Identification of early pericyte loss and vascular amyloidosis in Alzheimer’s disease retina

Donor eyes were obtained from two sources: (1) Alzheimer’s Disease Research Center (ADRC) Neuropathology Core at the Department of Pathology in the University of Southern California (USC, Los Angeles, CA; IRB protocol HS-042071) and (2) National Disease Research Interchange (NDRI, Philadelphia, PA; IRB exempt protocol EX-1055). Both USC-ADRC and NDRI maintain human tissue collection protocols approved by a managerial committee and subject to National Institutes of Health oversight. For a subset of patients and controls we also obtained brain specimens from USC-ADRC. The histological work at Cedars-Sinai Medical Center was performed under IRB protocols Pro00053412 and Pro00019393. Sixty-two postmortem retinas were collected from 29 clinically and neuropathologically confirmed AD patients (age mean ± SD: 81.38 ± 13.79; range 40–98 years; 20 females and 9 males with different disease severities), 11 age- and gender-matched MCI patients (age mean ± SD: 86.45 ± 6.87; range 80–93 years; 5 females and 6 males with different disease severities), and 22 CN individuals (age mean ± SD: 78.18 ± 8.86; range 58–95 years; 13 females and 9 males showing neither clinical cognitive impairment/dementia nor brain pathology). The entire human cohort information is listed in Table 1. The groups had no significant differences in age, sex, or post-mortem interval (PMI) hours. All samples were deidentified and could not be traced back to tissue donors.

The clinical and neuropathological reports provided by the USC ADRC Clinical Core included subjects’ neurological examinations, neuropsychological and cognitive tests, family history, and medication list; psychometric testing was performed by a trained psychometrist under the supervision of a licensed clinical neuropsychologist, following standard-of-care cognitive screening evaluations of patients in their respective neurology clinics, as previously described [21, 47]. NDRI reports provided the medical history of each subject. Most cognitive evaluations were performed annually, and, in most cases, less than 1 year prior to death. Cognitive testing scores from evaluations obtained closest to subjects’ death were used for this analysis. Two global indicators of cognitive status were used for clinical assessment: the Clinical Dementia Rating (CDR; 0 = Normal; 0.5 = Very Mild Dementia; 1 = Mild Dementia; 2 = Moderate Dementia; 3 = Severe Dementia) [60] and the Mini Mental State Examination (MMSE; normal cognition = 24–30, mild dementia = 20–23, moderate dementia = 10–19, severe dementia ≤ 9) [31]. In this study, the clinical diagnostic groups (AD, MCI, and CN) were determined by the source clinicians, based on a comprehensive battery of tests, including neurological examinations, neuropsychological evaluations, and the above-mentioned cognitive tests. For final diagnosis based on the neuropathological reports, the modified Consortium to Establish a Registry for Alzheimer’s Disease [65] was used as outlined in the National Institute on Aging (NIA)/Regan protocols with revision by the NIA and Alzheimer’s Association [39]. Aβ burden (diffuse, immature, or mature plaques), amyloid angiopathy, neuritic plaques, NFTs, neuropil threads, granulovacuolar degeneration, Lewy bodies, Hirano bodies, Pick bodies, balloon cells, neuronal loss, microvascular changes and gliosis pathology were assessed in multiple brain areas: hippocampus (CA1 and CA4), entorhinal cortex, frontal cortex, temporal lobe, parietal lobe, occipital lobe (primary visual cortex, area 17; visual association cortex, area 18), basal ganglia, brainstem (pons, midbrain), cerebellum and substantia nigra.

Amyloid plaques and tangles in the brain were evaluated using anti β-amyloid mAb clone 4G8, Thioflavin-S (ThioS), and Gallyas silver stain in formalin-fixed, paraffin-embedded tissues. Two neuropathologists provided scores based on independent observations of β-amyloid, NFT burden, and/or neuropil threads (0 = none; 1 = sparse 0–5; 3 = moderate 6–20; 5 = abundant/frequent 21–30 or above; N/A = not applicable), and an average of two readings was assigned to each individual. Final diagnosis included AD neuropathologic change (ADNC). Aβ plaque score was modified from Tal et al. (A0 = no Aβ or amyloid plaques; A1 = Thai phase 1 or 2; A2 = Thai phase 3; A3 = Thai phase 4 or 5) [74]. NFT stage was modified from Braak for silver-based histochemistry or p-tau IHC (B0 = No NFTs; B1 = Braak stage I or II; B2 = Braak stage III or IV; B3 = Braak stage V or VI) [15]. Neuritic plaque score was modified from CERAD (C0 = no neuritic plaques; C1 = CERAD score sparse; C2 = CERAD score moderate; C3 = CERAD score frequent) [58]. Neuronal loss, gliosis, granulovacuolar degeneration, Hirano bodies, Lewy bodies, Pick bodies, and balloon cells were evaluated (0 = absent; 1 = present) in multiple brain areas using hematoxylin and eosin (H&E) staining. Amyloid angiopathy was graded as follows: Grade I = amyloid restricted to a rim around normal/atrophic SMCs of vessels; Grade II = media replaced by amyloid and thicker than normal, but no evidence of blood leakage; Grade III = extensive amyloid deposition with focal vessel wall fragmentation and at least one focus of perivascular leakage; Grade IV = extensive amyloid deposition and fibrinoid necrosis, micro aneurysms, mural thrombi, lumen inflammation, and perivascular neuritis. For the correlation analyses against retinal parameters, we used the following CAA scoring system: no amyloid angiopathy was assigned ‘0’; grade I was assigned as ‘1’, grade I–II as ‘1.5’, grade II as ‘2’, and grade II–III as ‘2.5’.

