Identifying Hyperreflective Foci in Diabetic Retinopathy via VEGF-Induced Local Self-Renewal of CX3CR1⁺ Vitreous Resident Macrophages

The vitreous maintains transparency within the eye and possesses biochemical properties that inhibit cell migration and proliferation (1). The vitreous contains specialized tissue-resident macrophages (TRMs), also known as hyalocytes, which are embedded in the gel cortex and typically spread out in a single layer at the vitreoretinal interface (VRI) (1,2). Vitreous resident macrophages (VRMs) are derived from the bone marrow and demonstrate renewal dynamics during healthy aging (3). Accumulating evidence has revealed the involvement of VRMs in a few retinal disorders whereby their activation contributes to the pathogenesis (4,5).

Diabetic retinopathy (DR) is a growing cause of visual impairment worldwide and remains a serious problem (6). Continuous basic research and clinical studies have revealed that vascular endothelial growth factor (VEGF) plays a key role in diabetic macular edema (DME) and the stage progression to proliferative diabetic retinopathy (PDR), which can cause visual impairment (7–9). Various randomized clinical trials have shown that pharmacological VEGF suppression not only suppresses DME but also improves the stage of DR (10). However, the suppressive mechanism of anti-VEGF therapy on the DR stage is lacking.

While the retina is the obvious focal point for DR, it is known that the abnormalities in the vitreous also contribute to the pathogenesis, heralding the concept known as diabetic “vitreoretinopathy” or “vitreopathy” (11,12). Pathological observations and imaging techniques have confirmed alterations in VRI that might contribute to DME or fibrovascular membrane formation in PDR (1,13,14). Furthermore, recent advances in optical coherence tomography (OCT) have also revealed the clinical significance of hyperreflective foci (HRF), which appear to be cellular lesions consisting of macrophages, retinal pigment epithelium, or photoreceptors (15–19). HRF have been described at the VRI in DR, but their derivation and pathogenic significance remains ill-defined.

Monocytes and macrophages play crucial roles in tissue homeostasis as well as in various pathologies, where they are attractive therapeutic targets (20). Most tissues of a healthy individual contain TRMs (21), and it is widely accepted that their homeostasis relies on the constant replenishment from blood monocytes (22,23). Furthermore, during pathological conditions, the increase in TRMs is often attributed to an enhanced infiltration of circulating monocytes that express C-C chemokine receptor type 2 (CCR2), a chemokine receptor (20). While many TRM populations are derived from embryonic precursors that are seeded before birth and are maintained in adults by self-renewal (24), some findings suggest that certain subpopulations of TRMs retain a high capacity to proliferate on the site of inflammation or may even directly arise from local precursors other than the bone marrow in various diseases, such as atherosclerosis and obesity (25–27).

Monocytes/macrophages are known to be involved in the development of DR (28). Derived from hematopoietic precursors in the bone marrow, these circulating cells can infiltrate the retina from the peripheral blood and contribute to a building vascular and neuroinflammatory pathology (29). However, it has not been investigated whether the increase in cellular components at the VRI is due to infiltration from blood or local proliferation of VRMs in DR.

This clinical study was approved by the Kyushu University Hospital Institutional Ethics Committee (Protocol No. 26012, UMIN000017473) and was performed in accordance with the tenets of the Declaration of Helsinki. Written informed consent was obtained from all patients after a detailed explanation of the study. In addition to 15 healthy subjects, we prospectively enrolled 147 consecutive patients with diabetes at the Kyushu University Hospital between April 2018 and December 2018. All patients underwent OCT examination. We excluded eyes with any other ocular disease that may lead to microvascular disturbance in the retina or choroid (e.g., retinal vascular occlusion, age-related macular degeneration, and glaucoma). Furthermore, eyes with severe cataract, vitreous hemorrhage, or poor fixation were also excluded.

HRF can be observed in DME using OCT (15). We clinically evaluated HRF in healthy participants (control subjects; n = 15 eyes), subjects with diabetes but no DR (NDR) (n = 12 eyes), and DR patients with retinopathy at various stages (n = 128 eyes) using ultrahigh-resolution spectral domain (SD)-OCT (Kowa OCT Bi-μ; Kowa Company, Ltd., Nagoya, Japan) using two different methods. The SD-OCT provided cross-sectional images (B-scans) of the retinal structures with an axial resolution of 2.0 μm, at an image acquisition speed of 80,000 single axial scans (A-scans)/s. Each A-scan had a depth of 2.6 mm comprising 2,048 pixels, which provided a digital sampling depth of 1.3 μm/pixel. Each B-scan spanned 30° or ∼9 mm and consisted of 1,800 A-scans (30). In each case, the horizontal B-scan passing through the fovea and en face OCT imaging were performed. Two independent retina specialists (M.Y. and Y.K.) detected HRF at the VRI using B-scan images based on previously reported observation (15). Eyes with differing diagnoses were assessed by a third retina specialist (I.W.). The κ coefficient was 0.829 (95% CI 0.75–0.98; P < 0.0001) for the judgment of HRF. And after that, HRF at the VRI using horizontal B-scan images passing through the fovea of 112 eyes (14 NDR, 65 non-PDR [NPDR], and 33 PDR) without DME (retinal thickness <300 μm) were manually counted by two observers (M.Y. and Y.K.), as previously reported (31). The intraevaluator intraclass correlation coefficient (ICC) showed good agreement; the ICCs for M.Y. and Y.K. were 0.995 and 0.945. The interevaluator ICC showed good agreement at 0.957. Furthermore, HRF was quantified using our newly developed software from en face OCT images (3 × 3 mm) of the preceding 112 eyes. We also quantified HRF in seven PDR case subjects with DME (10 eyes) before and after 2 months of twice-delivered anti-VEGF therapy (ranibizumab).

