Longitudinal Assessment of Retinal Microvasculature in Preclinical Alzheimer's Disease

Abstract

Purpose: To investigate if changes in vessel density (VD) and the foveal avascular zone (FAZ) occur in the preclinical phase of Alzheimer's disease (pAD) over time.

Methods: Optical coherence tomography angiography (OCTA) was used to image VD and FAZ at baseline and for a follow-up period of 2 years. Positron emission tomography (PET) was used to determine the amyloid beta (Aβ) status of participants.

Results: The VD and FAZ of 148 participants (54% female) were analyzed at baseline and follow-up (mean time between measurements, 2.24 ± 0.35 years). The mean age of the participants was 68.3 ± 6.0 years at baseline and 70.3 ± 5.9 years at follow-up. Participants were divided into three groups: control group, participants who had negative Aβ status at both measurements (Aβ−, n = 116); converter group, participants who transitioned from negative to positive between baseline and follow-up (Aβ−+, n = 18); and participants who were consistently positive at both visits (Aβ++, n = 14). The VD of both Aβ+ groups demonstrated non-significant increases over time in both macula and optic nerve head (ONH) regions. The Aβ− group was found to be significantly higher in both ONH and macular regions. The VD of the Aβ++ group was significantly higher in the macula inner and outer rings compared to the Aβ−+ and Aβ− groups. No significant change was found in FAZ values over time.

Conclusions: Alterations in VD seem to manifest already in pAD, exhibiting distinct variations between the ONH and macula. Further longitudinal studies with a longer follow-up design and known amyloid pathology should be undertaken to validate these observations.

Alzheimer's disease (AD) is a neurological disorder that causes progressive decline in cognitive, functional, and behavioral abilities and ultimately leads to death.¹ The pathological hallmarks of AD include interneuronal plaques containing the protein amyloid beta (Aβ) and intraneuronal tangles characterized by hyperphosphorylated tau (Ptau).² The disease consists of three major stages now known as the Alzheimer's disease spectrum: (1) the preclinical phase, where patients are positive for known AD biomarkers (Aβ or Ptau), yet show no neurological symptoms; (2) the mild cognitive impairment phase, where mild cognitive dysfunction is observed but patients are still able to live independently; and (3) mild, moderate, and severe AD in which cognitive changes are progressive and patients depend on others for their daily life activities.³ 

Since the first description and characterization of Alzheimer's disease in 1906 by Alois Alzheimer, researchers are still struggling to find effective treatments.⁴ It is the most prevalent cause of dementia and holds a significant presence within the current global estimate of over 50 million individuals affected by dementia.^5,6 The diagnosis of AD makes use of biomarkers Aβ and/or Ptau.⁷ These biomarkers require expensive or invasive procedures such as positron emission tomography (PET) and lumbar puncture, which are not readily available in every community and are usually only undergone when an individual is referred to a memory clinic. Finding an inexpensive and patient friendly biomarker for the detection of AD could enable early detection, paving the way for initiation of treatment to prevent or slow down disease progression before symptoms occur. 

Embryologically related to the brain, the retina may hold promising, easily accessible, non-invasive, and inexpensive biomarkers potentially leading to a diagnosis of AD in the preclinical stage.^8–13 Optical coherence tomography angiography (OCTA), a widely used technique in ophthalmology, can visualize the microvasculature of the retina in a matter of seconds in a non- invasive manner.¹⁴ Biomarkers of interest such as vessel density (VD) in the superior capillary plexus (SCP), deep capillary plexus (DCP), and foveal avascular zone (FAZ) show promise based on evidence from studies that have demonstrated changes in vascular density in both the brain and retina in AD.^10,15 A number of cross-sectional studies have revealed decreases in the SCP and DCP in patients with mild cognitive impairment (MCI) and AD and either an enlargement or no change in the FAZ.^16–25 However, research on VD and FAZ in the preclinical stage remains sparse, with only three studies reporting inconsistent findings. VD was found to be both significantly lower²⁶ and higher²⁷ in two separate studies, and FAZ was found to be enlarged in two of the studies^26,28 yet unchanged in another.²⁷ 

Moreover, very few studies have covered longitudinal changes in any stage of the Alzheimer disease spectrum. One study demonstrated decreases in VD of the DCP and SCP along with an enlargement of FAZ over a 2-year period in amnestic MCI participants,²⁹ whereas another study showed no changes in VD or FAZ over a 3-year period in participants with pAD.³⁰ The aim of this study was to further explore changes in retinal VD and FAZ over time to investigate their potential as biomarkers for pAD. 