Donor eyes were collected within 7 h, on average, from time of death and were either preserved in Optisol-GS media (Bausch & Lomb, 50,006-OPT) and stored at 4 °C for less than 24 h, fresh frozen (snap; stored at − 80 °C), or punctured once and fixed in 10% neutral buffered formalin (NBF) or 2.5% Paraformaldehyde (PFA) and stored at 4 °C. Brain tissues (hippocampus; occipital lobe – primary visual cortex, area-17, and frontal cortex, area-9) from the same donors were snap frozen and stored at − 80 °C. Parts from the fresh-frozen brain tissues were fixed in 4% PFA for 16 h following dehydration in 30% sucrose/PBS. Brain tissues were cryosectioned (30 μm thick) and placed in phosphate-buffered saline 1x (PBS) with 0.01% sodium azide (Sigma-Aldrich) at 4 °C. Irrespective of the human donor eye source, USC-ADRC or NDRI, the same tissue collection and processing methods were applied.

Fresh-frozen eyes and eyes preserved in Optisol-GS were dissected with anterior chambers removed to create eyecups. Vitreous humor was thoroughly removed manually. Retinas were dissected out, detached from the choroid, and flatmounts were prepared [47]. By identifying the macula, optic disc, and blood vessels, the geometrical regions of the four retinal quadrants were defined with regard to the left and the right eye. Flatmount strips (2–3 mm in width) were dissected along the retinal quadrant margins to create four strips: superior-temporal—ST, inferior-temporal—TI; inferior-nasal—IN, and superior-nasal—NS, and were fixed in 2.5% PFA for cross-sectioning. In a subset of human eye donors, a second set of strips was prepared (5 mm in width) and stored at − 80 °C for protein analysis. Each strip was approximately 2–2.5 cm long from the optic disc to the ora serrata and included the central, mid, and far retinal areas. All the above stages were performed in cold PBS with 1 × Protease Inhibitor cocktail set I (Calbiochem 539,131). Eyes that were initially fixed in 10% NBF or 2.5% PFA were dissected to create eyecups, and the retinas were dissected free. Vitreous humor was thoroughly removed and flatmounts were prepared. As described above, a set of flatmount strips (ST, TI, IN, and NS) was dissected (2–3 mm in width), washed in PBS, and processed for retinal cross-sectioning.

Flatmount strips were initially embedded in paraffin using standard techniques, then rotated 90° horizontally and embedded in paraffin. The retinal strips were sectioned (7–10 µm thick) and placed on microscope slides that were treated with 3-Aminopropyltriethoxysilane (APES, Sigma A3648). Before immunohistochemistry, the sections were deparaffinized with 100% xylene twice (for 10 min each), rehydrated with decreasing concentrations of ethanol (100–70%), and then washed with distilled water followed by PBS.

We modified the retinal vascular isolation method to use on human retinal tissues and immuno-fluorescently label pericytes and amyloidosis (illustrated in Fig. 1a). This trypsin-induced retinal digestion and vascular network isolation technique was originally developed in 1993 [51] and subsequently modified by replacing trypsin with commercially available elastase [77]. Our modified protocol is as follows: retinal strips from human donors or mouse whole retinas preserved in PFA were first washed in lukewarm running distilled water overnight, then digested in 40 U/ml elastase solution (Merck Millipore, Burlington, MA) for 2 h at 37 °C. After digestion, tissues were incubated in activation solution (Tris buffer at pH 8.5) overnight for extensive digestion. The next day, retinas were transferred to superfrost microscope slides with 1 × PBS, then carefully cleaned with rat whisker to remove unwanted tissues under a dissecting microscope. After cleaning non-vascular tissues, 1 × PBS was applied three times to wash the isolated vascular tissues. When we were able to observe clean vascular tree on slides under dissecting microscope, tissues were mounted on slides carefully without dehydration, then incubated in blocking buffer (Dako #X0909) for 1 h at room temperature (RT). Primary antibodies were applied to the tissue after blocking, then incubated at 4 °C overnight as listed (antibody information provided in Table 2): 4G8/lectin/PDGFRβ, 6E10/lectin/PDGFRβ, 11A50-B10/lectin/PDGFRβ, 12F4/lectin/PDGFRβ. Tissues were then washed three times by PBS and incubated with secondary antibodies against each species (information provided in Table 2) for 2 h at RT. After rinsing with PBS three times, vascular trees were mounted by Prolog Gold antifade reagent with DAPI (Invitrogen #P36935). For quantification purposes, images were taken on a Carl Zeiss Axio Imager Z1 fluorescence microscope (Carl Zeiss MicroImaging, Inc.) equipped with ApoTome, AxioCam MRm, and AxioCam HRc cameras (for more details see “Stereological quantification” below). For representative images, Z-stack images were repeatedly captured at same tissue thickness using a Carl Zeiss 780 Confocal microscope (Carl Zeiss MicroImaging, Inc.). Routine controls were processed using identical protocols while omitting the primary antibody to assess nonspecific labeling. Representative images of all negative controls are shown in Supplementary Fig. 1, online resource.

