Integrated Retinal Amyloid Burden With Retinal Venular Tortuosity Predicts Verbal Memory Impairment

INTRODUCTION: Retinal imaging is a non-invasive tool to study retinal vasculature and neurodegeneration. Patients with mild cognitive impairment (MCI) and Alzheimer’s disease (AD)-related cognitive disorder exhibit both retinal vascular abnormalities and intraretinal accumulation of amyloid beta-protein (Aβ) plaque. Curcumin-enhanced retinal uorescence imaging (CRFI) was recently translated as a safe imaging tool for retinal Aβ plaque quantication, holding promise as an early-stage pathological biomarker of AD. In this exploratory study, we sought to determine whether retinal vascular features combined with retinal amyloid burden correlate with the neurocognitive status. METHODS: We used scanning laser ophthalmoscopy to assess quantitative CRFI in a cohort of patients with cognitive impairment that underwent standard neuropsychological testing. Retinal blood vessels were segmented in a predened circumpapillary region of interest. For each centerline, vessel tortuosity index, vessel inection index and branching angle was quantied. Additionally, we automatically quantied retinal amyloid count in the supero-temporal quadrant and its subregions: posterior pole, proximal mid-periphery, and distal mid-periphery. Investigators performing quantications were blinded to the subjects’ clinical characteristics. Linear regression models were used to assess the correlations between retinal vascular and amyloid parameters and cognitive domain Z-scores. RESULTS: In this pilot study, 34 subjects underwent automated retinal amyloid imaging and 29 subjects (55% female, mean age 64±6 years) had the combined retinal amyloid and peripapillary vascular analysis. Eleven subjects had normal cognition, 15 MCI, 2 probable AD and 1 non-AD dementia. CRFI was increased in individuals with impaired versus normal cognitive function (p 0.0012). Venous VTI was the most signicant vascular parameter that differ across levels of CDR. Branching angle correlated with amyloid count in the distal mid-periphery (p=0.03), whereas vessel inexion index correlated with posterior pole amyloid count (p=0.02). The combined proximal mid-periphery

as an early and core contributor to the development of AD (7-9), a chronic neurodegenerative condition and looming threat to public health (2,10). The currently recognized pathological markers of AD, amyloidbeta (Aβ), tau, and neurodegeneration (11), are the central focus of multiple ongoing investigations for diagnostic biomarkers (12)(13)(14)(15). Considering the emerging vascular hypothesis (8, 16-18), there is a critical need to incorporate vascular biomarkers into predictive models to allow for early and sensitive detection of burgeoning AD. Yet, imaging of the skull-shielded brain poses various limitations for widespread screening in the clinical setting. The retina is a central nervous system organ that exhibits many of the pathological hallmarks of AD (19)(20)(21)(22)(23)(24)(25)(26)(27)(28)(29)(30)(31) and is far more accessible for repeated and highresolution imaging (32)(33)(34)(35)(36)(37)(38)(39)(40). Dysfunctional pericytes in the blood-brain barrier (BBB) are signi cant contributors to the pathogenesis of vascular cognitive impairment, including cerebral small vessel and large vessel disease and AD (26,41). BBB pericyte injury is a predictor of apolipoprotein E (APOE) ε4associated cognitive decline (4). Whereas the BBB mediates cerebral Aβ deposition, disruption of the retinal-blood barrier via retinopathy was shown to predict cognitive decline (26, [42][43][44][45][46][47]. Post-mortem retinal vessels derived from patients with mild cognitive impairment (MCI) and AD exhibited early and progressive pericyte loss and Aβ accumulation inside retinal pericytes, correlating with similar AD pathology in the brain (26). Several studies demonstrated the linkage between cognitive deterioration and retinal vascular fractal dimensions, caliber, and tortuosity (43)(44)(45)(46)(47)(48)(49). Retinal arteriolar central re ex to vessel width ratio in digital retinal photographs was signi cantly higher in APOE ε4 allele carriers (46), hence the retina may allow for noninvasive monitoring of the effects of APOE ε4 on the cerebrovascular disease. Similarly, as targeting vascular risk factors is being considered in AD prevention trials (50), retinal vascular assessments could offer a window for assessing the response to various interventions.
Recent work has highlighted the promising utility of retinal uorescence imaging, an emerging technique capable of non-invasively imaging and quantifying retinal amyloid, the pathological marker of AD (20,21,32). Using this technique, our group previously identi ed a signi cant association between retinal amyloid count, especially in the proximal mid-periphery, and the severity of cognitive impairment and hippocampal volume (32,33). As the same retinal imaging modality also allows retinal vasculature analysis, we aimed to quantitatively examine both retinal vascular and retinal amyloid biomarkers in a cohort of subjects with cognitive decline. We sought to examine the relationship between retinal microvascular geometric features and retinal amyloid burden with global and domain-speci c cognitive scores.

