Decoding amyloid beta clearance systems at inner blood–retina barrier using three‐dimensional ex vivo retinal imaging in Alzheimer's disease

Abstract

INTRODUCTION

Impaired amyloid beta (Aβ) clearance contributes to sporadic Alzheimer's disease (AD). This study investigated retinal Aβ clearance involving neuronal, glial, and vascular interactions at the inner blood–retina barrier (iBRB), a functional analog of the blood–brain barrier (BBB).

METHODS

Retinal wholemounts from AD donors and controls were analyzed alongside transgenic amyloid precursor protein/presenilin 1 (APP‐PS1) and non‐carrier control mouse retinal cross‐sections using three‐ and two‐dimensional ex vivo imaging.

RESULTS

AD neuroretinas displayed increased larger Aβ42 deposits, microglial elongation, and substantial reductions in macroglial support and water channel expression. The uptake of soluble Aβ oligomers (SAβOs) by peripheral macrophage‐like, Aβ‐binding myeloid lineage cells was also diminished. In APP‐PS1 mice, elevated glia levels, alongside increased APP/Aβ expression, suggest gliosis and failures in clearance processes with disease progression.

DISCUSSION

Ex vivo three‐dimensional retinal imaging at the iBRB provides novel insights into Aβ clearance in AD, which is difficult to replicate in ex vivo brain studies at the BBB.

Highlights

• Impaired clearance mechanisms play a key role in sporadic AD.

• The iBRB serves as a functional analog to the BBB.

• At the iBRB, the glymphatic system and microglial phagocytosis help mitigate Aβ burden.

• Peripheral macrophage‐like myeloid lineage cells may aid SAβO clearance.

• The imaging plane (surface vs cross‐section) may affect AD pathogenesis findings.

1. BACKGROUND

Amyloid beta (Aβ) is produced through the sequential cleavage of amyloid precursor protein (APP) by β‐ and γ‐secretases, primarily in neurons, followed by brain endothelial cells and astrocytes. ¹ , ² , ³ Aβ peptides vary in length, with Aβ1‐40 and Aβ1‐42 being the most common isoforms found in senile plaques. ¹ , ² Aβ1‐40, which is soluble and less toxic, constitutes about 90% of Aβ in healthy brains, while Aβ1‐42, comprising less than 10%, is highly neurotoxic and predominant in Alzheimer's disease (AD) brains. ¹ , ² In AD brains, Aβ exists in various forms, including monomers, oligomers, and fibrils. Numerous studies have demonstrated that soluble Aβ oligomers (SAβOs) are the primary agents responsible for synaptic dysfunction and neurodegeneration in AD. ⁴ , ⁵ , ⁶ , ⁷ Furthermore, recent research suggests that SAβOs accumulate early in the disease and impair memory prior to plaque formation and that reducing their levels may reverse cognitive deficits. These findings identify SAβOs as one of the earliest pathogenic factors in AD. ⁸ , ⁹ , ¹⁰

It has been suggested that even minor impairments in Aβ clearance from the brain can lead to its accumulation, as effective clearance is crucial for maintaining Aβ homeostasis and preventing the toxic buildup of misfolded assemblies resulting from ongoing APP processing and Aβ generation. ¹¹ Similar to other brain metabolites, Aβ turnover partly relies on bulk flow through cerebrospinal fluid, crossing the BBB, perivascular circulation, and the glia–lymphatic (glymphatic) system. ¹¹ , ¹² Approximately 40% to 60% of Aβ produced in the brain diffuses into the bloodstream and is cleared through the peripheral system. ¹³ , ¹⁴ , ¹⁵ Aβ transport across the BBB is regulated by various transporters, whose expression is altered in AD. The low‐density lipoprotein receptor‐related protein 1 (LRP1), present in the endothelial cells and pericytes of the BBB, plays a key role in removing Aβ from the brain into the peripheral circulation. ¹ , ² , ¹⁶ Conversely, Aβ can enter the brain from the periphery via receptors like the receptor for advanced glycation end products (RAGE), which facilitate its transport across the BBB. ¹ , ² , ¹⁶ Additionally, innate immune cells are vital for defending the brain against toxic molecules such as Aβ, though their mechanisms appear inadequate in AD. ¹⁷ Circulating and infiltrating monocytes, which help clear Aβ deposits, are effective in the vascular system and parenchyma, respectively. Recent studies on AD and mild cognitive impairment (MCI) patients highlight that macrophage‐like Aβ‐binding monocytes with high phagocytic potential, both in the periphery and central nervous system (CNS), play a significant role in Aβ clearance. ¹⁸ , ¹⁹ , ²⁰ Microglia, the brain's resident immune cells, help maintain homeostasis and eliminate neurotoxic elements, but their efficiency in clearing Aβ declines with age.

The retina, an extension of the CNS, shares embryological origins and vasculature with the brain, utilizing similar molecules, growth factors, neurotransmitters, and cellular structures. ²¹ , ²² In both the retina and the brain, glial cells support neurons in a mutually beneficial relationship. ²³ Several retinal biomarkers for AD have been identified. For example, retinal thinning of the nerve fiber layer (NFL), which involves the axons of retinal ganglion cells, astrocytes, and Müller cells, has been consistently observed in AD patients using optical coherence tomography. ²⁴ , ²⁵ This NFL thinning correlates with disease severity and cognitive decline and can even be detected before cognitive symptoms manifest. ²⁶ , ²⁷ Moreover, retinal Aβ plaques have been found in both AD patients and animal models, suggesting their potential use for differential diagnosis. ²⁸ , ²⁹ , ³⁰ , ³¹ , ³² Retinal vascular changes, such as reduced vessel density, decreased perfusion, and increased vascular Aβ deposition, have also been noted in AD, potentially reflecting underlying cerebrovascular pathology and impaired Aβ clearance. ³³ , ³⁴

This study aimed to investigate impaired Aβ clearance systems involved in the pathogenesis of sporadic AD by examining the interactions among neuronal cells, glial cells, and retinal vasculature, particularly at the iBRB, using wholemount neuroretinas and three‐dimensional (3D) ex vivo imaging. These interactions cannot be effectively studied in brain tissues ex vivo, particularly at the BBB, because of the complex structure.

2. METHODS

2.1. Human donors

This study was approved by the Clinical Ethics Research Board of the University of British Columbia (H20‐02944) and strictly adhered to the Declaration of Helsinki and its later amendments or comparable ethical standards. Ten donor eyes from individuals diagnosed with AD were obtained from the Department of Pathology at Vancouver General Hospital, British Columbia, Canada. The severity of AD was confirmed neuropathologically at post mortem according to the National Institute on Aging‐Alzheimer's Association (NIA‐AA, 2012) criteria. ³⁵ Ten age‐matched donor eyes, provided for research purposes by the Eye Bank of British Columbia, were used as controls. Additionally, two donor eyes from male individuals under the age of 60 were included to assess age‐related structural or morphological alterations.

2.1.1. Neuroretina dissection

The posterior eyecup was prepared by careful dissection, with the cornea, lens, and iris being removed. The eyecup was then divided into four major quadrants (superior, temporal, inferior, and nasal) (Figure S1A), allowing most of the liquefied vitreous humor to drain. Each quadrant was further divided into three pie‐shaped sections, and the neuroretina was separated from the retinal pigment epithelium (RPE) and choroid. Any vitreous remnants were carefully removed using forceps. Next, 3.5‐mm‐diameter punches were taken from the mid‐peripheral and peripheral regions of the neuroretina, specifically from the temporal and superior quadrants, under a dissection microscope, ensuring no mechanical damage or tissue loss. The neuroretina punches were then washed in phosphate‐buffered saline (1× PBS, pH 7.2–7.4) before undergoing double or triple fluorescence staining.

2.1.2. Localization of selected protein markers in human neuroretina wholemounts

The primary antibodies used in various combinations included 12F4, specific to the 1‐42 amino acid residues of Aβ species; macroglial markers such as glial fibrillary acidic protein (GFAP), a type III intermediate filament protein characteristic of differentiated and mature astrocytes, and glutamine synthetase (GS), an enzyme expressed predominantly in Müller glia; aquaporin 4 (AQP4), a major CNS water channel; microglia/macrophage markers such as ionized calcium‐binding adapter molecule 1 (IBA1) and cluster of differentiation 68 (CD68); and the vascular endothelium marker Ulex europaeus agglutinin (UEA‐l). For the validation – such as distinguishing blood‐derived monocytes/macrophages from resident tissue microglia/macrophages – CD45 was used in combination with IBA1 and UEA‐l markers. To assess lysosomal activity, lysosomal‐associated membrane protein 1 (LAMP1) was used together with IBA1 and UEA‐l. As validations were based on a small sample size (N = 1 or 2 per group) from the same cohort, only qualitative assessments were performed. Table S1 summarizes the primary and secondary antibodies and their dilutions used for double and triple immunofluorescence staining. Negative control tissues were processed alongside experimental tissues, with primary antibodies omitted during treatment. All other reagents, including blocking solutions and secondary antibodies, were applied identically to ensure consistency.

RESEARCH IN CONTEXT

1. Systematic review: The authors conducted a literature search in PubMed and Web of Science. No studies have examined clearance impairments leading to pathological protein buildup at the iBRB – a functional analog of the BBB – in transgenic models or post mortem human retinas.

2. Interpretation: Wholemount neuroretina analyses suggested that the glymphatic system and microglia/macrophage phagocytosis serve as compensatory mechanisms for mitigating Aβ accumulation at the iBRB. Peripheral macrophage‐like myeloid lineage cells appeared to uptake SAβOs within the retinal vasculature. These mechanisms were largely disrupted in AD donors.

3. Future directions: Researchers should consider the plane of view, as demonstrated in wholemount neuroretinas and retinal cross‐sections, in the context of AD pathogenesis. Ex vivo 3D retinal imaging offers novel insights into retinal and BBB‐analog processes in AD. Future studies should include a larger sample size of both sexes and a diverse group of donors to explore the translational potential for diagnostic applications.

Neuroretinas were carefully placed in 24‐well plates containing 1× PBS and rinsed three times for 5 min each with gentle shaking. To detect Aβ, an 88% formic acid pretreatment was applied, followed by a 5‐min incubation at room temperature (RT). The samples were then rinsed three more times in 1× PBS (5 min each, with gentle shaking). Antigen retrieval was performed using one of two methods: either a 10‐min treatment with 0.05% proteinase K in Tris‐EDTA buffer (pH 8.0) at RT or heat‐induced retrieval in preheated citrate buffer (pH 6.0) for 4 min, followed by an additional 10 min at power level 800. Afterward, the samples underwent three additional 5‐min washes in 1× PBS with gentle shaking. Blocking was carried out by incubating the samples in blocking buffer (3% to 5% normal goat serum [NGS] in 0.3% Triton X‐100 in 1× PBS) for 20 to 30 min at RT. Primary antibodies, diluted in either 3% NGS blocking buffer or 1% bovine serum albumin (BSA), were applied for a 2‐h incubation at RT on a shaker, followed by an additional 48‐h incubation at 4°C with gentle agitation.