The following is a detailed description of the quantitative method. The newly developed software extracts HRF using 300 A-scans × 300 B-scans (3 × 3 mm) data acquired by ultrahigh-resolution SD-OCT. The first step was to perform noise reduction on the B-scans images using the Median filter. The second step was to extract the dot-like patterns. The dot patterns were extracted from the regions of the three-dimensional (3D) data where there were low-intensity regions around the high-intensity regions. In B-scan images, HRF and capillaries are perceived as similar shapes. Therefore, the software extracted a dot-like pattern in the 3D data. The third step was to exclude false positives other than HRF from all the extracted dot-like patterns. The dot-like pattern extracted in the second step contains a mixture of dots of various sizes, but we assumed that the HRF was ∼30 μm in size and excluded the others as false positives (15). To exclude false positives, the depth size was <15 μm or >30 μm, and the plane size was <10 μm or >40 μm. The difference in depth size and plane size is due to the different resolution of OCT. The ellipsoid zone and the retinal pigment epithelium were also recognized as similar dot-like patterns when they became blurred. Therefore, in this study, we used segmentation data and also excluded as false positives a region of 78 μm (1.3 μm/pixel × 60 pixels) from the lower surface of the retinal pigment epithelium to the inner retinal layer and a vitreous region of 60 μm from the inner limiting membrane. The last step was to measure the remaining dot-like pattern as HRF in VRI segmentation (Supplementary Fig. 1).

As a model of type 1 diabetes, heterozygous Akita (Ins2^Akita) on a C57BL/6 background, along with C57BL/6JJcl mice, were purchased from Japan SLC Company (Hamamatsu, Shizuoka, Japan). All mice were housed in institutional animal care facilities at Kyushu University (Protocol No. A29-263 and A29-51). Plasma glucose concentration was measured using an Accu-Chek glucometer (Roche Diagnostics, Basel, Switzerland). Kimba mice (trVEGF029), a model of human VEGF-induced retinal neovascularization (32,33) were provided by Prof. Elizabeth Rakoczy (University of Western Australia). Flt1 tk^−/− mice, a model lacking expression of VEGF receptor 1 (VEGFR1) (34) were generated by Prof. Masabumi Shibuya (University of Jobu). All mice had ad libitum access to water and standard laboratory chow and were housed in an air-conditioned room with a 12-h light and dark cycle. To create a type 1 diabetes model in which the retina had high VEGF expression, homozygous Kimba (vegf^+/+) mice were mated with heterozygous Akita (Ins2^+/−) mice to generate Akimba (Ins2^+/−vegf^+/−) mice. The blood glucose levels of these mice are shown in Supplementary Fig. 2 (35). Homozygous Kimba mice were mated with Ccr2^rfp/rfpCx3cr1^gfp/gfp mice to generate Ccr2^rfp/+Cx3cr1^gfp/+vegf^+/− mice. Homozygous Kimba mice were also mated with Flt1 tk^−/− mice (34) to generate Flt1 tk^−/−vegf^+/+ mice. All animal experiments were conducted in accordance with the Association for Research in Vision and Ophthalmology Statement for the Use of Animals in Ophthalmic and Vision Research and were approved by the Kyushu University Animal Care Committee (Protocol No. A29-263 and A29-51).

Mice were euthanized with an overdose of ketamine hydrochloride (Ketalar; Daiichi Sankyo Espha, Tokyo, Japan) and xylazine hydrochloride (Xylazine; Bayer, Puteaux, France). Both eyes were then enucleated and fixed with 4% paraformaldehyde (PFA) at 4°C for 30 min. The corneal limbus was incised circumferentially, and then the cornea, iris, lens, and choroid were removed from the eyeball. Subsequently, the retina was fixed with methanol at 4°C for 20 min. The retina was then incubated overnight with the primary antibodies and/or conjugated antibodies in 10% goat serum containing 1% Triton X-100 at 4°C, then washed three times for 15 min in PBS with Tween (PBST), followed by overnight incubation in the secondary antibodies in PBS at 4°C. After being rinsed three times for 15 min in PBST, a flat mount was prepared on a glass slide using a PermaFluor mount (TA-030-FM; Lab Vision Corporation, Fremont, CA). The flat mount was examined by fluorescence microscopy and digital images (500 μm × 500 μm) were captured using a fluorescence microscope (BZ-9000; KEYENCE, Tokyo, Japan) and a confocal microscope (A1; Nikon, Tokyo, Japan) with standardized illumination and contrast. A z stack of 55 photomicrographs (0.5 μm in thickness) was acquired, extending through 27 μm of the retinal structure. 3D reconstruction and analysis of z stacks was performed with NIS-Element (Nikon).

