Deciphering the Connection Between Microvascular Damage and Neurodegeneration in Early Diabetic Retinopathy

Diabetic retinopathy (DR) is a prevalent complication of diabetes and affects approximately one-third of people with diabetes globally (1). One of the earliest recognized pathologies in DR is the localized loss of small capillaries, resulting from the death of endothelial cells and pericytes (2,3). This process gives rise to acellular capillaries, also known as string or ghost vessels in the histopathology literature, consisting of the remaining vascular basement membrane (4). These acellular vessels first were described in human eyes with DR >50 years ago (5) and have also been extensively reported in various diabetes animal models (6–8). They are not perfused (9,10) and, therefore, can serve as a histological biomarker of nonperfusion.

In clinical practice, retinal vascular perfusion traditionally has been evaluated using fluorescent angiography. The advent of optical coherence tomography angiography (OCTA) has provided a less invasive alternative that can detect perfusion across all layers of the retinal vasculature, unlike fluorescent angiography, which is mainly limited to the superficial plexus (11). Although several OCTA studies have identified reduced retinal vasculature perfusion in patients with diabetes, it remains a matter of debate whether the earliest abnormalities manifest in the deeper capillary plexus (DVP) or the superficial capillary plexus (SVP) (reviewed in Supplementary Table 1).

Nonperfusion of the retinal vasculature typically leads to hypoxia and upregulation of vascular endothelial growth factor but can also cause neural cell death and retinal degeneration. Several studies have shown neural cell death and retinal tissue atrophy in patients with DR or retinal vascular occlusions (12–15). However, the widely accepted view that neurodegenerative changes in DR are a secondary consequence of the primary vascular damage has been questioned (16–18). There are several examples illustrating that the reverse (vascular dropout being secondary to neurodegeneration) can also occur. For instance, vessel density loss in the DVP has been observed in cases of retinal pigmentosa, which is well established to be primarily a neurodegenerative disease (19). Similarly, neuronal damage in glaucoma can also lead to vascular changes (20–22).

Furthermore, in patients with diabetes, subtle functional changes, such as those observed in electroretinograms (ERGs), have been detected before the onset of noticeable damage to the retinal vasculature (23). Regional variations in multifocal ERGs have also been found to predict future vascular lesions (24), suggesting that the earliest neuronal defects might occur independently of perfusion defects. Moreover, thinning of the nerve fiber layer has been demonstrated by optical coherence tomography and histologically in people with diabetes and no to minimal DR (25,26).

To investigate links between neural loss and capillary dropout during the earliest stages of retinopathy in human tissue, we collected postmortem eyes from anonymous donors with diabetes without clinical DR and quantified the number of acellular capillaries in their retinal vasculature. This allowed us to identify and study postmortem retinal tissue from donors with diabetes and with histological signs of early-stage DR, based on subtle microvascular defects, in the absence of a clinical diagnosis of DR in the donors.

The study has ethical approval (UK National Research Ethics, Integrated Research Approval System project identifier: 279162) and follows the tenets of the Declaration of Helsinki. All donors consented to participate in the study. Quality control of cross-sections was performed to exclude tissue with poor structural preservation. A total of 14 human postmortem eyes from 14 donors (n = 10 donors with diabetes) were included in the study and were obtained from Moorfields Eye Bank (London, U.K.) and Lions NSW Eye Bank (Sydney, Australia), as shown in Table 1. Eyes were fixed in 4% paraformaldehyde (in PBS).

Donors were anonymous and only limited clinical history information was available. This included whether the donor had any eye diseases (e.g., DR) and whether they had diabetes. To ensure the donors with diabetes did not have undiagnosed DR, we confirmed that the postmortem samples did not show signs of DR (as established by the Early Treatment Diabetic Retinopathy Study [27]), such as microaneurysms, hemorrhages (assessed in postmortem tissue by visual inspection), venous beading, and neovascularization (assessed in whole-mount vascular stains as described next).