Participants
This study was approved by the Cedars-Sinai Institutional Review Board. All subjects older than 40 years of age, presenting to our Neurology clinic with subjective cognitive decline and interested in undergoing retinal uorescence imaging were included in this cohort (Table 1). All subjects provided written informed consent prior to the commencement of the study.

Retinal Imaging
After ocular dilation, retinal imaging was performed with a confocal scanning ophthalmoscope (Retia TM , CenterVue SpA) that utilizes blue light for excitation of curcumin emission to obtain uorescent images of the retina, following a study design described in prior reports ( Figure 1A) (32,33). The investigators conducting the retinal image processing and quanti cations were blinded to the patients' clinical characteristics.

Retinal amyloid quanti cation
Retinal images were processed using an automated retinal uorescence measurement software system (NeuroVision Imaging, Inc.). A common region of interest (ROI) in the supero-temporal quadrant was applied with a eld of view of 50 degrees positioned on the image center, using fovea and optic nerve head centers as reference points to correct for eye rotation, with a zone around the fovea and optic nerve head masked, as previously reported (33). The ROI was further divided into three subregions: posterior pole, proximal mid-periphery and distal mid-periphery ( Figure 1B). Retinal amyloid count was quanti ed in the target ROI and the three speci ed subregions.

Retinal vascular quanti cation
From the same retinal fundus images, a ROI was de ned within a circumpapillary region centered on optic nerve head (ONH) and extended between 1.5 and 4 ONH radii (51) ( Figure 1C). Retinal vessels within the ROI were segmented using Frangi vesselness lter to generate a binary image (52). The vessels were classi ed into arteries and veins by a human observer based on the facts that retinal arteries are brighter in color and thinner in width compared to veins (53). For each vessel segment on the binary image, vessel endpoints were selected, and distance transformation was used to extract the vessel centerline. The extracted centerlines were smoothed using a cubic spline with a regularization parameter of 3 x 10-5. For each centerline, several geometric features including vessel tortuosity index (VTI), vessel in ection index, and branching angle were non-automatically quanti ed. VTI was calculated for each centerline based on combination of local and global centerline geometric variables as explained previously (54). Equation (1) shows formula for VTI, where is standard deviation of angle difference between lines tangent to each centerline pixel and a reference axis (i.e. x-axis). M is average ratio of centerline length to its chord length between pairs of in ection points including centerline endpoints. N is number of critical points where the rst derivative of the centerline vanishes. LA and Lc are the length of the vessel and its chord length, respectively. VTI is shown to provide good correspondence with human perception of tortuosity and is invariant to rigid transformations. Similar to other measures of tortuosity, VTI is unitless. Its minimum value is zero while it has no theoretical maximum as it can increase with twistedness of a vessel. Vessel in exion index was determined based on number of in ection points along the vessels. Mathematically, these are pixels where the second derivative of centerline vanishes. Vessel in exion index represents local changes in tortuosity of vessels and was found to be robust for ranking tortuosity of vessels with similar length (55). Branching angle of the vessels was calculated interactively using an open source tool GIMP 2.8.

Statistical Analysis
Descriptive statistics were calculated for patient demographics and clinical characteristics. Unless otherwise stated, data are expressed as mean ± standard deviation. Subjects were partitioned into three groups according to Clinical Dementia Rating (CDR; 0.5, questionable impairment; 1, mild cognitive impairment; 2, moderate cognitive impairment) (56). The subjects were also partitioned into groups according to the neuropsychometric diagnosis (normal cognition versus impaired cognition).
To produce combined indices of retinal vascular and amyloid measures, each variable was rst inspected for normality; any non-normal variables were then log transformed to produce a normal distribution. Each normalized variable was then standardized to a mean of 0 and a standard deviation of 1. While higher amyloid count was associated with worse cognitive function, higher venous vascular tortuosity index (VTI) values were associated with better cognitive function. To account for this inverted scale, the standardized values of venous VTI were multiplied by -1. Standardized variables were then summed to produce exploratory, combined index measures of retinal amyloid and retinal vascular features.
Differences in continuous variables between levels of CDR were assessed through one-way analysis of variance (ANOVA). Differences in continuous variables between diagnostic scores were assessed using Student's t-test. Linear regression was performed to assess the relationship between retinal vascular and retinal amyloid measures, as well as the relationship between combined retinal vascular and amyloid counts and cognitive parameters. All statistical analyses were performed using STATA v15.1 (StataCorp, College Station, TX) with an a priori signi cance level of 0.05.