Some eyes were embedded in optimal cutting temperature compound (4583; Sakura Finetech, Tokyo, Japan), and kept at −80°C until sectioning. Then, 30-μm-thick sections were cut with a cryostat and placed on glass slides. Samples were washed three times for 15 min in PBST and incubated overnight with the primary antibodies in 10% goat serum containing 0.5% Triton X-100 and 5% BSA at 4°C. The samples were washed three times for 15 min in PBST, incubated with the secondary antibodies in PBS for 1 h at room temperature, and washed three times for 15 min in PBST. DAPI was used to counterstain the nuclei, and samples were mounted on glass slides using a PermaFluor mount. The cryostat sections were examined by fluorescence microscopy, and digital images were recorded using a fluorescent microscope (BZ-9000; KEYENCE).

The retinal cryostat sections and flat mounts were incubated with the following antibodies: Alexa Fluor 647 anti-mouse F4/80 (1:200 dilution; MCA497A647; AbD Serotec, Kidlington, U.K.), FITC anti-mouse CD80 (B7-1) (1:40 dilution; 110-0801-81; eBioscience, San Diego, CA), FITC anti-mouse CD206 (1:100 dilution; 141703; BioLegend, San Diego, CA), anti-mouse CD31 antibody (1:100 dilution; 550274; BD Biosciences, San Diego, CA), anti-mouse CD169 (1:100 dilution; MCA947G; Bio-Rad, Hercules, CA), anti–Ki-67 (1:100 dilution; MA5-14520, Thermo Fisher Scientific, Waltham, MA), anti-human CD11b (1:100 dilution; MCA74GA; Bio-Rad), and Hoechst 33342 (1:1000 dilution; H3570; Molecular Probes, Eugene, OR).

Following vitrectomy, the preretinal membranes were harvested from four patients with PDR, fixed with 4% PFA at 4°C for 30 min, and permeated in methanol at 4°C for 20 min. The tissue was incubated overnight with anti-human CD11b (1:100 dilution; MCA74GA; Bio-Rad) and anti–Ki-67 (1:100 dilution; MA5-14520; Thermo Fisher Scientific) in 10% goat serum containing 1% Triton X-100 at 4°C, washed three times for 15 min in PBST, and then incubated overnight with Alexa Fluor 647 goat anti-rat secondary antibody (10 µg/mL; A-21247; Invitrogen) and Alexa Fluor 488 goat anti-rabbit secondary antibody (10 µg/ml; A-11008; Invitrogen) in PBS at 4°C. The tissue was incubated with a VectorTrue VIEW autofluorescence quenching kit (SP-8400; Vector Laboratories, Burlingame, CA) at room temperature for 5 min. After the nuclei were counterstained using DAPI, the vitreous samples were spread on glass slides using a 27-gauge needle, mounted, and viewed with fluorescence microscopy (BZ-9000; KEYENCE).

Eyes from Kimba mice were enucleated and fixed in 1% glutaraldehyde and 1% PFA in PBS for 24 h. Then, the eyes were carefully cut, and the posterior segments were postfixed in veronal acetate buffer osmium tetroxide (2%) for 24 h. The samples were dehydrated in ethanol and water and embedded in epon. Ultrathin sections were cut from the blocks with an ultramicrotome and mounted on copper grids. Samples were observed under an HT-7700 transmission electron microscope (TEM; Hitachi, Tokyo, Japan), as previously reported (3).

In observations using mice, the mice were anesthetized, and their pupils were dilated with 5% phenylephrine and 0.8% tropicamide. The mice were subsequently placed on the board in front of the OCT (Kowa OCT Bi-μ; Kowa Company). An averaged B-scan was taken along the meridian through the optic disc.

In order to label cells in the S-phase of the cell cycle, mice were injected intraperitoneally with BrdU (ab142567; Abcam, Cambridge, U.K.) at a dose of 50 mg/kg daily for 7 days (postnatal day [P]14–P20). BrdU was also dissolved with intraocular irrigating solution (Opeguard-MA; Senju Pharmaceutical, Osaka, Japan) at a final intravitreal concentration of 2 μg/μL. Mice were also injected with BrdU (1 μg/0.5 μL) intravitreally at 1 μg at P14 by the method as shown below.

Mice were euthanized with an overdose of pentobarbital at P21, and both eyes were enucleated. The eyes were fixed with 4% PFA at 4°C for 30 min. Then, the corneal limbus was incised circumferentially, and the cornea, iris, lens, and choroid were removed from the eyeball. Subsequently, the retina was fixed with methanol at 4°C for 20 min. The samples were treated with 4 N HCl for 45 min at 37°C, followed by 0.1 mol/L borate buffer (pH 8.5) for 15 min. Next, the samples were incubated overnight with anti-mouse BrdU (ab6326; Abcam) in 10% goat serum containing 1% Triton X-100 at 4°C. The tissue was washed three times for 15 min in PBST. The tissue samples were incubated overnight with goat anti-rat Alexa Fluor 488 in PBS at 4°C. The samples were rinsed three times for 15 min in PBST and were incubated overnight with anti-mouse F4/80 (1:200 dilution; MCA497A647; AbD Serotec) in 10% goat serum containing 1% Triton X-100 at 4°C. Flat mounts were prepared as described above.

