All animal work was reviewed and approved by the local Institutional Animal Care and Use Committee (IACUC; University of Oklahoma Health Sciences Center, Oklahoma City, OK, United States). Mice on the C57BL/6 background that were homozygous for an Igf1 floxed allele (Igf1^f/f) were purchased from Jackson Laboratories (line 016831). These mice have exon 4 of the Igf1 gene flanked by loxP sites, allowing for the excision of this entire exon via Cre recombinase. Transcripts of the altered Igf1 gene yield a protein that fails to bind the IGF-1 receptor. IGF-1 deficiency was induced in Igf1^f/f mice by adeno-associated virus (AAV8)-mediated expression of Cre recombinase in the liver at 4 months of age, as previously reported (Bailey-Downs et al., 2012; Toth et al., 2014). The AAV8 vector was acquired from the University of Pennsylvania Viral Vector Core (Penn Vector Core, Philadelphia, PA, United States¹). The thyroxine-binding globulin (TBG) promoter was used to restrict the expression of the AAV8 vector to hepatocytes. At 4 months of age, Igf1^f/f mice were randomly assigned to two groups and were administered approximately 1.3 × 10¹⁰ viral particles of AAV8-TBG-Cre or AAV8-TBG-eGFP via retro-orbital injection, as described (Tarantini et al., 2017c). The majority of circulating IGF-1 is produced in the liver. Since IGF-1 is critical for the development of many organ systems during adolescence, including the cardiovascular system, this model was used to specifically study the effects of adult-onset circulating IGF-1 deficiency (Toth et al., 2014). Animals were used for experiments at either approximately 1 year of age (12–14 months) or 2 years of age (24–27 months). Both male and female mice were used, however studies were not powered to evaluate the role of sex as a biological variable. Animals were housed in the Rodent Barrier Facility at OUHSC under specific pathogen-free barrier conditions, on a 12-h light/12-h dark cycle, with access to standard rodent chow (Purina Mills, Richmond, IN, United States) and water ad libitum. In-cage light levels during the light cycle were ∼30 lux.

To assess microvascular fragility, hypertension was induced in study animals at 12–14 months of age. AAV8-TBG-Cre and AAV8-TBG-eGFP mice were randomly assigned to either the “hypertensive” (HT) or “normotensive” (NT) groups. Hypertension was induced by a combination treatment with ω-nitro-_L-arginine-methyl ether [L-NAME (N5751, Millipore Sigma, St. Louis, MO, United States), 100 mg kg^–1 day^–1, in drinking water] and administration of angiotensin II [Ang II; s.c. via osmotic mini-pumps (Alzet Model 2006, 0.15 μL h^–1, 42 days; Durect Co, Cupertino, CA, United States)]. Pumps were filled with either saline or a solution of angiotensin II (Sigma Chemical Co., St. Louis, MO, United States) that delivered 1 μg min^–1 kg^–1 of angiotensin II for up to 28 days. Mini-pumps were surgically placed in isoflurane anesthetized mice in the subcutaneous space on the back of the animal. This was accomplished by making a small incision in the intrascapular region, blunt dissection of the subcutaneous space, and closure of the incision with surgical sutures using aseptic techniques. Animals were given an s.c. injection of sustained release Buprenorphine (ZooPharm, Fort Collins, CO, United States) to manage post-operative pain. Animal blood pressure was measured via the tail-cuff method using the CODA Non-Invasive Blood Pressure System (Kent Scientific Co., Torrington, CT, United States) as described (Toth et al., 2015b). Mice were placed in a restrainer on an animal warmer for the duration of the measurement to encourage tail vein dilation for accurate blood pressure measurements.

Mice underwent daily neurological examination to predict the presence of clinically manifest hemorrhages. The scoring system evaluates spontaneous activity, symmetry in limb movement, forelimb outstretching, climbing ability, body proprioception, and response to vibrissae touch. Scores were summed on an 18-point scale as a measure of neurological function. A decline in this neurological score correlates with cerebral hemorrhage development (Toth et al., 2015b). Mice were euthanized at a neurological score of 15 or lower, or when they reached day 28 post-hypertension surgery, whichever came first. Mice were euthanized via transcardial perfusion with ice-cold 1× phosphate buffered saline (1× PBS, 137 mM NaCl, 2.7 mM KCl 10 mM Na₂HPO₄, 1.8 mM KH₂PO₄, pH 7.4, pH 7.2) for 10 min under ketamine/xylazine (84/14 mg kg^–1) anesthesia.