Methods

The data were obtained from a cohort study that forms part of the European Medical Information Framework for Alzheimer's Disease (EMIF-AD), which has been previously described by Konijnenberg et al.³¹ Originally, 199 neurologically normal functioning monozygotic twins 60 years of age or older who underwent amyloid PET scans were recruited to participate in this substudy. Participants were selected from the Netherlands Twins register.^31,32 This study was approved by the medical ethics committee at Amsterdam UMC (METC Amsterdam number 2014.210) and strictly adhered to the tenets of the Declaration of Helsinki. All participants gave written informed consent before undergoing testing. Baseline retinal VD and FAZ measurements among the Aβ+ and Aβ− participants have previously been published by van de Kreeke et al.²⁷ These data were supplemented with VD and FAZ measurements taken at a 2-year follow-up interval among the same participants. In addition to Aβ PET data from 2015, we also had access to Aβ PET data from 2019. 

Cognitive Assessment

Cognitive assessment was based on a Clinical Dementia Rating (CDR) score of 0, Geriatric Depression Scale (GDS) of greater than 11, and a telephone interview for cognitive status-modified scores greater than 23. The Mini Mental State Examination (MMSE) was employed at baseline and at the 2-year follow-up to assess the cognitive status of participants. All participants scored higher than 25 at baseline, which is considered the cut-off for mild cognitive impairment.³³ 

Aβ Status Determination

Aβ status was established through PET scans performed in both 2015 and 2019 according to the previously described protocol in van de Kreeke et al.²⁷ and Konijnenberg et al.³¹ An Ingenuity TF PET/MRI scanner (Philips Medical Systems, Best, The Netherlands) was used at both baseline and follow-up. Analysis of PET scans was performed using constructed standardized uptake value ratio (SUVr) images and cerebellar gray matter as the reference region. Experienced nuclear physicans (BvB, EMvdG) conducted visual assessments of all SUVr images. Each image received a positive or negative rating based on predefined criteria established by the manufacturer (GE HealthCare, Chicago, IL, USA). Three distinct Aβ groups were established based on the PET scans in 2015 and 2019: Aβ− (negative at both 2015 and 2019), Aβ−+ (converters, who were negative at 2015 and positive at 2019), and Aβ++ (positive at both 2015 and 2019). 

Ophthalmic Examination

Comprehensive ocular assessments including best-corrected visual acuity obtained via refraction, intraocular pressure, slit-lamp examination, indirect fundoscopy, spectral-domain OCT (SD-OCT), OCTA, and fundus photography were performed on all participants at both baseline and follow-up. The eyes of all participants were dilated with tropicamide 0.5%. No additional ophthalmological pathology, other than floaters, affecting scan quality was detected at follow-up. At baseline, exclusion criteria included the presence of severe cataract, exudative macular degeneration, glaucoma, diabetic retinopathy, vascular occlusions, and refractive errors < −8 diopters (D) or > +5 D. This same exclusion criterion was also applied at follow-up. 

Optical Coherence Tomography Angiography

The ZEISS CIRRUS 5000 OCTA system (Carl Zeiss Meditec, Jena, Germany) was used to measure the VD of the SCP in the macula, the VD of the region around the optic nerve head (ONH) and the FAZ region over a 1-year period in both 2017/2018 and 2019/2020. The Early Treatment of Diabetic Retinopathy Study (ETDRS) grid was used to divide the retina into nine regions defined by two rings and was placed on the macula and ONH region (Fig. 1). The macular grid was divided into inner and outer rings. The inner ring had an inner radius of 1 mm and an outer radius of 3 mm, whereas for the outer ring these were 3 mm, and 6 mm, respectively. Only the outer ring was used in our analysis of the ONH region. We used the average of both the right and left eyes for each parameter due to the fact that FAZ and VD of the SCP have been shown to demonstrate a good interocular symmetry.^34,35 If the parameter of only one eye was available, then only that eye was used. 