Mice were injected intraperitoneally with BrdU at a dose of 50 mg/kg daily for 7 days (P14–P20). Mice were euthanized with an overdose of pentobarbital at P21. Both eyes were enucleated, and peripheral blood was collected from the tail vein. The retina was excised and disrupted with scissors in DMEM containing 10% FBS, collagenase type D (1 mg/mL), and DNase (100 μg/mL). The sample was shaken at 37°C for 30 min and passed through a 40-μm mesh. Erythrocytes were lysed with ammonium-chloride-potassium lysing buffer (10-548E; Lonza). Retinal and leukocyte single-cell suspensions were stained with BV421 anti-mouse CD11b (1:100 dilution; 101235; BioLegend), phycoerythrin anti-mouse CX3C motif chemokine receptor 1 [CX3CR1] (1:100 dilution; 149005; BioLegend), APC-Cy7 anti-mouse CD206 (1:100 dilution; 321119; BioLegend). BrdU and 7-amino-actinomycin D were stained according to the protocol included with the kit (559619; BD Pharmingen, San Diego, CA). Flow cytometry was performed using BD FACSCalibur (BD Biosciences, San Jose, CA). Data were analyzed using Cytobank (https://premium.cytobank.org). The gating strategy for BrdU⁺ CX3CR1⁺ cells is shown in Supplementary Fig. 3.

Total RNA was extracted from whole retinas at the selected time points using a MagDEA Dx SV RNA kit (E1330; Precision System Science, Pleasanton, CA), according to the manufacturer’s instructions. RNA concentrations were determined, and cDNA was synthesized using a First Strand cDNA Synthesis Kit (04896866001; Roche, Mannheim, Germany). Quantitative RT-PCR was performed and analyzed using TaqMan gene expression assays (Applied Biosystems, Foster City, CA) and a LightCycler 96 real-time PCR system (Roche, Pleasanton, CA). Each Light Cycler well was loaded with 20 μL of the PCR mixture containing 1 μL of primer. The reference numbers for the assays were as follows: Mm99999915_g1 (GAPDH), Mm00441242_m1 (Ccl2; MCP-1), Mm00445259_m1 (Il4; interleukin 4), Mm00446190_m1 (Il6; interleukin 6), Mm9999068_m1 (Tnf; tumor necrosis factor-α), Mm00432686_m1 (Csf1; macrophage colony-stimulating factor), Mm00435610_g1(Pgf; placental growth factor), Mm00456503_m1 (Angpt1; angiopoietin1), Mm00545822_m1 (Angpt2; angiopoietin2), Mm00441786_m1 (Tie1; Tie1), and Mm00443243_m1 (Tek; Tie2). GAPDH was used as an endogenous control. An initial step of 10 min at 95°C was used to activate the HotStart DNA polymerase, and then a two-step cycling program, including 45 cycles of 95°C for 20 s and 60°C for 40 s, was used for the TaqMan assays. The LightCycler 96 real-time PCR system software was used for detecting the probe, calculating the threshold cycles values, and additional analyses.

This study (UMIN000014724) was approved by the Institutional Review Board and performed in accordance with the ethical standards of the 1989 Declaration of Helsinki. Written informed consent was obtained from each participant. Vitreous samples (1,000–1,200 μL) were collected from patients who underwent vitrectomy because of idiopathic epiretinal membrane (iERM) or PDR at the Kyushu University Hospital. Samples with obvious bleeding were excluded. Vitreous samples were immediately centrifuged, and the supernatants were snap-frozen.

Bovine vitreous macrophages were isolated and cultivated in type I collagen-coated dishes poured DMEM (Nakalai Tesque, Kyoto, Japan) with 20% FBS (Invitrogen-Gibco, San Diego, CA), as previously reported (4). Hyalocytes obtained between passages 4 and 7 were used in the experiments.

In order to determine the extent of cellular proliferation of bovine hyalocytes, the incorporation of BrdU was quantified using a BrdU labeling and detection kit (11296736001; Roche, Indianapolis, IN), following the manufacturer’s instructions. Briefly, 1 × 10⁴ cells were cultured in DMEM (Nacalai Tesque) in 96-well μL plates at 37°C under 5% CO₂. Cells were washed 24 h after plating, and the media were replaced with 100 μL vitreous from ERM or PDR patients. At 48 h after stimulation, 10 μmol/L BrdU was added to each well. The cells were continuously cultured under the same conditions for 2 h, during which BrdU was incorporated into freshly synthesized DNA. Following fixation of cells, cellular DNA was partially digested by nuclease treatment. The incorporated BrdU was detected and quantified using a peroxidase-labeled antibody for BrdU. Peroxidase catalyzes the cleavage of the substrate, which produces a color reaction. The absorbance at 450 nm was directly correlated to the level of BrdU incorporated into cellular DNA and was measured using a microplate reader (Bio-Rad) to assess the extent of hyalocyte proliferation.