To assess vascular damage in the retina, fundus imaging and fluorescein angiography were performed using the Micron III system (Phoenix Research Laboratories, Pleasanton, CA, United States) as described (Koirala et al., 2013; Chakraborty et al., 2020). Mice were anesthetized with ketamine/xylazine (84, 7 mg kg^–1)) and eyes were dilated with 1% cyclopentolate eye drops (Family Medicine Pharmacy, University of Oklahoma Health Sciences Center, Oklahoma City, OK, United States). One drop of 2.5% Gonak (McKesson Medical Surgical, Richmond, VA, United States) was applied to each eye. Bright field images were collected, and then animals were injected intraperitoneally with 100 μL of 1% (w/v) fluorescein sodium (Sigma-Aldrich, St. Louis, MO, United States) for angiography. Angiogram images were captured using a GFP filter. All fundus images were captured using StreamPix software (Phoenix Research Labs, Bend, OR, United States). We observed repeated patterns of constriction in some retinal vessels, to semiquantitatively assess this phenotype, the number of affected vessels in each image was counted by an observer blinded to age and genotype.

Electroretinography (ERG) measures the electrical responses of various cell types in the retina, including the photoreceptors, inner retinal cells (bipolar and amacrine cells), and the ganglion cells, which are sensitive to hypoxia and structural damage to the neural retina. To assess functional consequences of accelerated microvascular aging associated with IGF-1 deficiency, full-field ERG measurements were performed in the experimental mice as described by Chakraborty et al. (2020). Following dark adaptation, mice were anesthetized with ketamine/xylazine and eyes were dilated using 1% cyclopentolate eye drops. ERGs were recorded with the Diagnosys Espion E3 ERG system (Diagnosys LLC, Lowell, MA, United States). Scotopic ERGs were recorded with a strobe flash stimulus of 157 cd-s m^–2 presented to the dark-adapted mouse followed by light-adaptation for 5 min at 29.03 cd m^–2. Photopic responses were recorded from 25 averaged flashes at 77 cd-s m^–2 for white light. Flicker ERGs were recorded for 30 s in response to a 10 Hz flicker stimulus.

Neurovascular coupling responses were tested in a subset of 2 year-old and young control mice (∼3–6 months of age) according to our previously reported protocol (Tarantini et al., 2017a,2018; Yabluchanskiy et al., 2020). Mice were anesthetized with isoflurane (4% for induction and 1–2% for maintenance during the surgery and measurements). A femoral artery cannula was placed in each animal to monitor and maintain blood pressure in the physiological range during the procedure (between 90 and 110 mmHg). Mice were then endotracheally intubated and ventilated (MousVent G500; Kent Scientific Co., Torrington, CT, United States). Rectal temperature was maintained at 37°C using a thermostatic heating pad (Kent Scientific Co., Torrington, CT, United States). Mice were placed in a stereotaxic frame (Leica Microsystems, Buffalo Grove, IL, United States) and a thinned-skull cranial window was prepared. To prepare the thinned-skull window, the scalp and periosteum were resected and the skull was thinned with a sterile scalpel blade. Nitrocellulose lacquer was then applied to the surface of the skull to allow for appropriate optics and light spreading. A laser speckle imager (Perimed, Järfälla, Sweden) was positioned 10 cm above the window. The change in CBF in response to whisker stimulation was measured by stimulating the whiskers on one side of the mouse for 30 s intervals and measuring the change in CBF of the contralateral whisker barrel cortex. The change in CBF is expressed as the percent increase from the baseline.

Blood was collected via puncture of the submandibular vein with a sterile lancet or 25G needle. The whole blood sample was allowed to coagulate for 20 min at room temperature and was then centrifuged at 2,500 × g for 20 min at 4°C. Serum was collected and stored at –80°C until use. IGF-1 concentration in the serum samples was measured by enzyme-linked immunosorbent assay (ELISA) (R&D Systems, Minneapolis, MN, United States) as previously reported (Toth et al., 2014). An IGF-1 control sample was included on each plate. Serum IGF-1 levels were reported in ng mL^–1.

