Retinal mid-peripheral capillary free zones are enlarged in cognitively unimpaired older adults at high risk for Alzheimer’s disease

Fifty-seven (57) eyes of 37 CU low-risk (age; mean: 66 years; range: 56–80 years; 18 males and 19 females) and 50 eyes of 38 CU high-risk older adults (age; mean: 64 years; range: 55–77 years; 11 males and 27 females) were involved in this current study (total of 107 eyes in 75 participants). There were no significant differences in age; t(73) = 1.19, p = 0.24 or proportion with respect to sex; X² (1, N = 75) = 3.07, p = 0.08 between the two groups. All participants had refractive errors of ≤  ± 5.00 DS (spherical equivalent; equivalent axial length of ~ 21–26 mm) to prevent significant differences in retinal magnification in the OCTA images as noted by the Bennett formula [38]. Also, all CU participants had Montreal Cognitive Assessment (MoCA) scores of ≥ 26 [39, 40] and Repeatable Battery for the Assessment of Neuropsychological Status Update (RBANS-U) Delayed Memory Index (DMI) scores of ≥ 85 [41, 42]. CU high-risk participants were defined as individuals with at least one allele of the APOE e4 gene and a first-degree family history of AD while CU low-risk participants were non-carriers for the APOE e4 allele and no first-degree family history of AD. The inclusion criteria for participants in the study were as follows: absence of or controlled hypertension (< 140/90) and hyperlipidemia and no systemic diabetes (HbA1c ≤ 7). The exclusion criteria for our study were as follows: unstable doses of antidepressants that have significant anticholinergic side effects; age-related macular degeneration; diabetic retinopathy; glaucoma; retinal ischemic conditions; large cataracts that will impede imaging; current intake of retino toxic drugs such as chloroquine, hydroxychloroquine, and cancer drugs; and other neurodegenerative diseases such as Parkinson’s disease or multiple sclerosis. Smoking history has previously been shown to have a significant effect on retinal vasculature [43]. There was no significant difference in the total years of smoking between the two groups, p = 0.22. All participants had best corrected visual acuity of ≥ 20/40 (~ LogMAR 0.30). An estimated sample size (N = 72) was computed with a GPower 3.1 calculator [44] using the following input parameters from a previous study in preclinical AD [30]: effect size (CU low risk vs. CU high risk; Cohen’s d) of 0.60, α of 0.05, and power of 0.80. Considering 107 eyes of 75 participants were involved in this study, our study was adequately powered to detect the differences in the OCTA parameters. The study adhered to the tenets of the Declaration of Helsinki, and informed consent from all subjects was obtained prior to the experimental data collection after the explanation of the nature and possible consequences of the study. The study was part of the Atlas of Retinal Imaging in Alzheimer’s Study (ARIAS; PJS served as principal investigator for ARIAS) which took place at the University of Rhode Island and Butler Hospital Memory and Aging Program, Providence, RI, between 2020 and 2022 and was approved by the BayCare Institutional Review Board (IRB).

All study participants underwent detailed neuropsychological evaluation with MoCA [39, 40] and RBANS-U [41, 42]. The MoCA test [39, 40] is a one-page, 30-point test administered in approximately 10 min. A score of 26 or over is considered CU. It assesses several cognitive domains including short-term memory, visuospatial ability, executive function, attention, concentration and working memory, language, and orientation to time and place. The total MoCA score was assessed for the participants.

The RBANS-U [41, 42] is a brief neuropsychological assessment battery that can be administered to adult patients aged 20–89 years old. The RBANS consists of 10 subtests, which give five scores (one for each of the five domains tested), including immediate memory (immediate list learning and immediate story), visuoperceptual abilities (figure copy and line orientation), language (naming and semantic fluency), attention (digit span forward and digit-symbol coding), and delayed memory (delayed list memory with recognition, delayed story memory, and delayed figure memory). It takes about 30 min to administer. The RBANS-U DMI scores were assessed for our study participants.

DNA was taken from the buccal samples of epithelial tissues collected from the inside of the cheek for each study participant. For reproducibility purposes, a total of 2 swabs were collected from each study participant. The sample was placed into a reagent tube, and both samples were inserted into the DNA analyzer cube (Spartan Bioscience Inc., Montreal, Canada). The test system integrated and automated DNA extraction using PCR-based amplification of the APOE target gene. The system had integrated controls for monitoring run performance and automatically informed the operator of any anomalies in the instrument or reagent. For the purpose of the study, samples were analyzed in real time, and the results were available to the investigator in less than 1-h. Low-risk participants had no allele of the APOE e4 gene (i.e., participants had e2 or e3 gene). High-risk participants had at least one allele of the APOE e4 gene (i.e., e2/e4, e3/e4, or e4/e4).