Figure 1.

 

ETDRS 6 × 6-mm grid placed over the ONH and macula.

Data Analysis

Data analysis was performed using both SPSS Statistics 28 (IBM, Chicago, IL, USA) and R (Rstudio 4.4.1; R Foundation for Statistical Computing). Participant data and parameters were checked in SPPS Statistics 28 and exported to R for statistical analysis. Linear mixed models were employed to analyze each parameter (VD macular inner ring, VD macular outer ring, VD ONH, and FAZ), with age, hypertension, hypercholesteremia, diabetes, and cardiovascular and vascular factors (angina, myocardial infarct, atherosclerosis, and transient ischemic attack) taken as co-variates. Additionally, the analysis accounted for data clustering within the twins pairs and missing data. Changes over time among the Aβ−, Aβ−+, and Aβ++ participants were shown by adding an interaction term with time in the model. Spaghetti plots were generated to illustrate the changes of the participants over time between 2017 and 2019, as well as trends between the parameters of Aβ+ and Aβ− participants. P < 0.05 (two-sided) was considered to be statistically significant. We additionally performed a sensitivity test on outliers with ±3 SD and ±2 SD difference in VD and FAZ values between the two measurements; however, this showed no difference in the changes over time or between the groups. 

Results

Participants from 2017 to 2018

Van de Kreeke et al. analyzed 124 participants in the baseline study.²⁷ 

Participants from 2019 to 2020

One hundred and ten participants returned for the follow-up visit. Unfortunately, due to the COVID-19 pandemic in 2020, during which the Netherlands experienced several lockdowns, many participants could not attend the hospital for ophthalmological testing. A further four participants were excluded from measurements due to extreme disturbance of the vitreous of both eyes which is known to affect the quality of VD and FAZ on OCTA. Therefore, 96 participants were included in our analysis for the follow-up. 

Total Included Participants and Missing Data

In total, 148 participants (69 twin pairs and 10 singletons) were included for longitudinal analysis. Due to the presence of missing data, only 74 participants had complete measurements at both time points. The remaining 74 participants had one measurement either at the baseline or at the follow-up visit, resulting in an unbalanced dataset. Furthermore, not all of the repeated measures of each parameter were available for each patient due to the fact that the quality of the OCTA scans was not good enough (missing at random data, MAR). In addition, some patients who were not present at baseline did return for the first follow-up measurement, and others who were present at baseline did not attend the first follow-up (missing completely at random data, MCAR). A clear overview of available measurements at baseline and follow-up for Aβ−, Aβ−+, and Aβ++ participants is presented in Table 1. A linear mixed model was employed, which provided the most robust way to accommodate missing data and to account for varying numbers of observations per participant.³⁶ 

Table 1.

 

Total Number of Participants Per Measurement Time Point and Number of Valid Measurements for Every Parameter Per Measurement Time Point

 

Participant Demographics

Patient demographics based on Aβ group are presented in Table 2. The Aβ− group was comprised of 116 individuals, the Aβ−+ group had 18 individuals, and the Aβ++ group had 14 individuals. Females accounted for 53% of the Aβ− group, 44% of the Aβ−+ group, and 71% of the Aβ++ group. Mean times between follow-ups for the groups were 2.33 ± 0.36 years, 2.32 ± 0.48 years, and 2.69 ± 0.53 years, respectively. Mean follow-up times between PET scans were 4.29 ± 0.39 years, 4.34 ± 0.54 years, and 4.45 ± 0.50 years, respectively. In addition, average MMSE scores and average signal strengths of the macula and ONH, as well as cardiovascular and vascular factors, at baseline and follow-up are also recorded in Table 2. 

Table 2.