Following primary antibody incubation, the wholemounts were rinsed four times in 1× PBS (5 min each, with gentle shaking). Secondary antibodies, including Alexa Fluor 546 IgG1, Alexa Fluor 546 IgG2a, Alexa Fluor 488, and DyLight 649, were diluted in 1× PBS in various combinations, added to each well, and incubated at RT for 45 min. The wholemounts were then washed 3 to 4 times in 1× PBS (10–15 min each, with gentle shaking). Nuclei were stained with 4′,6‐diamidino‐2‐phenylindole dihydrochloride (DAPI) (1:500) for 15 min, followed by four additional washes in 1× PBS (10–15 min each, with gentle shaking). Neuroretinas were then mounted onto glass slides with the inner limiting membrane (ILM) facing upward, using ProLong Antifade or Mowiol 4‐88 mounting medium, coverslipped with No. 1.5 thickness coverglass, and sealed for confocal microscopy.

Additionally, 6‐µm retinal cross‐sections from a different cohort of AD and age‐matched control cases (AD#11, AD#12, Control#11, and Control#12; Table S2) were examined using various markers to compare wholemount and cross‐sectional morphological and pathological changes associated with sporadic AD. The panel included markers for Aβ species (12F4, 6E10), macroglia (GFAP, GS), the AQP4 water channel, microglia/macrophages (IBA1), and the vascular marker UEA‐l (Table S1). Standard double and triple immunofluorescence staining, combined with confocal microscopy, was used for qualitative assessment.

2.1.3. Localization of SAβOs in human wholemount neuroretinas

A novel anti‐SAβO (E3) nanobody, generated from an alpaca immunized with solubilized human Aβ1‐42 peptide, was used for this purpose. ⁹ The E3 nanobody can traverse the BBB and has shown a distinct spatial distribution: SAβOs are primarily associated with neurons, while Aβ plaques are found in the extracellular matrix. ⁹

Wholemount neuroretinas from AD and control groups were used for nanobody‐antibody staining, utilizing Fluorescein (FAM)‐labeled E3 and 12F4, as well as near‐infrared (NIR)‐conjugated E3 and 12F4. A modified nanobody staining protocol was adapted to localize the nanoparticles (SAβOs) in the wholemount retinas. ³⁶ The original protocol, developed by Fang et al. (2018), enables a correlative light and electron microscopy approach that preserves tissue ultrastructure while allowing for the location of multiple molecular moieties, even deep within tissues.

Briefly, neuroretina punches were rinsed in 0.1 M PBS three times for 5 min each. Antigen retrieval was performed using 88% formic acid for 5 min at RT, followed by three 5‐min rinses in 0.1 M PBS. A glycine‐blocking buffer consisting of 0.1 M PBS, 0.1 M glycine, and 0.05% sodium azide was added to each sample and incubated for 20 min at RT. The primary antibody 12F4 (1:500) and nanobody E3 (1:100) were diluted in the glycine‐blocking buffer to co‐localize the 12F4⁺Aβ42 in the wholemount retinas. After adding the primary antibodies, the samples were light‐protected and incubated at 4°C for 48 h with gentle shaking. The samples were then rinsed in 0.1 M PBS three times for 5 min each. To visualize the 12F4⁺Aβ42, Alexa Fluor 546 IgG1 secondary antibody (diluted 1:400 in 0.1 M PBS) was added and incubated at RT for 45 min with gentle shaking. The samples were rinsed again in 0.1 M PBS three times for 5 min each. Hoechst nuclear dye (1:500 in 0.1 M PBS) was added and incubated at RT for 15 min, followed by three additional 5‐min PBS rinses. Finally, each retinal punch was mounted with the ILM facing up, using ProLong Antifade or Mowiol 4‐88 mounting medium, and covered with No. 1.5 thickness coverglass for confocal microscopy.

2.2. Animals

All animal experiments followed the guidelines and received approval from the University of British Columbia Animal Care Committee (A24‐0098) and the Biosafety Committee (B20‐0074). APP/PS1 double transgenic (Tg) mice (B6;C3‐Tg, strain 034829‐JAX), which express a chimeric mouse/human amyloid precursor protein (Mo/HuAPP695swe) and a mutant human presenilin 1 (PS1‐dE9), along with their non‐transgenic (NTg) siblings, were studied. These mutations are linked to early‐onset AD and amyloid plaque formation. ³⁷ Brain and eye samples were obtained from two groups of mice at two different age ranges: 3 to 4 months (younger) and 9 to 10 months (older). In total, 16 female mice were used, distributed across four groups (n = 4 mice per group) (Figure S1B).

2.2.1. Sample collection and preparation

Animals were euthanized, and brain and eye samples were collected as previously described. ³⁸ , ³⁹ The left hemisphere of the brain and the left eye were allocated for gene expression analysis, whereas the right hemisphere and the right eye were designated for protein expression studies.

2.2.2. Differentially expressed genes

As previously described, ³⁸ , ³⁹ total RNA was isolated from pooled neocortex‐hippocampus tissues and eye tissues (n = 4 per age group and model), followed by cDNA synthesis. Each reaction utilized approximately 10 ng cDNA. Gene expression analysis was performed for 22 target genes, with a minimum of three technical replicates per sample. RT‐qPCR, widely regarded as the gold standard for gene expression studies, was conducted using the 7500 Fast Real‐Time PCR System (Applied Biosystems). Glyceraldehyde‐3‐phosphate dehydrogenase (Gapdh) served as the reference gene for data normalization, and optimized primer pairs were selected for the experiments (Table S3).

2.2.3. Localization of selected protein markers in mouse brain sagittal and retinal cross‐sections

Mid‐sagittal brain and eye cross‐sections, 6 µm thick, were stained using double immunofluorescence protocols based on our previous studies. ⁴⁰ Both APP‐PS1 mice and their non‐carrier sibling controls (n = 4 per age group) were evaluated. Primary antibodies included 6E10, targeting APP and the 1 to 16 amino acid residues of Aβ species, as well as AQP4 (water channel); GFAP and GS (macroglia markers); and IBA1 (microglia/macrophage marker). To detect APP/Aβ peptides, an 88% formic acid pretreatment was applied, alongside standard antigen retrieval methods using either 0.05% proteinase K in Tris‐EDTA buffer (pH 8.0) or heat‐induced citrate buffer (pH 6.0). To validate 6E10⁺ APP labeling, we used a knockout‐validated APP antibody (rabbit monoclonal, Catalogue No. A17911) on brain and eye cross‐sections. The presence of SAβOs was also screened in brain and eye cross‐sections using an NIR‐conjugated E3 nanobody and a standard immunostaining protocol. As validations were based on a small sample size (N = 1 to 2 per group) from the same cohort, only qualitative assessments were performed. Table S1 summarizes the primary and secondary antibodies and their dilutions used for single or double immunofluorescence staining. Negative controls were processed in parallel, omitting the primary antibodies.

2.3. Fluorescence confocal microscopy and image analysis

All fluorescent images were captured using a Zeiss LSM 800 confocal microscope with ZEN 3.7 (Blue edition) software.

2.3.1. Imaging parameters

Various fluorophores, including Alexa Fluor 546 IgG1, Alexa Fluor 546 IgG2a, Alexa Fluor 488, DyLight 649, FAM, and NIR dyes, were used. Each fluorophore was imaged at its corresponding excitation and emission wavelengths. Confocal settings (e.g., laser wavelength, pinhole, master gain, scaling) were kept constant for each marker and its negative controls across all samples to ensure consistency and minimize variability. Regions with tissue damage, tearing, or other artifacts were excluded during imaging.

2.3.2. Magnifications and regions of interest

Z‐stack images (1‐µm slices) of human wholemount neuroretinas were taken at 200× magnification across five non‐overlapping fields (Figure S1A), focusing on the retinal ganglion cell layer (GCL). The GCL was identified by sequential imaging from the ILM/NFL through to the outer nuclear layer (ONL). The density and size of the DAPI‐labeled nuclei in the GCL, inner nuclear layer (INL), and ONL served as identifiers for these layers, while the inner and outer plexiform layers (IPL and OPL) were identified as DAPI‐ (e.g., cell) free zones. The Z‐stack images confirmed antibody penetration through the full thickness of the wholemounts, with labeling observed as deep as the ONL.

Mouse brain sagittal sections were imaged at 100× and 200× magnifications to analyze double‐labeling of 6E10‐GFAP and 6E10‐IBA1 markers, respectively. Non‐overlapping regions – four from the hippocampus (dentate gyrus [DG], cornu Ammonis 4 [CA4], CA3‐CA2, and CA1) and four from the neocortex (prefrontal, frontal, parietal, and occipital regions) – were captured. Mouse retinal cross‐sections were imaged at 200× magnification for double labeling of GFAP‐GS and 6E10‐AQP4, as well as single labeling of IBA1. For each retina, two central, two mid‐peripheral, and two peripheral regions were captured.

2.3.3. Semi‐quantitative analysis

The numbers of 6E10⁺ Aβ plaques (including diffuse‐type) and IBA1⁺ microglial cells were counted semi‐quantitatively in brain samples from APP‐PS1 and non‐carrier sibling mice. NIR‐E3⁺ macrophage‐like Aβ‐binding myeloid lineage cells were counted semi‐quantitatively in wholemount neuroretinas from AD and control donors. In addition, IBA1⁺ and CD68⁺ co‐localized microglial circular processes were counted in wholemount neuroretinas from AD and control donors. Cell counting was performed by two or more independent investigators who were masked to the group assignments.

2.3.4. Image analysis

It was performed using ImageJ software with appropriate plugins. High‐resolution TIFF files were used for both mouse brain samples (eight images per marker per mouse) and retina samples (six images per marker per mouse) and human wholemount neuroretinas (five orthogonal projections per marker per donor). Immunoreactivity was evaluated by two or more independent investigators, who were masked to the sample identities.

Briefly, individual channel TIFF files (preclinical data and FAM‐E3/12F4 nanobody staining) were opened in ImageJ. Although the mouse sagittal brain and human wholemount retina areas remained constant, the freehand selection tool was used to define the mouse retinal cross‐sectional area from the ILM to the outer limiting membrane. Areas outside this region were cleared using Edit>Clear Outside, and the selected area was recorded using the Analyze>Measure command. Next, images (brain or retina) were converted to black and white using Image>Color>Split Channels and duplicated using Image>Duplicate. The duplicated image was processed using Process>Noise>Remove Outliers, applying a constant radius (range: 10 to 15) and threshold setting (range: 30 to 50), depending on marker morphology or labeling pattern. The processed duplicate was then subtracted from the original black‐and‐white image using Process>Image Calculator, and the resulting image was converted into a binary format via Process>Binary>Make Binary. Finally, pixel values were recorded using Analyze>Histogram>List and normalized by dividing by the measured area to obtain standardized pixel values for each marker.

For human wholemount studies, the area (field of view) was kept constant across comparison groups. Protein markers – GFAP, GS, AQP4, and IBA1 – were evaluated using orthogonal projections to capture the optimized signals across the layers. TIFF orthogonal projection images were opened in ImageJ and converted to black and white using Image>Color>Split Channels. The threshold was manually adjusted for each image using Image>Adjust>Threshold to accurately capture the morphology. Pixel values were then obtained using the Analyze>Histogram>List command.