Prior to imaging, all participants were dilated with two drops of tropicamide (Mydriacyl 1%) per eye. There was a 15-min wait time from the administration of the dilation drops to image acquisition. All imaging procedures were completed for both the right eye and the left eye. We obtained 20 × 20°OCTA images consisting of 512 b-scans, 512 A-scans per b-scan, 12 µm spacing between the b-scans, and 5 frames averaged per each b-scan location of the central fovea and of paired major arterioles and venules with their surrounding capillaries inferior to the fovea (Spectralis HRA + OCT; Eye Explorer version 1.10.4.0; Heidelberg Engineering, Germany; Fig. 1) as done previously [22, 23]. The signal quality values (range for Spectralis = 0–40) of all our OCTA images from the vendor software were at least 20 to ensure good image quality. In addition, all scans were visually inspected for motion and shadowing artifacts. To increase our sample size, both eyes from participants were selected for the purpose of image analysis so far as they had good image quality. In cases, where only one eye had good image signal quality, only that eye was selected.

Raw OCTA images of the superficial vascular plexus (SVP) were exported from the HEYEX software as.tiff files (5.7 μm/pixel lateral resolution) into a custom programming software (MATLAB, Mathworks) where they were automatically cropped to eliminate the infrared component and include only the angiogram (Fig. 1a). The SVP was defined as the composite retinal vasculature from the inner limiting membrane to the inner plexiform layer/inner nuclear layer boundary [22, 23]. Images of the central foveal region and those of paired arterioles and venules inferior to the fovea were then automatically montaged using an image processing software (i2k Retina software) to generate a wider field of view of the paired vessels (Fig. 1a). The montaged images were then also exported into a custom programming software (MATLAB, Mathworks) for further processing. First, a vesselness filter was applied to the images [45, 46] to increase the probability of resolving a vessel at a specific location in the image when it is actually present versus noise or motion artifact (Fig. 1b). Next, Otsu thresholding method [47] was applied to the resultant image to reduce background noise (Fig. 1c) as have been done previously [23]. For all participants, we analyzed the mid-peripheral CFZs of the montaged images (Figs. 1 and 2) in an approximately linear pattern evenly along each sampled major arteriole or venule (first and second order branches). The average vessel distance from the fovea, vessel diameter, and vessel linear distance of sampling did not differ between the two groups, all p > 0.05 (Table 1).

For vessel density computation, only the 20 × 20° macular-centered OCTA images were processed (cropped, vesselness filtered, and Otsu thresholded) for further image analysis (Fig. 3).

Euclidean distances for the mid-peripheral CFZs, vessel distance from fovea, and vessel diameter were computed using similar formulas previously applied for CU young and older adults (Eqs. 1–3) [22, 23]. In this current paper, we also computed the linear distance of sampling (length of the vessel used for CFZ computation) for the mid-peripheral CFZs along the sampled arteriole or venule (Eq. 4). Briefly, we computed the mid-peripheral CFZ width in microns as the linear distance from the edge of an arteriole (periarteriole CFZ) or venule (perivenule CFZ) to the middle of the nearest capillary. The middle of the nearest capillary was used instead of the edge of the capillary because OCTA does not have enough lateral resolution to truly resolve the edge of a lumen of a capillary compared to other advanced retinal imaging modalities such as adaptive optics scanning laser ophthalmoscope (AOSLO). A custom MATLAB program automatically recorded to an Excel file the x and y coordinates from points evenly sampled perpendicular to an arteriole or venule and the middle of the nearest capillary (Fig. 2). The evenly sampled points in MATLAB included points perpendicular above and below a paired arteriole or venule as well as the center of the fovea.