 

Demographics of Participants Per Aβ Group

 

Vessel Density of the ONH

Changes Over Time Within the Aβ Groups

VD around the ONH increased slightly over the follow-up period for both the Aβ− group (β = 0.155, P < 0.01) and the Aβ−+ group (β = 0.157, P = 0.99) after correcting for age and the presence of hypertension, hypocholesteremia, diabetes, and cardiovascular/vascular parameters. VD also showed a near minimal increase for the Aβ++ group (β = 0.011, P = 0.31); however, this was not significant. Spaghetti plots were generated to illustrate the overall changes observed over time and are presented in Figure 2 for changes in ONH among the Aβ groups. 

Figure 2.

 

Changes in VD of the ONH between baseline and follow-up.

Differences Between the Aβ Groups

There were no significant differences observed between the VD values of the Aβ++ and Aβ−+ groups compared to the control group (Aβ−). A complete list of beta values, their 95% confidence intervals (CIs) and P values, and average VD values can be found in Table 3. 

Table 3.

 

Changes Over Time, Differences Between Groups, and Average VD for the Four Retinal Microvascular Parameters

 

VD Macular Inner and Outer Rings

Changes Over Time Within the Aβ Groups

An increase in the VD of both the macula inner and outer rings over the follow-up period was noted in all three groups; however, only the control group was found to be statistically significant (inner ring: β = 0.131, P = 0.03; outer ring: β = 0.122, P = 0.03) (Table 3, Figures 3 and 4). 

Figure 3.

 

Changes in VD of the macular inner ring between baseline and follow-up.

Figure 4.

 

Changes in VD of the macular outer ring between baseline and follow-up.

Differences Between the Aβ Groups

VD values of the Aβ++ group were found to be significantly higher compared to the Aβ− group in both inner and outer rings (inner ring: β = 0.554, P = 0.03; outer ring: β = 0.07, P < 0.01) (see Table 3). 

Foveal Avascular Zone

Changes Over Time Within the Aβ Groups

There were minimal changes over time for all three Aβ groups, and none of these were statistically significant (Figure 5). 

Figure 5.

 

Changes in FAZ values between baseline and follow-up.

Differences Between the Aβ Groups

The Aβ−+ group showed a slight significant difference compared to the control group (β = 0.027, P = 0.05), whereas the Aβ++ group was non-significant (β = 0.007, P = 0.71). 

Individual VD and FAZ Changes at the Patient Level

 Figures 2 to 5 show the general trends for VD and FAZ over time within the three Aβ groups. To complement these group-level trends, we also created scatterplots in order to visualize individual patient level changes in the VD and FAZ of those patients who had both measurements; these are presented in Figure 6. 

Figure 6.

 

Scatterplots of baseline versus follow-up measurements for ONH VD, macular inner and outer VD, and FAZ.

Discussion

This study demonstrated that VD appears to increase over an average 2-year period, with increases more prominent in Aβ-negative participants around the ONH compared to those who are positive for Aβ. Furthermore, our results show that the ONH and macula exhibited distinct variations between participants who were positive for Aβ for 4 years or longer (Aβ++) compared with the converter group (Aβ−+). In addition, differences were also observed in VD between the groups. VD in the Aβ++ group in the macula inner and outer rings remained significantly higher compared to VD in the Aβ− and Aβ−+ groups, whereas around the ONH there was no significant difference between the Aβ++ group and the other groups. 

These results provide intriguing insights into the relationship between amyloid pathology and changes in VD. The majority of studies have demonstrated a notable reduction in VD in the SCP among individuals diagnosed with AD compared to healthy controls.^{16,17,19,23,37} Chua et al.¹⁷ and Zhang et al.²⁵ also showed decreases in VD in MCI participants compared to controls, although this decrease was less prominent than that of AD participants in the study by Chua et al.¹⁷ In turn, Yoon et al.³⁷ demonstrated no difference in the VD of MCI participants compared to healthy controls. 

Our results regarding VD of the macula ETDRS inner and outer rings in the Aβ++ group appear to reflect the same increasing trend during the follow-up period as the baseline data reported by van de Kreeke et al.²⁷ and were significantly different compared to the other two groups. On the other hand, VD of the ONH in the Aβ++ group remained quite similar to that of the Aβ−+ group and did not significantly differ from the control group. 

