Volume 13, Issue 1 e12181
RESEARCH ARTICLE
Open Access

Retinal imaging demonstrates reduced capillary density in clinically unimpaired APOE ε4 gene carriers

Fanny M. Elahi

Corresponding Author

Fanny M. Elahi

Department of Neurology, Memory and Aging Center, Weill Institute for Neurosciences, San Francisco, University of California, San Francisco, California, USA

San Francisco Veterans Affairs Health Care System, San Francisco, California, USA

Correspondence

Fanny M. Elahi, Memory and Aging Center, Weill Institute for Neurosciences, 675 Nelson Rising Lane, Suite 190, San Francisco, CA 94158, USA.

E-mail: [email protected], [email protected]

Amir H. Kashani, Wilmer Eye Institute, Johns Hopkins School of Medicine, 600 N Wolfe St., Baltimore, MD 21287, USA.

E-mail: [email protected]

Search for more papers by this author
Senyo B. Ashimatey

Senyo B. Ashimatey

Department of Ophthalmology, USC Roski Eye Institute, Keck School of Medicine of the University of Southern California, Los Angeles, California, USA

Search for more papers by this author
Daniel J. Bennett

Daniel J. Bennett

Department of Neurology, Division of Neuroimmunology and Glial Biology, Weill Institute for Neurosciences, San Francisco, University of California, San Francisco, California, USA

Search for more papers by this author
Samantha M. Walters

Samantha M. Walters

Department of Neurology, Memory and Aging Center, Weill Institute for Neurosciences, San Francisco, University of California, San Francisco, California, USA

Search for more papers by this author
Renaud La Joie

Renaud La Joie

Department of Neurology, Memory and Aging Center, Weill Institute for Neurosciences, San Francisco, University of California, San Francisco, California, USA

Search for more papers by this author
Xuejuan Jiang

Xuejuan Jiang

Department of Ophthalmology, USC Roski Eye Institute, Keck School of Medicine of the University of Southern California, Los Angeles, California, USA

Search for more papers by this author
Amy Wolf

Amy Wolf

Department of Neurology, Memory and Aging Center, Weill Institute for Neurosciences, San Francisco, University of California, San Francisco, California, USA

Search for more papers by this author
Yann Cobigo

Yann Cobigo

Department of Neurology, Memory and Aging Center, Weill Institute for Neurosciences, San Francisco, University of California, San Francisco, California, USA

Search for more papers by this author
Adam M. Staffaroni

Adam M. Staffaroni

Department of Neurology, Memory and Aging Center, Weill Institute for Neurosciences, San Francisco, University of California, San Francisco, California, USA

Search for more papers by this author
Howie J. Rosen

Howie J. Rosen

Department of Neurology, Memory and Aging Center, Weill Institute for Neurosciences, San Francisco, University of California, San Francisco, California, USA

Search for more papers by this author
Bruce L. Miller

Bruce L. Miller

Department of Neurology, Memory and Aging Center, Weill Institute for Neurosciences, San Francisco, University of California, San Francisco, California, USA

Search for more papers by this author
Gil D. Rabinovici

Gil D. Rabinovici

Department of Neurology, Memory and Aging Center, Weill Institute for Neurosciences, San Francisco, University of California, San Francisco, California, USA

Department of Radiology and Biomedical Imaging, San Francisco, University of California, San Francisco, California, USA

Search for more papers by this author
Joel H. Kramer

Joel H. Kramer

Department of Neurology, Memory and Aging Center, Weill Institute for Neurosciences, San Francisco, University of California, San Francisco, California, USA

Search for more papers by this author
Ari J. Green

Ari J. Green

Department of Neurology, Division of Neuroimmunology and Glial Biology, Weill Institute for Neurosciences, San Francisco, University of California, San Francisco, California, USA

Department of Ophthalmology, San Francisco, University of California, San Francisco, California, USA

Search for more papers by this author
Amir H. Kashani

Corresponding Author

Amir H. Kashani

Department of Ophthalmology, USC Roski Eye Institute, Keck School of Medicine of the University of Southern California, Los Angeles, California, USA

USC Ginsberg Institute for Biomedical Therapeutics, Los Angeles, California, USA

Correspondence

Fanny M. Elahi, Memory and Aging Center, Weill Institute for Neurosciences, 675 Nelson Rising Lane, Suite 190, San Francisco, CA 94158, USA.

E-mail: [email protected], [email protected]

Amir H. Kashani, Wilmer Eye Institute, Johns Hopkins School of Medicine, 600 N Wolfe St., Baltimore, MD 21287, USA.

E-mail: [email protected]

Search for more papers by this author
First published: 11 May 2021
Citations: 5

Both F.M.E. and A.H.K. contributed equally to this work as senior authors.

Abstract

Introduction

Apolipoprotein E (APOE) ε4, the strongest non-Mendelian genetic risk factor for Alzheimer's disease (AD), has been shown to affect brain capillaries in mice, with potential implications for AD-related neurodegenerative disease. However, human brain capillaries cannot be directly visualized in vivo. We therefore used retinal imaging to test APOE ε4 effects on human central nervous system capillaries.