Results
Our study included a total of 34 subjects that presented to our Neurology clinic with cognitive concerns. Of those, 29 had retinal images of necessary quality to undergo both retinal amyloid and vascular analysis. Subject demographics and preexisting conditions are shown in Table 1. Mean MOCA was 26 (range 4-32) and median was 27. On formal neuropsychometric cognitive evaluation, 11 (37.93%) patients had normal cognition and 18 (62.06%) had impaired cognition (6 amnestic MCI, 9 multidomain MCI, 2 probable AD and 1 possible fronto-temporal lobar degeneration). Eleven subjects had a CDR of 0.5, 15 had a CDR of 1, and 3 had a CDR of 2.
Linear regression analyses revealed that venous branching angle correlated with distal mid-periphery amyloid count (p = 0.03) and arterial in exion index correlated with posterior pole amyloid count (p = 0.02). There were no associations between retinal vascular parameters and amyloid count in the proximal mid-periphery (Supplemental Table 1).
Analysis of retinal vascular and amyloid measures according to strata of cognitive function showed that retinal PMP amyloid count and total amyloid count were signi cantly higher in the cognitively impaired subjects compared to normal cognition (total: 343 ± 90 versus 247 ± 82, p = 0.04; PMP: 144 ± 52 versus 85 ± 32, p = 0.0012; Fig. 1D, E). There was no signi cant difference in venous branching angle (p = 0.98) or arterial VTI (P = 0.53) across levels of CDR, whereas arterial branching angle reached near signi cance (p = 0.066) (Fig. 1F). Venous VTI was signi cantly different across levels of CDR (mean ± SD of venous VTI values across increasing CDR categories: 0.13 ± 0.02, 0.13 ± 0.02, 0.09 ± 0.02; P = 0.026) (Fig. 1G). Given these group differences, and because of the independence of retinal vascular and retinal amyloid measures, the following combined amyloid-vascular indexes were calculated as exploratory variables: proximal mid-periphery amyloid count-venous VTI, total amyloid count-venous VTI, proximal midperiphery amyloid count-arterial branching angle, and total amyloid count-arterial branching angle. Oneway ANOVA revealed signi cant group differences in the VTI indices integrated with total retinal or PMP amyloid when compared according to CDR level ( Fig. 2A-D). The combined proximal mid-periphery amyloid-venous VTI index exhibited signi cant group differences when cognitively impaired were compared to cognitively normal subjects (0.49 ± 1.1 versus − 0.91 ± 1.4, p = 0.006; Fig. 2E). In this cohort, the combined total retinal amyloid-venous VTI index reached near signi cance of increased index in the cognitively impaired subjects (p = 0.09; Fig. 2F).
We next performed regression analyses to evaluate the correlations between retinal vascular geometric parameters and retinal amyloid counts with cognitive domain Z-scores. We found that venous branching angle correlated with WAIS-IV-digit span Z-score [beta