Equations 1–4 were then used to compute the Euclidean distances for the mid-peripheral CFZs (periarteriole and perivenule CFZ), vessel distance from the fovea, vessel diameter, and vessel linear distance of sampling, respectively, using the x and y coordinates written by MATLAB into Excel. The outcome of each equation produced an Euclidean distance in pixels. For the mid-peripheral CFZ width (Eq. 1), vessel diameter (Eq. 3), and vessel linear distance of sampling (Eq. 4), the Euclidean distances in pixels were then converted into microns by multiplying them by the micron-to-pixel ratio in the x and y directions, as computed from the vendor software fiducial marks (Heidelberg Engineering, Table 1). For the vessel distance from the fovea (Eq. 2, Table 1), each Euclidean distance in microns was divided by 300 to convert them into degrees (assuming 300 µm =  ~ 1°).where large vessel refers to a paired arteriole or venule, and capillary represents the middle of the nearest capillary (Fig. 2).where vessel top refers to the top of a paired arteriole or venule, and fovea represents the middle of the FAZ (Fig. 2).where vessel top/bottom refers to the top/bottom of a paired arteriole or venule (Fig. 2).where vessel top refers to the top of a paired arteriole or venule, and n refers to a series of evenly sampled neighboring coordinates (Fig. 2).

The area-finding tool (lasso tool) of the vendor software was used to delineate and compute the FAZ area in mm². FAZ effective diameter was then computed from the FAZ area values. The FAZ effective diameter in microns was defined as the diameter of a circle whose area was equivalent to the known FAZ areas; FAZ effective diameter = (4*FAZ area/π)^1/2, similar to that done for the CU young and older adults [22, 23].

Vessel density computation has been previously described [48]. Briefly, after the macular-centered 20 × 20° images have been processed (cropped, vesselness filtered, and Otsu thresholded; Fig. 3a) as described above, reversed contrast/negatives of those images were created (Fig. 3b). A customized MATLAB script was then used to count the number of white pixels in both the original processed (Fig. 3a; designated as vessels) and the reversed contrast images (Fig. 3b; designated as the non-perfused background). The number of pixels was then converted to mm² based on the micron‐to‐pixel ratio in the x and y directions, as computed from the fiducial marks acquired from the HEYEX software as done previously [22‐24]. The area of white pixels in mm² designated as vessels was then added to that of the background to compute the total area of the image which was expected to be ~ 36 mm². The vessel density was then computed as the ratio of the area of white pixels designated as vessels (Fig. 3a) to the total area of the image.

All statistical analyses were completed using statistical software (IBM SPSS Statistics for Windows, version 29; IBM Corp., Armonk, NY, USA). All values were descriptively presented as mean ± SD. Considering a sample size of 107 eyes with kurtosis and skewness of ≤  ± 3.50 for all our outcome variables, normality was assumed, and parametric tests were performed for the data analyses. We previously found the perivenule CFZ to be significantly positively associated with vessel distance from the fovea and vessel diameter in CU young adults [22]. In this current study, vessel distance from the fovea, vessel diameter, and vessel linear distance of sampling served as covariates when comparing the periarteriole and perivenule CFZ between the two groups of participants (Table 1). An independent sample t test was performed to compare the covariates (Table 1) between the two groups. Since the covariates did not significantly differ between the two groups (Table 1), the periarteriole and perivenule CFZ width was compared between the high-risk and low-risk participants using an independent sample t test. A similar independent sample t test was used to compare the FAZ size, FAZ effective diameter, and vessel density between the two groups. Cohen’s d was used to measure the effect size. A paired sample t test was used to compare the periarteriole versus perivenule CFZ in both the low-risk and high-risk CU older adults. A logistic regression model combining the periarteriole and perivenule CFZ (mid-peripheral CFZs) was initially used to provide predictive/probability values. A receiver operating characteristic (ROC) model with these predictive/probability values was then used to assess the sensitivity, specificity, and area under the curve (AUC) to distinguish between high-risk and low-risk participants. Another logistic regression model combining the mid-peripheral CFZs, FAZ effective diameter, and vessel density was also utilized to provide predictive/probability values. FAZ effective diameter was chosen instead of FAZ size because we found the former to be less variable than the latter in our previous study [23]. A second ROC model with these predictive/probability values assessed the sensitivity, specificity, and AUC to distinguish between high-risk and low-risk participants. In both ROC scenarios, a Youden’s index that maximizes sensitivity and moderates specificity was chosen to create cutoffs. Such a Youden’s index was chosen because the goal of our study is to develop a screening test, and hence, such a test/model should have good sensitivity even if it has moderate specificity. The following AUC classification was used for our study; 0.5–0.6 = unsatisfactory, 0.6–0.7 = satisfactory, 0.7–0.8 = good, 0.8–0.9 = very good, and 0.9–1 = excellent. A p-value of < 0.05 was considered statistically significant.

