Macrophage-Like Cell Density and Distribution in Relation to Vascular Leakage in Diabetes

Abstract

Purpose

To investigate the relationship between macrophage-like cell (MLC) density and vascular leakage in the macula, an indicator of blood–retinal barrier (BRB) breakdown, in diabetic retinopathy (DR), and to understand whether MLCs cluster within areas of vascular leakage.

Design

Cross-sectional imaging analysis.

Subjects

A total of 161 eyes from 152 diabetic participants with various stages of DR.

Methods

Fluorescein angiography (FA) was used to delineate and quantify macular leakage and OCT angiography was used to visualize MLCs using an averaged 3-μm OCT slab above the vitreoretinal interface. In a subgroup of 36 eyes, MLC density maps were overlaid onto FA images to assess the spatial distribution of MLCs relative to leakage zones and their perimeter.

Main Outcome Measures

Correlation between MLC density and macular leakage, and the spatial distribution of MLCs in relation to leakage zones.

Results

A nonlinear relationship was observed between MLC density and macular leakage, captured by a quadratic model (r² = 0.246, P < 0.001). Macrophage-like cell density increased with leakage, peaking between 50% and 60%, but then declined in eyes with more extensive leakage (>60%). Macrophage-like cells were more likely to cluster near leakage boundaries (P < 0.01), without difference in MLC density inside- versus outside-leakage areas (P = 0.08).

Conclusions

The nonlinear relationship between MLC density and macular leakage in DR, along with MLC accumulation around the borders of leakage, suggest a nuanced role for these cells in DR and BRB breakdown. These findings also suggest a complex interplay among different macrophage populations and may reflect changes in MLC function or migration to deeper retinal layers with advancing leakage.

Financial Disclosure(s)

Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.

Our current understanding of diabetic retinopathy (DR) suggests a more complex pathogenesis beyond the traditional view of purely microvascular disease. Emerging evidence highlights the pivotal role of inflammation in the onset and progression of DR.1, 2, 3 Hyperglycemia not only induces direct neuronal and vascular injury through various biochemical pathways, but also fosters a chronic proinflammatory environment, triggering secondary immune cell responses within the retinal tissues.^3,4 In the diabetic retina, microglia, the tissue-resident macrophages, are considered the primary responders to changes in the microenvironment, and their goal is to maintain tissue homeostasis.5, 6, 7 Upon activation, microglia acquire a proinflammatory phenotype, releasing cytokines and neurotoxic substances, initiating a cascade of events that includes leukostasis and culminates in the recruitment of circulating macrophages.^3,7 Dysregulation of this cascade in diabetes exacerbates vascular permeability and neuronal damage, perpetuating a vicious cycle of inflammation and neuronal injury.⁷

A number of in vitro studies and in vivo investigations of animal models and human postmortem specimens have shown a significant increase in both the number and activity of microglial cells in DR.^5,6,8 Recent advances in retinal imaging, such as OCT and adaptive optics scanning laser ophthalmoscopy, enabled in vivo visualization of macrophage-like cells (MLCs) on the vitreoretinal interface.^9,10 Our group and others have shown that MLC density increases with advanced DR stages, especially severe nonproliferative DR (NPDR) and proliferative DR (PDR) and that the density of these cells correlates with macular edema.11, 12, 13, 14, 15 These findings suggested that MLCs may serve as a potential imaging biomarker for DR severity and inflammatory status.

Our previous work characterized MLC distribution relative to the underlying vasculature, revealing a notable tendency for MLCs to cluster on blood vessels.¹⁰ This finding aligns with histological studies in human donor eyes, where cells expressing markers for macrophages and activated microglia were identified in a perivascular distribution near microaneurysms, hemorrhages, and neovascularization.⁵ While the precise role of these cells in DR remains unclear, their association with areas of vascular damage suggests that they could play a role in modulating vascular leakage and hence retinal edema.

Given the role of chronic inflammation and leukostasis in endothelial damage and subsequent blood–retinal barrier (BRB) disruption, we hypothesized that MLC density would increase in proportion to the extent of BRB breakdown. To test this hypothesis, we investigated whether MLC density correlates with the degree of fluorescein leakage in the macula in diabetic eyes, a well-established indicator of BRB breakdown. Additionally, we assessed the spatial relationship between MLCs and zones of leakage to determine whether MLCs cluster specifically around leaking vessels or within zones of macular leakage. Through these analyses, we seek to provide insights into the role of MLCs in BRB breakdown and their involvement in the inflammatory processes driving DR.