The double-transgenic B6.Cg-Tg (APP_SWE/PS1_∆E9)85Dbo/Mmjax hemizygous (ADTg) mice strain (MMRRC stock #34832-JAX|APP/PS1) and their non-Tg littermates (as WT control non-AD) were used for retinal vascular isolation experiments. All mice are on the genetic background of B6. Mice were purchased from MMRRC and later bred and maintained at Cedars-Sinai Medical Center. The mouse experiments were conducted in accordance with Cedars-Sinai Medical Center Institutional Animal Care and Use Committee (IACUC) guidelines under an approved protocol. We used a total of nine 8.5-month-old mice (all males) divided into three groups: perfused WT (n = 3), perfused ADTg (n = 3), and non-perfused ADTg (n = 3) mice. Animals were deeply anesthetized under Ketamine/Xylazine (40–50 mg/kg) before being euthanized either by transcardial perfusion (0.9% ice-cold sodium chloride supplemented with 0.5 mM EDTA) or cervical dislocation (non-perfused group). Eyes were dissected and the retinas were immediately isolated. Using a 25-gauge needle, a hole is poked in the cornea and an incision is made along the ora serrata to remove the lens and cornea-iris. Next, a small incision is made in the sclera-choroid layers toward the optic nerve and using fine forceps, sclera and choroid is gently separated from the retina, which is cleanly snipped at its base from the optic nerve. Care is taken to isolate whole retina undamaged to preserve vasculature network. Following isolation, retinas were fixed in 4% PFA for 7 days. Retinas were then processed for retinal vascular isolation and immunofluorescent staining as described above.

Frozen human retinal flatmount strips from the temporal hemisphere (ST, TI) were weighed and placed in a tube with cold homogenization buffer [Tris/EDTA buffer pH 9 (DAKO, S2368), 1% Triton X-100 (Sigma, T8787), 0.1% NaN₃ (Sigma, 438456) and 1 × Protease Inhibitor cocktail set I (Calbiochem 539131)], then homogenized by sonication (Qsonica Sonicator M-Tip, Amplitude 4, 6 W, for 90 s; sonication pulse was stopped every 15 s to allow the cell suspension to cool down for 10 s). The ultrasonic probe positioned inside the tube was placed in ice water. Next, retinal strip homogenates were incubated for 1 h at 98 °C in a water bath. After determination of the protein concentration (Thermo Fisher Scientific), retinal Aβ_1–40 was determined using an anti-human Aβ_1–40 end-specific sandwich ELISA kit (Thermo Fisher, KHB3481).

After deparaffinization, retinal cross-sections were treated with antigen retrieval solution at 98 °C for 1 h (PH 6.1; Dako #S1699) and washed in PBS. Retinal sections were then incubated in blocking buffer (Dako #X0909), followed by primary antibody incubation (information provided in Table 2) overnight in 4 °C with the following combinations: PDGFRβ (1:200)/lectin (1:200)/11A50-B10 (1:200), PDGFRβ (1:200)/lectin (1:200)/12F4 (1:200), CD31 (1:50)/JRF/cAβ 40/28 #8152 (1:2000), LRP-1 (1:200)/PDGFRβ (1:200)/lectin (1:200), cleaved caspase-3 (1:200)/PDGFRβ (1:200)/lectin (1:200). Alexa Fluor 488-conjugated tomato lectin was used to visualize blood vessel cells. Retinal sections were then washed three times by PBS and incubated with secondary antibodies against each species (1:200, information provided in Table 2) for 2 h at RT. After rinsing with PBS for three times, sections were mounted with Prolong Gold antifade reagent with DAPI (Thermo Fisher #P36935). Images were repeatedly captured at the same focal planes with the same exposure time using a Carl Zeiss Axio Imager Z1 fluorescence microscope (Carl Zeiss MicroImaging, Inc.) equipped with ApoTome, AxioCam MRm, and AxioCam HRc cameras. Images were captured at 20 ×, 40 ×, and 63 × objectives for different purposes (for more details see “Stereological quantification” below). Routine controls were processed using identical protocols while omitting the primary antibody to assess nonspecific labeling. Representative images of negative controls are shown in Supplementary Fig. 1, online resource.

Fixed brain sections and retinal cross-sections after deparaffinization were treated with target retrieval solution (pH 6.1; S1699, DAKO) at 98 °C for 1 h and washed with PBS. In addition, treatment with 70% formic acid (ACROS) for 10 min at RT was performed on brain sections and retinal cross-sections before staining for Aβ. Peroxidase-based immunostaining was performed. For antibodies’ list and dilutions, see Table 2. Prior to peroxidase-based immunostaining, the tissues were treated with 3% H₂O₂ for 10 min, and two staining protocols were used: (1) Vectastain Elite ABC HRP kit (Vector, PK-6102, Peroxidase Mouse IgG) according to manufacturer’s instructions or (2) All Dako reagents protocol. Following the treatment with formic acid, the tissues were washed with wash buffer (Dako S3006) for 1 h, then treated with H₂O₂ and rinsed with wash buffer. Primary antibody (Ab) was diluted with background reducing components (Dako S3022) and incubated with the tissues for 1 h at 37 °C for JRF/cAβ 40/28 # 8152, or overnight at 4 °C for 11A50–B10 (Aβ₄₀) mAbs. Tissues were rinsed twice with wash buffer on a shaker and incubated for 30 min at 37 °C with secondary Ab (goat anti mouse ab HRP conjugated, DAKO Envision K4000), then were rinsed again with wash buffer. For both protocols, diaminobenzidine (DAB) substrate was used (DAKO K3468). Counterstaining with hematoxylin was performed followed by mounting with Faramount aqueous mounting medium (Dako, S3025). Routine controls were processed using identical protocols while omitting the primary antibodies to assess nonspecific labeling. Representative images of negative controls are shown in Supplementary Fig. 1, online resource.