In low-risk CU older adults, the periarteriole CFZ width (71.3 ± 7.07 µm; range = 58.1–92.1 µm; 95% CI for mean = 69.5–73.2 µm; SEM = 0.94) was significantly larger than the perivenule CFZ width (57.3 ± 6.40 µm; range = 47.4–86.3 µm; 95% CI for mean = 55.6–59.0 µm; SEM = 0.85), t(56) = 13.1, p < 0.001 (Fig. 4). Similarly in high-risk CU older adults, the periarteriole CFZ width (75.8 ± 9.19 µm; range = 60.0–111 µm; 95% CI for mean = 73.2–78.4 µm; SEM = 1.30) was significantly larger than the perivenule CFZ width (60.4 ± 8.55 µm; range = 48.2–93.2 µm; 95% CI for mean = 58.0–62.8 µm; SEM = 1.21), t(49) = 13.3, p < 0.001 (Fig. 4).

The mean periarteriole CFZ width of the high-risk CU older adults (75.8 ± 9.19 µm) was significantly larger than that of the low-risk CU older adults (71.3 ± 7.07 µm), t(105) =  − 2.85, p = 0.005, with a medium effect size (Cohen’s d) = 0.55 (Figs. 4, 5, and 6). The mean periarteriole CFZ width (75.8 µm) of the high-risk CU older adults was outside the 95% CI of the low-risk CU adults (69.5–73.2 µm). There was no overlap between the 95% CI of the mean periarteriole CFZ width of the high-risk group (73.2–78.4 µm) and that of the low-risk group (69.5–73.2 µm). Similarly, the mean perivenule CFZ width of the high-risk CU older adults (60.4 ± 8.55 µm) was significantly greater than that of the low-risk CU older adults (57.3 ± 6.40 µm), t(105) =  − 2.15, p = 0.034, with a small effect size (Cohens’ d) = 0.42 (Figs. 4, 5, and 6). Also, the mean perivenule CFZ width of the high-risk CU older adults (60.4 µm) was outside the 95% CI of the low-risk CU (55.6–59.0 µm). However, there was an overlap between the 95% CI of the mean perivenule CFZ width of the high-risk group (58.0–62.8 µm) and that of the low-risk group (55.6–59.0 µm).

Processed OCTA images in MATLAB are shown in Fig. 6 to demonstrate an example of large mid-peripheral CFZs in a high-risk CU older adult versus a low-risk CU older adult.

The FAZ area of the high-risk CU older adults (0.345 ± 0.142 mm²) did not significantly differ from that of the low-risk CU older adults (0.361 ± 0.115 mm²), t(105) = 0.64, p = 0.52. Similarly, the FAZ effective diameter for the high-risk CU older adults (648 ± 139 µm) did not significantly differ from the low-risk CU older adults (669 ± 110 µm), t(105) = 0.86, p = 0.39. Vessel density also did not differ between the high-risk group (0.507 ± 0.039) and the low-risk group (0.497 ± 0.041), t(105) =  − 1.27, p = 0.21.

The ROC model combining the periarteriole and perivenule CFZ width (mid-peripheral CFZs) was statistically significant with an AUC = 0.65 (95% CI = 0.55–0.76), p = 0.006 (Fig. 7). A Youden’s index of 0.244 (which maximizes sensitivity and moderates specificity) was chosen to create a cutoff predictive value of 0.423. The cutoff predictive value corresponded to a periarteriole CFZ width of ≥ 71.2 µm and perivenule CFZ width of ≥ 56.8 µm for the high-risk positive state with a sensitivity of 70% and a specificity of 54%.

Interestingly, an ROC model which combined the mid-peripheral CFZs, FAZ effective diameter, and vessel density yielded a better model which was also statistically significant with an AUC = 0.70 (95% CI = 0.60–0.80), p < 0.001 (Fig. 7). A Youden’s index of 0.314 was chosen to create a cutoff predictive value of 0.443. The cutoff predictive value corresponded to a periarteriole CFZ width of ≥ 72.2 µm, perivenule CFZ width of ≥ 66.6 µm, FAZ effective diameter of ≥ 731 µm (FAZ size of ≥ 0.42 mm²), and vessel density of ≥ 0.47 for the high-risk positive state with a sensitivity of 70% and a specificity of 61%.