Whole eyes were collected from transcardially perfused mice and fixed in EM grade 4% paraformaldehyde (PFA) for 4 h at 4°C. Eyes were paraffin-embedded and sectioned at 6 μm thickness onto glass slides. After deparaffinization, slides underwent antigen retrieval in a solution of 10 mM citrate buffer (pH 6.0) for 20 min in a vegetable steamer and were then allowed to cool for 20 min at room temperature. Slides were then pre-treated with 1% sodium borohydride for 90 s, followed by three water washes and three 1× PBS washes. Blocking was performed in a blocking solution containing 5% BSA, 1% fish gelatin, 2% donkey serum, and 0.5% Triton in 1× PBS for 1 h at room temperature, followed by incubation in antibody overnight at 4°C in a humidity chamber (antibodies are listed in Table 1). The slides were then washed four times in 1× PBS for 10 min, incubated in secondary antibodies at a concentration of 1:500 for 1–2 h at room temperature, washed four additional times in 1× PBS, and then mounted with Prolong Diamond with 4’,6-diamidino-2-phenylindole (DAPI) (Thermo Fisher Scientific, Waltham, MA, United States) and coverslipped. An Olympus BX62 microscope with a 20× air, 40× air, or 100× oil objective (Olympus Life Science, Waltham, MA, United States) was used for fluorescent imaging. To semiquantitatively assess the presence of GFAP labeling of gliosis, each image was scored by an observer blinded to age and genotype. Scores were assigned as follows: 0: no gliosis (GFAP labeling in endfeet only), 1: mild gliosis (very little sign of GFAP penetrating into other retinal layers), 2: moderate/intermittent gliosis, 3: elevated gliosis, 4: very high levels of gliosis.

Brains and eyes were collected from transcardially perfused mice (described above) and fixed in 4% paraformaldehyde for 48 or 4 h, respectively, at 4°C. Brains and eyes were stored in PBS at 4°C until they were embedded in paraffin. Serial coronal brain sections were cut at 8 μm thickness, yielding approximately 1,000 sections per brain. Every twelfth slide from each brain was stained with 3,3-diaminobenzidine (DAB) (Vector Laboratories/Maravai LifeSciences, San Diego, CA, United States) to reveal hemorrhages, and counterstained with Gill’s No. 1 hematoxylin (Millipore Sigma, St. Louis, MO, United States) to show brain structure. DAB reacts with endogenous peroxidases in red blood cells generating a dark brown product that allows for easy and precise detection of extravasated blood in the brain parenchyma. All stained sections were imaged at 10× (for brains) or 20× (for retinas) using the VS120-L100-W Virtual Slide Microscope (Olympus Life Science, Waltham, MA, United States). Images were then analyzed by an observer blinded to genotype, group, and age. ImageJ 1.52p (NIH, Bethesda, MD, United States) software was used to quantify and measure the size of hemorrhages. Images were color deconvolved and thresholded uniformly; then, the pixel intensity integrated density was measured on the selected bleed area using a protocol and ImageJ macro described in (Nyúl-Tóth et al., 2020). Identified hemorrhages were then mapped to specific brain regions by comparing images to the Allen Mouse Brain Atlas.² A similar workflow was followed for identifying microbleeds in eyes. Eyes were serially sectioned at 6 μm thickness, yielding approximately 200 sections per eye. Every sixth section was stained with hematoxylin and DAB. A brain section adjacent to one with a bleed was included in each batch of retinal staining to serve as a positive control for the DAB. However, no microbleeds were identified in the eye sections so bleeds could not be counted, measured, or mapped to regions of the eye.

Central retinal sections (through the optic nerve) were stained with hematoxylin and eosin (H&E), and then imaged at 20× on a VS120-L100-W Virtual Slide Microscope. Spidergrams were generated by measuring outer nuclear layer (ONL), outer plexiform layer (OPL), and inner nuclear layer (INL) thickness at defined distances from the optic nerve head. Values were captured from three sections per eye, by an observer blinded to age and genotype and at least three eyes per genotype/age/group were measured. Thickness measurements were made using Adobe Photoshop CS6.

Statistical analyses were performed using Graphpad Prism version 9.2.0. Differences between groups were analyzed by two-tailed unpaired t-tests (to compare two groups), one-way ANOVA with Tukey’s post hoc comparison (to compare more than two groups), or two-way ANOVA with Tukey’s post hoc comparison (in cases where there were two different variables). Differences between survival curves were assessed by Log-rank (Mantel–Cox) test. When data were not normally distributed, the Mann–Whitney test was used.