Methods

This was a prospective observational study with a cross-sectional analysis of patients with diabetes mellitus (DM) who were examined in the Department of Ophthalmology at Northwestern University in Chicago, Illinois, between October 2021 and May 2023. This study was approved by the Institutional Review Board of Northwestern University, followed the tenets of the Declaration of Helsinki, and adhered to the Health Insurance Portability and Accountability Act regulations. Written informed consent was obtained from all participants before any study procedures were conducted.

Study Sample

This study prospectively enrolled participants aged ≥18 years with a diagnosis of type 1 or type 2 DM during the study period. We recruited 161 eyes of 152 participants with a range of DR severity, including no apparent DR (DM-no DR), mild NPDR, moderate NPDR, severe NPDR, treatment-naive PDR, and quiescent PDR (characterized by regressed PDR with no signs of active neovascularization). Diabetic retinopathy severity was graded according to the International Clinical Diabetic Retinopathy Disease Severity Scale using ultra-widefield color fundus photographs (Optomap Panoramic 200; Optos PLC).¹⁶

The exclusion criteria included participants with additional ocular disorders that could confound imaging or outcomes, including significant lens or media opacities, glaucoma, and other retinal pathologies such as age-related macular degeneration, retinal vascular occlusion. Eyes with center-involving diabetic macular edema—defined as central subfield thickness >300 μm on OCT—were also excluded to avoid confounding from cystoid fluid accumulation during leakage analysis. Additionally, participants with a history of intravitreal injections or panretinal photocoagulation within the last 6 months, or any ocular surgery within the last 3 months, were excluded.

Fluorescein Angiography Imaging

All participants underwent ultra-widefield fluorescein angiography (FA) using a standard protocol on Optos California device (Optomap Panoramic 200, Optos PLC). After pupil dilation and intravenous fluorescein dye injection, ultra-widefield FA images centered on the fovea were captured between 0:00 to 8:00 min postdye injection.

Fluorescein Angiography Image Analysis

Ultra-widefield FA images were analyzed using the proprietary OptosAdvanced software (Optos PLC, version 4.4.33.107911) to identify fluorescein leakage within the macula. Early-phase FA images were identified between 0:30 to 1:00 min postinjection, when the dye had filled the most peripheral vessels. Late phase FA images (5:00–7:30 min postdye injection) were reviewed sequentially, and the frame showing peak leakage (maximum lateral extent) without appreciable expansion in subsequent images was selected to quantify macular leakage. Leakage was defined as any region of increased hyperfluorescence compared with an early-phase image, with signal intensity greater than or equal to that of adjacent vessels, and excluding areas where the increase in intensity can be attributed to choroidal fluorescence.¹⁷ Early hyperfluorescent structures that did not leak, such as staining laser scars and nonleaking microaneurysms, were not considered leakage. Similarly, preretinal neovascularization, identified as early bright hyperfluorescence with extensive, bright late leakage, was not considered in the area of leakage.

We quantified the leakage within the macula by digitally overlaying the ETDRS 9-field grid on late phase angiographic frames, centering the mask on the fovea.¹⁷ The ETDRS grid mask on the proprietary software consists of 3 concentric circles, representing inner, middle, and outer zones of the macula. Using the software's “circle” tool, we manually traced the outlines of the outermost circle and the middle circle on the ETDRS grid to identify leakage in the entire macula and within the central 3.5-mm zone, respectively. The corresponding areas for each circle were automatically calculated in square millimeters (mm²) by the software. The digital mask was then removed to ensure the gridlines did not obstruct leakage measurements. Using the region-of-interest–free annotation tool, graders identified and manually delineated any regions of leakage within the selected circles of the ETDRS grid. The software then automatically calculated the sum of the outlined leakage areas. The percentage of overall and central macular leakage was calculated by dividing the combined leakage areas within the outer and middle circle of the ETDRS grid by the area of the respective grid circles and multiplying by 100. Figure 1 demonstrates these steps.

Figure 1.

Quantification of vascular leakage. A, The ETDRS 9-field grid is digitally overlaid on late phase angiographic frames, centering the mask on the fovea. B, Using the software's “circle” tool, area within the outermost circle and the middle circle on the ETDRS grid are outlined to identify leakage in the entire macula (blue circle) and within the central 3.5-mm zone (red circle), respectively. C and D, Using the ROI-free annotation tool, regions of leakage within the selected circles are manually delineated (green). The automatically generated areas for each circle and the sum of measured leakage are shown in blue, red, and green boxes, respectively. The total percentage of overall macular leakage and central leakage is displayed in white boxes. ROI = region of interest.