Methods

We collected retinal optical coherence tomography angiography, cognitive testing, and brain imaging in research participants and built statistical models to test genotype–phenotype associations.

Results

Our analyses demonstrate lower retinal capillary densities in early disease, in cognitively normal APOE ε4 gene carriers. Furthermore, through regression modeling with a measure of brain perfusion (arterial spin labeling), we provide support for the relevance of these findings to cerebral vasculature.

Discussion

These results suggest that APOE ε4 affects capillary health in humans and that retinal capillary measures could serve as surrogates for brain capillaries, providing an opportunity to study microangiopathic contributions to neurodegenerative disorders directly in humans.

1 INTRODUCTION

It has become increasingly clear that Alzheimer's disease (AD) is not strictly a disorder of neuronal dysfunction but that changes in vasculature could play an important role in disease pathogenesis.1-6 The most prevalent genetic risk for AD remains the ε4 allelic variant of apolipoprotein E (APOE), a gene systemically involved in lipid metabolism and vascular health, including cerebral vasculature.7-10

In autopsy studies, vascular changes are among the most common co-pathologies in subjects harboring APOE ε4 and AD neuropathological changes.11-13 Genetic studies, including pleiotropy analyses, suggest shared genetic risk among vascular disease, lipid metabolism, and AD.4 APOE ε4 is a prominent genetic example of this shared vulnerability.5 The exact mechanisms by which APOE ε4 contributes to AD pathogenesis remains an active area of research. Studies in model systems14, 15 and human epidemiological investigations16 suggest that APOE ε4-mediated neurodegeneration is related to vascular pathology.5 It remains unclear, however, which vessels are most prominently affected by APOE ε4 in humans and how the pathological changes contribute to AD neuropathology. The largest community-based neuropathological study of APOE ε4 reported a higher prevalence of large-vessel atherosclerosis in APOE ε4 gene carriers.17 However, due to technical limitations, most historical human autopsy studies have been limited to investigating larger caliber vessels. Recent studies in mouse models harboring human APOE ε4 alleles demonstrate capillary pathology, along with blood-brain barrier (BBB) leakage and dysfunction of the neurovascular unit (NVU).14, 15, 18, 19

APOE ε4 may be affecting capillaries via endothelial or pericytic changes, but a more likely possibility is multicellular dysfunction of the BBB and NVU, and dysregulated clearance of toxic metabolic byproducts of brain function. On average, APOE ε4 carriers, especially female carriers, have a higher cerebral amyloid burden. Sex-dependent differences in amyloidopathy and other neuropathological components of AD are an active area of research. APOE ε4-associated amyloidopathy, in specific, may in part be related to deficiencies in cerebral clearance mechanisms.20 Recently, oligomeric amyloid has been shown to cause capillary disease (constriction) via an effect on vascular pericytes,21 master regulators of capillary function. Importantly, more than a decade prior to these studies, Salloway et al. reported a dose-dependent APOE ε4 effect on the microvascular basement membrane of the frontal cortex in 35 AD brains.22 These pathological changes could be reactive or causative. However, the combination of these studies with mechanistic interventions in model systems,14, 15 and human epidemiological studies,16 provides strong support for vascular contributions to APOE ε4-associated neurodegenerative disease, and likely one that affects microvasculature.

RESEARCH IN CONTEXT

  1. Systematic review: The authors reviewed the available scientific literature on PubMed for articles examining retinal biomarkers in cognitively normal apolipoprotein E (APOE) ε4. While abundant studies examine retinal “biomarkers” for clinically impaired individuals with Alzheimer's disease (AD), we did not find any studies investigating APOE ε4-associated retinal vascular changes in preclinical disease stage. In this study, we investigated the microangiopathy associated with APOE ε4 carrier status and demonstrate the relevance of retinal findings by showing associations with brain perfusion. The association of APOE ε4 with capillary rarefaction was not dependent on amyloid burden.

  2. Interpretation: We show that APOE ε4 is associated with age-dependent capillary rarefaction. Given the importance of APOE ε4 for AD, this work strongly suggests that small vessel disease may occur early, in the preclinical stage of disease, and play an important role in the risk of AD at least in APOE ε4 carriers.

  3. Future directions: Longitudinal studies of retinal and brain imaging in APOE ε4 cohorts, including individuals who convert from cognitively normal to cognitively impaired, will help decipher causality of vascular abnormalities. Studies in model systems will also help elucidate mechanisms.