Anesthetized mice were injected with either BrdU (1 μg/0.5 μL), intraocular irrigating solution (Opeguard-MA; Senju Pharmaceutical, Osaka, Japan), anti-VEGF drug (aflibercept; 40 mg/mL; 0.5 µL) (Eylea; Regeneron Pharmaceuticals, Tarrytown, NY, and Bayer, Basel, Switzerland), or human IgG1 (0.5 µL; bingo bio, Hayward, CA), and clodronate encapsulated liposomes (0.5 µL; Clodrosome; Encapsula Nanosciences LLC, Brentwood, TN), or plain liposomes for control (0.5 µL; Encapsome, Encapsula Nanosciences LLC). These were injected intravitreally at ∼1 mm from the corneal limbus at P14 and P17 using a Hamilton syringe (33-gauge needle; 65460-02; Hamilton Co., Reno, NV). The eyes were enucleated at P21.

Univariate analyses were performed using JMP 13.0 software (SAS Institute, Cary, NC). Continuous variables were analyzed using the Wilcoxon rank sum test, Student t test, or the Tukey-Kramer test, as appropriate. Categorical variables were assessed using the χ² test. Differences between the experimental groups were considered statistically significant or highly significant when the P probability value was <0.05 or <0.01, respectively.

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

OCT B-scan imaging of the VRI revealed that HRF were significantly more frequent in DR subjects compared with healthy subjects and individuals with diabetes but NDR (healthy, 26.7%; NDR, 25.0%; and DR, 64.8%; χ² = 8.2; P = 0.004) (Fig. 1A–C and Table 1). The number of HRF at the vitreous interface was increased as the stage of DR progresses using B-scan image (Fig. 1D and E). En face quantitative imaging also confirmed an increase in HRF at the VRI with DR progression (Fig. 1F). The HRF at the VRI were significantly more prominent in PDR than in NDR or NPDR (Fig. 1G). Taken together, these observations suggest an increase in the number of HRF at the VRI during DR progression.

Next, we investigated whether anti-VEGF therapy could impact the number of HRF at the VRI in DR patients. As one exemplar, we examined HRF at the VRI in a PDR patient with DME who received anti-VEGF therapy (ranibizumab) (Fig. 2A). Overall, OCT B-scan images showed reduced macular thickness in all DME patients after anti-VEGF therapy (Fig. 2B, upper panel). Quantitative analysis with en face OCT showed that the abundance of HRF at the VRI after anti-VEGF treatment was significantly reduced, suggesting that anti-VEGF therapy decreases the number of HRF at the VRI in DR patients (Fig. 2B, lower panel, and C).

HRF in OCT have been suggested to be lesions with cellular components consisting of macrophages, retinal pigment epithelium, or photoreceptors (15–19). Since DR has an established proinflammatory etiology, we hypothesized that HRF occurring at the VRI could be derived, at least in part, from VRMs (2). To investigate the number of macrophages at the VRI in DR, immunostaining with F4/80 was performed on retinas of Akimba mice. F4/80⁺ macrophages at the VRI of Akimba mice were increased significantly compared with wild-type (WT) mice. Next, we performed immunostaining with F4/80 in Akita and Kimba mice to determine whether the increase in macrophages at the VRI of Akimba mice was due to VEGF or diabetes. VRI in Kimba mice, but not in Akita mice, had significantly more F4/80⁺ macrophages than WT mice. Further, there was no difference in the number of F4/80⁺ macrophages between Kimba and Akimba mice (Fig. 3A and B), indicating that VEGF-A, but not diabetic status, affected the number of macrophages at the VRI. As reported previously (33), the overexpression of human VEGF in Kimba mice is short and transient between P7 and P20. We published this time-course data in our previous study (36). Compared with 10-day-old mice, the number of VRMs was significantly higher in 3-week-old Kimba mice, with the maximum level of VEGF expression in the retina (36) (Fig. 3C and D). After that, the number of VRMs was significantly reduced in 8-week-old mice compared with 3-week-old mice in accordance with the VEGF level (37) (Fig. 3C and D). Since VRMs are located on the vitreous side of the internal limiting membrane at an average distance of 50 μm from the inner surface of the retina (37), we identified the vertical structure in the retinal flat mount of Kimba mice using confocal images. 3D images revealed the presence of F4/80⁺ macrophages on the surface of the Kimba retina (Fig. 3E). TEM also showed that these ameboidal-shaped cells assumed many forms and possessed lysosome-like granules, mitochondria, and micropinocytic vesicles, presenting the characteristics of vitreous VRMs on the inner limiting membrane (Fig. 3F). Immunohistochemistry also showed most F4/80⁺ macrophages at the VRI expressed CD31 in Kimba and Akimba mice, suggesting activated macrophages in VEGF-A overexpressing retinas (38) (Fig. 3A). These data suggest that high levels of retinal VEGF could increase the number of F4/80⁺ macrophages at the VRI.