Notably, the software automatically transformed all manually selected regions to a stereographic projection of the spherical retina by factoring in the magnification artifact of the 3-dimensional retina onto a 2-dimensional image, thereby ensuring accurate measurements in mm².¹⁸

Before manually delineating leakage on all study eyes, 2 graders (E.O.Z. and N.D.) independently measured macular leakage in 10 eyes as a pilot sample to ensure reliability of the quantitative method. After confirming strong agreement with an intergrader intraclass correlation coefficient of 0.948, one of the graders (E.O.Z.) quantified leakage in all study eyes.

OCT Angiography Imaging

OCT angiography (OCTA) images were acquired using the RTVue-XR Avanti system (Optovue) with split-spectrum amplitude-decorrelation angiography software (version 2017.1.0.151) over a 3 × 3 mm (304 × 304 pixels) area centered on the fovea.¹⁹ Each participant underwent multiple OCTA scans (mean: 5.4, range: 4–7) to ensure image quality. OCT angiography images with a quality index (Q-score) >6, a signal strength index >50, and without shadowing or motion artifacts were included.

OCTA Image Analysis and MLC Quantification

Images were processed as previously described.^11,12,15 Briefly, images were segmented into the superficial capillary plexus (SCP, internal limiting membrane [ILM] to 10 μm above from the inner plexiform layer), deep capillary plexus (9 μm above the inner plexiform layer to 10 μm above the outer plexiform layer), MLC (structural scan 3 μm above the ILM to the ILM), and retinal nerve fiber layer slabs (structural scan, ILM to 28 μm below the ILM). Images were registered using SCP slab via the Register Virtual Stack Slices Plugin (Feature Extraction Model = Rigid, Registration Model = Elastic) in FIJI, a distribution of ImageJ (National Institutes of Health). Then, the saved transformation matrix was applied to the deep capillary plexus slab, MLC slab, and retinal nerve fiber layer slab using the Transform Virtual Stack Slices Plugin. We averaged the transformed image stacks from each layer using the Z-project plugin to optimize image quality and increase the signal-to-noise ratio.

Macrophage-like cells were identified and quantified from the averaged MLC slab using our previously published custom FIJI macro.^11,12 This macro utilizes a combination of noise reduction and signal enhancement for automated MLC identification using both MaxEntropy automated thresholding followed by Bernsen local thresholding. Once masks from the macro were created, the Analyze Particles function was used to quantify MLC density, defined as the number of identified cells per mm². The MLC densities were then corrected for axial length as previously described.²⁰ Macrophage-like cell quantification was performed by 2 graders (J.B.B. and C.J.H.) who were masked to participant's demographics, DR severity, and FA leakage data. There was strong agreement between MLC graders (intraclass correlation coefficient = 0.932). Macrophage-like cell densities were averaged between graders to obtain the final axial length-corrected MLC density.

Overlaying FA and OCTA Images for Spatial Mapping of MLC and Macular Leakage

To visualize and analyze the spatial relationship between MLCs and macular leakage, MLC density maps were overlaid onto FA images in a subgroup of patients with central leakage ranging from 10% to 80%. Cases with minimal or extensive leakage were excluded from the analysis to prevent skewed results, as MLCs would predominantly appear either entirely within or outside the leakage areas. Additionally, eyes were excluded if the spatial orientation or angles of FA and OCTA images differed significantly, preventing accurate alignment. After these exclusions, 36 eyes were included for spatial relation analysis. The workflow of this process is outlined below and illustrated in Figure 2.

Figure 2.

Example image demonstrating the workflow for spatial mapping of MLCs and vascular leakage. A, OCT angiography maps of MLC density are overlaid onto the SCP and manually aligned with early-phase FA images. B, Late-phase FA images, colored to enhance contrast, are aligned with the early-phase FA-OCTA composite images. C, After creating a 3-layered composite image, the OCTA layer is initially set to full transparency to allow for the redelineation of areas of angiographic leakage (outlined in green). D, The transparency of the OCTA layer is then removed to reveal the final composite image (shown in blue square), showing the SCP layer with MLCs and the outlined leakage areas (green). FA = fluorescein angiography; MLC = macrophage-like cell; OCTA = OCT angiography; SCP = superficial capillary plexus.

The first step involved overlaying MLC density maps onto the SCP enface image. The SCP-MLC overlay images were then aligned with early-phase FA images to ensure accurate alignment, since the leaking fluorescein in late-phase FA images could obscure vessel details. The spatial transformation established between early FA and SCP-MLC overlay was subsequently applied to the late-phase FA images, where leakage was outlined in green. This process resulted in final composite images that revealed the spatial distribution of MLC density in relation to leakage areas in the macula. Overlays and spatial analyses of composite images were independently performed by 2 researchers (E.O.Z and K.E.).