Despite strong evidence that APOE ε4 is associated with microangiopathy,14, 15 and that this pathology may be an important contributor to AD pathogenesis23-25 (reviewed in Sweeney et al.26), it is currently not technically feasible to individually visualize brain microvessels in living humans. The retina is an extension of central nervous system (CNS) tissue, constituted by neurons and glia, and the retinal microvasculature can be visualized in vivo at a high resolution with optical coherence tomography angiography (OCTA).27, 28 OCTA is a noninvasive, rapid, and US Food and Drug Administration (FDA) approved imaging modality that can provide high resolution images of retinal capillaries by detecting the variable scatter of light from moving red blood cells compared to non-moving neuroretinal tissue.29-31 Abnormal retinal capillary function has been reported in subjects with early or subclinical retinal vascular diseases, including diabetes32 and hypertension.33 In patients with advanced AD, OCTA has demonstrated significantly lower retinal capillary density.34 In the current study, we leveraged the established sensitivity of OCTA for subclinical capillary pathology to investigate APOE ε4-associated capillary disease in asymptomatic gene carriers at risk for AD. Subsequently, we investigated the specific association of retinal capillary measures with arterial spin labeling (ASL) magnetic resonance imaging (MRI), a measurement of cerebral blood flow (CBF), as a means of testing the relevance of retinal vessel measures to global brain perfusion.

2 METHODS

2.1 Study participants and design

Consecutive enrollment of research participants from January 11, 2018 to April 8, 2019 was undertaken from ongoing longitudinal studies of brain aging and neurodegenerative disease at the Memory and Aging Center at University of California San Francisco (UCSF; National Institute of Health [NIH]-funded studies: Aging and Cognition, Chronic Inflammation, MarkVCID studies, and Alzheimer's Disease Research Center; and Larry L. Hillblom Foundation Aging Network study) into our prospective study (N = 272) in the UCSF Visual Neurodiagnostics Laboratory. All study participants had a measure of functional impairment (informant Clinical Dementia Rating [CDR]), extensive neuropsychological testing, physical examination, and demographic information, while the majority had vascular disease risk factor measures (blood pressure; body mass index [BMI]; history of hyperlipidemia, hypertension, and diabetes) and brain imaging (Figure 1). Exclusion criteria of parent studies also applied to this study, including active or uncontrolled psychiatric disease such as psychosis, brain tumor, or history of brain surgery; in addition, we required absence of any clinically significant ophthalmological abnormalities, such as glaucoma, visually significant cataract, macular degeneration, or cystoid macular edema. Given our interest in associations of APOE ε4 with microvascular changes prior to neurodegenerative symptom presentation, our primary analyses were limited to the subgroup of cognitively normal participants (n = 75) defined as those individuals who had cognitive test scores within normal range for age and level of educational attainment and no study partner concern (CDR = 0). All study participants provided informed consent and the study protocols were approved by the UCSF Human Research Protection Program and Institutional Review Board. Research was performed in accordance with the Code of Ethics of the World Health Organization. Compared to participants without genomic DNA (n = 66), participants with genomic DNA for APOE genotype (N = 93) were more likely to be 75 years and older (P = .01), but similar in sex, race, years of education, and history of hypertension and diabetes (P > .05 for all).

Details are in the caption following the image
Study design. Study design flowchart, including subject recruitment (blue box) and censorship (gray box). Green boxes represent data used for primary analyses of apolipoprotein E ε4 genetic associations with retinal capillary morphometric measures; yellow boxes represent available data used in secondary analyses such as structural brain imaging, perfusion imaging, and molecular neuroimaging

2.2 Optical coherence tomography angiography and retinal capillary density quantification

OCTA images were acquired with a commercially available and FDA-approved Spectral Domain (SD) OCTA system—Cirrus HD-OCT (Carl Zeiss Meditec, Inc.). This system uses a light source with wavelength 840 nm and acquires 68,000 A-scans per second. The highest resolution 3 × 3 mm2 scan, which has 245 B-scans in the horizontal plane, were used for all analyses. The axial and transverse resolution of the scan pattern are 5 and 15 μm, respectively. Each B-scan was consecutively repeated at least three times at acquisition and the scan with the best image quality from either eye of a subject was selected for analysis. OCTA scans were performed at UCSF and reviewed for quality by an experienced OCTA technician at USC who was masked to APOE genotype and to all clinical measures, and had no knowledge of the results of brain imaging and neurocognitive assessments performed at UCSF. Only retinal scans with signal strength greater than 7 and otherwise good quality, including centeredness of the scan on the fovea and minimal to no motion artifacts, were used for analyses (n = 159; Figure 1). A previously validated semiautomated software program written in MATLAB (R2018b; MathWorks, Inc.) was used to analyze OCTA images.35-38 To avoid the potential confounding effect of projection artifacts, only the superficial retinal layer of the image was segmented and used for our primary analysis. The superficial retinal layer includes the retinal nerve fiber layer, ganglion cell layer, and inner plexiform layer. We and others have previously described several measures of retinal capillary density and morphology.35, 36 In this study, we report the most simple and direct measures of retinal capillary density, termed vessel area density (VAD) and vessel skeleton density (VSD). VAD is derived from a binarized raw OCTA image of the superficial retinal layer and represents the total area of the image that is occupied by blood flow signal.36 VSD is computed from a binarized OCTA image in which the vessels are reduced to a single pixel thickness and then divided by the total area of the image. VSD provides a measure of the length of vessels present within an image.36 Both methods of assessing capillary density are complementary. Examples of OCTA image and capillary density metrics are shown in Figure 2. Additional details on the OCTA metrics can be found elsewhere.27, 32, 35, 36, 38