For 12F4⁺Aβ42 depositions appearing as clumps in human wholemounts, a different protocol was applied. A representative Z‐stack slice image was selected at the GCL level, excluding surface‐level 12F4⁺Aβ42 signals that might reflect clearance processes. The selected image was opened in ImageJ and converted to black and white using Image>Color>Split Channels. The threshold was manually adjusted using Image>Adjust>Threshold to accurately capture 12F4⁺Aβ42 clump‐like depositions. Clump size was then defined using Analyze>Analyze Particles, and the number and area of clumps (in pixel units) were recorded by applying a constant size and circularity range across all samples.

2.4. Statistical analysis

2.4.1. Differential gene expression in mouse samples

Relative expression levels were determined using the comparative cycle threshold (Ct) method. ⁴¹ Ct values were obtained using constant threshold and baseline settings across all samples and target genes. Normalized Ct values were compared across animal groups and time points. The Shapiro–Wilk test was initially used to assess normality. As a result of pooling, Ct values were normally distributed for each target gene. Therefore, a two‐way ANOVA (two factors: model and age group) with Bonferroni correction for multiple comparisons was performed to analyze differences among the comparison groups. Target genes were considered differentially expressed if they exhibited a statistically significant (p < 0.05) ≥2‐fold intergroup difference.

2.4.2. Protein expression analysis in mouse samples

Pixel data, normalized to area, were screened for outliers using the ROUT method (Q = 1%) across the four comparison groups: Tg younger, Tg older, NTg younger, and NTg older. The Shapiro–Wilk test was performed to assess normality in the cleaned dataset and to determine whether the data followed a normal or non‐normal distribution within each group. Immunoreactivity was analyzed using either the Brown–Forsythe and Welch ANOVA with Dunnett's T3 multiple comparisons test or the Kruskal–Wallis test with Dunn's multiple comparisons test, as appropriate. For subgroup analysis (central vs peripheral), two central and two peripheral regions were analyzed, excluding two mid‐region pixel values for each retina.

2.4.3. Protein expression analysis in human wholemounts

Pixel data were screened for outliers using the ROUT method (Q = 1%) for AD and control groups. The Shapiro–Wilk test was performed to assess normality in the cleaned data. Differences in wholemount neuroretina findings between AD and control donors were compared using an unpaired t‐test (Student's or Welch's corrected t‐test depending on variance equality) or a Mann–Whitney U test for non‐normally distributed data.

All statistical analyses and graph generations were performed using GraphPad Prism 10.3.0 (GraphPad Software Inc., San Diego, CA, USA).

3. RESULTS

3.1. Evidence from retrospective human post‐mortem samples

3.1.1. Demographics of the human samples studied for wholemount neuroretinas

The demographic details of the samples used are summarized in Table S2. Eight out of 10 AD cases had a high likelihood of dementia. Only two had pure, severe AD neuropathology at the time of death, which occurred in their 60s. The remaining cases had one or more coexisting primary or additional pathologies alongside AD, including Lewy body dementia, frontotemporal lobar degeneration, cerebrovascular disease, cerebral amyloid angiopathy, and others. Most of the control eyes were obtained from donors who died of cancer, and none showed evidence of neurodegenerative diseases.

The specific wholemount neuroretina regions, protein markers used for double and triple staining, number of Z‐stack images analyzed for each combination, sample sizes, mean ages (in years), and gender distributions for the AD and control groups are summarized in Table S4. A total of 70 Z‐stacks for 12F4⁺Aβ42, 65 Z‐stacks for IBA1⁺ microglia/macrophages, 70 Z‐stacks for GFAP⁺ astroglia, 35 Z‐stacks for GS⁺ Müller glia, 55 Z‐stacks for AQP⁺ water channels, and 15 Z‐stacks for CD68⁺ lysosomal compartments – spanning from the ILM to the ONL – were investigated for each group. An unpaired t‐test revealed no statistically significant difference in mean ages between the AD and control groups for each experimental comparison. This lack of significant age difference confirms that the groups were appropriately age‐matched, reducing the potential for age‐related confounding and ensuring that observed effects are more likely attributable to disease‐specific factors rather than age.

3.1.2. Aβ expression pattern in neuroretinas of AD and control subjects from superficial perspective

The imaging planes (surface vs cross‐section) of a control neuroretina (Figure 1A) were used to illustrate the spatial differences in UEA‐I⁺ vasculature, demonstrating how wholemount neuroretina preparations allow for detailed visualization of morphological and pathological changes at the iBRB. Additionally, it is important to consider that in the wholemount neuroretina, labeling patterns varied across layers (e.g., 12F4⁺Aβ42, AQP4⁺ water channels) depending on spatial distribution, as well as the localization of glial cells and associated vascular structures.

FIGURE 1.

Differences in 12F4⁺Aβ42 expression between AD and control neuroretinas. (A) Imaging planes: surface versus cross‐section showing differences in UEA‐l⁺ vasculature in a control retina. (B) Superficial lower‐magnification (10× objective) views of 12F4⁺ Aβ42 expression in a control and an AD donor, respectively. White arrowheads in control retina suggest drainage path‐like patterns. (C) A younger control neuroretina showing weaker 12F4⁺Aβ42 labeling and polarized AQP4⁺ water channels at ILM, above superficial vascular plexus. (D, E) Co‐localization of 12F4⁺Aβ42 with AQP4⁺ water channels and UEA‐l⁺ vessels in control and AD neuroretinas, respectively. White arrowheads in control neuroretina suggest 12F4⁺Aβ42 drainage path‐like patterns; white dashed circle in AD neuroretina suggests 12F4⁺Aβ42 deposition in clumps. (F, G) Co‐localization of 12F4⁺Aβ42 with IBA1⁺ microglia/macrophages (indicated by white arrowheads and zoomed‐in white dashed boxes) in control and AD neuroretinas, respectively. In the control retina, white dashed lines outline blood vessels, suggesting Aβ drainage path‐like patterns through peri‐ and paravascular spaces. In contrast, in the AD retina, white dashed lines outline blood vessels, suggesting defective Aβ drainage around blood vessels, resulting in Aβ clump formation (Video S9). (H, I) Three‐dimensional views of Aβ expression around larger retinal blood vessels (white dashed lines) in control and AD neuroretinas, respectively. White stars in panels (F) and (H) suggest powder‐like soluble forms of 12F4⁺Aβ42, while white arrowheads in panel (I) show neuritic type‐like 12F4⁺Aβ42 plaques. (J, K) The number of 12F4⁺Aβ42 clumps and the clump area, measured in pixel units, were compared between AD and control groups. Mean pixel values for individual donors are overlaid on the violin plots. (L) Retinal cross‐sections from control and AD donor eyes showing intraneuronal inclusions of 6E10⁺ APP/Aβ peptides in both groups, with 6E10⁺ Aβ plaques found only in AD. Statistical comparisons were performed using an unpaired t‐test; p values < 0.05 were considered statistically significant and are indicated. (Scale bar: 20 µm; sample size: N = 10 per group for 12F4 and IBA1 double labeling, and N = 4 per group for 12F4, AQP4, and UEA‐I triple labeling). Aβ, amyloid beta; AD, Alzheimer's disease; APP, amyloid precursor protein; AQP4, aquaporin 4; ILM, inner limiting membrane; UEA, Ulex europaeus agglutinin.

All samples were analyzed for Aβ expression. A low‐magnification view (10× objective) of the superficial layers revealed differences in 12F4 labeling patterns between control and AD neuroretinas (Figure 1B). In the AD neuroretina, 12F4⁺Aβ42 appeared as clump‐like deposits, whereas the matched control neuroretina showed 12F4⁺Aβ42 localization along drainage path‐like patterns (indicated by white arrowheads). In contrast, the neuroretina of a younger control exhibited comparatively weaker 12F4⁺Aβ42 signals (Figure 1C), suggesting more effective clearance systems in the younger group compared to both the AD retina and the age‐matched control.

In the younger control neuroretina, co‐localization of AQP4 with 12F4 and UEA‐l markers revealed that AQP4⁺ water channels were primarily localized at the ILM, facing the vitreous humor (Figure 1C; Video S1). Co‐localization of 12F4 with AQP4 and UEA‐I markers in the ILM and superficial layers is shown in Figure 1D,E for control and AD neuroretinas, respectively. In control tissue, AQP4⁺ water channels were distributed at the ILM–NFL, alongside 12F4⁺Aβ42 displaying drainage‐path‐like patterns (indicated by white arrowheads). In contrast, the AD retina exhibited marked degeneration of AQP4⁺ water channels at the ILM–NFL, co‐localizing with clump‐like 12F4⁺Aβ42 deposits (indicated by a white dashed circle).

Figure 1F,G illustrate the co‐localization of 12F4 with IBA1 in both control and AD neuroretinas, focusing on the level of the GCL and deeper layers. A 3D view of 12F4⁺Aβ42 deposition in control and AD neuroretinas is shown in Figure 1H,I, respectively. In controls, 12F4⁺Aβ42 frequently colocalized within IBA1⁺ microglia/macrophages encasing large retinal blood vessels—a pattern largely diminished in AD (white arrowheads and zoomed‐in dashed white boxes in Figure 1F,G, respectively). Additionally, 12F4⁺Aβ42 deposits were observed within the peri‐ and paravascular spaces of large blood vessels in controls (white dashed lines outline the blood vessels in Figure 1F,H). These deposits appeared as a fine, powder‐like distribution, likely representing soluble Aβ forms (white stars in Figure 1F,H), suggesting the involvement of the peri‐ and paravascular clearance system. In contrast, 12F4 labeling in AD neuroretinas appeared as unevenly distributed clumps, possibly corresponding to insoluble Aβ forms, clustered around blood vessels in the GCL and deeper layers (Figure 1G,I; white dashed lines outline the vessels). The number of 12F4⁺Aβ42 clumps (Figure 1J; mean ± SD: 24.46 ± 19.7 for AD and 5.61 ± 7.8 for controls) and the clump area in pixel units (Figure 1K; 0.824 ± 0.69 for AD and 0.098 ± 0.11 for controls) were significantly higher in AD samples compared to controls (p < 0.0001 for both comparisons, unpaired t‐test).

We have included videos featuring Z‐stack flythroughs to illustrate the drainage path‐like patterns of 12F4⁺Aβ42 in relation to AQP4⁺ water channels and UEA‐I⁺ vessels (Video S2 for control; Video S3, and additionally Video S4 for AD), as well as the peri‐ and paravascular localization of 12F4⁺ Aβ42 peptides with IBA1⁺ microglia/macrophages across retinal layers in both control and AD neuroretinas (Videos S5, S6, and additionally Video S7 for controls; Videos S8, S9, and additionally Video S10 for AD cases).

Additionally, we used retinal cross‐sections to see differences in imaging planes for localizing Aβ species across retinal layers. Using the 12F4 antibody, we did not detect intraneuronal inclusions in retinal cross‐sections from either group. In contrast, a few 12F4⁺Aβ42 plaques were found in the AD retina, particularly within the IPL–ONL regions (data not shown). Conversely, 6E10⁺ APP/Aβ peptides appeared as intraneuronal inclusions, predominantly within the GCL and INL, in both control and AD retinas (Figure 1L), consistent with findings in APP‐PS1 and control mice (Section 3.2.1). In addition, a few 6E10⁺Aβ plaques were observed only in AD retinas (yellow arrowhead).