Analyses of a retinal whole mount from an AD donor retina that was pre-stained with anti-Aβ₄₂ mAb (12F4) and a high-sensitivity immunoperoxidase-based system with 3,3′ Diaminobenzidine (DAB) substrate chromogen were performed using transmission electron microscopy. Stained tissues were processed for electron microscopic imaging; the samples were dehydrated in serially graded ethanol and then infiltrated in Eponate 12 (Ted Pella, Inc. Redding, CA, USA) prior to embedding between two acetate sheets. Ultrathin sections of retina were cut into cross sections at a thickness of 70 nm, examined on a JEOL JEM 2100 (JEOL USA, Peabody, MA, USA), and photographed with the Orius SC1000B digital camera (Gatan, Pleasanton, CA, USA). Images were processed and colorized using Adobe Photoshop CS4 (Adobe Inc., San Jose, CA, USA).

Formalin-fixed paraffin-embedded retinal cross-sections after deparaffinization were washed with PBS and then incubated with Proteinase-K (Recombinant PCR grade, 15 µg/ml in 10 mM Tris/HCL pH 7.6; Roche Diagnostics GmbH; 03115836001) at 37 °C for 20 min. Next, slides were washed with PBS and incubated with TUNEL reaction mixture (50 µl on each slide; Roche Diagnostics GmbH; 11684795910) at 37 °C for 60 min, in a humidified chamber in dark (the samples were covered with parafilm to ensure a homogeneous spread of TUNEL reaction and to avoid evaporation loss). Afterward, slides were washed with PBS and fluorescent-based immunostaining was performed using blocking solution (DAKO X0909) for 45 min at RT. The tissues were incubated with primary antibody, goat anti PDGFRβ, overnight at 4 °C, then the secondary antibody, donkey anti goat Alexa 647, was applied for 1 h at RT. Then, the samples were washed with PBS and covered with ProLong™ Gold antifade mounting media with DAPI (Molecular Probes; #P36935). Negative and positive controls were included (see Supplementary Fig. 1, online resource) in this experimental setup: for TUNEL negative control the retinal tissues were incubated with only 50 µl of TUNEL label solution (without the TUNEL Enzyme solution-terminal transferase) instead of TUNEL reaction mixture. For TUNEL positive control the retinal tissues were incubated with DNase I (1000 U/ml in 50 mM Tris–HCL, pH 7.5; Worthington Biochemical Corp. Code D) to induce DNA strand breaks, prior to labeling procedure. The retinal tissue sections were then evaluated under fluorescent microscope.

For Fig. 1 of isolated retinal blood vessels, quantification was performed from 5 AD donors and 5 age- and sex-matched CN controls. The fluorescence of specific signals was captured using the same setting and exposure time for each image and human donor, with a Z-stack of 10 µm thickness using Axio Imager Z1 microscope (with motorized Z-drive) with AxioCam MRm monochrome camera ver. 3.0 (at a resolution of 1388 × 1040 pixels, 6.45 µm × 6.45 µm pixel size, dynamic range of > 1:2200 that delivers low-noise images due to Peltier-cooled sensor). Images were captured at 40 × objective, at respective resolution of 0.25 µm. Fifteen images were taken randomly from each region of central, mid-, and far-peripheral retina (five from each region) per subject. Acquired images were converted to grayscale and standardized to baseline using a histogram-based threshold in the NIH ImageJ software (version 1.52o). For each biomarker, total area of immunoreactivity was determined using the same threshold percentage from the baseline in ImageJ (with same percentage threshold setting for all diagnostic groups). The images were then subjected to particle analysis for lectin and Aβ to determine IR area. Pericyte number was based on 15 images, averaging the number in each microscopic visual field (covering 1.8 × 10⁴ µm² area), per human donor. We used the grid mode in ImageJ to manually count the number of pericytes. The ratio of Aβ to lectin was calculated by dividing Aβ IR area by lectin IR area in each of the 15 images (described above) and averaging the values per human donor. The sum of Aβ IR area from an identical number of randomly selected pericytes (n = 10) from each human donor was used to calculate Aβ in pericytes. An identical region of interest was used for the standardized histogram-based threshold technique and subjected to particle analysis.

For Figs. 2, 3, 4, 5, and 6 with analysis of retinal cross-sections and quantifications of PDGFRβ, vascular Aβ₄₂, vascular Aβ₄₀, Aβ₄₀, LRP-1, cleaved caspase-3 and TUNEL, images were also acquired at the same setting and exposure time for each experiment, using the Axio Imager Z1 microscope, as described above. Images were captured at either 20 × or 40 × objectives, at respective resolutions of 0.5 and 0.25 µm. Three images were taken from central and far-peripheral retina and four images were taken from mid-peripheral retina (as shown in Fig. 2a, b). For each biomarker, the total area of immunoreactivity was determined using the same threshold percentage from the baseline in ImageJ (with same percentage threshold setting for all images), then subjected to particle analysis for each biomarker to determine their area or area percentage. For vascular PDGFRβ, vascular Aβ₄₂ and Aβ₄₀, and vascular LRP-1, area of blood vessels was chosen to acquire positive immunoreactive (IR) area percentage. For total retinal Aβ₄₀ and total LRP-1 area, we chose the whole retina and documented total IR area of each biomarkers. Quantification of cleaved caspase-3⁺ and TUNEL⁺ pericytes was performed by randomly choosing 10–15 pericytes from each human donor, followed by manually counting using the grid in ImageJ. Then a percentage of cleaved caspase-3⁺ or TUNEL⁺ pericytes was calculated.