Details are in the caption following the image
Single subject comparison of optical coherence tomography angiography (OCTA) scans in a carrier versus noncarrier. Retinal capillary density maps derived from OCTA images for an apolipoprotein E (APOE) ε4 carrier and non-carrier. The first column shows raw OCTA images where white pixels represent retinal vessels within a 3 × 3 mm2 region surrounding the naturally occurring foveal avascular zone of the retina. The second and third columns represent pseudo-colored capillary density heat maps of the same images using either the vessel skeleton density or vessel area density metrics. Hot colors depict regions of higher capillary density. Rows represent a single subject comparison of capillary density between a heterozygote for APOE ε4 (bottom row) versus a noncarrier (top row), matched by demographic and vascular risk factors. Both are females, aged 72 years, with body mass index of 21 kg/m 2. The APOE ε4 noncarrier had numerous focal regions of greater retinal capillary density (red) throughout the retina as well as a smaller foveal avascular zone compared to the APOE ε4 carrier

2.3 Brain imaging acquisition and processing

MRI scans were acquired on a 3T MRI scanner Prisma Fit (Siemens AG). Volumetric MPRAGE sequences at UCSF were used to acquire T1-weighted images of the entire brain (sagittal slice orientation; slice thickness = 1.0 mm; slices per slab = 160; in-plane resolution = 1.0 × 1.0 mm; matrix = 240 × 256; TR = 2,300 ms; TE = 2.9 ms; TI = 900 ms; flip angle = 9°).39 Pseudo-continuous ASL (pcASL) images were acquired using the recommendations described by Alsop et al.40 pcASL-MRI was performed with TR/TE = 4050/34 ms, labeling duration of 1500 ms, and postlabeling delay of 1800 ms. Each slice is 3 mm thick with a 2.5 × 2.5 mm2 in-plane resolution. Quality control included visual inspection of all images. Images with excessive motion or image artifact were excluded. T1-weighted images underwent bias field correction using an N3 algorithm, and the segmentation was performed using SPM12 (Wellcome Trust Center for Neuroimaging, London, UK, http://www.fil.ion.ucl.ac.uk/spm) unified segmentation.41 A group template was generated from the segmented gray and white matter tissues and cerebrospinal fluid by non-linear registration template generation using large deformation diffeomorphic metric mapping framework.42 Native subjects space gray and white matter were normalized, modulated, and smoothed in the group template. The applied smoothing used a Gaussian kernel with 8 mm full width half maximum. Every step of the transformation was carefully inspected from the native space to the group template. For statistical purposes, linear and non-linear transformations between the group template space and International Consortium for Brain Mapping (ICBM) was applied.43 Quantification of volumes in specific brain regions at each time point was accomplished by transforming a standard parcellation atlas44 into ICBM space and summing all modulated gray matter within each parcellated region. Total intracranial volume (TIV) was estimated for each subject in Montreal Neurological Institute (MNI) space.45 Gray matter volume (GMV) and TIV are reported in cm3. ASL was processed to obtain partial volume corrected maps of gray matter perfusion as previously described.46-48 Frames of the ASL acquisition were corrected for motion, coregistered with the first frame (M0) using FSL (FMRIB Software Library),49 and differential perfusion images were created by subtracting unlabeled frames from adjacent labeled frames and averaging these subtraction images.50 Susceptibility artifacts along the phase-encoding direction were corrected in the M0 frame and perfusion map using Advanced Normalization Tools image registration (ANTs SyN)51 restricted to the coronal axis. An automatic quality control process removed tagged/untagged pairs of frames when the relative root mean square (RMS) distance value between two consecutive frames was higher than 0.5 mm. The participant was dropped if this RMS value was higher than 1 mm. CBF was calculated by applying the Buxton kinetic model to the perfusion map.52, 53 Partial volume correction was based on the tissue segmentation maps from MPRAGE using the transformation matrix from T1 to M0.46, 54 All CBF images were visually inspected in the native and MNI spaces. Poor quality images that were out of the field of view or contained large susceptibility or motion artifacts were removed from the study.

2.4 Molecular neuroimaging (amyloid positron emission tomography)

Amyloid-positron emission tomography (PET) results on 43 subjects were used to assess amyloid-associated differences in retinal microvasculature. PET was acquired with either 11C-Pittsburgh compound B (PiB, acquired on a Siemens Biograph PET/CT scanner, n = 9, or a ECAT EXACT HR PET scanner, n = 1) or 18F-florbetapir (on a GE Discovery STE/VCT PET-CT, n = 33). PiB pre-processing and acquisition followed previously described methods.51 Briefly, we used PiB-PET data acquired between 50 and 70 minutes post injection as four 5-minute frames; frames were realigned, averaged, coregistered to the closest available MRI, and scaled using the cerebellar cortex as a reference region (based on FreeSurfer segmentation) to obtain standardized uptake value ratio (SUVR) images. Florbetapir was acquired between 50 and 70 minutes post injection as four 5-minute frames and processed following Alzheimer's Disease Neuroimaging Initiative procedures (http://adni.loni.usc.edu/methods/pet-analysis-method/pet-analysis/). Frames were smoothed to a final 8 mm isotropic resolution, realigned, averaged, coregistered to the closest available MRI, and scaled using the whole cerebellum (gray and white matter) as a reference region to obtain SUVR images. For both tracers, FreeSurfer segmentation of the cortical mantle was used to define a cortical composite region encompassing frontal, cingulate, parietal, and lateral temporal areas from which the average SUVR value was extracted as a measure of neocortical amyloid beta burden. To determine amyloid status, tracer- and preprocessing-SUVR thresholds derived from previous publications were applied (1.21 for PiB,55 and 1.11 for florbetapir56). These thresholds have been validated compared to a post mortem gold standard.55-58