Measurement of MLC Density in Relation to Areas of Macular Leakage

The final composite images, overlaying MLC density maps onto the outlined leakage boundaries, were processed in MATLAB (The MathWorks) using Image Processing and Signal Processing Toolbox functions. First, the outlined leakage boundaries, marked in light green as shown in Figure 2, were detected using the “bwconncomp” function. A similar procedure was followed for the red-colored marked MLC to identify the number of MLCs and their spatial locations. For each distinct leakage area within the macula, a mask was generated to differentiate MLCs located inside and outside the leakage. Macrophage-like cell densities were then calculated separately for the regions inside and outside the leakage for each subject (Fig 3).

Figure 3.

Workflow illustrating the calculation of MLC density relative to leakage areas in an example eye. A,Color Coding: key regions in the composite image (e.g., MLCs in red, leakage boundaries in green) were color-coded to enable accurate color masking and reliable extraction of regions of interest for downstream analysis. B,Cell Detection: the red channel image (from A., representing MLCs) was converted to a binary image where all cells were identified using a connected component algorithm (bwconncomp) on MATLAB. Detected cells were marked with blue asterisk. C,Leakage Area Mask: the green channel image (from A) was converted to a binary image where all distinct leakage areas were identified using connected component algorithm. For each distinct leakage area, a spatial mask was generated and displayed with a transparent gray overlay. D,Labeling Cell Location: combining B and C area masks, we distinguished cells located inside (labeled with blue asterisk) vs. those outside the leakage areas (labeled with yellow asterisk). Please note that each asterisk marks a single MLC but may not fully cover the red-labeled cell. For large or irregular shaped MLCs, the asterisk is positioned centrally and may not visually cover the entire extent of the cell, which can create the impression of multiple red cells around a single marker. Nonetheless, all red-labeled MLCs are accurately accounted for as either within (blue) or outside (yellow) the leakage area based on their position. E–H: Calculation of Normalized Cell Density:E, after marking all cells as either inside or outside the leakage area, the number of cells was counted for both regions; F, leakage area masks were used to calculate the total leakage area and the nonleakage area; G, Three cell densities were calculated: inside the leakage, outside the leakage, and overall density (cells per mm² in each respective region); and H. the cell densities (inside and outside-leakage) were divided by the overall cell density to obtain normalized densities, which allowed us to compare cell densities between different subjects, independent of their absolute cell count and leakage area size. MLC = macrophage-like cell.

To enable comparison of MLC densities across eyes with varying cell counts and leakage extents, we calculated normalized MLC densities. For each eye, we measured 3 quantities: (1) the mean MLC density across the entire image, (2) the density within the leakage region, and (3) the density outside the leakage region.

To make these regional densities comparable—both within and across subjects—we expressed them as ratios relative to the eye-wide mean. Specifically, the density inside the leakage region was divided by the overall density, and the same was done for the nonleakage region. By definition, one of the regional values will exceed 1 (indicating a higher-than-average density concentrated in that region) and the other will fall below 1 (indicating relative depletion), because the 2 regions partition the entire image. This normalization step eliminates the confounding effects of absolute cell count and leakage area size, allowing direct comparison of MLC enrichment or depletion patterns across eyes (Fig 3H).

Measurement of MLC Distance to Leakage Boundary

For each MLC, the minimum distance between the MLC boundary and the leakage boundary was calculated in MATLAB. Leakage areas vary in shape and the expected distance between an arbitrary point and the leakage boundary depends strongly on the geometry of the leakage region.

To eliminate the bias due to leakage geometry, we performed a simulation of random point distributions for each eye with various leakage geometries. Briefly, we overlaid a fine spatial grid on each image and computed the average expected distance from uniformly distributed grid nodes to the leakage boundary—separately for regions inside and outside the leakage. Each MLC's observed distance to the boundary was then divided by this expected value and defined as the “Observed-to-Simulated Distance Ratio” (Fig 4). Values >1 indicate that a cell sits farther from the boundary than would be expected by chance, whereas values <1 indicate enrichment near the boundary. Since this method removes the purely geometric component, the resulting metrics allow for meaningful comparison of boundary proximity both within an eye and across eyes that have very differently shaped leakage areas.

Figure 4.