For vascular markers of Aβ₄₂, Aβ₄₀, and PDGFRβ, analysis was performed separately for longitudinal blood vessels and vertical blood vessels. Retinal cross-sections in this study were cut sagittally from flatmount strips, hence blood vessels were categorized by the shape of lectin stain: either as vertical blood vessels (≥ 10 µm in diameter) or longitudinal blood vessels (~ 10 µm in diameter). Note: for vertical blood vessels, the vascular wall area (determined by lectin) was selected for analysis, while excluding the blood vessel lumen. For longitudinal blood vessels, the total blood vessel including lumen and wall were selected for quantitative analysis. Dotted eclipse or rectangle frames were added to the representative images to highlight the area of quantification for both vertical blood vessels and longitudinal blood vessels.

GraphPad Prism 8.1.2 (GraphPad Software) was used for analyses. A comparison of three or more groups was performed using one-way ANOVA followed by Sidak’s multiple comparison post-hoc test of paired groups. Groups with two independent variables/factors were analyzed by two-way ANOVA followed by Sidak’s multiple comparison test to further understand interaction between the two independent variables. Two-group comparisons were analyzed using a two-tailed unpaired Student t test. The statistical association between two or more variables was determined by Pearson’s correlation coefficient (r) test (Gaussian-distributed variables; GraphPad Prism). Pearson’s r indicates direction and strength of the linear relationship between two variables. Required sample sizes for two group (differential mean) comparisons were calculated using the nQUERY t test model, assuming a two-sided α level of 0.05, 80% power, and unequal variances, with the means and common standard deviations for the different parameters. Results are expressed as mean ± standard error of the mean (SEM). P value less than 0.05 is considered significant.

To exclusively investigate the extent of retinal microvascular amyloidosis and possible pericyte degeneration in AD without interference from other retinal tissues, we enzymatically digested retinas, preserved solely the vascular network [51, 77], and subsequently conducted fluorescent immunostaining for blood vessels (lectin), PDGFRβ, and different types of Aβ (Fig. 1; extended data in Supplementary Fig. 2, online resource). Our modified method for human retinal vascular isolation and immunofluorescence is illustrated in Fig. 1a. This approach was performed on postmortem retinas isolated from a cohort of age- and sex-matched human subjects with AD diagnosis (avg. age 79.20 ± 10.9 years, 3 females/2 males, CAA score 1.7 ± 0.27) and CN controls (avg. age 75.60 ± 5.63 years, 2 females/3 males, no known CAA). Intense deposits of Aβ₄₂ were visible in AD retinal microvasculature including colocalization with lectin, as compared to CN retina (Fig. 1b).

Vascular Aβ₄₂ accumulation was also detected inside retinal pericytes in AD but not in CN (Fig. 1c). This is further supported by immunostaining with other antibodies against Aβ, including 11A50–B10 (Aβ₄₀), 6E10, and 4G8 (Fig. 1d, e, Supplementary Fig. 2a–d, online resource). In addition, a substantial decrease in PDGFRβ expression was observed in retinal microvasculature from AD as compared to CN controls (Fig. 1f; see lectin/PDGFRβ co-labeling in yellow). A quantitative analysis of retinal microvascular pericyte count per microscopic visual field (1.8 × 10⁴ µm²) revealed a significant 37% pericyte loss in AD compared to controls (Fig. 1g). By quantification of Aβ-immunoreactive area normalized for the lectin-positive vascular area, a significant 5.6-fold increase of Aβ deposition in retinal microvasculature was measured in AD vs. CN (Fig. 1h). Moreover, a substantial 8.7-fold increase of Aβ-immunoreactive area within retinal pericytes was detected (Fig. 1i). Further, Aβ deposits were identified inside degenerated, acellular retinal capillaries that appear to lose lectin expression (Supplementary Fig. 2c, online resource).

Next, we validated the presence of Aβ deposition in isolated retinal blood vessel walls in double-transgenic murine models of AD (ADTg). Performing the blood perfusion procedure prior to retinal vascular extraction allowed us to exclude the contribution of circulating Aβ in the blood. A comparison between perfused and non-perfused ADTg mice and their non-transgenic littermates (WT) revealed that regardless of blood perfusion, there were substantial amounts of retinal vascular Aβ deposits in ADTg mice (Supplementary Fig. 3, online resource).

Retinal vascular pathology was further investigated in cross-sections prepared and analyzed from a larger cohort of 46 human eye donors with pre-mortem diagnosis of AD (n = 21), MCI (n = 11), or CN (n = 14). There were no significant differences in mean age, sex, or PMI between the three diagnostic groups (for more details see Supplementary Tables 1–2, online resource). Histological samples from this cohort were prepared through dissection of retinal strips (2 mm) from four quadrants (superior-temporal—ST, inferior-temporal—TI, inferior-nasal—IN, and superior-nasal—NS) spanning from the optic disc to the ora serrata, processed into paraffin-embedded cross-sections, and immunostained (Fig. 2a, b).