3.1.3. Disrupted SAβO clearance in Alzheimer's eye

We utilized a novel anti‐SAβO E3 nanobody in conjunction with a modified nanobody staining protocol to visualize SAβO distribution. The expression patterns of the E3 nanobody, along with the 12F4 antibody, are shown in Figure 2. SAβO⁺ signals, detected by the NIR‐E3 nanobody, were predominantly localized within retinal blood vessels, demonstrating the nanobody's ability to cross the iBRB and bind macrophage‐like, Aβ‐binding myeloid lineage cells – a capability not observed with the 12F4 antibody (Figure 2A–E; Videos S11–S14). Video S11 demonstrates E3 nanobody labeling within blood vessels using the vascular marker UEA‐l. Labeling of NIR‐E3⁺ SAβOs across the retinal vasculature is illustrated alongside a schematic diagram (Figure 2B). A significant difference was observed between control (mean ± SD: 6.08 ± 3.7) and AD (1.71 ± 2.1) samples, with counts of NIR‐E3⁺ macrophage‐like, Aβ‐binding myeloid lineage cells (yellow arrowheads) significantly higher in controls (p < 0.0001, unpaired t‐test) (Figure 2C). Erythrocytes (pink arrowheads), identified by their biconcave disk shape and lack of nuclear staining, did not exhibit any positive signal for the NIR‐E3 nanobody.

FIGURE 2.

Localization of SAβOs in human wholemount neuroretinas. An E3 nanobody conjugated with NIR (655 nm, yellow) and FAM (495 nm, green) dyes was used alongside the 12F4 antibody (546 nm, red) to stain wholemount peripheral retinas from human AD and control donor eyes, following an ex vivo nanobody staining protocol. (A) Two‐dimensional images of control and AD retinas stained for NIR‐E3 and 12F4 markers demonstrate NIR‐E3⁺ SAβO uptake by macrophage‐like, Aβ‐binding myeloid lineage cells (yellow arrowheads) within retinal blood vessels in superficial vascular plexus. In contrast, erythrocytes (pink arrowheads, identified by their biconcave disk shape) did not show SAβOs uptake. (B) SAβOs were detected throughout the retinal vasculature (a schematic of the retinal vasculature is included), regardless of vessel size. (C) Differences in NIR‐E3⁺ cell counts between control and AD retinas were analyzed using an unpaired t‐test, with mean counts for individual donors superimposed on the violin plots. (D, E) Three‐dimensional visualization of SAβOs within retinal blood vessels in control and AD retinas, respectively. (F) Aβ species at the retina–vitreous humor interface stained positive for FAM‐E3 and 12F4 markers. (G) Immunoreactivity, measured in pixels for the FAM‐E3 nanobody and 12F4 antibody, was compared between AD and control groups using the Mann–Whiteny U test. P values < 0.05 were considered statistically significant and are indicated. (Scale bar: 20 µm; sample size: N = 7 per group for NIR‐E3 and 12F4 double labeling, and N = 3 per group for FAM‐E3 and 12F4 double labeling). Aβ, amyloid‐beta; AD, Alzheimer's disease; FAM, fluorescein; NIR, near‐infrared; SAβO, soluble Aβ oligomer.

Figure S2 illustrates the presence of NIR‐E3⁺ SAβOs within both the superficial (Figure S2A) and deep (Figure S2B) retinal vascular plexuses. Higher‐magnification 2D images (Figure S2C–E) further demonstrate the uptake of NIR‐E3⁺ SAβOs, possibly by peripheral monocytes (indicated by white dashed circles), based on their size, nuclear morphology, and intravascular localization. A corresponding 3D image and Video S13 are provided alongside schematic cartoons (Figure 2SF; white arrowheads and dashed circles). Due to the lack of commercially available nanobodies capable of penetrating the BRB to specifically label peripheral monocytes, we were unable to confirm their identity using a definitive marker. However, our findings suggest a potential role for peripheral monocytes in the recruitment of SAβOs – a process that appears to be significantly diminished in AD neuroretinas.

A small subset of samples (N = 3 per group) was used for double staining with FAM‐labeled E3 nanobody and Alexa Fluor 546‐labeled 12F4 antibody. Both markers confirmed their localization at the retina–vitreous humor interface (Figure 2F; Videos S15 and S16). In control samples, a flow‐like nanoparticle distribution was observed at this interface. In contrast, these patterns were either diminished or disrupted in AD retinas, resulting in a significant reduction in the FAM‐E3 signal (p = 0.0186, Mann–Whitney U test) (Figure 2G). Similar results were obtained using the Alexa Fluor 546‐labeled 12F4 antibody, with a comparable signal reduction in AD retinas (p = 0.0408, Mann–Whitney U test) (Figure 2G). However, quantitative analysis was limited to three AD and three control neuroretinas due to technical challenges in preserving the retina–vitreous interface from dissection through mounting.

Figure S3 provides additional evidence of SAβO labeling within retinal blood vessels, illustrating differences among a negative control (Figure S3A), a younger control (Figure S3B), and an age‐matched control (Figure S3C). Three‐dimensional views of SAβOs⁺ cells within retinal blood vessels revealed no discernible differences in the staining patterns of NIR‐ (Figure S3D) and FAM‐ (Figure S3E) dye‐conjugated to E3 nanobodies between control and AD neuroretinas. Aβ42, labeled by both the E3 nanobody and the 12F4 antibody, was detected at the retina–vitreous humor interface in both control and AD retinas (Figure S3F,G), with larger Aβ42 species prominently observed in AD neuroretinas (Figure S3G).

3.1.4. Microglia morphology in neuroretinas of AD and control donors

Differences in Aβ expression patterns between AD and control neuroretinas suggest potential differences in Aβ clearance between the two groups. Microglia, the resident macrophages of the CNS, serve as the first line of defense against pathogens and injury in the brain. Reduced Aβ uptake capacity by microglia has been proposed as a major contributor to AD pathogenesis. ⁴² , ⁴³ To explore this possibility, we analyzed all samples to compare microglial expression patterns between AD and control neuroretinas (Figures 3 and S4).

FIGURE 3.

Differences in microglia morphology between AD and control neuroretinas. (A) Low‐magnification (10× objective) superficial views of IBA1⁺ microglia/macrophages in control and AD neuroretinas (infiltrated IBA⁺ cells are indicated by white arrowheads). (B) Co‐localization of 12F4⁺Aβ42 with IBA1⁺ cells (yellow arrowheads) in an AD neuroretina suggests hyalocyte infiltration, based on their spatial localization at ILM. (C) Three‐dimensional renderings of microglial expression in control and AD neuroretinas. White arrowheads in control retina indicate microglia processes containing circular structures, while yellow arrowheads in AD retina indicate rod‐shaped or elongated morphology. (D) IBA1‐CD68 double labeling is shown for a younger control, an age‐matched control, and an AD donor. Yellow arrowheads indicate IBA1 and CD68 co‐localization in both control and AD retinas. (E) The immunoreactivity of IBA1⁺ cells, quantified in pixels, was compared between AD and control groups using an unpaired t‐test. Mean pixel values for individual donors are overlaid on violin plots. (F) The total number of IBA1⁺CD68⁺ co‐localized circular structures was compared between AD and control groups using the Mann–Whiteny U test. (G) Triple labeling of IBA1, LAMP1, and UEA‐l markers is shown for a control and an AD donor, with no colorization observed between IBA1 and LAMP1. (H) Triple labeling of IBA1, CD45, and UEA‐I markers shows sparse IBA1⁺ cells with CD45⁺ cytoplasmic tail labeling (yellow arrowheads) in both control and AD retinas. (I) Retinal cross‐sections from AD and control donors stained for IBA1 and UEA‐l show microglial processes with rounded structures in controls, predominantly spanning from the NFL to the IPL. In AD retinas, microglia exhibit a rod‐shaped or elongated morphology within the NFL–GCL, IPL, and OPL. A p value < 0.05 was considered statistically significant and is indicated. (Scale bar: 20 µm; sample size: N = 10 per group for 12F4 and IBA1 double labeling, and N = 3 per group for IBA1 and CD68 double labeling). Aβ, amyloid beta; AD, Alzheimer's disease; GCL, ganglion cell layer; ILM, inner limiting membrane; IPL, inner plexiform layer; NFL, nerve fiber layer; OPL, outer plexiform layer; UEA, Ulex europaeus agglutinin.

Low‐power images of the superficial layers in wholemount neuroretinas revealed notable differences in microglial distribution and morphology (Figure 3A, 10× objective; Figure S4A,B, 2.5× objective). Specifically, in AD neuroretinas, infiltration of IBA1⁺ cells was detected in the ILM–NFL region, as indicated by white arrowheads in Figure 3A. Some of these infiltrated IBA1⁺ cells, possibly hyalocytes – vitreous‐resident macrophages – also showed positive signals for 12F4⁺Aβ42, suggesting their involvement in Aβ phagocytosis (Figure 3B, yellow arrowheads). Video S17 further supports the notion that these infiltrating IBA1⁺ cells are most likely hyalocytes based on their spatial localization.

High‐magnification 3D views revealed distinct microglial morphologies between AD and control neuroretinas (Figure 3C). In control samples, microglial processes were predominantly rounded, resembling phagocytic cup‐like structures (white arrowheads), and were concentrated in the NFL–IPL, particularly around larger retinal vessels (Videos S5–S7). In contrast, microglia in AD neuroretinas appeared rod‐shaped or elongated (yellow arrowheads), with retracted or absent phagocytic processes, suggesting dystrophic morphologies within the NFL–IPL (Videos S8–S10). Quantitative analysis of immunoreactivity, measured as normalized pixel counts, revealed significantly higher labeling in AD neuroretinas compared with controls (p = 0.0036, unpaired t‐test) (Figure 3E). This increase is attributed to resident microglia/macrophages and infiltrating IBA1⁺ cells.

To further investigate phagocytic activity, a subset of samples (N = 3 per group) was analyzed using CD68, a marker for lysosomal/endosomal compartments of microglia/macrophages (Figures 3D and S4C,D). Representative images include a younger control, an age‐matched control, and an AD donor (yellow arrowheads denote IBA1⁺CD68⁺ cells). The number of co‐labeled microglial processes – potentially suggestive of phagocytic activity – was assessed semi‐quantitatively and yielded no significant difference between AD and control retinas (p = 0.5969, Mann–Whitney U test; Figure 3F). Figure S4 illustrates CD68 and IBA1 co‐localization in the ILM and superficial layers (Figure S4C), where yellow arrowheads denote IBA1⁺CD68⁺ cells and white dashed circles indicate IBA1⁺CD68⁻ cells, as well as around blood vessels (Figure S4D), with white dashed lines outlining vessels and yellow arrowheads indicating co‐localization. These findings suggest that co‐labeling is particularly prominent around large vessels, regardless of disease status. Videos S18 and S19 illustrate the IBA1 and CD68 co‐localization across retinal layers in control and AD neuroretinas, respectively. In AD samples, most infiltrating IBA1⁺ cells were negative for CD68, further supporting their identity as hyalocytes, consistent with prior literature. ⁴⁴ , ⁴⁵

For additional validation, we used IBA1, LAMP1, and UEA‐l markers to assess lysosomal activity (Figures 3G and S4E) and IBA1, CD45, and UEA‐l markers to evaluate the presence of blood‐derived macrophages (Figures 3H and S4F) in both groups. Minimal co‐localization of LAMP1 and IBA1 was observed in control neuroretina, but not in the AD sample. Interestingly, co‐localization of LAMP1 and UEA‐1 was detected in the superficial vascular plexus of the control neuroretina, suggesting lysosomal activity in UEA‐l⁺ vascular endothelial cells (Figure S4E, yellow arrowheads; Video S20). CD45, a surface marker of hematopoietic cells, was not detected in the wholemount neuroretinas of either group. However, a few IBA1⁺ cells encasing vessel walls exhibited cytoplasmic CD45 labeling, possibly resulting from cleavage of the CD45 cytoplasmic tail during phagocyte activation (yellow arrowheads in Figures 3H and S4F). ⁴⁶ , ⁴⁷ Videos S21 and S22 show resident microglial morphology and lack of CD45 labeling, along with UEA‐1⁺ vessel labeling, in control and AD retinas, respectively.