The central region of the eye was dissected and trimmed to include the optic disk, fovea, and some nasal and some temporal retinal periphery, as shown in Supplementary Figure 2. Dissected samples were stained with rhodamine-labeled Ulex europaeus agglutinin I (UEA; catalog no. RL-1062, Vector Laboratories; 1:500) diluted in whole-mount blocking buffer (1% FBS, 3% Triton X–100, 0.5% Tween 20, and 0.2% sodium azide in PBS) overnight 4°C. After obtaining satisfactory whole-mount images (described in Immunohistochemistry of Paraffin Sections), samples were paraffin embedded and 6-µm sections were cut naso-temporally.

Each sample yielded approximately 800 paraffin sections. Three consecutive sections were chosen every 100th section, meaning approximately 18 sections from each donor were analyzed. Staining was performed as previously described (9), using heat-induced antigen retrieval (90% glycerol, 10% 10 mmol/L sodium citrate pH 6, 120°C, 20 min). Blocking buffer was 1% BSA, 0.5% Triton, and 0.2% sodium azide in PBS. Primary antibodies were diluted in blocking buffer and incubated at 4°C overnight. Table 2 lists the details of primary antibodies. Secondary antibodies were used at 1:200 dilution and stained in Hoechst solution (catalog no. H33258; Sigma-Aldrich) diluted at 1:10,000.

Retinal whole mounts stained with rhodamine-UEA were imaged using an Olympus SZX16 stereoscope (Tokyo, Japan) to generate an overview of the vasculature to evaluate vascular abnormality. Zeiss Axioskop 2 Upright Fluorescence Microscope (Carl Zeiss Microscopy) or Invitrogen EVOS FL Auto 2 (Thermo Fisher) was used to visualize capillaries in cross-sections. Regions of interest were then imaged with a Zeiss LSM 700 with a ×40 objective. For consistency, six z-stacks with interval of 1 μm were taken for each region of interest.

Adobe Photoshop CS6 (Adobe Systems, Inc.) was used to generate whole-mount panorama images with the built-in “Load Files into Stacks” and “Auto-Blend Layers” functions followed by manually stitching regions together. For cross-sections, the built-in “Panorama” function was used. Manual alignment and adjustment were performed when needed.

Capillaries were counted and assigned to a vasculature complex based on collagen IV and Hoechst staining. A collagen IV-positive, UAE-negative profile was considered a ghost vessel (9). Normal and ghost vessels in the different vascular plexuses were counted (nonmasked) by three researchers separately. Capillary dropout was expressed as the percentage of ghost vessels divided by total vessels. Plexus segmentation criteria were as previously published (28). Regions where retinal vasculature plexuses cannot be distinguished, for instance, around the optic nerve head, were excluded from analysis.

To reveal the spacing profile of capillaries in the DVP, the nearest neighbor distance (NND) was measured based on collagen IV staining. The NNS method has been used widely to quantify cell and capillary spacing profiles (29,30). Here, we measured the NND of each capillary residing in the inner nuclear layer (INL) in cross-sections, using Fiji (National Institutes of Health; research Resource Identifier: SCR_002285). For each donor, the NND was measured every hundredth section and at least five sections were examined, resulting in >1,000 measurements per donor. To correct for possible tissue shrinkage during processing, retinal pigment epithelium (RPE) nuclei height was used as a reference. At least 15 measurements were performed per donor. The NND values of each donor were normalized to their reference accordingly. Data were presented as fold change to the reference.

In cross-sections, nuclei were counted in a circle around capillaries with the radius of half of the mean NND, using a LSM700 laser scanning confocal microscope (Carl Zeiss Microscopy; ×40 objective). Five z-stacks of 1-µm steps were scanned in each field and analyzed. Random distribution simulation of the same sample size, mean, and SD were generated using Microsoft Excel.