The differences in microvasculature between the macula and ONH could explain the disparity of changes found in VD in these two areas. The macula exhibits a higher metabolic demand within the retina, serving to meet the metabolic requirements of this region, whereas the ONH and peripheral retina possesses a comparatively lower microvascular supply.³⁸ If we take the hypothesis of AD proposed by van de Kreeke et al.²⁷ and suggested by Biron et al.,³⁹ one could infer that the increased inflammatory response in the preclinical phase causes increased blood flow to the macula and, as a result, an increase in VD to maintain adequate perfusion and metabolic support during early disease progression. On the other hand, the peripapillary area surrounding the ONH consists of a less dense microvascular supply where this compensatory mechanism is not necessary. Furthermore, a recent study by López-Cuenca et al.⁴¹ that explored the effect of a genetic risk (APOE ε4 carrier) of developing AD, hypertension, and hypercholesteremia on VD observed differences in trends between VD measurements pertaining to these genetic and cardiovascular risk factors in the ONH and macular regions. Ma et al.⁴² also found evidence of changes in the macular and peripapillary regions both at baseline and follow-up among APOE ε4 carriers; however, the two regions both showed significant decreases, and the trends did not differ from each other. López-Cuenca et al.⁴¹ proposed that the reduced number of anatomical anastomoses in the peripapillary region might make it more prone to vascular dysfunction, explaining the observed differences. This could imply that genetic risk factors could also influence microvascular alterations, although this is currently not considered in the research framework of the National Institute on Aging and Alzheimer's Association. 

Another interesting observation is that, of the 10 participants excluded in our analysis due to glaucoma, five of them were Aβ+. Recently, an association between the development of Alzheimer's disease and glaucoma has been proposed due to a possible shared genetic susceptibility to the APOE protein; the accumulation of Aβ and tau proteins damages the lateral geniculate body neurons connecting the ONH to the brain and causes thinning of the retinal nerve fiber layer.⁴⁴ In a systematic review, eight cohort and case studies comprised of 6870 AD cases were investigated, and individuals with glaucoma were 1.52 times more likely to develop AD.⁴⁵ Additionally, primary open-angle glaucoma was found to be associated with an increased risk of AD.⁴⁰ Given that there appears to be a difference in VD trends among the ONH and macula in Aβ++ individuals, further studies in this area would be warranted. 

No significant changes were observed in the FAZ, either throughout the follow-up period or between the three groups. An important consideration to take into account is the physiological difference in FAZ areas across the population. Shiihara et al.⁴⁶ demonstrated a relatively large range of FAZ values ranging from 0.073 to 0.565 mm² among a healthy population of 70 adults. Eldaly et al.⁴⁷ further demonstrated morphological changes of FAZ among the healthy population. 

Another factor that can affect FAZ measurements is floaters or shadows. From our own clinical experience using the CIRRUS 5000, floaters or shadows can sometimes give the impression of a large FAZ. If there are too many shadows or floaters in the eye being measured, it can cause a great difference in FAZ values. For example, in the current study population, the macula scans of a few participants were measured twice due to the presence of a shadow and gave extremely different values. In one of the participants, we measured a FAZ area of 0.93 mm², and this scan was repeated without the shadow obstructing the image. When the shadow was not present, which is evident in the en face image, the FAZ area of the same participant was measured as 0.26 mm², which is a much more accurate representation. Signal strength has been known to affect the quality and interpretation of VD and FAZ measurements. Lim et al.⁴⁸ studied the effect of different signal strengths from 7 to 10 on VD and FAZ in 292 eyes of males and 154 eyes of females. They found that VD and FAZ measurements increased significantly with each increase of signal strength from 7 to 9, highlighting the importance of a high signal strength for accurate interpretation. The use of high signal strengths, coupled with good en face inspection of the scan for the presence of troubling media opacity, should therefore always be considered. 