Workflow illustrating labeling of leakage areas and MLCs, calculation of cell-to-leakage boundary distances, and simulation of uniformly distributed points in an example eye. A, The labeled image shows each distinct leakage area and all MLCs in this image. B, For each cell, the minimum distance from the cell boundary to the nearest leakage boundary was calculated. For illustration, a representative sample of 40 cells is shown: blue lines indicate distances for cells inside-leakage areas, and yellow lines indicate distances for cells outside. C, A zoomed-in view of (B), illustrating the minimum distances between the cells and the leakage boundaries. D, To account for geometric biases, 10 000 uniformly distributed points were simulated for each image. A color map demonstrates the simulated points, with brighter colors indicating greater distance from the leakage boundary and darker colors indicating closer proximity. For each simulated point, the minimum distance to the leakage boundary (green) was computed for points inside and outside-leakage areas. The observed-to-simulated distance ratio was then calculated by dividing each observed MLC—boundary distance by the average simulated distance—yielding a proximity metric that reflects how close a cell is to the leakage boundary compared to a random point, independent of leakage shape, size, or image boundaries. MLC = macrophage-like cell.

The observed-to-simulated distance ratio was recorded for each MLC in the subgroup of 36 subjects. To confirm that this rescaling process did not distort the original biological signal, we compared the raw distance values with the observed-to-simulated ratios. As each value in an eye for both inside- and outside-leakage areas is divided by a constant (ie, mean distance of random points), the raw MLC distances and their rescaled values show a perfect rank correlation (Spearman's rho = 1) within each respective region. Thus, this method rescaled the data without significantly altering the underlying biological signal.

Statistical Analysis

Statistical analysis was performed using IBM SPSS Software v28.0 (IBM SPSS Statistics; IBM Corporation), and statistical significance was set at P < 0.05. The Shapiro–Wilk test was used to determine the normality of the data distribution. Demographic and clinical data variables are expressed as means with standard deviations, medians and ranges, or as total counts with percentages. Visual inspection of scatterplots was employed to assess the relationship between MLC density and the amount of macular leakage. Quadratic fitting was applied to model the nonlinear trend between MLC density and leakage, with subgroup analysis to further explore the relationship. Central leakage severity was categorized into no (0%), low (>0% and <20%), moderate (≥20% and <60%), and high (≥60%) leakage groups for subgroup analyses. These thresholds were chosen based on the ETDRS grading of DR on FA images which classified subfields of the angiograms containing roughly 20% to 25% leakage as the lower bound for a moderate fluorescein leakage grade, and 60% leakage as the lower bound for a severe fluorescein leakage grade.¹⁷ For subgroup comparisons of continuous data, independent samples t-tests or Mann–Whitney U tests were used depending on the data distribution, and the Kruskal–Wallis test was applied for multiple subgroups with non-normal data.

The intraclass correlation coefficient scores were calculated to test for interrater reliability based on a single-rating, absolute agreement, 2-way mixed-effects model and interpreted using ranges suggested by general guidelines:²¹ <0.5, poor; 0.5 to 0.75, moderate; 0.75 to 0.9, good, and values >0.90, excellent agreement.

Results

A total of 161 eyes from 152 subjects were included in the study (mean age, 58.5 ± 13.2, 53% female). The demographics and clinical characteristics of the study population are summarized in Table 1.

Table 1.

Patient Demographics and Clinical Characteristics

Subjects (Eyes), n	152 (161 Eyes)
Age (yrs), mean ± SD	58.5 ± 13.2
Sex (female), n (%)	80 (52.6)
DM type (type 1), n (%)	30 (19.7)
DM duration (years), median (range)	19 (1–57)
HbA1c (%), mean ± SD	7.3 ± 1.3
Axial length (mm), mean ± SD	23.8 ± 1.2
DR stage
No DR, n (%)	43 (26.7)
Mild NPDR, n (%)	26 (16.1)
Moderate NPDR, n (%)	30 (18.6)
Severe NPDR, n (%)	15 (9.3)
PDR, n (%)	47 (29.2)
MLC density (cells/mm2), median (range)	12.3 (1.5–70.2)
Overall macular leakage (%), median (range)	11.9 (0.0–94.1)
Central leakage (%), median (range)	14.5 (0.0–91.4)

DM = diabetes mellitus; DR = diabetic retinopathy; HbA1c = glycosylated hemoglobin; MLC = macrophage-like cell; NPDR = nonproliferative diabetic retinopathy; PDR = proliferative diabetic retinopathy; SD = standard deviation.