2.5 Statistical analyses

The primary analysis investigated the influence of APOE ε4 genotype on retinal capillary density using analysis of covariance (ANCOVA) controlling for the effects of age, sex, and quantified vascular risk factors (systolic blood pressure [SBP] and BMI). To better understand the relationship between age and capillary density in our model, we investigated age as a predictor of capillary density in APOE ε4 carriers and non-carriers independently, using separate linear models without the addition of covariates. Finally, we used a multivariate hierarchical linear regression approach to understand the predictive value added by basic demographics (model 1); basic demographics and vascular risk factors (model 2); and basic demographics, vascular risk factors, and APOE gene status (model 3). Only the eye with highest OCTA quality from each participant was selected for analysis. Analyses were repeated in the other eye to ensure that a selection bias was not introduced. Statistical significance was defined at an alpha level of P = .05.

In secondary analyses, we aimed to demonstrate a correlation between OCTA measures of retinal vascular density, and MRI measures of global CBF, as a proof of concept that the two measurements are related. All subjects with retinal and brain imaging, regardless of phenotype and degree of impairment, were included in the analysis. Multivariate linear regression models were used to assess the relation between retinal capillary density and CBF, total gray matter volumes (GMV), white matter volumes (WMV), and mean hippocampal volumes (HIP, averaged between the left and right hippocampus), controlling for TIV, and in the case of CBF, also controlling for sex. Due to the similarities of aging effect, an important variable that influences the dynamics of retinal and CBF and neurodegeneration, age was not compensated in this analysis. For similar reasons, vascular risk factors were also not compensated.

To ensure that missing data did not alter results and our inferences, we re-ran statistical models using imputed values for Mini-Mental Status Examination (MMSE), heart rate, systolic and diastolic blood pressure, and BMI. We imputed missing values using the least squares prediction method.

3 RESULTS

3.1 Demographics

Participant characteristics are summarized in Table 1. No significant differences were found in age or vascular risk factors between APOE ε4 carriers and noncarriers (P = .27). In addition, there were no significant differences with respect to sex and educational attainment between groups. Measures of cognitive function, and functional independence (CDR) were similar as well (Table 1). Thirty-two percent (24/75) of the cognitively normal study cohort were APOE ε4 carriers (2 Ɛ4/Ɛ4, 2 Ɛ2 Ɛ4, and 20 Ɛ3/Ɛ4).

TABLE 1. Summary of demographic and basic clinical data by APOE status in the primary analysis cohort
APOE ε4 negative APOE ε4 positive P-value
n 51 24 n/a
Age - median [range] 75 [37–90] 72 [26–83] .474
Female - n (%) 33 (65) 12 (50) .337
Education years - mean (SD) 17.4 (2.2) 17.0 (2.1) .432
missing - n (%) 3 (6) 3 (12)
Body mass index - mean (SD) 25.5 (3.8) 23.8 (2.8) .067
missing - n (%) 8 (16) 3 (12)
Systolic blood pressure - mean (SD) 131.1 (15.4) 136.9 (15.6) .164
missing - n (%) 9 (18) 3 (12)
Diastolic blood pressure - mean (SD) 73.7 (9.1) 73.3 (6.9) .858
missing - n (%) 9 (18) 3 (12)
Heart rate - mean (SD) 69.4 (9.8) 69.5 (13.4) .987
missing - n (%) 9 (18) 3 (12)
Clinical Dementia Rating - mean (SD) 0.0 (0.0) 0.0 (0.0) n/a
Mini-Mental State Examination score - mean [range] 29 [26–30] 30 [25–30] .521
missing - n (%) 10 (20) 3 (12)
Structural brain imaging analysis - included/not included (%) 43/8 (84/16) 17/7 (71/29) .293
Cerebral blood flow analysis - included/not included (%) 18/33 (35/65) 12/12 (50/50) .337
Molecular neuroimaging analysis - included/not included (%) 18/33 (35/65) 7/17 (29/71) .793
  • Notes: Summary statistics of the demographic, medical information, and secondary analysis group inclusion for the subjects with known APOE ε4 gene status. Normally distributed variables are characterized by mean and standard deviation (mean [SD)]), while skewed variables are characterized by median and range (median [range]). P-values represent t test for normally distributed variables and Mann-Whitney U tests for skewed variables. Within the APOE ε4 positive group, two participants were homozygote for the APOE ε4 allele.
  • Abbreviations: APOE, apolipoprotein E; SD, standard deviation.