To test our hypothesis that HRF at the VRI are indeed macrophages, we treated Kimba mice with clodronate liposomes (dichloromethylene diphosphonate liposomes [Cl2MDP-LIP]). Intravitreal Cl2MDP-LIP injection showed a significant reduction in F4/80⁺ cells at the VRI compared with the control liposomes in mice (Fig. 4A and B). A few HRF-like lesions could be detected in C57BL/6J WT mice using OCT, whereas more HRF were observed in Kimba mice (Fig. 4C). Intravitreous Cl2MDP-LIP treatment in Kimba mice also significantly reduced HRF in the vitreous, but not in the retina, compared with the treatment of control liposomes (Fig. 4D and E), indicating that a part of HRF at the VRI could be vitreous macrophages.

Immunostaining to investigate the phenotype of VEGF-enhanced vitreous F4/80⁺ macrophages showed costaining with CD169, which is a marker of TRMs in the skin, liver, spleen, and choroid (24,39,40). Furthermore in Kimba mice, we identified the VRMs expressing CD206, a representative marker of M2-like macrophages, but not CD80 (positive control is shown in Supplementary Fig. 4), a representative marker of M1-like macrophages (41–43) (Fig. 5).

To further explore the phenotype, we produced Ccr2^rfp/+Cx3cr1^gfp/+ and Ccr2^rfp/+Cx3cr1^gfp/+vegf^+/− mice. Vitreous cells of both types of mice showed CX3CR1 expression, but not CCR2, at the VRI, confirming VRMs (Fig. 6A). CCR2⁺ cells, classical circulating monocytes, were observed in the retinas, but not at the VRI of Ccr2^rfp/+Cx3cr1^gfp/+vegf^+/− mice (Fig. 6B). Confirmation of VEGF-mediated increase in CX3CR1⁺ VRMs motivated us to investigate local self-renewal in the vitreous local microenvironment. To explore local proliferation, we detected the capture of BrdU by daily intraperitoneal injection of BrdU for 7 days from P14. Flow cytometry revealed BrdU incorporation in 31.4% of CD206⁺ and CX3CR⁺ cells in Kimba mice retinas (Fig. 6C). Conversely, there was no capture of BrdU in CX3CR1⁺ monocytes in the peripheral blood. In immunohistochemistry, BrdU was incorporated in 50.6% of F4/80⁺ VRMs in Kimba mice (Fig. 6D). Furthermore, retinas from Kimba mice were stained with Ki-67. Immunohistochemistry showed Ki-67 staining in a part of the F4/80⁺ VRMs (Fig. 6E), indicating proliferation of these macrophage cells in VEGF-overexpressing retinas. To rule out the possibility of macrophage proliferation in the blood, BrdU was injected into the vitreous, and its uptake was also examined. BrdU incorporation was also observed on day 14 after intravitreous BrdU injection (Fig. 6F), indicating the local proliferation in the vitreous. We next examined the retinal expression levels of Ccl2, Il4, Il6, Csf1, Pgf, and Tnf, which have been reported to induce local proliferation of TRMs in other nonvitreoretinal tissues (27,43,44). Expression of all genes, except Il4, was increased in Kimba mice compared with WT mice (Fig. 6G). Il4 expression was undetectable in mouse retinas. Furthermore, in vitro experiments with PDR patient-derived vitreous demonstrated a significant induction of BrdU uptake in VRMs compared with vitreous from patients with iERM or no-vitreous control subjects (Fig. 6G). These data indicate that the vitreous from PDR patients can cause the proliferation of VRMs. Lastly, to investigate whether local proliferation of macrophages occurs in patients with PDR, immunohistochemistry was performed with Ki-67 using epiretinal membranes from PDR patients. We observed 3.3 ± 0.4 Ki-67⁺ and CD11b⁺ cells per image (Fig. 6H), suggesting local proliferation of macrophages in PDR patients.

Intravitreal treatment with the anti-VEGF drug, aflibercept, significantly reduced the number of VRMs and the proportion of BrdU uptake compared with IgG treatment in Kimba mice (Fig. 7A–C). OCT also confirmed that treatment with intravitreal aflibercept significantly reduced the HRF at the VRI compared with the treatment with IgG (Fig. 7D and E). To investigate whether the reduction in TRMs in VEGF inhibition was directly affected by the R1 receptor expressed on macrophages, vegfr1 tk^−/− mice were crossed with Kimba mice. However, there were no differences in the number of VRMs between vegf^+/+ mice, Kimba, and vegfr1 tk^−/−vegf^+/+ mice. These data suggest that VEGF-induced local proliferation of VRMs was not mediated by VEGFR1 (Fig. 7F and G).

The clinical significance of HRF has been reported in DR, especially in DME (15,16), although the precise nature of these lesions at the VRI has not been investigated. This prospective study revealed that HRF at the VRI were clearly observable in DR and that they increased with the progression of disease stage. Significantly, HRF were reduced by anti-VEGF therapy, suggesting that the elevated levels of VEGF known to occur in many patients with progressive DR may be linked to formation of these lesions at the VRI.