Taking these points into consideration, we believe that FAZ would not make a good biomarker to track changes over time in preclinical Alzheimer's disease. This could potentially account for the diverse findings of FAZ values reported in the literature. Three studies have documented an enlarged FAZ. Bulut et al.¹⁶ reported an enlarged FAZ in AD participants compared to healthy controls. Wu et al.²¹ observed a progressive enlargement from healthy controls to MCI rather than to AD, and O'Bryhim et al.²⁸ demonstrated an enlarged FAZ in preclinical AD participants. Conversely, four other studies found no difference in FAZ in AD, MCI, or pAD.^17,25,27,37 

The above-mentioned issue concerning shadows and floaters also applies to VD. If there are too many floaters, VD can appear less than what it is in reality, as vessel detail is lost on the scan. However, when repeated, the value becomes slightly higher. Therefore, caution should be taken when interpreting images where shadowing or excessive floaters are present. 

The discrepancies in VD and FAZ observed in the literature could be attributed to the use of different diagnosing criteria and biomarkers for the AD spectrum. Some studies have chosen only clinical criteria as their basis, whereas others have used magnetic resonance imaging, amyloid PET scans, or Aβ42 and tau cerebrospinal fluid (CSF) biomarkers, or even a combination, to define their populations. To date, the three groups that have studied preclinical changes in VD and FAZ used CSF, amyloid PET, or a combination of these biomarkers. Corradetti et al.²⁶ showed a decrease in VD and increase in FAZ among participants by using tau and Aβ42 CSF biomarkers. O'Bryhim et al.²⁸ reported no difference in VD at either baseline or follow-up; however, they did observe an increase in FAZ at baseline using a mix of amyloid PET or CSF biomarkers compared to our increase in VD both at baseline and follow-up among Aβ PET++ participants. Whereas CSF and Aβ PET biomarkers show excellent sensitivity for diagnosis,⁴⁹ CSF biomarkers reflect biochemical changes of protein production and clearance, and amyloid PET reflects plaque brain changes and damage to the brain over time.⁴³ Whether or not this affects retinal microvascular measurements in such an early stage of the disease remains unknown, and working to standardize this process would greatly enhance the reliability and comparability of research findings. 

This study possesses several notable strengths. First, it incorporated longitudinal Aβ data, enabling us to track changes within our cohort over an extended period of time. Consequently, we were able to more accurately categorize participants into three distinct groups: those who remained negative, those who converted, and those who were positive for 4 years or longer. In addition, we had a larger number of Aβ+ participants included in our study (n = 32) compared to the baseline study (n = 13), which increased the statistical power of this study. 

The current study is, however, not without its limitations. The follow-up results were limited to a duration of only 2 years, which is a relatively brief time frame. Moreover, it is important to note that we utilized a macula ETDRS grid overlay on the ONH instead of the official ONH software, which was not available during the time the scans were taken. 

To conclude, this longitudinal cohort study provides an indication that the VD of the SCP remains stable over a 2-year period in participants who are Aβ++ for longer than 4 years in contrast to healthy controls and converters. However, there appears to be a distinctly higher VD among Aβ++ participants in the macula inner and outer ring. OCTA offers a simple, non-invasive way to image and record changes in VD, reinforcing the notion that it could be effective as a potential biomarker to detect AD before cognitive decline is present, although the exact timing for these changes begin still remains unclear today. Future studies investigating VD as a potential biomarker for preclinical AD should focus on longer periods of time for which the participants’ Aβ status is known, to be able to more accurately track these changes in the preclinical phase. Improved software and algorithms now available will enable not only SCP to be accurately explored but also other microvascular areas that could have the potential to reveal new microvascular retinal biomarkers for preclinical AD. 

Acknowledgments

Supported by grants from the European Union/European Federation of Pharmaceutical Industries and Associations (EU/EFPIA) Innovative Medicines Initiative Joint Undertaking (115372); GE HealthCare (PET tracer); and Optina Diagnostics. The sponsors had no role in the design or conduct of this research. 

Disclosure: K.R Curro, Optina Diagnostics (F); R.M.A van Nispen, Janssen-Cilag NV (C); A. den Braber, None; E.M van de Giessen, None; J.A. van de Kreeke, None; H.S. Tan, None; P.-J. Visser, None; F.H. Bouwman, Optos (F), Optina Diagnostics (F), Roche (C), Biogen (C); F.D. Verbraak, None