3.2 APOE ε4 is associated with decreased capillary density in cognitively normal older adults

Controlling for comorbidities and vascular risk factors, group comparisons of APOE ε4 carriers versus noncarriers revealed significantly lower capillary density in carriers with large effect sizes for both measures of density: VAD (F[1,62] = 16.99, P = .0001; partial eta squared = 0.22; Figure 3.A.) and VSD (F[1,62] = 11.52, P = .001; partial eta squared = 0.16; Figure 3.B.). Reduced capillary density between APOE ε4 carriers and noncarriers could be directly visualized comparing participants matched by age, sex, and BMI (Figure 2). Importantly, we found an age-dependent effect on retinal capillary density in APOE ε4 carriers (Figure 4).

Details are in the caption following the image
Retinal capillary density by apolipoprotein E (APOE) ε4 status. A comparison of APOE ε4 carriers (red) and non-carriers (blue) on residual measures of retinal capillary density (vessel area density [A] and vessel skeletal density [B]) after controlling for age, sex, systolic blood pressure, and body mass index
Details are in the caption following the image
Association of age with retinal capillary densities. Association of two measures of retinal capillary density (vessel area density [A] and vessel skeleton density [B]) in apolipoprotein E (APOE) ε4 carriers (red) and noncarriers (blue)

Next, hierarchical regression models were built to demonstrate the effect of APOE ε4 on retinal capillaries beyond other vascular risk factors (Figure 5). Model 1 included demographic factors of age and sex as predictors of capillary density. In model 2, vascular risk factors were added to the demographic factors included in model 1, and in model 3, APOE ε4 status was added to the variables in model 2. Results support the importance of the APOE ε4 effect on vasculature, with a trivial increase in model fit with the addition of vascular risk factor to age (comparison of models 1 and 2) and a substantive increase with the inclusion of APOE ε4 in model 3. Specifically, for VAD, R2 of .10 in model 2 increased to .29 in model 3, and for VSD, R2 of 0.10 in model 2 increased to 0.24 in model 3. To determine the unique contribution of APOE ε4 to capillary density, the coefficient of determination of model 2 was deducted from that of model 3 to demonstrate that an APOE ε4 allele uniquely explains up 18% (P = .0001) and 14% (P = .001) of the variability in the VAD and VSD, respectively (Figure 5).

Details are in the caption following the image
Hierarchical regression models to assess effect of apolipoprotein E (APOE) ε4 on retinal capillary densities. Step-wise assessment of the added value of APOE ε4 status in predicting two measures of retinal capillary density (vessel area density [A] and vessel skeleton density [B]) in cognitively normal individuals. A hierarchical regression model was built using (i) age and sex; (ii) age, sex, and quantifiable vascular risk factors (body mass index and systolic blood pressure); and (iii) age, sex, vascular risk factors, and APOE ε4 carrier status. APOE ε4 versus non-APOE ε4 genotype uniquely explained ≈18% of the variance in vessel area density and ≈14% of the variance in Vessel Skeletal Density. VAD , vessel area density; VSD , vessel skeleton density; R= model fit

As noted earlier, primary analyses were limited to cognitively unimpaired individuals and only retinal scans passing stringent quality-based inclusion criteria were included in the analyses. Therefore, to rule out bias by censorship, analyses were cross-validated by including data from all subjects irrespective of the CDR score, and re-running analyses using the unselected eye in the primary cohort, including eyes that were not originally used for analysis due to quality-based exclusion. Similar findings were noted when participants with functional impairment (CDR score > 0) were included (Figure S1 in supporting information), as well as in analyses using the alternate eye (Figure S2 in supporting information).

Of note, to ensure that our inferences were not impacted by missing data, we confirmed all models in a complete, imputed dataset.

3.3 Associations of retinal capillary density with neuroimaging measures

We found an association between retinal capillary density and CBF (VAD [β = 0.36, P = .01], VSD [β = 0.35, P = .01]; Figure 6). The correlation of retinal capillary density with gray matter volume (VAD [β = 0.12, P = .1]; VSD [β = 0.12, P = .1]), white matter volume (VAD [β = 0.00, P = .9]; VSD [β = 0.03, P = .7]) and hippocampal volume (VAD [β = 0.12, P = .2]; VSD [β = 0.12, P = .2]) were smaller in magnitude and not significant (Figure 6). Controlling for the effects of vascular risk factors in addition to sex yielded similar correlation magnitudes for the CBF analysis, with reduced P-values VAD (r = 0.35, P = .03) and VSD (r = 0.32,P = .05). As expected, however, when age was included as a covariate, the magnitude of the retinal vessel density correlation with CBF was small and non-significant (VAD [r = 0.2, P = .16] and VSD [r = 0.15,P = .14]). For retina-brain imaging associations we included our entire cohort, regardless of cognitive status (Table S1 in supporting information).