Data sets were assessed for normal distribution using the Shapiro-Wilk test with a significance level of 0.05. To compare two groups, a two-tailed Student t test (for normal distributed data) or nonparametric Mann-Whitney test was used. A one-way ANOVA test was used to compare the means of more than two groups, Dunnett post hoc analysis for the means of control and other groups, and Tukey post hoc analysis for the means of any two groups. P values in the ANOVA test were adjusted. P values <0.05 were considered statistically significant. The DR case patient served as a positive control, and the one sample t test was used to compare values from the DR sample with mean values of other groups of interest. For linear regression, the Pearson correlation coefficient was calculated to assess the potential relationship between data sets. A linear relationship was assumed when the slope of the best fit model was significantly different from zero. All data are presented as mean ± SD unless otherwise stated. Sex of the donors was not considered a factor in the statistical analysis of the data.

The data sets generated during and/or analyzed during the present study are available from the corresponding author upon reasonable request.

To gain an overview of potential nonperfusion in the retinal vasculature of postmortem retina from donors with diabetes, we visualized endothelial cells in retinal whole mounts, using fluorescently labeled UEA. Figure 1A–C is a representative sample of a whole-mount image from a patient with diabetes. It shows small regions of endothelial cell loss in the peripheral retina (outside the standard 6 × 6 mm screening area of OCTA, which would not usually be captured during standard clinical assessment).

Eight of 11 retinal whole mounts from donors with diabetes did not show any obvious loss of capillaries in this readout and appeared very similar to those of control eyes (Table 1). However, because it is difficult to directly measure the loss of individual capillaries (because they are no longer there), we developed a more sensitive method to measure retinal vasculature damage. One of our previous studies showed that basement membrane of nonperfused vessels is preserved for years despite the loss of the endothelial lining (9). Therefore, we costained retinal sections with UAE and an antibody against collagen IV, visualizing vascular basement membrane to identify acellular capillaries (Fig. 2 and Supplementary Figure 3), providing a quantifiable surrogate readout of loss of perfusion.

A total of 14 retinas (from 14 donors; Table 1) were chosen for quantitative analysis (n = 3 control donors, 10 donors with diabetes and no signs of DR, and 1 donor with DR). The retinas had different amounts of acellular capillaries (Fig. 3A). Based on the frequency of acellular capillaries, two subgroups clearly emerged in the group with diabetes. The subgroup with diabetes and no dropout (DNDO) presented a level of capillary dropout (1.69% ± 0.37%) below the overall mean of the whole group with diabetes (3.82% ± 1.89%; Fig. 3A, dashed line) and similar to that of control eyes. In contrast, the subgroup with diabetes and capillary dropout (DDO) had a notably higher incidence of acellular capillaries (5.23% ± 0.57%). As assessed by one-way ANOVA test with post hoc analysis, the means of the control group and the two subgroups with diabetes were significantly different from one another (P < 0.001). Whereas the DNDO subgroup appeared to have only a slightly elevated degree of vessel loss compared with control eyes (P = 0.04), there was clearly more vessel loss in the DDO (P < 0.0001 when compared with control eyes or the DNDO subgroup). In the sample with clinical DR, the incidence of vessel loss was considerably increased (29.83%), indicating substantial vascular damage (P < 0.001 compared with any other group).

To investigate whether acellularity leads to morphological changes of the vascular basement membrane, we measured the inner diameter of acellular and normal capillary basement membranes. Data were normalized to the size of RPE nuclei of each donor to correct for potential tissue shrinkage during processing. Results showed that capillary basement membrane profiles are reduced in diameter (by ∼50%) after endothelial cell loss in retinas from all donor groups (Fig. 3B). Furthermore, a mild dilation of normal capillary profiles in the DDO subgroup was observed (1.46-fold; P = 0.0051), whereas the diameter of the acellular capillaries remained comparative to those of control eye (P = 0.72). In addition, the DR eye presented an almost twofold dilation in both acellular (1.67-fold; P = 0.457) and normal (1.85-fold; P = 0.0046) capillaries compared with the control group.