In addition to wholemounts, retinal cross‐sections were examined to assess microglial morphology. In control retinas, IBA1⁺ microglial processes appeared mostly rounded, spanning from the NFL to the IPL (Figure 3I). In contrast, microglia in AD retinas exhibited rod‐shaped or elongated morphology, primarily localized to the NFL–GCL, IPL, and OPL (Figure 3I).

Additionally, in control neuroretinas, beyond the prominent co‐localization of 12F4⁺Aβ42 with IBA1⁺ signals along blood vessel walls (Figure 1F), co‐localization of 12F4⁺Aβ42 with IBA1⁺ phagosome‐like structures was also observed at the superficial layers (Figure S4G). However, some 12F4⁺Aβ42 signals did not co‐localize with IBA1⁺ cells (Figure S4H), suggesting that lysosomal activity may also originate from the superficial vasculature, as illustrated in Figure S4E and Video S20 for the control neuroretina. These Aβ clearance processes are not readily captured in brain or eye cross‐sectional studies. In our previous work comparing wholemounts with cross‐sections, we identified IBA1⁺ blood‐derived phagosome‐like structures co‐labeled with 6E10⁺ APP/Aβ peptides in ergothioneine‐treated 5XFAD mice, in contrast to untreated 5XFAD mice. ⁴⁰

3.1.5. Macroglia morphology in AD and control neuroretinas

Macroglial (GFAP⁺ and GS⁺) morphology in the wholemounts is illustrated in relation to retinal blood vessels (Figure 4A–E). According to the literature, GFAP labels astrocytes and reactive Müller cells; however, under normal conditions, GFAP expression in Müller cells is restricted to their endfeet. In contrast, GS is a robust marker for Müller glia and is expressed throughout the entire cell. UEA‐l labels the endothelial cells of retinal blood vessels.

FIGURE 4.

Differences in macroglia morphology between AD and control neuroretinas. (A–C) Macroglia (GFAP⁺ and GS⁺) associated with retinal blood vessels (UEA‐1⁺) in ILM superficial layers are illustrated for a younger control, an age‐matched control, and an AD donor, respectively. (D, E) Three‐dimensional views of macroglial morphology and blood vessels are presented for control and AD neuroretinas, respectively. (F, G) To illustrate differences in imaging planes, retinal cross‐sections from AD and control donors stained with GFAP, GS, and UEA‐l markers are shown. (H) An artery and a vein, along with associated macroglia located between the ILM and GCL of a control eye, are distinguished by their round and flattened lumens, respectively. (I, J) Immunoreactivity, measured in pixels, is shown for both GFAP and GS markers. An unpaired t‐test was used for the comparisons. Mean pixel values for individual donors are superimposed on the violin plots. P values < 0.05 were considered statistically significant and are indicated (scale bar: 20 µm; sample size: N = 7 per group for GFAP, GS, and UEA‐I triple labeling). AD, Alzheimer's disease; GCL, ganglion cell layer; GFAP, glial fibrillary acidic protein; GS, glutamine synthetase; ILM, inner limiting membrane; UEA, Ulex europaeus agglutinin.

Our comprehensive assessment of wholemount Z‐stacks from control and AD neuroretinas revealed a range of differences in the superficial macroglial structures. Like younger controls (Figure 4A; Video S23), age‐matched controls showed preserved macroglial structures (Figure 4B,D; Video S24), whereas AD neuroretinas exhibited prominent degenerative changes (Figure 4C,E; Video S25). Müller glial endfeet were tightly attached to larger blood vessels in controls, similar to those observed in younger controls (merged images in Figure 4A,B,D), whereas these attachments appeared degraded in AD neuroretinas (merged images in Figure 4C,E). Importantly, AD retinas displayed a wide spectrum of degenerative changes, ranging from bead‐like formations or clasmatodendrosis to a near‐complete absence of macroglia (Videos S25 and S26).

Additionally, to assess differences in imaging planes, retinal cross‐sections from human AD donors and age‐matched controls were examined using GFAP, GS, and UEA‐I markers (Figure 4F,G, respectively). Similar to the wholemounts, control samples exhibited intact macroglial morphology, with UEA‐I⁺ blood vessels prominently within ILM–GCL. In contrast, GS⁺ Müller glia and UEA‐I⁺ endothelial cells showed marked degenerative changes, and GFAP⁺ astroglia showed relatively reduced expression in AD – supporting the wholemount findings that macroglial degeneration is a prominent feature of AD.

Figure 4H illustrates the labeling patterns of GFAP, GS, and UEA‐I markers around an artery with a rounded lumen and a vein with a flattened lumen in a retinal cross‐section from an older control, based on their characteristic appearances as described in the literature.

Pixel intensity measurements of immunoreactivity revealed a significant reduction in GFAP (p = 0.0025, unpaired t‐test) and GS (p = 0.0015, unpaired t‐test) expression in AD neuroretinas (Figure 4I,J, respectively), consistent with our qualitative assessment of human retinal cross‐sections. This suggests degeneration of both astrocytes and Müller glia, potentially linked to ocular glymphatic dysfunction. Although reports on wholemount GFAP expression in human AD versus control neuroretinas are limited, our earlier study similarly demonstrated significant depletion of GFAP expression in 5xFAD mice wholemounts compared with matched C57BL/6J controls, further supporting ocular glymphatic dysfunction. ⁴⁰

3.1.6. Membrane distribution of AQP4 water channels along with GFAP⁺ astroglia

AQP4 is the principal water channel in the neuropil of the CNS, primarily expressed in astrocytes but also found in endothelial cells. ⁴⁸ , ⁴⁹ It exhibits a polarized distribution in brain astrocytes and retinal Müller cells, with enrichment in the endfeet membrane that contacts brain microvessels, the subarachnoid space, or the vitreous humor and retinal blood vessels. A lower but significant presence is also observed in non‐endfeet membrane, including astrocytic processes that ensheath glutamatergic synapses. ⁴⁸ , ⁴⁹

Since AQP4 water channels contribute to glymphatic Aβ clearance, ⁵⁰ , ⁵¹ we investigated their distribution in AD and control wholemount neuroretinas, along with GFAP⁺ astroglia and UEA‐l⁺ vessels, using 3D imaging. Co‐localization of GFAP⁺ astroglia and AQP4⁺ water channels was primarily observed in the ILM and superficial layers of both control and AD donor neuroretinas (Figure 5A–E, indicated by white dashed circles and Videos S27 and S28). The non‐endfeet distribution of AQP4⁺ water channels was largely preserved in age‐matched controls (Figure 5B,D), resembling that of younger controls (Figures 5A and S5A), but was notably degenerated in AD retinas (Figure 5C,E). The densest expression of AQP4⁺ water channels at the endfeet of Müller glia, which are tightly attached to the blood vessel wall, was confirmed by AQP4, GS, and UEA‐l, as well as GS, GFAP, and UEA‐l triple staining in a young control neuroretina (white dashed circles, Figure 5F,G, respectively). Quantification of immunoreactivity, measured in pixels for AQP4 and GFAP, revealed significantly higher values in controls compared to AD retinas (p < 0.0001 and p = 0.0013, respectively, unpaired t‐test) (Figure 5H,I, respectively).

FIGURE 5.

Distribution of AQP4 water channels along with GFAP⁺ astroglia between AD and control neuroretinas. (A–C) The expression patterns of AQP4⁺ water channels, GFAP⁺ astroglia, and UEA‐l⁺ vessels are shown at the ILM superficial layers for a younger control, an age‐matched control, and an AD, respectively. (D, E) Three‐dimensional views of control and AD neuroretinas, respectively. (F, G) Densest expression of AQP4⁺ water channels at endfeet of Müller cells, particularly at the vessel wall, was confirmed using the Müller cell‐specific marker GS, along with GFAP and UEA‐l, in a younger control retina. This arrangement resembled a “sewer top” structure facilitating water movement (white dashed circles). (H, I) Immunoreactivity, measured in pixels, is shown for both AQP4 and GFAP, respectively. An unpaired t‐test was used for statistical analysis. Mean pixel values for individual donors are superimposed on the violin plots. (J, K) Retinal cross‐sections from a control and an AD donor, respectively, illustrate AQP4⁺ water channel distribution, suggesting a polarized arrangement at the Müller glia endfeet‐like structures in controls and degeneration with loss of polarity in the AD retina. (L) An older control wholemount neuroretina at the ILM level shows the peri‐ and paravascular localization of 12F4⁺Aβ42 along with the densest AQP4⁺ water channels. White dashed lines outline an adjacent artery and vein, indicated within the white dashed box at the GCL level based on the lumen size. White arrows suggest possible direction of 12F4⁺Aβ42 drainage from an artery to a vein. P values < 0.05 were considered statistically significant and are indicated. (Scale bar: 20 µm; sample size: N = 7 per group for AQP4, GFAP, and UEA‐I triple labeling). Aβ, amyloid beta; AD, Alzheimer's disease; AQP4, aquaporin 4; GCL, ganglion cell layer; GFAP, glial fibrillary acidic protein; GS, glutamine synthetase; ILM, inner limiting membrane; UEA, Ulex europaeus agglutinin.

Additionally, retinal cross‐sections from AD and control neuroretinas were examined for AQP4, GFAP, and UEA‐I markers to validate differences observed across imaging planes. In controls, AQP4⁺ water channels exhibited a polarized appearance, possibly localized to Müller cell endfeet‐like structures within the ILM–NFL region (Figures 5J and S5B). In contrast, AD retinal cross‐sections exhibited both a depolarized arrangement and degeneration of AQP4⁺ channels in the same region (Figures 5K and S5B). Such AQP4⁺ water channel appearance was not observed in any of our preclinical data, despite a relatively larger sample size (Section 3.2.1).

Ex vivo imaging of a control wholemount neuroretina illustrated the peri‐ and paravascular localization of 12F4⁺Aβ42 in both an artery and a vein at the ILM (Figure 5L; white dashed lines outline the adjacent vessels, and white arrows suggest the possible direction of the flow path). 12F4⁺Aβ42 was particularly localized along vessel walls, where glial cell endfeet are tightly attached and enriched with AQP4⁺ water channels, as illustrated in Figure 5F,G. The distinction between artery and vein was based on lumen size, as shown at the GCL level (white dashed box). Videos S29 and S30 provide a layer‐by‐layer fly‐through of the staining patterns described earlier.