The mean overall macular leakage for the study population was 11.9%, ranging from 0.0% to 94.1%, and the mean central macular leakage was 14.5% (0.0%–91.4%). Visual inspection of the scatterplot showing the association between central leakage and MLC density revealed a nonlinear pattern, with MLC density peaking at 50% and 60% leakage, followed by a decrease thereafter. This relationship was represented by a quadratic fit (R² = 0.246, P < 0.001) and subgroup analysis, shown in Figure 5. Macrophage-like cell density was significantly higher in eyes with low and moderate central leakage compared with eyes with no leakage in the macula (P = 0.024, P < 0.001, respectively), and in eyes with moderate leakage compared with eyes with low leakage (P = 0.026). Notably, no significant difference in MLC density was observed between eyes with high leakage (>60%) and those with no leakage in the macula (P = 0.856). To investigate the spatial relationship between MLCs and areas of macular leakage, a subgroup of 36 eyes with a mean central leakage of 38.9% (12.4%–79.6%) was further analyzed. The mean normalized MLC density across 36 subjects was 1.13 ± 0.5 and 0.94 ± 0.22 cells/mm², inside and outside the leakage areas, respectively. Although normalized MLC densities were 20% higher inside the leakage areas compared to outside, the difference was not statistically significant (P = 0.08). The scatterplot in Figure 6 demonstrates that the distribution of MLCs is well centered along the unity line, suggesting a random distribution of MLCs inside versus outside the leakage areas.

Figure 5.

Macrophage-like cell density demonstrated a positive trend with increasing leakage severity up to 60%, followed by a decrease in MLC density in eyes with >60% leakage. Box-whisker plots illustrate MLC density across leakage groups, with asterisks denoting significant differences between groups of no leakage and low/moderate leakage (∗P < 0.001, ∗∗P = 0.024, respectively), and between low and moderate leakage (∗∗∗P = 0.026). The scatter plot shows the relationship between central leakage % and MLC density, with the final quadratic fit overlaid (r² = 0.246, P < 0.001). MLC = macrophage-like cell.

Figure 6.

Distribution of MLC density in relation to leakage areas. A, Box-and-whisker plots comparing normalized MLC density inside and outside-leakage areas. The 2-sample t-test revealed no significant difference between the groups (P = 0.08). B, Scatter plot showing normalized MLC density versus leakage percentage for both inside and outside the leakage areas. The data points are centered around the unity line, indicating a random distribution, with outliers primarily showing higher MLC density inside the leakage area at the lower end of the leakage spectrum. C, Example of MLC distribution in a case with 27% leakage, where cells are predominantly clustered outside the leakage area. D, Example of MLC distribution in a case with 34% leakage, showing MLCs primarily clustered inside the leakage area. These contrasting examples illustrate the variability in MLC distribution, even in eyes with comparable leakage levels, further supporting the notion of random distribution, independent of leakage areas. MLC = macrophage-like cell.

Next, the spatial distribution of MLCs relative to the boundaries of macular leakage was analyzed. The mean observed-to-simulated distance ratio was 0.89 ± 0.20 inside the leakage areas and 0.79 ± 0.23 outside. Each group was compared against a unit mean of 1, reflecting the null hypothesis where MLCs are uniformly distributed. Values <1 indicated that MLCs were closer to leakage boundaries than expected in a uniform distribution, with statistically significant results for both inside and outside-leakage areas (P < 0.01 for each). However, when comparing the observed-to-simulated distance ratios between inside and outside-leakage areas, no significant difference was observed (P = 0.05). This suggests that MLCs were more likely to cluster near the boundaries of macular leakage, regardless of whether they were located inside or outside the leakage zone, as visually represented in Figure 7.

Figure 7.

Macrophage-like cell localization in relation to leakage boundaries. A, Box-and-whisker plots showing the observed-to-simulated distance ratios for MLCs located inside- and outside-leakage areas. The mean (±SD) observed-to-simulated distance ratios were 0.89 ± 0.20 for the inside group and 0.79 ± 0.23 for the outside group. Both groups were compared against a unit mean of 1, representing the null hypothesis of a uniform MLC distribution. The analysis demonstrated that MLCs were significantly closer to leakage boundaries compared to a fully uniform distribution (P < 0.01 for both groups). B, Scatter plot comparing observed-to-simulated distance ratios for inside and outside groups. Despite the overall trend of MLCs clustering near leakage boundaries, no significant difference was found between the inside and outside groups (P = 0.05), suggesting that MLC proximity to leakage boundaries is independent of their location relative to the leakage areas. MLC = macrophage-like cell; SD = standard deviation.