It is worth noting that in the DNDO subgroup, no obvious vascular lesions were detected in the whole mount, likely because, in the whole-mount images, the DVP is more difficult to see than the SVP. Therefore, we assessed the DVP and SVP separately in cross-sections. In all donor eyes examined, we found that vessel loss in the DVP was always more pronounced than in the SVP (Fig. 3C). On average, differences in interplexus vessel loss was not significant in control eye (1.38% vs. 0.75%; P = 0.07) or in the DNDO subgroup (2.02% vs. 1.36%; P = 0.08), but they were significant in the DDO subgroup (7.77% vs. 2.67%; P < 0.0001) and the case with DR (15.28% vs. 44.38%; P < 0.001) (Fig. 3C). Intergroup differences of vessel loss were not significant between control eyes and the DNDO subgroup (P = 0.13 in SVP; P = 0.07 in DVP). But there was a significant increase in the DDO subgroup (P < 0.001 in both plexuses), with a steeper increase observed in the DVP. Vessel loss in the DR case was always significantly higher in both plexuses when compared with any other group at any plexus (P < 0.001) (Fig. 3D).

Because we found that vessel loss in the DVP was always higher than in the SVP, we tested for potential correlation between them. Figure 3E reveals a linear positive association between the DVP and the SVP (R² = 0.972; P < 0.0001). We could identify a cluster formed by the control group and the DNDO subgroup in Fig. 3E (bottom left plot), indicating low level of capillary dropout in both plexuses, from which the DDO subgroup forms a clearly separate cluster. Farther away, at the top right corner, is the DR case, with a considerably higher level of vessel loss in both plexuses. All data were within the 99% CI area (Fig. 3E, red region) from the best fit model (Figure 3E, solid line), with the formula y = 2.94x − 0.84. This means with every 1% increase of vessel loss at the SVP, loss in the DVP will increase by approximately 3%.

To test how the integrity of the gliovascular unit is affected by the capillary dropout during the earliest stages of DR, we studied sections from the DDO subgroup by immunohistochemistry using antibodies against glial markers, aquaporin 4 (AQP4), cellular retinaldehyde–binding protein (CRALBP), glutamine synthetase (GS), and glial fibrillary acidic protein (GFAP), visualizing glial proteins (Fig. 4; corresponding stains on control retina are shown in Supplementary Figure 4).

GFAP expression was usually confined to retinal astrocytes and was either not upregulated or predominantly affected by microvascular dropout in eyes with diabetes (Fig. 4A). Glutamine synthetase was highly expressed in Müller cell endfeet at the inner limiting membrane and the outer limiting membrane, and processes in the outer plexiform layer, which were not tightly associated with vessels (Fig. 4B). CRALBP was found in the Müller cell soma, processes in the outer plexiform layer and the RPE (Fig. 4C). CRALBP expression around Müller cell endfeet was visible in both normal (Fig. 4C, zoomed in images, arrows) and acellular (Fig. 4C, zoomed in images, arrowheads) capillaries; however, no obvious difference in staining intensity was noticed. AQP4 was expressed prominently at endfeet (Fig. 4D). The staining was variable around capillaries (Fig. 4D, zoomed in images, arrowheads), but there was no statistically significant difference between perfused and nonperfused capillaries (−0.29%; P = 0.88) based on fluorescence intensity quantification.

It is well established that ischemia can lead to retinal atrophy. This is exemplified in the DR case in our study, where, in a large area of retinal vascular nonperfusion, the retinal ganglion cell layer and the INL (except for Müller cells) are absent (Supplementary Figure 5). On the other hand, the local impact of capillary dropout (and presumed loss of oxygen supply) on the survival of neural cells—specifically neurons in the vicinity of acellular DVP capillaries—is, however, less understood.