Initially, we assessed retinal vascular PDGFRβ expression by fluorescent immunostaining in lectin⁺ blood vessels. We classified and analyzed two types of blood vessels by shape and size: longitudinal (~ 10 µm in diameter) and vertical (≥ 10 µm in diameter). The examination of small-size longitudinal vessels allowed for analysis of PDGFRβ⁺ pericytes that exist in capillaries and pericytic venules, while excluding vSMCs in larger-size vessels. The separate analysis of vertical vessels covered both PDGFRβ-expressing pericytes and vSMCs. We observed a notable decrease of PDGFRβ signal in both retinal longitudinal and vertical blood vessels in MCI, which was further exacerbated in AD (Fig. 2c, d, Supplementary Fig. 4a, b, online resource). Stereological analysis of percent retinal PDGFRβ area in retinal cross-sections is shown in a subset of age- and sex-matched AD, MCI, and CN subjects (n = 38, avg. age ± SD: AD = 81.2 ± 15.3, MCI = 86.3 ± 6.2 and CN = 78.1 ± 10.4). Data indicate a significant (38%) early loss of retinal PDGFRβ in vertical vessels of MCI as compared to CN controls, whereas a more profound reduction (72%) of vertical vascular PDGFRβ was detected in AD retina (Fig. 2e). To evaluate the relationship between retinal PDGFRβ and CAA scores, we applied Pearson’s correlation coefficient (r) analysis between the two parameters in a cohort of cognitively impaired individuals. We found a significant inverse relationship between retinal PDGFRβ levels and brain CAA score in MCI and AD (Fig. 2f), suggesting that retinal vascular changes in the form of PDGFRβ loss may predict amyloid angiopathy severity in the brains of these patients.

To measure changes in retinal PDGFRβ distribution across the four retinal quadrants, we analyzed PDGFRβ area coverage in human subjects that were stratified by their clinical diagnosis and for each quadrant separately (vertical vessels in Fig. 2g–j, longitudinal vessels in Supplementary Fig. 5a–e, online resource; for comparisons between the four quadrants see Supplementary Fig. 5g, online resource). Our analysis indicated that the temporal hemiretina (ST and TI) had early substantial decreases in vertical vascular PDGFRβ in MCI (Fig. 2i, j), whereas the percentage of PDGFRβ area loss was only significant at later disease stages in the nasal hemiretinal quadrants (NS and IN), as seen in AD (Fig. 2g, h). Consistent with the vascular impairment seen in vertical vessels, early and progressive loss of PDGFRβ⁺ pericytes residing along longitudinal capillaries and post-capillary venules was detected in MCI and AD (Supplementary Fig. 5a–e, online resource). A significant inverse association with neuropathological CAA scores was also identified among individuals with MCI and AD (Supplementary Fig. 5f, online resource).

Moreover, in subjects with neuropathological reports (n = 20), retinal PDGFRβ loss inversely correlated with brain Aβ plaques (NP, DP, IP) and NTs, as summarized in Fig. 2k, l heat-map (extended data on Pearson’s r correlations for pre-defined brain regions in Supplementary Table 3, online resource). In particular, significant correlations between retinal vascular PDGFRβ and brain neuritic plaques were detected for overall brain severity score, as well as separately for the hippocampus, entorhinal cortex, and visual association cortex (Fig. 2k, l). In a subset of subjects where MMSE scores were available, we identified a significant correlation between retinal PDGFRβ loss and cognitive impairment (Fig. 2m). While PDGFRβ in most of retinal quadrants significantly correlated with MMSE scores, the most significant correlation was found with the superior retina (Fig. 2m; extended data on Pearson’s r correlations between MMSE scores against retinal PDGFRβ, per each retinal subregion, are provided in Supplementary Table 5, online resource).

Given that our group and others have demonstrated the existence of retinal Aβ deposits in AD patients [3, 25, 46, 47, 50], our next question was whether vascular PDGFRβ loss is associated with increased vascular Aβ deposition in postmortem retinas from MCI and AD patients. To this end, we studied retinal vascular Aβ₄₂ pathology in a cohort of age- and sex-matched human eye donors (n = 31, avg. ± SD age: AD = 82.8 ± 18.4, MCI = 87.8 ± 5.5, CN = 78.8 ± 10.3; Fig. 3). We analyzed percent 12F4⁺Aβ₄₂-IR area separately for vertical and longitudinal retinal vessels (Fig. 3a–d); the two types of blood vessels were classified as detailed above. The AD retina displays substantially more vascular Aβ₄₂ as compared to both MCI and CN retinas (Fig. 3a–b; additional image panels in Supplementary Fig. 6a, b, online resource). In our quantitative IHC analyses, to avoid signal from circulating blood Aβ, in the vertical vessel analyses we quantify the immunoreactive area by selecting the vascular wall region and excluding the lumen area (see example of dotted eclipse frames in Fig. 3a). Analysis of vertical vascular Aβ₄₂ confirmed a significant increase in the retina of MCI and AD compared to CN controls (Fig. 3c). Analysis of retinal longitudinal vessels also indicated a significant increase in Aβ₄₂ burden in AD compared to MCI and CN controls (Fig. 3d), representing accumulation of both circulating and vascular Aβ₄₂.