Figure S5 further illustrates a spectrum of morphological changes in GFAP and AQP4 labeling across control cases (Figure S5C–E) and AD cases (Figure S5F–J). A 70‐year‐old female control showed some GFAP⁺ reactive astrocytes and reduced non‐endfeet AQP4⁺ water channels (Figure S5E), while two female AD cases in their 80s exhibited GFAP⁺ severe reactive astrogliosis and markedly depolarized or degenerated AQP4⁺ water channels (Figure S5I,J). Notably, two sporadic AD cases in their 60s – both presenting severe AD neuropathology without other comorbidities – exhibited marked degeneration of both endfeet and non‐endfeet AQP4⁺ water channels, along with GFAP⁺ astroglia degeneration (Figure S5H), suggesting a potential link between early‐onset sporadic AD and glymphatic system dysfunction.

3.2. Evidence from preclinical transgenic mouse models

3.2.1. Localization of APP/Aβ peptides and glial cells in the retina of 9‐month‐old APP‐PS1 mice

The localization of APP/Aβ peptides and AQP4⁺ water channels across central and peripheral retinal cross‐sections from 9‐month‐old APP‐PS1 mice is shown in Figure 6A. Intraneuronal inclusion was predominantly observed in the GCL, followed by the INL, as indicated by the yellow dashed regions. The co‐localization of the macroglial markers GFAP and GS in central and peripheral retinal cross‐sections of 9‐month‐old APP‐PS1 mice is shown in Figure 6C. The endfeet of GS⁺ Müller cells, along with GFAP⁺ astrocytes, were primarily localized to the ILM and NFL. Regarding IBA1⁺ microglia, staining appeared relatively weak in the retinal cross‐sections regardless of mouse model; however, their processes were detectable in the NFL–GCL, IPL, and OPL as illustrated in Figure 6E for 9‐month‐old APP‐PS1 mice.

FIGURE 6.

Localization of APP/Aβ peptides and glial cells in retina of 9‐month‐old APP‐PS1 mice. (A) Co‐localization of 6E10⁺ APP/Aβ peptides with AQP4⁺ membrane water channels, (C) co‐localization of macroglia markers GFAP⁺ astrocytes and GS⁺ Müller cells, and (E) IBA1⁺ microglia were demonstrated across central and peripheral regions of retina of 9‐month‐old APP‐PS1 mice. (B, D, F) Quantification of immunoreactivity (measured in normalized pixels) was performed for 6E10⁺ APP/Aβ peptides, AQP4⁺ water channels, GFAP⁺ astrocytes, GS⁺ Müller cells, and IBA1⁺ microglia across the entire retina. Data are presented as violin plots comparing 3‐ and 9‐month‐old APP‐PS1 mice and controls. Statistical significance was assessed using the non‐parametric Kruskal–Wallis test followed by Dunn's multiple comparisons test. (G) A knockout‐validated rabbit monoclonal APP antibody confirmed the intraneuronal inclusions of APP peptides in both the central and peripheral retinal regions (indicated by yellow dashed boxes) in 9‐month‐old APP‐PS1 mice. (H) Double immunostaining with 6E10 and AQP4 shows the distribution of AQP4⁺ water channels in the hippocampus and neocortex, illustrating similarities in the imaging planes between retinal cross‐sections and brain sagittal sections, particularly regarding AQP4⁺ water channels (yellow arrowheads) in 9‐month‐old APP‐PS1 mice. White arrowheads indicate 6E10⁺ Aβ plaques. (I) NIR‐E3⁺ SAβOs were detected in a 3‐month‐old APP‐PS1 mouse along the endothelial cell layer of blood vessel lumens near the optic nerve (white arrowheads), with some oligomers taken up by peripheral myeloid lineage cells, possibly monocytes, based on their characteristic horseshoe‐shaped nuclei (zoomed‐in images shown in white dashed boxes and arrowheads). (J) Co‐localization of 6E10⁺ APP/Aβ peptides with IBA1⁺ macrophages in the superficial retinal layers suggests the presence of phagosome‐like structures in a 3‐month‐old APP‐PS1 mouse (yellow arrowheads). (K) A heatmap illustrates differential gene expression in pooled eye tissues comparing APP‐PS1 and control mice at 3 and 9 months of age, including age‐related changes within each group. Significantly upregulated genes are shown with corresponding p values and fold changes. Two‐way ANOVA with Bonferroni‐corrected multiple comparisons was used for statistical analysis. P values < 0.05 were considered statistically significant. (Scale bar: 20 µm; sample size: N = 4 per age group and model for 6E10 and AQP4 double labeling, GFAP and GS double labeling, and IBA1 single staining). Aβ, amyloid beta; APP, amyloid precursor protein; AQP4, aquaporin 4; GFAP, glial fibrillary acidic protein; GS, glutamine synthetase; NIR, near‐infrared; SAβO, soluble Aβ oligomer.

Pixel quantification was performed using six regions of interest per retinal cross‐section to represent the entire retina. Table S5 summarizes the immunoreactivity of each marker, measured in normalized pixel values. Expression of 6E10⁺ APP/Aβ peptide was significantly elevated in the retinas of 9‐month‐old APP‐PS1 mice compared to both age‐matched control mice (p < 0.0001) and their younger counterparts (p < 0.0001). In contrast, 9‐month‐old control mice exhibited a significant decrease in APP/Aβ levels compared to 3‐month‐old controls (p = 0.0032) (Figure 6B). Interestingly, AQP4 expression significantly increased with aging in the retinas of both APP‐PS1 (p < 0.0001) and control mice (p = 0.0019) (Figure 6B). GFAP expression was significantly lower in 3‐month‐old APP‐PS1 mice compared to their age‐matched controls (p = 0.041) (Figure 6D). However, GFAP immunoreactivity showed a significant increase in 9‐month‐old APP‐PS1 mice (p = 0.0003), whereas GS immunoreactivity significantly decreased in 9‐month‐old control mice compared to their 3‐month‐old counterparts (p = 0.0007) (Figure 6D). Additionally, IBA1 expression significantly increased with aging in the retinas of both APP‐PS1 mice (p < 0.0001) and control mice (p < 0.0001) (Figure 6F).

To validate 6E10⁺ APP peptide presence in mouse tissues, we repeated staining with a rabbit monoclonal, knockout‐validated APP antibody, revealing a similar expression pattern in retinal cross‐sections of both APP‐PS1 (Figure 6G, yellow dashed regions) and control (Figure S6C, white dashed regions) mice. AQP4⁺ water channel localization in the hippocampus and neocortex of a 9‐month‐old APP‐PS1 mouse, alongside 6E10⁺ Aβ plaques (Figure 6H, white arrowheads), displaying similar AQP4 expression patterns in both retinal and brain cross‐sections (yellow arrowheads, Figure 6A,H). We further examined the presence of SAβOs in the retinal cross‐sections using the NIR‐E3 nanobody (Figure 6I). SAβOs were primarily observed in endothelial cells lining the blood vessel lumens surrounding the optic nerve (white arrowheads, left), with some also taken up by myeloid lineage cells (white arrowheads, right). The horseshoe‐shaped nucleus (indicated in the zoomed‐in white dashed box) suggests monocyte involvement in SAβO uptake, consistent with findings from human neuroretina wholemounts (Figures 2 and S2). Notably, SAβOs were not observed as intraneuronal inclusions; tiny puncta were sparsely scattered across the retinal layers. NIR‐E3⁺ labeling is relatively weak in 9‐month‐old APP‐PS1 mice compared to the other three groups.

Furthermore, wholemount neuroretina from a 3‐month‐old APP‐PS1 mouse revealed IBA1⁺ phagosome‐like structures co‐labeled with 6E10⁺ APP/Aβ peptides in the superficial ILM–NFL region (Figure 6J). These structures are not discernible in cross‐sectional imaging, as previously demonstrated in our view‐plane comparison of 5xFAD and C57BL/6J mice. ⁴⁰ The lack of wholemount preparations for the respective animals – due to one eye being used for cross‐sectional studies and the other for gene expression analyses – precluded confirmation of these phagosome‐like structures across all groups using markers CD68 or LAMP1.

Figure S6 summarizes the 6E10 and AQP4 labeling across the four comparison groups for central (Figure S6A) and peripheral (Figure S6B) retinas, respectively. Pixel quantification of 6E10 and AQP4 expression was conducted separately for central (Figure S6D) and peripheral (Figure S6E) regions, excluding the mid‐retinal region. This subgroup analysis revealed a significant increase in 6E10⁺ APP/Aβ peptide levels in the central retina of 9‐month‐old APP‐PS1 mice compared to their younger counterparts (p = 0.0018). In the peripheral retina, 6E10⁺ APP/Aβ peptide levels were significantly higher in 9‐month‐old APP‐PS1 mice compared to age‐matched controls (p = 0.0186). Other comparisons – either between models or across age groups – were not statistically significant. In contrast, AQP4 expression showed a significant age‐related increase in 9‐month‐old APP‐PS1 mice in both the central (p = 0.0348) and peripheral (p = 0.0006) retinas. For control mice, AQP4 expression significantly increased only in the peripheral retina at 9 months of age (p = 0.0146).

Co‐localization of GFAP and GS across the comparison groups for the central and peripheral retinas is shown in Figure S6F,G, respectively. Pixel quantification for these regions is presented in Figure S6H,I. In the central retina, aging was associated with a significant increase in GFAP expression in APP‐PS1 mice (p = 0.0442) and a significant decrease in GS expression in control mice (p = 0.0285) compared to their younger counterparts. No significant changes were observed for either marker in the peripheral retina.

IBA1 staining patterns in the central and peripheral retinas are shown in Figure S6J,K, respectively, with pixel quantification provided in Figure S6L. Both APP‐PS1 and control mice exhibited significant age‐related increases in IBA1 expression in both the central (p = 0.0258 and p = 0.0221, respectively) and peripheral (p = 0.0035 and p = 0.0043, respectively) retinas compared to their younger counterparts.

Table S6 summarizes the relative expression levels of 22 target mRNAs based on RT‐qPCR data. Both mouse App and human PSEN1 mRNA levels were significantly upregulated in pooled eye tissues from both 3‐ and 9‐month‐old APP‐PS1 mice compared to age‐matched controls. A heatmap (Figure 6K) illustrates the differential gene expression patterns between APP‐PS1 and control mice at 3 and 9 months of age, as well as age‐related differences within each group. Notably, the microglial markers Aif1 (2.4‐fold, p < 0.0001) and Trem2 (2.3‐fold, p = 0.0016) were significantly upregulated in 9‐month‐old APP‐PS1 mice compared to 3‐month‐old counterparts.