Details are in the caption following the image
Association of retinal capillary densities with brain imaging measures. Association of two measures of retinal capillary density (vessel area density and vessel skeleton density) with cerebral blood flow, total gray matter volume, white matter volume, and hippocampal volume. All volumetric brain imaging measures were adjusted for total intracranial volume and cerebral blood flow was adjusted for both total intracranial volume and sex. Only cerebral blood flow was significantly associated with measures of retinal capillary density. These analyses include all subjects, regardless of cognitive status and apolipoprotein E (APOE) ε4 status (blue = noncarrier, red = APOE ε4 carrier, and black = unknown gene status)

3.4 Retinal capillary densities and cerebral amyloid

To investigate the independence of APOE ε4 vascular effects from its known association with cerebral amyloidopathy, a gold-standard biomarker of AD pathology, we analyzed differences in retinal capillary densities in subjects with positive and negative amyloid PET scans in addition to analyses of continuous measures of cerebral amyloid (SUVR) with retinal capillary densities. We did not find significant differences in retinal capillary measures between amyloid positive (n = 16) and negative groups (n = 27; U = 186, P = .45 for VAD; U = 0.183, P = .41 for VSD). Adjusting for potential demographic confounders (age and sex) using a Kruskal-Wallis test did not change the findings (χ2[1] = 0.57 P = .45 for VAD; χ2[1] = 0.68 P = .41 for VSD; Figure 7A). Out of the total of 43 subjects, 32 had known APOE genotype (Table S1). Furthermore, regression of retinal VAD and VSD measures with a continuous measure of cerebral amyloid burden (18F-florbetapir SUVR, n = 33), controlling for age and sex, demonstrated no significant associations (VAD [r = 0.13, P = .49] and VSD [r = 0.12,P = .42]; Figure 7B).

Details are in the caption following the image
A, Retinal capillary density by cerebral amyloid group. A comparison of two measures of retinal capillary density (vessel area density and vessel skeleton density) between amyloid positive and amyloid negative subgroups controlling for age and sex. Amyloid status was determined by either 18F-florbetapir or 11C-Pittsburgh compound B standardized uptake value ratios. Analysis included all participants with available amyloid status, regardless of cognitive status and apolipoprotein E (APOE) ε4 status (blue = non-carrier, red = APOE ε4-carrier, black = unknown gene status). No significant relationships were found between amyloid status and retinal capillary density. B, Retinal capillary density regressions with cerebral amyloid positron emission tomography signal. Association of two measures of retinal capillary density (vessel area density and vessel skeleton density) with 18F-florbetapir standardized uptake value ratios (SUVR). Plots on the left present raw values, while plots on the right present residuals after controlling for age and sex. Analyses included all patients with available tracer uptake ratios, regardless of cognitive status or APOE ε4 status (blue = non0carrier, red = APOE ε4-carrier, black = unknown gene status)

4 DISCUSSION

APOE ε4 genotype is the strongest common genetic risk factor for sporadic AD.59 Understanding mechanisms by which APOE ε4 contributes to AD could provide novel therapeutic avenues and help with the understanding of AD pathogenesis.60-62 In model systems, APOE ε4 has been shown to affect microvascular density and function.14, 15 The goal of this study was to test whether the noted effects of APOE ε4 on microvasculature in model systems translate to humans, and whether the effects can be detected, early, in cognitively asymptomatic individuals, upstream of clinically overt symptomatic neurodegenerative disease. Because brain microvasculature cannot be directly investigated in vivo in humans, retinal capillary density was used as a surrogate for brain capillaries. After controlling for systemic vascular risk factors and demographics, APOE ε4 carriers had lower retinal capillary density compared to non-carriers. Moreover, in light of interest in the characterization of early, asymptomatic disease stages, subjects with mild cognitive impairment or known neurodegenerative disease were excluded from primary analyses.

Genes interact with an individual's environment throughout the lifespan to give rise to phenotypes. Therefore, to assess whether APOE ε4 capillary effects are post-developmental and age dependent, linear regressions with age were conducted, demonstrating significant associations with retinal capillary density in APOE ε4 carriers. Because we used the retina as surrogate for the brain, we investigated the association of retinal capillary density with ASL, a non-invasive vascular brain imaging method that might reflect effects of capillary-level pathology. In this analysis we demonstrated that across all disease categories, from at risk asymptomatic individuals to those with dementia, retinal capillary density was significantly and specifically associated with brain perfusion—with higher retinal capillary density being associated with higher brain perfusion. Consistent with these findings, in mouse models of APOE ε4, brain capillary densities were found to be associated with ASL signal.15 This suggests that retinal capillaries may be a valuable surrogate for brain capillaries. Of note, when all other risk factors are matched, the capillary differences could be observed on a single subject level. This is a strength of OCTA-derived capillary maps that are increasingly used for clinical monitoring of retinal vascular disease.