Next, investigation of the potential association between retinal vascular Aβ₄₂ burden and respective CAA scores suggested a significant correlation, albeit in a limited cohort (Supplementary Fig. 6c, online resource). Additionally, retinal vascular Aβ₄₂ load had a significant correlation with cerebral Aβ plaque burden (Fig. 3e), and moreover, a strong, inverse correlation with retinal PDGFRβ (Fig. 3f). Figure 3g demonstrates the gradual PDGFRβ loss concomitant with increased Aβ₄₂ burden in retinas isolated from MCI and AD patients relative to CN controls (for extended representative images see Supplementary Fig. 7a–f, online resource). Higher magnification fluorescent images show Aβ₄₂ deposits inside residual retinal vascular PDGFRβ⁺ cells, with increased co-localization in MCI vs. AD (Fig. 3h; extended representative images in Supplementary Fig. 6d, e, online resource). TEM analysis in retinal vertical sections from AD patients reveals Aβ₄₂ deposits in multiple locations near blood vessels and within pericytes (Fig. 3i, j). Retinal Aβ₄₂ deposits were found perivascular in close proximity to pericytes (Fig. 3i, left), in the lumen adjacent to an endothelial cell (Fig. 3i, right), and inside pericytes (Fig. 3j).

Since Aβ₄₀ is the major alloform type deposited in cerebral blood vessels [34], we further studied its distribution in retinal blood vessels in a cohort of eye donors from age- and gender-matched individuals with diagnosis of AD, MCI, or CN (n = 36, avg. age ± SD: AD = 81.8 ± 14.8, MCI = 86.3 ± 6.2 and CN = 78.1 ± 10.4; see Fig. 4 and extended data in Supplementary Figs. 8, 9, online resource). Initial analysis of vascular Aβ₄₀ burden in paired retinas and brains utilizing peroxidase-based DAB staining with a specific antibody recognizing the C-terminal amino acid sequence of Aβ₄₀ (JRF/cAβ40/28; courtesy of Janssen Pharmaceutica) revealed an increase in retinal vascular Aβ₄₀ deposition in MCI and AD compared to CN controls (Fig. 4a–c). This was in agreement with our findings following application of a commercially available antibody (11A50–B10) recognizing the C-terminal sequence of Aβ₄₀ (Supplementary Fig. 8a, online resource). Intriguingly, strong signal of Aβ₄₀ was observed in the tunica media (Fig. 4c and Supplementary Fig. 8a, online resource), although deposits were also detected in tunica adventitia and intima (Fig. 4b, c and Supplementary Fig. 8a, online resource, see arrows).

Immunofluorescent staining using 11A50-B10 and JRF/cAβ40/28 antibodies demonstrated an extensive retinal vascular Aβ₄₀ burden in both vertical and longitudinal vessels in AD (Fig. 4d–f; extended representative images in Supplementary Figs. 8b and 9a, b, online resource). Quantitative analysis of vascular 11A50–B10⁺Aβ₄₀ immunoreactivity indicated substantial 7- to 12-fold increases in retinal vertical (vessel wall with lumen area excluded) and longitudinal vessels (representing both Aβ in circulating blood and vessel walls) in AD compared to CN (Fig. 4g, h). A non-significant trend was noted in MCI vs. CN controls and between MCI and AD groups. In an additional subset of human donors, analysis of retinal vascular Aβ₄₀ using JRF/cAβ40/28 antibody verified significant increases in retinas from MCI and AD compared to CN controls (n = 14; Supplementary Fig. 6c, online resource).

Similar to retinal vascular Aβ₄₂, retinal vascular Aβ₄₀ burden significantly and inversely correlated with retinal PDGFRβ (Fig. 4i) and also tightly and directly correlated with retinal vascular Aβ₄₂ burden (Fig. 4j). Moreover, retinal vascular Aβ₄₀ was associated with entorhinal cortex parenchymal CAA in a subset of human donors where CAA scores are available (Supplementary Fig. 8d, online resource). Notably, increased retinal vascular Aβ₄₀ burden positively correlated with elevated brain neuritic plaques (NPs), especially in the entorhinal cortex (Fig. 4k, l; for more details see Supplementary Table 4, online resource). The strongest correlation was observed with NTs in the visual association cortex (A-18). In addition, immature plaques in the primary visual cortex significantly associated with retinal vascular Aβ₄₀ (Fig. 4k, l; Supplementary Table 4, online resource). Similar to vascular Aβ₄₂, vascular Aβ₄₀ was also detected in PDGFRβ⁺ cells (Supplementary Fig. 8e, online resource). Stratification of human subjects based on clinical AD/MCI vs. CN diagnosis revealed a significantly lower PDGFRβ and significantly higher Aβ₄₀ and Aβ₄₂ levels in AD/MCI retinal vessels (Fig. 4m–o). These findings highlight the potential to distinguish between the diagnostic groups using retinal vascular parameters, and especially vascular PDGFRβ and Aβ₄₀ (Fig. 4n).

To evaluate the overall retinal Aβ₄₀ burden, including abluminal deposits outside blood vessels, we quantified Aβ₄₀ levels and mapped Aβ₄₀-IR area in all four quadrants (ST, TI, IN, NS), central/mid/far (C/M/F) geometrical subregions, and inner vs. outer cellular layers of the neurosensory retina (Fig. 5). We initially measured retinal Aβ_1–40 peptide levels in protein homogenates isolated from fresh-frozen donor eyes in an additional cohort of age- and gender-matched subjects (n = 11; 6 AD human eye donors: avg. ± SD age = 79.33 ± 17.6 years, 4 females and 2 males, and 5 CN controls: avg. age 75.4 ± 4.93 years, 3 females and 2 males). Quantitative ELISA revealed a significant increase in retinal Aβ_1–40 concentrations in AD compared to CN (Fig. 5a). Immunofluorescence staining in a larger cohort (n = 36) using 11A50–B10 antibody confirmed a substantial elevation of retinal Aβ₄₀ load in AD compared to MCI and CN (Fig. 5b, data were normalized per retinal thickness; for raw data see Supplementary Fig. 10f, online resource). Separate analyses of retinal Aβ₄₀ burden per quadrant indicated consistently higher Aβ₄₀ load in AD compared to MCI and CN, especially in TI quadrant (Supplementary Fig. 10a–d, online resource, for normalized data per retinal thickness; Supplementary Fig. 10 g–j, online resource, for raw data). As expected, a significant positive correlation was noted between total Aβ₄₀ and vascular Aβ₄₀ in the human retina (Supplementary Fig. 10e, online resource).