3.2.2. Localization of APP/Aβ peptides and glial cells in neocortex‐hippocampus of 9‐month‐old APP‐PS1 mice

The distribution of Aβ plaques and co‐localization with astroglia and microglia in the hippocampus and neocortex of 9‐month‐old APP‐PS1 mice are shown in Figure 7A,B, respectively. The number of 6E10⁺ Aβ plaques significantly increased (p < 0.0001) in both regions of 9‐month‐old APP‐PS1 mice compared to 3‐month‐old counterparts and age‐matched controls (Figure 7F; Tables S7 and S8). In the hippocampus, 3‐month‐old APP‐PS1 mice showed a significantly higher GFAP expression than their age‐matched counterparts (p = 0.0022). However, with aging, GFAP⁺ astroglia showed a significant reduction in the hippocampus of 9‐month‐old APP‐PS1 mice (p = 0.0004) (Figure 7H; Table S7). In contrast, IBA1⁺ microglial cell counts (p = 0.0099) (Figure 7G; Table S7) and pixel counts (p = 0.0374) (Figure 7I; Table S7) were significantly increased in the hippocampus of 9‐month‐old controls compared to 3‐month‐old controls, suggesting a role for microglia in mitigating APP/Aβ peptide buildup, similar to findings observed in the aging control retina (Figure 6F). Aside from the 6E10⁺ Aβ plaques, astroglial and microglial expression in the neocortex showed no significant differences between APP‐PS1 mice and controls at either time point (Table S8).

FIGURE 7.

Localization of APP/Aβ peptides and glial cells in the neocortex‐hippocampus of 9‐month‐old APP‐PS1 mice. Co‐localization of 6E10⁺ Aβ plaques (yellow arrowheads) with (A) GFAP⁺ astroglia and (B) IBA1⁺ microglia was observed exclusively in the hippocampus and neocortex regions of 9‐month‐old APP‐PS1 mice. (C) A rabbit monoclonal knockout‐validated APP antibody confirmed intraneuronal inclusions of APP peptides (white arrows), predominantly in the neocortex, with minimal or no APP inclusions observed in the hippocampal CA regions. APP⁺ dystrophic neurites (green arrows) were uniquely identified in 9‐month‐old APP‐PS1 mice. (D) NIR‐E3⁺ SAβOs in a 3‐month‐old APP‐PS1 mouse were primarily detected within the endothelial cell layers of blood vessel lumens at meninges and appeared as small deposits within parenchyma (white arrowheads). (E) Intraneuronal inclusions of 6E10⁺ APP peptides (white arrows) were observed in neocortex of 9‐month‐old APP‐PS1 mice. (F–I) Semi‐quantification of 6E10⁺ Aβ plaques (including diffuse‐type) and IBA1⁺ microglia, along with immunoreactivity quantified in pixels for GFAP⁺ astroglia and IBA1⁺ microglia, is presented as violin plots comparing 3‐ and 9‐month‐old APP‐PS1 mice and controls in both the hippocampus and neocortex. Statistical significance was assessed using either the Brown–Forsythe and Welch ANOVA tests with Dunnett's T3 multiple comparisons or the Kruskal–Wallis test with Dunn's multiple comparisons, as appropriate. (J) A heatmap illustrates differential gene expression in pooled neocortex‐hippocampus tissues, comparing APP‐PS1 and control mice at 3 and 9 months of age, including age‐related changes within each group. Significantly upregulated genes are shown with corresponding p values and fold changes. Two‐way ANOVA with Bonferroni‐corrected multiple comparisons was used for statistical analysis. P values < 0.05 were considered statistically significant. (Scale bar: 20 µm; sample size: N = 4 per age group and model for 6E10 and GFAP, and 6E10 and IBA1 double labeling). Aβ, amyloid beta; APP, amyloid precursor protein; GFAP, glial fibrillary acidic protein; NIR, near‐infrared; SAβO, soluble Aβ oligomer.

Figure S7 summarizes the double labeling of 6E10 and GFAP or IBA1 across comparison groups in the hippocampus and neocortex. Co‐localization of 6E10⁺ Aβ plaques with GFAP⁺ astrocytes (yellow arrowheads) was observed exclusively in 9‐month‐old APP‐PS1 mice (Figure S7A). Similarly, co‐localization of 6E10⁺ Aβ plaques with IBA1⁺ microglia (yellow arrowheads) was detected only in the hippocampus (Figure S7B) and neocortex (Figure S7C) of 9‐month‐old APP‐PS1 mice. To a lesser extent, diffuse‐type 6E10⁺ Aβ plaques were also found in 3‐month‐old APP‐PS1 mice (white arrowhead in Figure S7B).

Although we focused solely on 6E10⁺ Aβ plaques for quantification, intraneuronal inclusions of 6E10⁺ APP/Aβ peptides were also identified within cortical neurons (Figure 7E, white arrows), consistent with the intraneuronal inclusions observed in retinal cross‐sections (Figure 6A). To ensure that the observed signals were not due to background staining or cross‐reactivity from mouse antibodies, we employed a knockout‐validated rabbit monoclonal APP antibody. Intraneuronal APP⁺ inclusions were detected in the neocortex of all animal groups but were absent in the hippocampus (Figures 7C and S7D,E). Strong APP⁺ intraneuronal inclusions were detected in 3‐month‐old APP‐PS1 mice, while APP⁺ dystrophic neurites were observed in the neocortex of 9‐month‐old APP‐PS1 mice. In control mice, APP immunoreactivity decreased with age. This is noteworthy, as AD is characterized by neuritic plaques containing dystrophic neurites, ⁵² likely resulting from impaired clearance of intraneuronal APP peptides, as demonstrated in the neocortex of 9‐month‐old APP‐PS1 mice (Figures 7C and S7E). Interestingly, our recent work in 9‐month‐old apolipoprotein E (ApoE) knockout mice also showed strong immunoreactivity for APP⁺ intraneuronal inclusions in both the hippocampus and cortex, supporting the role of ApoE in APP clearance. ³⁹

Additionally, we used the NIR‐E3 nanobody to detect SAβOs in the neocortex and hippocampus. NIR‐E3 labeling was predominantly observed in endothelial cells lining blood vessel lumens, particularly in the leptomeninges, while small, dot‐like, and sparsely distributed NIR‐E3⁺ SAβOs were detected throughout the parenchyma (Figure 7D). Similar to the retina, NIR‐E3⁺ labeling in the leptomeninges of 9‐month‐old APP‐PS1 mice was relatively weak compared to other groups.

Supplementary Table S9 summarizes the relative expression levels of 22 target genes screened in pooled neocortex‐hippocampus tissue using RT‐qPCR. Both mouse App and human PSEN1 mRNA levels were significantly upregulated in 3‐ and 9‐month‐old APP‐PS1 mice compared to age‐matched controls (Table S9). A heatmap (Figure 7J) illustrates the differential expression patterns between APP‐PS1 and control mice at both ages, as well as age‐related differences within each group. Notably, Trem2 (3.2‐fold, p < 0.0001) and Aqp4 (2‐fold, p < 0.0001) were significantly upregulated in 9‐month‐old APP‐PS1 mice compared to their 3‐month‐old counterparts (Figure 7J).

APP‐PS1 mice express chimeric mouse/human APP (Mo/HuAPP695swe) and mutant human PSEN1 (PS1‐dE9) genes, which drive Aβ overproduction. We confirmed the upregulation of these gene transcripts in both neocortex‐hippocampus and eye tissues at both time points compared to age‐matched control mice (Tables S6 and S9). The expression patterns of the remaining genes were comprehensively summarized in heatmaps for both the neocortex‐hippocampus and eye tissues (Figures 6K and 7J, respectively). Apart from the transgenes, we observed a significant upregulation of Trem2 mRNA in both brain and eye tissues of APP/PS1 mice as the disease progressed. Trem2 is a key regulator of microglial adaptation to neurodegeneration, mediating the balance between protective phagocytosis and pro‐inflammatory responses. ⁵³ Downregulation of Trem2 in microglia has been shown to markedly exacerbate AD‐related neuropathology in APP/PS1 mice, including increased Aβ deposition, gliosis, neuroinflammation, and neuronal and synaptic loss, which are accompanied by cognitive decline. ⁵⁴ Authors ⁵⁴ state that Trem2 expression is elevated in APP/PS1 mice as early as 3 months of age and continues to increase with aging, supporting its proposed neuroprotective role.

While differentially expressed mRNAs correlate significantly better with their protein products than non‐differentially expressed mRNAs, ⁵⁵ as we have shown between Aif1 mRNA levels and IBA1⁺ immunoreactivity in eye tissue of aging APP‐PS1 mice, discordance between mRNA and protein abundance can still occur, as demonstrated by the mismatch between GFAP mRNA levels and GFAP⁺ immunoreactivity in the same tissue. Such discrepancies are often attributed to post‐transcriptional and translational regulation (e.g., microRNAs), temporal decoupling, compensatory evolutionary mechanisms, and measurement limitations. ⁵⁶ , ⁵⁷

4. DISCUSSION

An imbalance between the production and clearance of pathological proteins is a key contributor to AD, particularly in sporadic form. In familial AD, mutations in APP, PSEN1, and PSEN2 drive disease progression by increasing Aβ production. ⁵⁸ Conversely, sporadic AD is strongly associated with risk genes like APOE and TREM2, which affect clearance and promote pathological protein accumulation. ⁵⁸ Wholemount neuroretinas offer a valuable model to study interactions among neurons, glial cells, and the vasculature, particularly at the iBRB–interactions that cannot be effectively demonstrated at the BBB in brain cross‐sections ex vivo.

In this study, we observed 12F4⁺Aβ42 drainage path‐like patterns in control neuroretinas that were largely disrupted in AD, suggesting the formation of 12F4⁺Aβ42 aggregates, primarily around retinal blood vessels. Over the past decade, retinal Aβ deposition in human AD and control neuroretinas has been reported using various antibodies targeting Aβ oligomers, fibrils, plaques, and non‐invasive imaging probes. ²³ , ³⁴ , ⁵⁹ , ⁶⁰ , ⁶¹ Recent in vivo mouse studies have revealed substantial peri‐ and para‐vascular spaces surrounding pial arteries and penetrating arterioles, ⁵⁰ , ⁵¹ , ⁶² , ⁶³ , ⁶⁴ which were largely unrecognized before the advent of high‐resolution imaging, as they are not visible in fixed tissues. To our knowledge, this is the first ex vivo study to demonstrate a defective 12F4⁺Aβ42 clearance system at the iBRB, marked by increased 124⁺Aβ42 clump‐like deposition around blood vessels in AD (Figure 1), along with reduced SAβO uptake (Figure 2), microglial elongation (Figure 3), macroglial degeneration (Figure 4), and AQP4 depletion (Figure 5). In contrast, control neuroretinas showed 12F4⁺Aβ42 drainage path‐like patterns at the ILM–NFL and along peri‐ and para‐vascular spaces, where glymphatic structures were preserved and microglia displayed rounded processes. These findings are supported by 2D and 3D imaging of wholemount neuroretinas (Figure 1; Videos S2–S10). Due to the limitations of ex vivo 3D imaging, both “peri” and “para‐vascular” terms were used to describe 12F4⁺Aβ42 peptide localization around blood vessels.

Compared to insoluble Aβ deposits, Aβ oligomers are considered the most toxic components in AD. ¹ , ⁶⁵ We demonstrated SAβO⁺ labeling within the retinal vasculature in controls, which was significantly diminished in AD cases (Figure 2; Videos S11–S14). The SAβO antibody was developed against human Aβ1‐42 residues. Extensive in vitro and in vivo characterization has confirmed that the E3 nanobody crosses the BBB and binds selectively to SAβOs and Aβ plaques in a preclinical AD mouse model. ⁹ The NIR‐ and FAM‐labeled E3 nanobody, approximately 10 times smaller than conventional antibodies, effectively penetrates the iBRB, suggesting SAβO uptake by peripheral macrophage‐like, Aβ‐binding myeloid lineage cells (Figure S2; Video S13).