To study the effects of IGF-1 deficiency on the development of vascular pathologies in the central nervous system, we used a mouse line in which exon 4 of the IGF-1 gene is floxed (Igf1^f/f). We knocked down circulating IGF-1 levels in young adult mice (4 months of age) by injection of an AAV carrying a liver-specific promoter driving Cre recombinase expression [TBG-Cre-AAV8, here referred to as IGF-1 KD (“knockdown”)]. Control animals (also Igf1^f/f) received AAV8-TBG-eGFP at 4 months of age (here referred to as control). The liver is the primary source of paracrine IGF-1 (Ungvari and Csiszar, 2012; Blum et al., 2018), and knocking down IGF-1 production in hepatocytes causes a significant decrease in serum levels of IGF-1. Using ELISAs on serum harvested at 1 year of age, we confirmed that this AAV-mediated approach results in almost complete elimination of circulating IGF-1 levels in our knockdown animals (Figure 1A). We elicited CMHs at 1 year of age by inducing hypertension using a well-established paradigm (Toth et al., 2015b; Tarantini et al., 2017c) wherein Ang II is delivered via osmotic mini-pump and L-NAME is delivered in drinking water (Figure 1B). Normotensive control mice did not receive Ang II or L-NAME. To track the onset of neurological signs of CMHs, mice underwent daily neurological scoring which involved assessment of six different criteria on an 18-point scale. As expected, hypertensive IGF-1 KD mice experienced an earlier onset of neurological signs of CMHs (defined as a drop in neurological score from 18 to 17 or below) compared to control animals (Figures 1C,D). No normotensive animals (either IGF-1 KD or control) exhibited a change in neurological score. Tissues were collected when mice exhibited a neurological score of 15 or lower (or at 28 days after insertion of the mini-pumps). Consistent with their earlier onset of neurological signs of CMHs, hypertensive IGF-1 KD mice were removed from the study and euthanized at a significantly earlier time than control mice due to severe neurological decline (Figure 1E). CMHs were identified histologically by DAB staining (reddish brown color, Figure 1G) which reacts with endogenous peroxidases in red blood cells. We calculated the mean number of bleeds per brain section and found that hypertensive IGF-1 KD exhibited significantly more bleeds per section than hypertensive control animals (Figures 1F,G).

To further characterize CMHs, we measured the area of all identified bleeds. The cumulative size distribution of all detected bleeds is plotted in Figure 2B. IGF-1 deficient hypertensive mice had a greater number of bleeds of all sizes than hypertensive control mice but the overall size of bleeds in IGF-1 KD was significantly smaller than in controls (Figures 2A,B). Each identified bleed was mapped to its brain region using the online Allen Mouse Brain Atlas (last accessed 10/01/2021).³ Each bleed was given a broad identifying region (shown in Figure 2C) as well as a specific brain region. In the hypertensive IGF-1 knockdown animals, the largest fraction of CMHs occurred in the cortex (42%), although CMHs were detected throughout the brain (examples shown in Figures 2D–J, black arrows indicate especially small or hard to see bleeds). In control animals the largest fraction of CMHs occurred in the brainstem and white matter (33 and 25%, respectively), though very few bleeds were detected in control brains overall. Some of the CMHs that occurred can be traced back to an adjacent arteriole or capillary in the brain (Figures 2D–J, red arrowheads indicate vessels). These findings demonstrate that hypertensive IGF-1 knockdown animals have an earlier onset and increased number of CMHs in the brain compared to hypertensive control animals, confirming that IGF-1 deficiency exacerbates microvascular fragility in the mouse brain, mimicking the aging phenotype.