Our current observation of Aβ₄₀ distribution predominantly in the inner retina and previous studies describing inner retinal pathology in AD, including thinning or RGC degeneration [6, 13, 18], prompt our analysis of Aβ₄₀ burden in inner vs. outer retinal layers. To this end, we separated the inner retina (from inner limiting membrane to inner nuclear layer) and the outer retina (from outer plexiform layer to outer limiting membrane), as illustrated in Fig. 5c. This analysis revealed that the vast majority of retinal Aβ₄₀ deposition (> 90%) is found in the inner as compared to the outer layers across all diagnostic groups (Fig. 5d), with evidence for propagation to the outer retina (~ 5%) in AD (for extended data on Aβ₄₀ mapping in inner vs. outer retina with a separate analysis for each retinal quadrant see Supplementary Fig. 11a–j, online resource). Further evaluation of Aβ₄₀ immunoreactivity in retinal C/M/F subregions indicated a significantly elevated burden in the central- vs. far-peripheral retina of AD, with similar but non-significant trends of increase in MCI (Fig. 5e; see comparative analysis of the four quadrants in Supplementary Fig. 11 k, online resource). A summary of retinal Aβ₄₀ burden analyzed in four quadrants, three geometrical subregions, and inner vs. outer layers is illustrated by a color-coded pie graph (Fig. 5f).

The feasibility to separate between the MCI/AD and CN clinical groups by total retinal Aβ₄₀ burden was further assessed. This analysis showed some overlap between the populations but indicated a significantly greater than fivefold increase in the retina of AD/MCI vs. CN controls (Fig. 5g). Finally, to examine possible associations between total retinal Aβ₄₀ burden and other retinal and brain parameters, Pearson’s (r) correlations were calculated (Fig. 5h–j; Supplementary Fig. 11l, online resource). These analyses demonstrated a significant inverse correlation with retinal vascular PDGFRβ and non-significant correlations with brain NP (Fig. 5h), CAA scores (Fig. 5i), and cognitive status by MMSE (Fig. 5j; for correlations with different retinal quadrants see Supplementary Table 6, online resource). Nonetheless, a significant association was detected between retinal Aβ₄₀ burden and NP in the entorhinal cortex (Fig. 5h).

In murine models of AD, vascular LRP-1 was shown to be expressed by pericytes, mediate the clearance of brain-parenchymal Aβ via blood vessels, and affect cerebral amyloid deposition [69]. To evaluate LRP-1 and vascular LRP-1 expression in the human retina, we analyzed retinal cross-sections from a cohort of 18 subjects with AD, MCI, and CN (n = 6 subjects per each diagnostic group). Representative microscopic images demonstrated reduced vascular LRP-1 expression along with marked vascular PDGFRβ loss in postmortem retinas from AD as compared with CN control (Fig. 6a–d; extended images for separate channels in Supplementary Fig. 12a, b, online resource). A quantitative IHC analysis indicated a non-significant trend of decreased total retinal LRP-1 immunoreactivity in AD compared to CN controls (Fig. 6e), with no difference between levels of LRP-1 in MCI vs. CN controls. Evaluation of vascular LRP-1 expression revealed a significant 32% decrease in AD compared to CN (Fig. 6f). Retinal vascular LRP-1 significantly correlated with retinal vascular PDGFRβ in this cohort (Fig. 6g), yet showed a non-significant trend of association with retinal Aβ₄₀ burden (Fig. 6h).

To investigate whether the findings of retinal PDGFRβ and pericyte loss in MCI and AD are due to apoptotic cell death, we evaluated two markers of apoptotic cells in this cohort. First, we immunolabelled cleaved caspase-3 and investigated apoptosis of pericytes in small blood vessels (Fig. 6i–o). Representative microscopic images show a frequent occurrence of cleaved caspase-3⁺ in pericyte nuclei in postmortem retinas from MCI and AD patients as compared to CN controls (Fig. 6j–l vs. i; extended images for separate channels in Supplementary Fig. 13a–d, online resource). Quantification of cleaved caspase-3⁺ pericyte number confirmed an early retinal pericyte apoptosis in MCI, which was on average higher in AD (Fig. 6m). Cleaved caspase-3 in retinal pericytes inversely and strongly correlated with retinal PDGFRβ and positively with retinal Aβ₄₀ burden (Fig. 6n, o). To further validate apoptosis of retinal pericytes during AD progression, fluorescent TUNEL assay was utilized on the same cohort (representative microscopic images in Fig. 6p–t; extended images for separate channels in Supplementary Fig. 14a–d, online resource). Analysis of TUNEL⁺ pericyte count indicated increased apoptosis of retinal pericytes in MCI and more significantly in AD (Fig. 6u), and a significant inverse correlation with retinal PDGFRβ (Fig. 6v).