Discussion
This exploratory investigation of retinal uorescence imaging reveals that the combination of retinal amyloid deposits and venous tortuosity index could predict cognitive impairment, and especially verbal memory loss. Further, this study suggests that retinal proximal mid-periphery amyloid count is a comprehensive indicator of AD-related cognitive functioning. In this cohort, we found that retinal vascular features generally did not correlate with retinal amyloid deposition in the proximal mid-periphery but rather with deposits in the distal mid-periphery and posterior pole regions. Overall, by quantitatively assessing two key measures, vascular disease and amyloid burden, we showed the utility of retinal uorescence imaging as a potential outcome measure in AD clinical trials.
Microvascular damage as a critical initiator of AD pathology is increasingly recognized (5,9,57). Vascular dysregulation leading to cerebral amyloid accumulation, and the link between cerebrovascular disease and dementia, are explained by several mechanisms (4, 5, 58). Pericyte loss and de cient vascular platelet-derived growth factor receptor-β signaling were identi ed in both retinal and cerebral vasculature in subjects with MCI and AD (4,26). Prior reports demonstrated that retinal vasculature may be used as a biomarker of early or preclinical AD (59), and retinal microvascular abnormalities in AD have been demonstrated using various retinal vasculature imaging modalities (e.g. retinal fundus photography (46, 60, 61), optical coherence tomography angiography (62, 63), high-frequency icker light stimulation (49,64), and retinal function imaging (62)). Conversely, den Haan et al (65) showed that retinal vascular measures did not differ between patients with AD and control participants, and venular tortuosity was smaller in subjects with greater white matter disease burden. Tortuosity is a common arteriolar and venular feature found to be associated with vascular disease and aging. We noted a signi cant difference (p = 0.026) in venous tortuosity between patients with low (0.5) and high (2) CDR, as patients with higher CDR had lower venous tortuosity indices. Our results are in agreement with a prior study (66) that found less tortuous venules in AD patients compared to controls without cognitive decline (p = 0.0244), in a similarly analyzed retinal region that included 0.5-to-2.0 disc diameters away from the optic disc margin. In addition, a study by Williams et al. (67) showed that subjects with lower arteriolar tortuosity were more likely to have AD after appropriate adjustment, whereas no signi cant variations in venular tortuosity were detected. In contrast to our ndings, another study that explored the association between retinal vascular tortuosity and cognitive impairment showed that subjects with increased arteriolar and venular tortuosity were more likely to have AD when compared with cognitively intact controls (60). The disparity in these ndings could be explained by the various methods and software that were used to calculate vessel tortuosity and/or by including different retinal regions of interest in the vascular analysis.
Previous investigations have also shown a strong relationship between retinal vasculopathy and brain amyloid deposition (28), providing compelling evidence for the vascular hypothesis of AD. Shara et al.
(28) evaluated the relationship between retinal vascular status (vessel diameter, tortuosity and spatialspectral texture measures) using hyperspectral retinal imaging and CNS amyloid status assessed with (18) F-orbetaben positron-emission tomography. They found that retinal venules of amyloid-positive subjects showed a higher mean tortuosity compared with the amyloid-negative subjects. This pilot study suggested that the inclusion of metrics related to retinal vasculature and surrounding tissue-related texture could improve the discrimination performance of the cerebral amyloid status (28).
As both retinal amyloid accumulation and retinal vascular pathology (26, [32][33][34][35] are reported in patients with early or preclinical AD, we explored the interplay between retinal vascular geometric measures and retinal amyloid burden using retinal uorescence imaging. Prior studies showed that amyloid quanti cation in the retinal proximal mid-periphery may best re ect cerebral AD pathology, as it correlates with cognitive performance and hippocampal volume (33,35). In this cohort, we found that retinal vascular features correlate with amyloid deposition in the posterior pole and retinal distal mid-periphery, but not the proximal mid-periphery. A possible explanation is that this investigation measured physical features of the retinal vasculature (e.g., branching angle and tortuosity index), and did not assess functional endpoints. In addition, our quantitative vascular analysis could not target the smaller retinal blood vessels. The mechanisms driving vascular remodeling and amyloid deposition may occur at different rates, leading to the appearance of these clinical signs at different stages in disease progression. The investigation of subjects with mainly mild cognitive impairment in our cohort may explain why venous VTI was lower in subjects with worse cognition, and why the other arterial or venous vascular parameters did not show any signi cant differences across cognitive strata. This hypothesis is supported by the lack of association between retinal vascular features and most of the neuropsychometric cognitive scores in our cohort. Whereas retinal vascular measures did not correlate with any cognitive measures except for attention and concentration, total and proximal mid-periphery retinal amyloid correlated with verbal memory measures, and distal mid-periphery amyloid count correlated with non-verbal memory. Interestingly, in this cohort with early cognitive disorders, subjects with higher amyloid count and worse cognition had lower retinal venular tortuosity. The only combined index that discriminated between individuals with impaired cognition and normal cognition was proximal mid-periphery amyloid count-venular VTI. This combined index was also associated with verbal memory and the 'mental component' summary of psychological functioning (SF-36 Mental Component Score). This later nding re ects the association with the AD-related psychological and emotional functioning. This appears to represent an exclusive contribution, as physical functioning status (SF-36 Physical Component Score) was not associated with amyloid or vascular retinal markers.
In a prior study, two or more retinal vascular abnormalities were associated in a dose-response manner with an increased risk of disabling dementia (47). It is possible that combined amyloid-vascular indexes are better discriminators of AD, with potential for use as outcome measures in AD trials.
Our ndings underscore the potential value of the exploratory amyloid-vascular indexes presented herein.
Future investigations are warranted to explore the clinical utility of retinal uorescence imaging in concert with combined amyloid-vascular index measures. More comprehensive cohort studies including a larger sample size, a greater range of disease severity among the participants, and age-matched controls could help to elucidate the stage at which retinal amyloid and/or retinal venular versus arterial impairments begin to develop, while further developing clinical tools for the detection and characterization of AD.
Given the cost and technical requirements of gold-standard methods for assessing cerebral amyloid deposition and vascular pathology, further validation of these retinal imaging methods could potentially yield greater accessibility to testing, thus facilitating more extensive clinical trials as well as improved detection of AD.