To evaluate the potential influence of image edge effects on boundary-distance calculations, we conducted a subgroup analysis limited to 21 eyes in which the majority of the leakage area was contained within the OCTA field of view. The distribution patterns remained consistent with the full cohort: MLCs both inside and outside the leakage region were still slightly closer to the leakage boundary compared to a random point distribution (0.89 ± 0.23 for inside leakage, 0.84 ± 0.26 for outside leakage). However, statistical significance was reduced (P = 0.047 for inside; P = 0.011 for outside), likely due to reduced sample size.

Discussion

This study builds upon previous work demonstrating higher numbers of MLCs in eyes with advancing DR.11, 12, 13, 14, 15 Our aim was to further explore the role of MLCs in vascular leakage, an indicator of BRB breakdown in DR. Our findings demonstrated a significant positive correlation between MLC density and macular leakage, affirming the role of MLCs in the inflammatory processes that are related to BRB breakdown. However, while we anticipated a linear relationship between MLC density and macular leakage, the data revealed a more complex nonlinear association. Specifically, MLC density peaked at moderate levels of leakage (50%–60%) and subsequently declined at higher levels of leakage.

Interestingly, this nonlinear relationship contrasts with findings from a recent study investigating Behçet uveitis, where MLC density exhibited a direct linear correlation with vascular leakage.²² In that study, higher MLC densities were consistently associated with increased fluorescein leakage, indicating that MLC recruitment intensified proportionally with worsening vascular inflammation.²² The stronger linear relationship observed in Behçet disease likely reflects the fundamentally different nature of the 2 diseases. Behçet disease, a systemic vasculitis, involves more acute and purely inflammatory responses compared with the chronic, multifactorial nature of DR. As our results suggest, the role of inflammation in DR may be more nuanced, particularly in advanced stages.

Renal dysfunction in diabetes shares many similarities with retinopathy, particularly how macrophages contribute to tissue damage within a hyperglycemic environment.²³ In diabetic nephropathy, macrophage infiltration has been found to correlate with proteinuria, an indicator of BRB breakdown, much like macular leakage and BRB breakdown in DR.^23,24 Studies of kidney biopsies in patients with type 2 DM revealed that glomerular macrophages transiently increase in number during moderate glomerulosclerosis but remain low at mild and advanced stages.²⁵ Eardley et al²⁶ also demonstrated that macrophage density increased in proportion to albumin leakage in the early to moderate stages of nephropathy, but this association became less pronounced in advanced renal disease. They observed that interstitial macrophage numbers no longer correlated directly with albuminuria as the disease progressed, suggesting that other factors, such as hypoxia or fibrosis, may play a more significant role in advanced stages of nephropathy.²⁶ These findings parallel our observations in DR, where MLC density peaks at moderate levels of macular leakage but declines at more advanced stages. This pattern may indicate that macrophage activity becomes less relevant as structural damage progresses, or that saturation of the inflammatory response limits further immune cell recruitment. In advanced stages of DR, factors such as ischemia may interfere with immune cell extravasation from the bloodstream.²⁷ While in early to moderate stages of DR, the presence of MLCs likely reflects an initial immune response, in later stages with extensive vessel loss, the reduction in immune cell extravasation may contribute to the apparent decrease in MLC density.

An alternative explanation for the observed decline in MLC density at higher levels of macular leakage could be related to the limitations of current imaging technology, which only allows visualization of MLCs at the vitreoretinal interface. Although immune cell extravasation primarily occurs in postcapillary venules, confined to the superficial retina, there is evidence of macrophage migration into deeper retinal layers.^5,27, 28, 29 In a histopathological study of donor eyes with DR and macular edema, Zeng et al⁵ demonstrated that microglia—the resident macrophages of the retina—were significantly increased, extending into the outer plexiform and outer nuclear layers, the photoreceptor cell layer, and even the subretinal space. This raises the possibility that, rather than a true decrease in MLC density, our results may reflect the relocation of these cells to deeper retinal layers, which is not captured with the current study protocol. Such relocation suggests that in the presence of severe vascular damage, MLCs may migrate away from the vitreoretinal interface to areas of more extensive injury deeper in the retina.^5,28,29 Further supporting this idea, studies in diabetic animal models and patients with DR have observed the recruitment of macrophages to outer retinal layers, including the retinal pigment epithelium.^5,30, 31, 32 A recent study by Lee et al³² found that the increased accumulation of macrophages in the retinal pigment epithelium layer corresponded with locations of tight junction disruption, a key factor in the breakdown of the outer BRB. As vascular leakage in DR results from a combination of breakdown of both inner and outer BRBs,³³ the quantity and localization of macrophages may vary depending on the relative involvement and severity of dysfunction of these barriers at different stages of DR.

