immediately adjacent outer segments. The function of cross-connecting apical processes, if confirmed, remains to be determined. Anderson et al. speculated that the elaborate ensheathment of outer segments increased the surface area for molecular, metabolic, or electrolytic interactions, a concept supported by subsequent research.¹⁶ In addition to proteins responsible for transferring and cycling retinoids for phototransduction,^19,42 apical processes also express glucose and monocarboxylate transporters (GLUT1, MCT1, and MCT3) that regulate glucose and lactate exchange with photoreceptors,^43,44 as well as inwardly rectifying K+ channels important for fluid regulation.⁴⁵ Exchanges of this nature may tolerate nonselectivity of apical processes contacting photoreceptors. 

We estimated that apical processes constitute roughly 70% of the cell's surface area. Automated calculations, as first reported here, may be refined in the future. Automation is sensitive to features like processes that touch and appear fused, which is common in this specimen due to compaction. Nevertheless, our estimated ratio of apical to basal membrane (3.2:1) is very close to a 3:1 ratio previously determined for rat RPE by single-section electron microscopy and direct assay of cell-surface proteins.⁴⁶ The substantial coverage of these specialized structures underscore their critical role in mammalian retinal physiology, particularly in subretinal fluid management and nutrient exchange.^6,47 

The Subretinal Space Contains a High Proportion of Interphotoreceptor Matrix

We found that the interphotoreceptor matrix is the majority component of the subretinal space (see Fig. 4), occupying nearly 40% of this compartment. This specialized and hydrophilic extracellular matrix creates unique microenvironments around cone and rod photoreceptors. Comprised of glycoproteins, proteoglycans, and insoluble glycoconjugates, the interphotoreceptor matrix critically regulates nutrient and retinoid transport and acts as a buffer for the intense molecular activity of phototransduction.⁴⁸ To our knowledge, our study is the first to attempt volumetric quantitation of components in the subretinal space, including interphotoreceptor matrix. Because our tissue block did not extend to the external limiting membrane, we could not assess the entire subretinal space, so more and larger samples are needed. An estimate of interphotoreceptor matrix volume is valuable for at least two reasons. First, this volume can be used to estimate the concentration of molecules released from surrounding cells into it. Second, the human retina appears to have qualitatively more space between the outer segments than common laboratory rodents, by inspection of many publications with electron micrographs.^49,50 Animal models are increasingly studied for potential noninvasive imaging measures of photoreceptor energy utilization, which is dynamically regulated by light in relation to tissue hydration. In the dark, the distance between the external limiting membrane and RPE shrinks,^51–53 and interphotoreceptor matrix proteins redistribute.^48,54 These experiments, if combined with SBFSEM, may provide new insight about the function of the interphotoreceptor matrix and explain mechanisms behind subretinal fluid resorption. 

Organization of Nuclear Pores and Chromatin

Our reconstructions revealed a distinctive array of nuclear pores. These stable protein complexes are responsible for importing transcription factors in, and exporting mRNAs, tRNAs, and ribosomes out, of the nucleus. Nuclear pore complexes are altered in cellular aging and neurodegenerative diseases, like Alzheimer’s disease and amyotrophic lateral sclerosis.^55–57 A complete 3D reconstruction showed that RPE cell nuclear pore complexes are arranged in orderly rows that follow the underlying euchromatin distribution (see Fig. 7). Nuclear pore clustering and patterning has been previously reported,²⁷ and our reconstruction suggests a larger scale organization across the entire nuclear envelope.³⁹ Whereas the functional significance of this unusual patterning is unclear, volume EM of additional samples may clarify if this feature varies with age, sex, or retinal location. 