To determine which neurons might be affected by a nonperfused capillary in the DVP, we measured the NND (i.e., the distance between neighboring capillaries), which was constant across groups (Fig. 5A). Furthermore, distribution analysis of NND revealed consistency between the control group and diabetes subgroups. Although the peak of DR NND did not shift, it was clearly less than in other groups (Fig. 5B). This can be further described by the regularity index (RI), defined as the ratio of the mean NND over its SD. Despite consistent NND across all groups, the RI of the DR case had a statistically significant RI reduction (P < 0.05), whereas there was no difference among the control group and diabetes subgroups (Fig. 5C).

To assess whether NND changed with respect to retinal eccentricity, multiple sections with varying distance to the optic disc–fovea axis were analyzed for each group. As shown in Fig. 5D, NND plotted against the distance to the optic disc–fovea axis showed no changes in the superior–inferior axis (DR and DNDO subgroups not shown).

Having determined the NND as a definition of DVP zone of influence, we quantified the number of nuclei within a circular area (with the NND × 0.5 as its radius) from each capillary in the DVP. An example is shown in Fig. 6A. Results showed a comparable number of cells around normal and nonperfused capillaries in DNDO subgroup (12.27 ± 1.80 vs. 12.39 ± 1.65; P = 0.9994) and DR (8.42 ± 2.33 vs. 8.35 ± 1.33; P > 0.999). However, compared with control eyes, there was a general loss of 7% INL cells in the DDO subgroup (13.25 ± 2.27 vs. 12.33 ± 1.71; P = 0.0457), which increased to 37% in DR (13.25 ± 2.27 vs. 8.38 ± 1.89; P < 0.0001). Taken together, these results suggest that in patients with diabetes and no DR, there is a subtle but statistically significant loss of INL cells, which was much more pronounced in the DR case. Importantly, this effect was pan-retinal and independent of local capillary dropout (Fig. 6B).

Cells that contribute to the ERG b-wave may be affected in patients with diabetes and rodent models with diabetes. Therefore, we used antibodies raised against parvalbumin (PV) and protein kinase C-α (PKCα) to label horizontal (Fig. 6C) and bipolar (Fig. 6E) cells, respectively. Quantifying these interneurons within deeper capillary zones of influence showed results in line with our previous finding that cell loss is independent of local capillary dropout (Fig. 6D and F). Of note, in the DDO subgroup, a subtle loss of approximately 10% PV plus horizontal cells (3.63 ± 1.07 vs. 3.13 ± 0.74; P = 0.31) and PKCα plus bipolar cells (4.75 ± 1.39 vs. 4.27 ± 0.65; P = 0.51) compared with the control group was identified. Although this difference was not statistically significant, it represents an interesting trend because this cell loss was much more profound in the DR case, in which approximately 60% of both interneuron types were lost compared with that in control eyes (PV plus cells: 3.63 ± 1.07 vs. 1.36 ± 0.84; PKCα plus cells: 4.75 ± 1.39 vs. 1.97 ± 1.09; P < 0.0001 for both) (Fig. 6D and F).

Here, we established capillary loss in the deeper retinal vasculature plexus as a novel and sensitive histological biomarker for the earliest stages of DR in human postmortem tissue, which is likely to occur before a clinical diagnosis (typically based on visible changes in the SVP) and was remarkably common (n = 6 of 10) in our randomly selected group of donors with diabetes. This finding aligns with those of many OCTA studies of people with diabetes without DR, showing more pronounced perfusion deficiencies in the DVP compared with the SVP (reviewed in Supplemental Table 1). Histopathological studies have also described a higher incidence of more advanced vascular pathology, such as microaneurysms within the INL (31,32).