The precise cellular identity and pathogenesis of HRF has been uncertain, although clinical observations in DR often show cellular inclusions at the VRI that resemble HRF-like lesions (1,2,14). This study provides the first evidence identifying HRF as being linked to macrophage activation in PDR. Nevertheless, the molecular mechanism(s) linking VRMs and DR and whether they contribute to HRF remain unknown. In the present investigation, immunohistochemistry and flow cytometric studies showed increased proliferation and numbers of VRM at the VRI of VEGF-overexpressing retinas in mice, but not in the retinas of diabetic mice. OCT of VEGF-overexpressing mouse retinas also showed an increase in HRF at the VRI, which is consistent with our OCT findings from DR patients. The number of HRF was reduced by Cl2MDP-LIP depletion of phagocytes, indicating that the HRF at VRI are phagocytes in the vitreous, presumably resident macrophages (hyalocytes). These data suggest that changes in the microenvironment triggered by VEGF upregulation in DR induces local proliferation of VRMs at the VRI, which can be observed as HRF in OCT. Furthermore, our in vitro data also confirmed that vitreous samples from PDR patients were sufficient to induce VRM proliferation (Fig. 6), indicating that these specialized macrophages are influenced by PDR vitreous (Fig. 8). Diabetic vitreopathy is an established concept, and our data support the concept that VRMs are significant in PDR pathogenesis, especially at the VRI (11,12) (Fig. 8).

The current study has revealed that retinal VEGF could play a role in local macrophage proliferation at the VRI in addition to inflammatory cell infiltration, angiogenesis, and vascular hyperpermeability (7,8,45). Because VRMs can be a source of various cytokines (2), and the pathological environment of VRI is important not only for progression to DME but also for progression to PDR (1,13), proliferation of this macrophage subpopulation may contribute to various pathologies in PDR, such as fibrovascular membrane formation (2,4,5). However, our current data do not prove the impact of VTM proliferation on the progression to PDR, and further studies are needed to address this issue.

The later stages of DR, especially PDR, have been reported to increase not only VEGF but also various proinflammatory cytokines in the eye (36,46). In this study, VRM proliferation at the VRI was observed in Kimba mice overexpressing VEGF in the photoreceptors. Although VEGF expression started from P5 in the transgenic mice, administration of anti-VEGF drugs from P14 was enough to suppress the local proliferation of VRMs. This suggests that local proliferation is induced in a VEGF-dependent manner despite the possible upregulation of other cytokines. Almost no F4/80⁺ macrophages were observed in 10-day-old Kimba mice, whereas 3- or 8-week-old WT mice had a few F4/80⁺ cells. Furthermore, Xu et al. (23) previously reported the presence of a few BrdU⁺ cells coexpressing F4/80 in normal mouse retina. This further supports that VRM proliferation might partially include localized activation of immune cells associated with eye development.

VEGF binds VEGFR1 and R2, and macrophages express VEGFR1 (also called Flt1), but not R2 (47). Pharmacological VEGF inhibition blocked VRM proliferation in VEGF-overexpressing retinas, whereas genetic VEGFR1 did not. These data indicate that VEGF-induced VRM self-renewal is not a direct effect on macrophages via VEGFR1. In this model, we also observed upregulation of various cytokines, which are reported to be able to induce macrophage proliferation (27,43,44). Therefore, VEGF-induced VRM self-renewal could be due to the secondary effect associated with vascular change and/or cytokine upregulation via VEGFR2, which contribute to driving vascular leakage and DME (7). Moreover, VEGFR1 tyrosine kinase signaling is known to play an important role in macrophage infiltration and migration (48). However, vegfr1 tk^−/−vegf^+/+ mice showed no reduction in F4/80⁺ macrophages at VRI, suggesting that VEGF-induced accumulation of F4/80⁺ macrophages at the VRI is not affected directly via VEGFR1-mediated macrophage recruitment. Furthermore, VEGF-overexpressing retinas showed significant upregulation of transcripts relating to the angiopoietin family, including Ang1, Ang2, and Tie1 but not Tie2 (Supplementary Fig. 5). Such growth factor signaling has been reported to induce macrophage trafficking (49).

The impact of anti-VEGF therapy on DME and DR has been reported clinically. A post hoc analysis showed improvement of the DR stage during anti-VEGF therapy (10). However, the detailed suppressive mechanism is unknown. We previously showed that intravitreal anti-VEGF therapy suppresses leukocyte trafficking in the retina (45). In the current study, we observed that intravitreal anti-VEGF drug injection blocked the local proliferation of VRMs. These inhibitory effects may be a candidate mechanism for improving DR stage by anti-VEGF therapy in addition to glycemic control (10).