Surface Area Expanded by Basolateral Infoldings

Our direct measurements indicate that the absorptive area provided by basolateral infoldings is 10-fold expanded over a flat surface, a milestone finding for a sparse literature on these structures. Loss of basolateral infoldings, with the accumulation of extracellular matrix material that either replaces or incorporates them, is a hallmark of retinal aging.^23,50,58 Human eyes at older ages have been illustrated with qualitatively reduced or absent infoldings.^59–61 Our study represents the first attempt to quantify surface area, at any age. Because many channels and transporters localize to the RPE basal surface,^44,62–64 age changes suggest a reduction of essential absorption and transfer capacity. Using both volume and single-section electron microscopy in the mouse, Hayes et al. recently defined three sublayers of basolateral infoldings and demonstrated their dynamism in response to osmotic challenge via low solute concentration.²² These authors also articulated a model of how basolateral infoldings, together with capacitive calcium flux from nearby endoplasmic reticulum and mitochondria, could regulate transcellular water transport via activity at calcium-gated chloride channels.²² 

Implications for Investigating RPE Cell Biology via Clinical Imaging

We extend our previous conclusion that RPE organelles are vertically stacked in partially overlapping cushions (see Fig. 5). Two organelle classes (electron-dense organelles and mitochondria) combine to create three cushions of distinct composition in the cell body; and melanosomes in the apical processes make a fourth. In the current study, which used the plasma membrane and actin belt as anchoring landmarks, we identified a narrow zone in the apical cell body with mitochondria and relatively few electron-dense organelles. This zone may pertain to a hyporeflective band recently shown to be sensitive to light exposure and mitochondrial respiration in humans and mice.⁶⁵ Organelle cushions are also seen by other technologies. Structured illumination microscopy shows that autofluorescent lipofuscin and melanolipofuscin are concentrated in the apical three fourths of the cell body.¹² A circular arrangement of these organelles around the nucleus (see Fig. 2G, Fig. 5) may contribute to a distinctive autofluorescence pattern revealed in vivo by adaptive optics assisted imaging.^66,67 

Our data affirm a distinctive layer of mitochondria in basal RPE cell body that extends upward at lower density into the apical half (see Figs. 5C, 5D).¹⁰ Although this bottom-heavy distribution was seen in the 1960s, the numerosity and dense packing are appreciated best with recent volume electron microscopy.^10,68,69 We can suggest by reference to other cell types that high concentrations of both calcium signals and reactive oxygen species are possible functional consequences of this arrangement. First, in cells of muscles and pancreatic acini, which are also anisotropic like RPE, mitochondria form packed layers. These sequester calcium ions at high levels near the plasma membrane, a configuration called a “mitochondrial firewall.”⁷⁰ We speculate that the RPE's distinctive mitochondrial layer also serves this purpose for the basolaterally located chloride channels important in fluid balance.^64,71 Our demonstration of abundant endoplasmic reticulum converges with the previously reported strong signal for the calcium-binding protein calreticulin in an RPE cell line,²⁴ to underscore the importance of calcium signaling. Second, mitochondria are considered the source of 90% of reactive oxygen species in mammalian cells, due to inefficiencies in the electron transport chain.^72,73 RPE mitochondria could thus represent a high-impact target of antioxidant therapies, long studied for AMD, due to their proximity to drusen, a rich source of peroxidizable lipids. 

We confirmed that RPE-specific spindle-shaped melanosomes are numerous within apical processes adjacent to the outer segments (see Fig. 3).¹⁰ In the current sample, compaction brings electron-dense organelles inside apical processes closer to the cell body than in vivo, where they might be confused with intracellular organelles,⁷⁴ if viewed with less magnification and resolution than achieved here. RPE melanosomes concentrate calcium to remarkably high levels, as evidenced by experimental metabolic tracer and channel pharmacology studies.^63,75 Calcium is a required second messenger in the detection, transduction, and synaptic transfer of light stimuli by photoreceptors. A potential role of melanosomes in photoreceptor calcium regulation remains to be determined. Melanosomes are in vivo optical coherence tomography (OCT) reflectors, as experimentally proven in the frog,⁷⁶ mouse,¹³ and ground squirrels.⁷⁷ The majority of the reflectors in the human RPE are lipofuscin, melanolipofuscin, and mitochondria. All these membrane-bounded organelles could contribute to shadowing of the choroid in healthy eyes. Further, within the cell, anterior melanosomes and electron-dense organelles may shadow the posterior mitochondria.¹³ 