One possible explanation for the increased occurrence of vascular damage in the DVP is that these vessels are more distal than vessels in the SVP. A small perfusion dysfunction in the SVP may have a more profound impact on the downstream vessels in the DVP (all of which receive their supply from the SVP). An alternative explanation might be that the only glia component contributing to gliovascular unit in the DVP comes from Müller cells, whereas in the SVP retinal astrocytes are also present. In our study, we have shown altered distribution of AQP4 in Müller cell endfeet, possibly leading to water homeostasis defects in DR. This may be related to a compromised blood-retina barrier associated with DR (33), contributing to vascular leakage and edema. However, in our tissue cohort with diabetes, immunohistochemistry with antibodies against human IgG did not reveal any signs of serum leakage into the retina (data not shown).

Another widely described feature of DR is increased vessel diameter. This matches our finding of wider perfused capillaries in the early DR group and the DR case. OCTA measurements in patients have also shown increased vessel calibers in DR (34,35). The caliber increase we observed may be explained by a thickening of vessel basement membrane, as in diabetic animal models (36,37). Alternatively, loss of pericytes described in human diabetic or DR eyes may lead to vessel dilation (3,32).

Our data also identify neural cell loss in the INL as an early neurodegenerative change, which could be a consequence of reduced perfusion of the DVP. Oxygen tension in the INL is low (38) and the DVP may be an evolutionary adaptation to provide the INL with oxygen. Perfusion defects in the DVP may lead to excessive hypoxia and, eventually, cell loss in the INL. Interestingly, baseline DVP nonperfusion (assessed by OCTA) in non-proliferative diabetic retinopathy patients can predict DR progression with high accuracy (39,40). Moreover, it was shown that the parafoveal vessel density in the DVP is the parameter most robustly associated with the clinical stage of nonproliferative DR (41). Not only does this finding highlight the potential use case of microvascular changes in the DVP as a biomarker to predict the DR progression, it also indicates a potential functional interaction.

However, the lack of spatial correlation between localized capillary and neural loss observed in our study seems to exclude the simple mechanism of localized hypoxia causing neural cell death in the vicinity of acellular capillaries. Remarkably, we detected subtle and diffuse INL cell loss in the DDO as well as in the DNDO subgroups. One possible interpretation of this finding is that vascular defects are not the cause of neural defects at the earliest stages of DR. However, it cannot be excluded that vascular dysfunction (currently not detectable in postmortem tissue), such as impaired autoregulation of perfusion, may cause the diffuse cell loss in the INL we observed.

Alternatively, diabetes may have a direct impact on the neural retina independently of, or possibly even before, vascular damage (42–45), or the neuronal damage might drive the vascular changes. However, the diffuse neural damage observed in our study is subtle, and it may be the case that an area of nonperfusion is required to reach a certain threshold size to cause more severe, local atrophy of the inner retina (e.g., as described in our DR case). As noninvasive in vivo clinical imaging improves (46), longitudinal studies may provide further insights about the temporal and causal relationship between vascular and neural dysfunction in DR.

In summary, our findings show strong evidence for early retinotopic vascular and neurodegenerative changes, which are meaningful for early DR detection. DVP changes and INL atrophy might be incorporated into newer DR classification systems to reap the benefits of early diagnosis and strengthening measures to improve metabolic control in patients with early DR changes.

Acknowledgments. The authors thank Meaghan O’Neill for technical assistance and Almas Dawood for help with initial processing of the DR eye.

Funding. This study was supported by Diabetes UK (grant 13/0004748), a Santen Pharmaceutical PhD studentship, and the National Institute for Health Research Biomedical Research Centre at Moorfields Eye Hospital NHS Foundation Trust and UCL Institute of Ophthalmology.

Duality of Interest. No potential conflicts of interest relevant to this article are reported.

Author Contributions. Q.Y. performed the analysis. Q.Y., M.Y., A.O.-B., and J.M. contributed data or analysis tools. Q.Y., M.Y., and J.M. collected the data. Q.Y. and M.F. wrote the manuscript. M.Z. collected and prepared tissue. C.E., A.T., and M.F. conceived and designed the analysis. M.F. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.