Bolz et al. (15) reported that intraretinal HRF could be a sign of early breakdown of the blood-retinal barrier in DME. Our observation using OCT revealed significantly more HRF at the VRI in DR case subjects than in healthy subjects. Our experimental study with intravitreous administration of Cl2MDP-LIP confirmed macrophage depletion at the VRI, but not in the retina, which led to a concomitant decrease in HRF at the VRI. This result strongly suggests that HRF at the VRI are, at least in part, formed by macrophages responding to high retinal VEGF expression, although the concordance between HRF and VRI has not been directly proven. We also observed an increase in HRF at the VRI during the progression of DR. This was consistent with the increased macrophage fraction at the VRI when exposed to VEGF for a longer time in mice. Recently, high-resolution imaging techniques, such as adaptive optics scanning laser ophthalmoscopy, have been used to detect macrophage-like morphology in DR patients (50). Scoles et al. (51) also suggested that hyperreflective dots in the adaptive optics scanning laser ophthalmoscopy that did not move or change in appearance for some months in healthy individuals were presumably resident cells. HRF at the VRI can be a biomarker of disease progression in DR. However, this conclusion requires longer continuous prospective observations during DR progression in the same patient. Furthermore, the cellular nature of intraretinal HRF is not yet elucidated.

In this study, the conclusion was that VEGF affected HRF or VRI more than hyperglycemia; however, the duration of hyperglycemia in these mice could be still too short to evaluate the effect of diabetes on HRF or VRI at the VRI. Therefore, our current data do not deny the impact of hyperglycemia on HRF or VRI. Kimba mice were used to examine the impact of VEGF on vitreous macrophages; however, the model did not show retinal angiogenesis toward the vitreous, indicating that this mouse is not a DR model despite high VEGF expression in the retina (32,33,36). Future studies using other diabetic models may support our observation of VRM proliferation at the VRI in DR in spite of lack of appropriate PDR animal models. Although phagocyte fractions other than hyalocytes have not been identified at the VRI (1), Cl2MDP-LIP eliminates not only macrophages but also other subfractions of phagocytes; therefore, we cannot deny the possibility that cells other than VRM were also eliminated by this treatment. Moreover, our FACS investigation of Ki-67⁺ cells was performed using retinal samples because it was difficult to harvest only the VRI region, although immunostaining could confirm many Ki-67⁺ cells at the VRI. Our clinical observation had various limitations inherent in any study with a limited sample size. The study of HRF was performed in a limited area of 3 × 3 mm in the macula, despite DR occurring throughout the retina. Therefore, a wider field of view using OCT is required, because DR also shows vascular abnormalities in peripheral lesions.

In conclusion, our observation from patients as well as mouse models suggests that VEGF overexpression induces local proliferation of VRMs at the VRI in PDR and that HRF at the VRI could be followed through OCT imaging in patients. This biomarker may be a better diagnostic and therapeutic target for PDR progression as well as a risk assessment for fibrovascular scar formation.

Acknowledgments. The authors are grateful to Prof. Elizabeth Rakoczy (University of Western Australia) for gifting Kimba mice. The authors also thank Masayo Etou, Mitsuhiro Kurata, Mikako Kumano, Yuka Matsutani, and Takako Iwasaki (Kyushu University) for their technical assistance. The authors would like to thank Editage (www.editage.com) for English language editing.

Funding. This study was supported by grants from Japan Society for the Promotion of Science (JSPS) Grants-in-Aid for Scientific Research National Institute of Informatics (KAKENHI) no. 17K11456 (S.N.), no. 20K09829 (S.N.), and no. 21K16897 (M.Y.), the Charitable Trust Fund for Ophthalmic Research in Commemoration of Santen Pharmaceutical’s Founder (S.N.), and The Foundation for The Advancement of Clinical Medicine, Fukuoka, Japan (S.N.).

Duality of Interest. This study was also supported by the Takeda Science Foundation, Bayer Retina Award (S.N.), Novartis Pharmaceuticals research grants (S.N.), Alcon research grants (S.N.), and grants from Kowa Company, Ltd. This research was conducted in collaboration with Kyushu University and Kowa Company, Ltd. S.N. and K.-H.S. received research support fees from Kowa Company, Ltd. M.M. is an employee of Kowa Company, Ltd. Kyushu University and Kowa Company, Ltd. have a patent (Patent No. 6860884) related to software that quantifies HRF from en face OCT images. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. M.Y. and S.N. were involved in data analysis. M.Y., S.N., I.W., M.A., Y.K., and K.I. performed the in vivo experiments. M.Y., S.N., I.W., Y.K., and M.M. performed OCT in patients and mice. M.Y., S.N., and T.M. and designed the experiments. M.Y. and M.Shir. performed immunohistochemistry, FACS, and PCR. M.Y., S.N., and A.W.S. wrote the manuscript. M.Y. and T.N. conducted the in vitro experiments. M.Y. and T.H. performed TEM. S.N. conceived the project. S.N. harvested the clinical samples with vitrectomy. T.M., Y.M., T.H., T.I., M.Shib., A.W.S., and K.-H.S. revised the manuscript. W.S. and R.Y. generated Ccr2r^fp/rfpCx3cr1^gfp/gfp mice. M.Shib. generated Flt1 tk^−/− mice. S.N. 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.