The goal of our study was not to test the spectrum of ophthalmological effects of APOE, which can be complex, but rather test its association with capillary changes in an organ directly related to the brain and accessible to direct imaging. To this end, we found that the effect sizes of group differences in retinal capillary density measures were substantial, such that up to 14% of the variance in retinal capillary density can be attributed to APOE genotype, independent of vascular risk factors. Work by Koronyo et al. suggests that it may be possible for amyloid to accumulate in the retina in the setting of cerebral amyloidopathy in AD.63 However, the prevalence of retinal amyloidopathy and disease stage at which this may occur remain to be determined. In our current study, we have low suspicion that amyloidopathy is driving the differences noted in capillary densities and the associations of retinal capillary densities with CBF. To address this, we compared retinal capillary densities in a group of individuals with positive versus negative amyloid PET scans and found no differences in retinal capillary densities between the two groups. Nor did we find continuous associations of amyloid PET signal with retinal capillary densities. This suggests that capillary changes can occur early and independent of AD neuropathological biomarkers in APOE ε4 carriers. However, it should be noted that we did not have power to test possible interactions between amyloid and APOE ε4 on retinal capillaries, as amyloid imaging was only available for a subset of the sample. Additionally, in the case of amyloid, recent studies suggest pathogenic effects from oligomeric amyloid, rather than deposited plaques,64 and PET-based analyses reported here are reflecting the presence of amyloid fibrils rather than amyloid oligomers. Additional limitations of our study included a cross-sectional study design and lack of AD biomarkers on all participants. Ultimately, longitudinal studies including all stages of clinical symptomatology and multi-modal biomarkers would shed further light on the relation of retinal vasculature in APOE ε4 gene carriers and non-carriers. In addition, inclusion of additional homozygote gene carriers to test potential dose effects, as well as other genetic risk factors for AD and other neurodegenerative disorders, would further help elucidate the biological bases of observed differences.

Overall, these findings provide the first report of lower retinal capillary densities in cognitively normal APOE ε4 carriers, and the relevance of these findings to brain microvasculature with associations of retinal capillary densities with a measure of CBF. This study provides insights into ways in which retinal vascular changes can be used as a surrogate measure for the study of vascular contributions to neurodegenerative disorders, starting from an at-risk stage. It is believed that therapeutic success in AD and other neurodegenerative disorders will require early interventions, with focus shifting from stages of overt clinical disease to at-risk asymptomatic or preclinical disease marked by the earliest detectable abnormalities in biomarkers.65 This shift in perspective led to the drafting of biomarker-driven AD research diagnostic criteria that is based on A/T/(N) biomarkers (amyloid, tau, neurodegeneration).66 However, mechanistic understanding of the many ways in which vascular disease can contribute to AD-associated neurodegeneration, cognitive impairment, and development of proteinopathies remains an important and active area of reseach.67-72 To this end, better in vivo measures of vascular disease for incorporation into modeling disease progression and clinical symptomatology are urgently needed. If the success of interventions depends on early detection of disease, the in vivo markers of underlying pathology should capture all pathological processes associated with disease, and components such as vascular disease, including microangiopathy,73-75 should be included. The capillary densities captured on OCTA could potentially contribute to microangiopathy measures in the studies of AD and related neurodegenerative disorders. Ultimately, impactful treatments for neurodegenerative disorders may require combinatorial therapies, addressing the multitude of age-associated pathologies that contribute to disease states and eventually functional decline and impairment. Our results indicate that retinal capillary measures merit further investigations as biomarkers of cerebral small vessel disease.

ACKNOWLEDGMENTS

We would like to thank patients and their families whose help and participation made this work possible. We also thank the National Centralized Repository for Alzheimer Disease and Related Dementias (NCRAD), for sample storage, which receives government support under a cooperative agreement grant (U24 AG21886) awarded by the National Institute on Aging (NIA). We thank Dr. William Jagust for conducting PiB scans. Additionally, we thank contributors who collected samples used in this study, in addition to Drs. Eliana Marisa Ramos and Giovanni Coppola and Ms. Anna Karydas for assistance in obtaining genotypes, as well as the funding sources mentioned below for grants that have supported our work.

This study was supported by the following grants: Larry L. Hillblom start-up (2019A012SUP) and American Academy of Neurology fellowship to F.M.E.; UH2NS100614, UH3NS100614, R01EY030564, and unrestricted funding from Research to Prevent Blindness to the USC Department of Ophthalmology to A.H.K.; That Man May See and Hellman family funding to A.J.G., UH2NS100608, UH3NS100608, and Larry L. Hillblom Network Grant (2014-A-004-NET) to J.H.K.; NIH-NIA ADRC (P50 AG023501) and PPG (P01AG019724) to B.L.M., R01 (AG045611) to G.D.R., R01(s) (AG032289, AG048234), an Alzheimer's Association fellowship (AARF-16-443577) to R.L.J., and Larry L. Hillblom fellowship grant (2018-A-025-FEL) and NIH-NIA (K23AG061253) to A.M.S.

    CONFLICTS OF INTEREST

    The authors have no conflicts of interest to disclose with respect to the content presented in this manuscript. A.H.K. and F.M.E. note that they have received equipment from Carl Zeiss Meditec Inc in support of this study. In addition, A.H.K. has received grants, honoraria, and financial support from Carl Zeiss Meditec. G.D.R. receives research support from Avid Radiopharmaceuticals, Eli Lilly, GE Healthcare, and Life Molecular Imaging. G.D.R. has also received honoraria as a consultant for Axon Neurosciences, Eisai, and Merck, and speaking honoraria from GE Healthcare.