A recent study of 359 plasma samples reported lower concentrations of Aβ oligomers in individuals with subjective cognitive decline and AD compared to healthy controls, linking impaired clearance to APOE4 status. ⁶⁶ Growing evidence supports a role for Aβ‐binding monocytes in Aβ clearance, ¹⁹ , ²⁰ with reduced Aβ uptake by blood monocytes observed during aging and more prominently in AD. ¹⁸ Huang et al. ¹⁹ developed a flow cytometry assay to identify CD14⁺CD16⁺ monocytes that bind to Aβ in the blood, revealing their macrophage‐like phenotype with high Aβ phagocytic potential. Using confocal microscopy, they localized Aβ peptides on the cell surface and within the cytoplasm. Notably, surface Aβ levels on monocytes were significantly lower in patients with MCI and AD, indicating impaired brain Aβ clearance and suggesting these cells as potential biomarkers for AD diagnosis and therapeutic monitoring.

In our study, employing the E3 nanobody combined with a modified nanobody staining protocol ³⁶ and confocal microscopy, we similarly demonstrated reduced uptake of SAβOs by macrophage‐like, Aβ‐binding myeloid lineage cells within the retinal vasculature of AD wholemount neuroretinas. The retina‐vitreous humor interface, where flow‐like nanoparticles are typically observed, appeared intact in controls but showed larger Aβ42 species or reduced flow‐like patterns in AD neuroretinas (Figures 2F and S3F,G; Videos S15 and S16). This aligns with our prior study showing enhanced APP/Aβ efflux at the wholemount surface in ergothioneine‐treated versus untreated 5xFAD mice. ⁴⁰

Microglia respond to brain tissue damage from aging and neurodegeneration by altering their morphology and transcriptome to clear debris and restore homeostasis. ⁵³ While microglial phenotypes have been extensively studied in rodent AD models and human AD brain tissue, research on retinal microglia in AD – particularly in humans – remains limited. ²³ , ⁶⁰ , ⁶⁷ , ⁶⁸ , ⁶⁹ , ⁷⁰ In our recent study using 3D imaging and machine learning, we found significantly fewer but larger microglia in human AD wholemounts. ⁶⁷ Building on these findings with a larger sample, we observed significantly higher IBA1 expression in AD mid‐peripheral regions, likely due to infiltrating IBA1⁺ cells (Figures 3 and S4; Videos S17 and S19). Despite some infiltrating or amoeboid‐shaped IBA1⁺ cells in the ILM and superficial layers and a few resting IBA1⁺ microglia in deeper layers (OPL), 2D and 3D analyses revealed predominantly rod‐shaped or elongated microglia with dystrophic features in AD, mainly from the NFL to OPL. Similar morphologies were reported in post mortem AD cases in the 1900s and have re‐emerged in recent brain injury models. ⁷¹ In contrast, IBA1⁺ microglia/macrophages in controls exhibited rounded processes, mainly located in the NFL to IPL. Limited CD68 or LAMP1 co‐localization with IBA1⁺ cells in small subgroup analyses underscores the need for larger cohorts or alternative markers to assess phagocytic activity at the iBRB.

Neuroimaging and fluid transport studies emphasize the glymphatic–lymphatic system's role in clearing Aβ and other proteins. ⁵⁰ , ⁵¹ , ⁷² As anti‐Aβ trials falter, targeting waste clearance provides a broad, transporter‐independent approach to mitigating protein aggregation. Evidence suggests that Aβ initiates – but does not directly cause – neuronal loss and cognitive decline in AD. Using a rodent model, Wang et al. ⁷² demonstrated that the ocular glymphatic pathway, reliant on the glial water channel AQP4, removes metabolites and Aβ tracers from the retina and vitreous humor via an ocular‐cranial pressure gradient through the optic nerve and meningeal lymphatics. The iBRB, structurally and functionally similar to the BBB, ⁷³ , ⁷⁴ , ⁷⁵ consists of non‐fenestrated endothelial cells, pericytes, and astroglial/Müller cell endfeet enriched with AQP4. Loss of perivascular AQP4 contributes to age‐related vulnerability to protein misaggregation, such as Aβ, in AD. ⁷⁶ While in vitro and transgenic models have advanced AD research, their relevance to humans remains uncertain, as demonstrated by structural differences in the ILM–GCL region between human and mouse retinas observed in this study. Notably, AD neuroretinas exhibited marked degeneration of macroglia (GFAP⁺ astrocytes and GS⁺ Müller cells) and AQP4⁺ water channels, in contrast to their preservation in controls, as seen in wholemounts and supported by human retinal cross‐sections. 

FIGURE 8.

Schematic illustration depicting a proposed model of multiple Aβ clearance pathways, potentially incorporating findings observed through ex vivo 3D imaging at iBRB. (A) Illustration of connection between eye and brain via optic nerve. (B, C) Flat‐mounted neuroretina, illustrating retinal vasculature at superficial plexus. (D) Vascular transport of soluble Aβ as bulk flow. (E) Ocular glymphatic drainage pathway for soluble Aβ bulk flow. (F) Uptake and degradation of SAβO by peripheral monocytes, and phagocytosis of insoluble Aβ deposits by resident microglia/macrophages. This figure was created in BioRender (https://BioRender.com/b64q914). Aβ, amyloid beta; iBRB, inner blood–retina barrier; SAβO, soluble Aβ oligomer; LRP1, low‐density lipoprotein receptor‐related protein 1; RAGE, receptor for advanced glycation end products.

In our preclinical 2D analyses, intraneuronal inclusion of 6E10⁺ APP/Aβ peptides, along with elevated GFAP, AQP4, and IBA1 expression, was observed in the retinas (all regions) of aging APP‐PS1 mice – consistent with findings from human AD retinal cross‐sections. ²³ , ⁶⁰ Koronyo et al. ⁶⁰ reported increased IBA1⁺ microgliosis and GFAP⁺ or S100β⁺ macrogliosis – markers of astrocytes and Müller glia – surrounding retinal Aβ deposits in MCI and AD patients. These glial responses correlated with Aβ burden, implicating Aβ‐induced glial activation. ²³ , ⁶⁰ In contrast, non‐carrier sibling controls showed reduced 6E10⁺ APP/Aβ and increased IBA1⁺ microglia, consistent with our previous findings in age‐matched C57BL/6J mice. ³⁹ APP‐specific antibody validation confirmed lower APP in cortical neurons of non‐carrier sibling controls, unlike APP⁺ dystrophic neurites in aging APP‐PS1 mice, suggesting failures in APP/Aβ peptide clearance.

We acknowledge the limitations of our study. As it relied on ex vivo imaging, we could not assess Aβ clearance or drainage through functional assays with Aβ tracers or evaluate the effects of AQP4 inhibition – approaches typically used in in vivo imaging, mainly in animal models. We also lacked vessel markers to distinguish veins from arteries in adult wholemounts using fluorescence microscopy. Additionally, due to the limited sample size, we did not analyze the findings concerning comorbidities in AD donors. Age‐matched control donor eyes were obtained through an eye bank, with no documented history of AD or related dementias. The small sample size also prevented us from addressing diversity, equity, and inclusion (DEI) in the study's design, execution, and interpretation.

For the animal studies, limitations included a small sample size, a focus on female mice, and analysis at only two time points (3 and 9 months). The sample size was determined using the resource equation method. Beyond age and genetic predisposition, female sex is a non‐modifiable risk factor for AD; thus, we focused on female mice. We analyzed 3‐ and 9‐month time points to assess early disease changes; however, this was not a true longitudinal study, as different mouse cohorts were used.

Overall, our schematic illustration (Figure 8) succinctly depicts a proposed model of multiple Aβ clearance pathways at the iBRB. The imaging view plane (surface vs cross‐section) may significantly influence findings, as demonstrated by differences observed in human wholemounts and mouse retinal cross‐sections in the context of AD. Wholemounts revealed layer‐specific labeling and notable variability, especially within the AD group. Nevertheless, our comprehensive wholemount investigations identified compensatory interactions between the glymphatic system and microglia/macrophage phagocytosis in mitigating Aβ deposition, as well as the involvement of peripheral macrophage‐like Aβ‐binding myeloid lineage cells in SAβO uptake in controls. These processes appeared largely disrupted in AD donors. These findings were revealed through ex vivo 3D retinal imaging, employed for the first time to study this process, particularly at the iBRB, a functional analog of the BBB.

AUTHOR CONTRIBUTIONS

Conceptualization: Printha Wijesinghe and Joanne A. Matsubara. Project administration: Printha Wijesinghe and Joanne A. Matsubara. Funding acquisition: Printha Wijesinghe, Wellington Pham, and Joanne A. Matsubara. Data curation: Printha Wijesinghe, Jeanne Xi, Ian R. Mackenzie, Veronica Hirsch‐Reinshagen, and Ging‐Yuek Robin Hsiung Formal analysis: Printha Wijesinghe. Investigation: Printha Wijesinghe, Amir Hosseini, Matthew Campbell, Shivani Tejpal, Justin Haynes, Jeanne Xi, Ian R. Mackenzie, Veronica Hirsch‐Reinshagen, Ging‐Yuek Robin Hsiung, Benjamin W. Spiller, and Brian E. Wadzinski. Methodology: Printha Wijesinghe, Amir Hosseini, Matthew Campbell, Shivani Tejpal, Justin Haynes, Jeanne Xi, Ian R. Mackenzie, Veronica Hirsch‐Reinshagen, Ging‐Yuek Robin Hsiung, Benjamin W. Spiller, and Brian E. Wadzinski. Validation: Ian R. Mackenzie, Veronica Hirsch‐Reinshagen, Ging‐Yuek Robin Hsiung, Benjamin W. Spiller, Brian E. Wadzinski, Wellington Pham, and Joanne A. Matsubara. Visualization: Printha Wijesinghe, Amir Hosseini, and Matthew Campbell. Writing – original draft and major revision: Printha Wijesinghe. Writing – review and editing: Amir Hosseini, Matthew Campbell, Shivani Tejpal, Justin Haynes, Jeanne Xi, Ian R. Mackenzie, Veronica Hirsch‐Reinshagen, Ging‐Yuek Robin Hsiung, Benjamin W. Spiller, Brian E. Wadzinski, Wellington Pham, and Joanne A. Matsubara. Resources: Wellington Pham and Joanne A. Matsubara. Supervision: Printha Wijesinghe and Joanne A. Matsubara.

CONFLICT OF INTEREST STATEMENT

The authors declare that the research was conducted in the absence of any commercial or financial relationship that could be construed as a potential conflict of interest. BWS and BEW are co‐founders and owners of Turkey Creek Biotechnology (TCB). TCB was not involved in this work.

CONSENT STATEMENT

We confirm that informed consent was obtained from all human subjects.

Supporting information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Wijesinghe P, Hosseini A, Campbell M, et al. Decoding amyloid beta clearance systems at inner blood–retina barrier using three‐dimensional ex vivo retinal imaging in Alzheimer's disease. Alzheimer's Dement. 2025;21:e70592. 10.1002/alz.70592

DATA AVAILABILITY STATEMENT

The raw data supporting the conclusions of this manuscript will be made available by the authors upon request.