The second part of our study investigated the spatial distribution of MLCs in relation to the boundaries of macular leakage. Although the exact identity of MLCs in human eyes remains uncertain, studies using transgenic mice suggest that these cells are predominantly microglia, with smaller populations of vitreous hyalocytes and perivascular macrophages under normal conditions.³⁴ During active inflammation, classical monocytes and monocyte-derived circulatory macrophages also contribute to the observed MLC population.³⁴ Based on this, we hypothesized that blood-derived macrophages would accumulate more densely within areas of leakage, where BRB breakdown and inflammatory signals are most pronounced. Contrary to our expectations, MLCs were randomly distributed inside and outside zones of leakage, with a slight preponderance of MLCs closer to the leakage borders rather than at the center. The observed random distribution likely reflects the complex dynamics between different macrophage populations in DR, where both reparative and pathological processes are at play.³⁵ While resident microglia proliferate and spread throughout the retina to maintain homeostasis and respond to ongoing stress, recruited pathological macrophages, often derived from circulating monocytes, might target specific injury site and accumulate more at the sites of BRB injury.²⁹ The heterogeneity in MLC types and their respective roles likely complicates their spatial distribution, which may appear random when viewed as a whole. The observed tendency of MLCs to be closer to leakage boundaries, whether inside or outside of the leakage areas, may suggest a dynamic recruitment process occurring at the interface between damaged and relatively healthier tissue.³⁶ This boundary might be where reparative processes, such as those mediated by microglia and perivascular macrophages, are actively mitigating further retinal vascular damage.^36,37

This study has several limitations that we would like to acknowledge. First, the cross-sectional design limits our ability to infer causality between MLC density and macular leakage. While previous evidence from animal models of DR suggests that macrophages play a dual role^28,31,32—contributing to hyperglycemia-induced BRB breakdown and being recruited to the retina after BRB disruption—the temporal relationship between MLC recruitment and BRB breakdown in human eyes remains unclear. Longitudinal studies would be necessary to better understand this relationship over time. Second, the decline in MLC density observed in eyes with high macular leakage was based on a relatively small subgroup (n = 9), which may limit statistical robustness and could potentially exaggerate the nonlinear trend suggested by the quadratic model. Alternative explanations for this apparent decline should also be considered, including saturation of the inflammatory response, migration of MLCs to deeper retinal layers, or image artifacts that may impair cell detection at high leakage levels. The use of current imaging technology may have limited our ability to fully capture the distribution of MLCs, particularly in deeper retinal layers where significant dynamic changes and migration might occur.^5,28,32 Furthermore, distinguishing between activated resident microglia and circulatory macrophages on OCTA was not possible, as both adopt similar morphologies upon activation.^11,22,29 This potentially complicates the interpretation of MLC dynamics and their specific roles in DR. Third, spatial analysis could not be applied to eyes with minimal or widespread leakage, as MLCs in these extremes were either almost entirely outside or entirely within the leakage regions, precluding meaningful comparison. While this approach ensured internal consistency of the spatial metrics, it may reduce the generalizability of findings to early or advanced disease states. Also, boundary-distance calculations may be influenced by edge effects, particularly in eyes where leakage extended beyond the OCTA field of view. Although our random point simulation approach helps correct for this by applying the same spatial constraints to both real and simulated data, some residual bias may remain. To address this, we performed a sensitivity analysis limited to eyes with leakage contained within the OCTA field, which confirmed similar spatial trends, though with slightly lower statistical significance due to reduced sample size. Lastly, manual tracing of leakage boundaries introduces inherent subjectivity, and the dynamic nature of fluorescein leakage over time may lead to variability in boundary delineation depending on the selected frame. These limitations might be overcome by development of automated tools in future studies.

In conclusion, this study reveals a nonlinear relationship between MLC density and macular leakage in DR, differing from the linear patterns seen in other inflammatory diseases like Behçet uveitis.²² The MLC density peaks at moderate leakage levels, followed by a decline at higher levels of leakage, which may suggest a shift in MLC function or their relocation to deeper retinal layers with advancing leakage. Additionally, the preponderance of MLC at the perimeter of leakage areas indicates a complex interaction between resident and recruited macrophages, possibly driven by changing microenvironmental signals. Future research should focus on elucidating these dynamics through longitudinal studies and advanced imaging approaches that capture the full range of their dynamic movement in the entire retina.

Manuscript no. XOPS-D-25-00006.