To assess whether mice in which CMHs developed in the brain also exhibited bleeds in the retina, we serially sectioned eyes collected from the same 1-year-old IGF-1 knockdown and control hypertensive and normotensive animals used for the experiments presented in Figures 1, 2. Every 6th retinal section was stained with DAB to detect extravasated red blood cells (RBCs) and counterstained with hematoxylin. Our focus in these studies was identifying bleeds originating from the retinal vasculature. While the choroidal vasculature also supplies nutrients and oxygen to the retina, choroidal vessels are anatomically different from retinal and brain vessels (for example, choriocapillaris vessels are fenestrated and do not participate in the blood–retina barrier). In addition, the pigmented nature of the mouse choroid masks any DAB staining. The DAB stained retinal sections were scored by multiple blinded observers for the presence of any sign of bleed in the retina. However, no evidence of bleeds was detected in any retinal section examined in any region of the retina from any of the groups (example images are shown in Figures 3A,B). To confirm that this was not a staining artifact, brain sections adjacent to those in which cerebral bleeds had previously been detected were included in each retinal staining experiment as positive controls (Figure 3C). To verify that the retina carries receptors for IGF-1, we performed immunofluorescence labeling for IGF-1 receptor (red, Figures 3D,E). IGF-1 receptor is localized throughout the retina, and is particularly enriched in the retinal pigment epithelium (RPE) layer, in the two synaptic layers (inner and outer plexiform layers), and in blood vessels (white arrows). The labeling pattern for IGF-1 receptor is similar in IGF-1 KD and control retinas and in hypertensive and normotensive retinas (Figure 3D). IGF-1 receptor labeling in the brain with pronounced vascular labeling (white arrows) is shown in Figure 3E. These findings suggest that the retina is not as susceptible as the brain to the development of CMHs.

As part of our general evaluation of the effects of circulating IGF-1 KD in the retina, we also undertook morphometric measurements to assess retinal degeneration (representative images shown in Figures 4A,B). One-year-old IGF-1 knockdown mice (regardless of whether they were hypertensive or normotensive) did not exhibit thinning of the ONL (photoreceptors, Figure 4C), INL (retinal interneurons, Figure 4E), or OPL (the layer where photoreceptor terminals form synapses with bipolar and horizontal cells, Figure 4D) compared to age-matched controls and young (3 month old) controls.

We next undertook a series of experiments designed to evaluate whether circulating IGF-1 deficiency had any general effects on the retina in aged mice (∼2 years of age). Similar to the case at 1 year of age, we observed no signs of retinal degeneration in IGF-1 KD or age-matched controls compared to young control mice (young control group replotted from Figure 4 for ease of comparison) (Figures 5A–D). Incipient photoreceptor degeneration is often preceded by mislocalization of photoreceptor outer segment proteins such as rod and cone opsins, however, we observed no signs of this in 1 or 2-year-old IGF-1 KD animals (or age-matched controls, Supplementary Figure 1). Because retinal function does not correlate directly with retinal degeneration, and it has been well established that there is significant age-related loss of retinal function without significant structural degeneration (Li et al., 2001; Gresh et al., 2003; Samuel et al., 2011; Ferdous et al., 2021), we also conducted scotopic and photopic electroretinography on 2-year-old animals to assess rod and cone function. As expected, 2-year old mice (both IGF-1 knockdown and control) exhibited significant reductions in both dark adapted (rods, Figure 5F) and light-adapted (cones, Figures 5G,H) responses compared to young (3-month-old) controls. These findings are consistent with prior studies in C57BL/6 showing significant age-related loss in rod and cone function with little or no photoreceptor degeneration (Li et al., 2001; Gresh et al., 2003; Samuel et al., 2011; Ferdous et al., 2021). IGF-1 KD mice were partially protected from age-related decreases in rod and cone function compared to control mice (Figures 5F–H), exhibiting scotopic a- and b- wave amplitudes and photopic b-wave amplitudes that were slightly higher than those in age-matched controls.

These data were in contrast to findings in the brain, wherein circulating IGF-1 deficiency led to signs of neuronal dysfunction such as gait defects and cognitive decline as a result of impaired neurovascular coupling when assessed at 1 year of age (Toth et al., 2014; Tarantini et al., 2017c). Thus, we asked whether neurovascular coupling in the brain was also impaired in IGF-1 KD animals at 2 years of age. Measurements of CBF changes in the somatosensory cortex in response to whisker stimulation were performed on aged 2-year-old IGF-1 KD animals, age-matched controls, and young (3–6 month) controls. Both groups of 2-year-old mice exhibited significantly reduced neurovascular coupling compared to young mice (Figure 5E). IGF-1 KD mice had mean values slightly less than age-matched controls, however the difference did not achieve statistical significance. These data are consistent with the view that genetic IGF-1 deficiency at younger ages promotes accelerated cerebromicrovascular aging, which mimics the effects of age-related decline in circulating IGF-1 manifested at later ages in wild type mice.