Our new data can launch studies of how membrane specializations contribute to OCT imaging. Importantly, the fully realized apical processes resemble closely the interdigitation zone (IZ),⁷⁸ a feathery reflectivity emanating inwardly from the RPE-basal lamina-Bruch's membrane band and clearly representing more than just the particulate melanosomes therein. The principles by which apical processes add to reflectivity needs further research and may involve redirection of light by the extensive lipid-aqueous interface at cellular membranes. Changes at the apical RPE, including the IZ, have been associated with delayed rod mediated dark adaptation in aging and AMD.⁷⁹ This functional measure of photoreceptor sustenance is rate-limited by retinoid transfer and is thus expected to be impacted by changes at this interface. Basolateral infoldings are relevant to a hyporeflective band visualized in visible light OCT,^76,80 prototype ultra-high-resolution OCT,⁶⁰ and adaptive optics OCT (AO-OCT),⁸¹ in healthy eyes. A hyporeflective band separating hyper-reflective RPE cell body and Bruch's membrane in young adults, attributed to basolateral infoldings, was invisible in persons at mid-life. Intriguingly this band appeared in aged persons and appeared thick in AMD, and thought to represent basal laminar deposits.⁶⁰ The prospects of observing basolateral infoldings over the lifespan and in relation to fluid balance through in vivo imaging are thus good. 

Implications for Management of Retinal Disease

Our data have immediate relevance to clinical management. First, we show that apical processes greatly outnumber outer segments and may contain specialized junctions. Physical interdigitation and adhesion are important passive mechanisms to keep the neuroretina in place.^47,82–84 The neurosensory retina detaches from the RPE in disease and is surgically detached during the delivery of viral vectors for gene therapy. Previous histological studies reported attached remnants of cone matrix sheaths at both outer segments and RPE,^82,85 with breakage close to the RPE.⁸⁶ Clearly there are many points of contact between RPE and outer segments that must be re-established as part of reattachment and successful visual rehabilitation.⁸⁷ Interventions causing a significant disruption will thus require time to re-establish.^88,89 Second, the extent of membrane specializations potentially involved in fluid balance is vast (over 90% of total surface area) and is accompanied by ultrastructural signs of intense calcium signaling capacity. Active mechanisms to keep the retina in place and efflux metabolites include fluid pumping toward the choroid to dehydrate the subretinal space.^90,91 Aberrations in subretinal fluid dynamics and nutrient exchange processes are implicated in conditions like AMD, retinal vascular diseases, and retinal dystrophies.^1,92,93 These extensive membrane specializations could be leveraged for the development of targeted drug delivery systems.⁹⁴ 

Strengths, Limitations, Future Directions, and Conclusions

Strengths of this study include a well-preserved human specimen with an attached retina, and the use of deep learning tools to allow an unprecedentedly comprehensive view of a key retinal cell. Limitations are the focus on one cell thus limiting generalization, the presence of compaction artifact, and the use of volumetric imaging methods with non-isotropic voxels. Of note is that our tissue sample was acquired from an organ donor in an intensive care unit who had suffered irreversible brain damage and likely disruption of circadian rhythm, of importance for photoreceptor disc renewal, and overall health.^95,96 Future studies will extend these methods to other regions in this specimen, other human specimens, and relevant animal models. In conclusion, deep learning assisted volumetric reconstruction of RPE newly illuminate its 3D morphology, expanding knowledge of previously known outer retinal functions and highlighting new areas worthy of further study. 

Acknowledgments

The authors thank Zachary Arsenault for valuable contributions to the deep learning segmentation task. 

Supported by National Institutes of Health (NIH) Grant R01EY028282, and P51 OD010425 from the NIH Office of Research Infrastructure Program to the Washington National Primate Research Center, and EY01730 to the Vision Research Core at the University of Washington (D.M.D.); institutional support to the Department of Ophthalmology and Visual Sciences from Research to Prevent Blindness, Inc., and EyeSight Foundation of Alabama (C.A.C.). 

Disclosure: M. Lindel, None; D. Kar, None; A. Sedova, None; Y.J. Kim, None; O.S. Packer, None; U. Schmidt-Erfurth, None; K.R. Sloan, None; M. Marsh, None; D.M. Dacey, None; C.A. Curcio, Heidelberg Engineering (F), Genentech/Hoffman LaRoche (F), Apellis (C), Astellas (C), Boehringer Ingelheim (C), Character Biosciences (C), Osanni (C), Annexon (C); A. Pollreisz, Zeiss Meditec (F), Hoffman LaRoche (F), Genentech/Hoffman LaRoche (C), Bayer (C), AbbVie (C), Oertli Instruments (C)