Limitations
Our pilot study is limited by a small sample size and heterogeneity. Due to limited numbers, we could not adjust for the presence of traditional vascular risk factors or the presence of retinopathy, which are known contributors to retinal vascular geometric changes. Given the heterogeneity in sample size across study groups and small numbers of probable AD patients, further con rmation of these preliminary ndings is needed.

Conclusions
This work reports a pilot retinal imaging-based investigation of the interaction between two reported biomarkers of Alzheimer's disease (AD) -vasculopathy and amyloidopathy -in a small cohort of subjects mostly with mild cognitive impairment. This integrated approach holds the potential advantage of assessing both a hypothesized driver of AD, vascular impairment, as well as the hallmark pathological marker of AD, amyloid deposition. These biomarkers were quantitatively examined in the retina, a central nervous system organ that allows for direct, non-invasive, and high-resolution imaging. We found that the combined amyloid count in the retinal proximal mid-periphery and retinal venular tortuosity correlated with verbal memory and the mental component of the psychological functioning SF-36. Our quantitative assessments of these two key measures may validate retinal uorescence imaging as a useful outcome parameter in AD clinical trials.

Supplementary Tables
Supplemental Table 1 is not available with this version. Figure 1 Study timeline of brain and retinal imaging followed by sectoral amyloid and vascular analysis. Study design scheme illustrating that subject underwent baseline brain imaging and neuropsycological evaluation, followed by retinal uorescence imaging after 4 days of daily oral curcumin intake (A). Illustration of the region of interest in the right eye supero-temporal retinal quadrant and its three subregions, which were used for quantifying retinal amyloid counts (B). Illustration of the region of interest used for the retinal vascular analysis. The red circle indicates the center of optic nerve head and the smallest yellow circle shows optic nerve head area. The two larger circles demonstrate the region of interest for vascular analysis which are 1.5-and 4-times the diameter of the optic disc. The branching angle and tortuosity of vessels within the region of interest were calculated. Arteries and veins are outlined by red and blue lines, respectively (C). Graphs illustrating differences in total amyloid (D) and proximal amyloid counts (E) when strati ed by cognitive status. Graphs illustrating the differences between arterial branching angle (F) and venous tortuosity index (G) when strati ed by CDR.

Figures
Abbreviations: MRI, magnetic resonance imaging; PP, posterior pole; PMP, proximal mid-periphery; DMP, distal mid-periphery; ODD, optic disc diameter; CDR, Clinical Dementia Rating; VTI, vessel tortuosity index Combined retinal amyloid and vascular parameters in patients strati ed by cognitive scores. Graphs illustrating the differences in the combined proximal mid-periphery amyloid -arterial branching angle index (A), total amyloid -arterial branching angle index (B), proximal mid-periphery amyloid -venous tortuosity index (C) and total amyloid -venous tortuosity index (D) when strati ed by CDR score. Graphs illustrating the differences between the combined proximal mid periphery amyloid -venous tortuosity index (E) and total amyloid-venous tortuosity index (F) when strati ed by the cognitive status. Bar graphs are showing mean and deviation (* P < .05, ** P < .01 by two-tailed paired Student t test). Abbreviations: VTI, vessel tortuosity index; ABA, arterial branching angle; PMP, proximal mid-periphery; CDR, Clinical Dementia Rating.