Global IGF-1 knockout retinas exhibit loss of synapses in the OPL. To evaluate whether there are retinal synaptic defects in mice with adult-onset circulating IGF-1 deficiency, we labeled retinas from IGF-1 KD and control animals for VGLUT1 (photoreceptor and bipolar cell terminals) and SV2 (all presynaptic terminals, Figure 6A, magenta). Aged (2-year-old) IGF-1 KD and WT control animals exhibited a thinner layer of photoreceptor terminals in the OPL compared to young control mice, a phenotype that was more pronounced in the peripheral retina than in the central retina (Figure 6A). The OPL contains a mix of rod spherules and cone pedicles, but the terminals we observed in the 2-year-old animals exhibited morphological signs of cone pedicles. To verify this, we co-labeled retinal sections with the cone marker cone arrestin (CARR, yellow, Figure 6B) and SV2 (magenta, Figure 6B). In young control mice, cone terminals (white terminals are co-labeled with SV2/VGLUT1, and white arrows) and rod terminals (magenta label only, arrowheads highlight examples) are present, but the majority of terminals in aged retinas are co-labeled for CARR and SV2 indicating they originate from cones. To help evaluate whether second-order neurons were present at these synapses, we co-labeled retinal sections for SV2 (magenta, Figure 6C) and the horizontal cell marker calbindin (yellow, Figure 6C). Examination of high magnification images showed that horizontal cell process were properly present in 2 year old IGF-1 KD and control mice (Figure 6C bottom, arrows). We performed similar analyses on 1-year-old IGF-1 KD and control animals (Supplementary Figure 2) but found no abnormalities. Combined, these data suggest that circulating IGF-1 deficiency does not exacerbate age-related defects in retinal structure or function.

To further analyze retinal phenotypes associated with circulating IGF-1 deficiency, we performed in vivo fluorescein angiography on both IGF-1 KD and control mice (all normotensive) at 1 and 2 years of age. On fluorescein angiograms, we frequently observed patterns of repeated vascular constriction, referred to as “sausage on a string” phenotype (Jacobsen et al., 2002) in one or more retinal vessels in IGF-1 KD mice (Figures 7A,B, arrows). At 1 year of age, 5/6 IGF-1 KD, and only 2/5 age-matched controls exhibited this phenotype. This phenomenon persisted at 2 years of age: 7/12 IGF-1 KD mice had the severe constriction phenotype while only 1/5 age matched controls had the phenotype (Figure 7B). Young control animals did not exhibit this phenotype (Figure 7C).

We proceeded to evaluate other signs of stress in the retina. We used immunofluorescence to evaluate the expression of retinal glial fibrillary acidic protein (GFAP), an intermediate filament protein normally restricted to the endfeet of Müller glial cells (Figures 8A–D). Increased expression of GFAP leads to staining along the Müller cell body toward the outer layers of the retina, and is a marker of gliosis and retinal stress. At 1 year of age, we observed some minor signs of glial cell activation in IGF-1 KD retinas (Figure 8A, arrowheads). This phenomenon was more pronounced at 2 years of age, wherein significantly increased GFAP labeling was frequently observed in IGF-1 KD retinas compared to age-matched control retinas (Figure 8B). We next scored GFAP-labeled retinas on a scale of 0–4 (0 = no gliosis, 4 = severe gliosis) (Figure 8D). At 1 year of age, gliosis was minor or not detected in all groups. At 2 years of age, there is a high degree of eye-to-eye variability, but even so, IGF-1 KD eyes had a higher median gliosis score than age-matched controls, and 3/9 IGF-1 KD eyes exhibited a high degree of gliosis (score 3 or higher) compared to 0/7 control eyes. To ask whether this gliosis was accompanied by increased oxidative stress, we labeled retinas with a marker of lipid peroxidative stress, 4-hydroxynonenal (4-HNE). However, 4-HNE expression was similar in IGF-1 KD and control mice at 1 and 2 years of age (Figures 8E,G), though staining was modestly increased in aged retinas compared to young controls (Figure 8F). These findings indicate that circulating IGF-1 deficiency leads to abnormalities in the retinal vasculature and retinal gliosis.