Background
Globally, individuals with Alzheimer’s disease (AD) dementia, mild cognitive impairment (MCI) due to AD, and preclinical AD are estimated at 32, 69, and 315 million, respectively. Altogether, these constitute 416 million people across the AD continuum or 22% of all persons aged 50 and above [
1]. AD is the most common cause of dementia in the elderly, and a progressive neurodegenerative disease affecting approximately 6.7 million Americans aged 65 and older, and projected to reach 13.8 million by 2060 [
2]. It is in the top 10 leading causes of death in America and has no proven preventative or curative interventions [
3]. Early diagnostics are critical for the development of effective therapies. The pathophysiological process of AD occurs decades before symptoms of dementia emerge [
4‐
9]. It is therefore critical to develop non-invasive/cost-efficient biomarkers to aid in the early detection and interventions to prevent or delay dementia onset. While positron emission tomography (PET) and cerebrospinal fluid (CSF) assessment via lumbar puncture have the greatest utility, they are not routinely used because they are expensive and invasive. While rapid advances in blood-based biomarkers will likely become part of the normal clinical diagnostic pathway within the next few years [
10‐
13], there is still the need for other non-invasive biomarkers.
There is increasing evidence that there are retinal manifestations of AD; the foveal avascular zone (FAZ) area is enlarged; retinal vessel density, central macular, and choroidal thickness are reduced in individuals with a genetic risk (apolipoprotein E; APOE e4) and first-degree family history of AD [
14‐
17]. The eye and brain are anatomically, embryologically, and physiologically linked. The retinal ganglion cells (RGCs) are similar to the cerebral cortex neurons, and the cerebral small vessels are similar to retinal vessels [
18,
19]. The human retina is an easily accessible part of the central nervous system (CNS) and an ideal target for the identification of AD risk biomarkers.
The FAZ and vessel density areas proposed as retinal vascular biomarkers for early AD detection all have limitations that decrease their effect size or clinical relevance for early disease detection [
20‐
24]. One issue with using the FAZ area as a biomarker for AD is that it is limited to a few or a single layer of capillaries. Moreover, it may become saturated with increasing disease severity and not increase further in size, even with the loss of capillaries [
22,
23]. FAZ area is highly variable in cognitively unimpaired (CU) older adults [
23,
25‐
28], and studies have shown that the FAZ area is not consistently significantly larger in an AD cohort than a control cohort [
29‐
32]. Also, vessel density measurements in AD patients using optical coherence tomography angiography (OCTA) [
29,
33‐
35] are influenced by noise in the image, along with variable anatomic features such as vessel diameter [
20,
21,
23].
In previous studies, we characterized and quantified the metrics of tissue oxygenation in the retina of young and older CU participants in the form of periarteriole and perivenule capillary free zones (mid-peripheral CFZs) [
22,
23]. The mid-peripheral CFZs represent the maximum distance that oxygen and nutrients must diffuse to reach the retinal neurons with larger distances indicating potential ischemia [
22,
23]. There is a breakdown of the inner retinal blood barrier, pericyte loss, and capillary non-perfusion or dropout in AD [
36,
37] leading to potential enlargement of the mid-peripheral CFZs around the arterioles and venules in the retina of AD patients.
The goal of the current study was twofold: (1) determine if the mid-peripheral CFZ is a more robust biomarker for early AD risk detection compared to FAZ and vessel density measurements and (2) assess whether a model of combined metrics for mid-peripheral CFZ, FAZ, and vessel density will better distinguish between low-risk and high-risk CU older adults for AD compared to a model of the mid-peripheral CFZs alone. Thus, this proof-of-concept study explored a novel non-invasive, inexpensive retinal vascular biomarker and a model of retinal vascular metrics for their potential to assist with early AD risk detection and disease monitoring. Such multimodal measures of retinal abnormalities may facilitate large-scale screening of older adults and referral of at-risk individuals by point-of-care clinicians to neurologists and neuropsychologists for detailed cognitive health/biomarker assessment.
Discussion
In our current study of high-risk for AD defined by the presence of at least one APOE e4 allele and a first-degree family history of AD, we found statistically significant larger periarteriole and perivenule CFZs (mid-peripheral CFZs) in the high-risk CU older adults compared to the low-risk CU older adults. FAZ and vessel density metrics did not significantly differ between these two groups. The moderate effect size for the periarteriole CFZ shows that it has better potential to serve as a future clinical biomarker than the small effect size we found for the perivenule CFZ. A statistically significant satisfactory ROC model including the mid-peripheral CFZs distinguished between the low- and high-risk CU older adults, which was modestly better with increased specificity when a multimodal ROC model combined the mid-peripheral CFZs with other retinal vascular metrics (FAZ effective diameter and vessel density) for a good AUC to distinguish between the two groups. Our data provide cutoff predictive values for periarteriole and perivenule CFZ widths for the high-risk positive state in this multimodal model to yield a sensitivity of 70% and a specificity of 61%.
The inner retinal blood barrier essentially controls nutrient flow to the neural retinal; specifically, the inner retinal neurons [
49]. A mid-peripheral CFZ represents the distance that oxygen, and nutrients must diffuse to reach the neural retina, with larger distances indicating potential ischemia [
22,
23]. In addition, the mid-peripheral CFZs have an anatomical resemblance to the perivascular spaces seen in the brain parenchyma (Virchow-Robin spaces). These spaces play an important role in nutrient distribution and may be a key element of the recently described glymphatic pathway; a network of perivascular spaces involved in the removal of cerebral solutes and cell byproducts such as beta-amyloid (Aβ) and tau [
50‐
52]. In AD, breakdown of the inner retinal blood barrier, pericyte loss, and capillary non-perfusion or dropout occur [
36,
37] leading to potential enlargement of the mid-peripheral CFZs around arterioles and venules in the retina. Also, there is dilatation of the cerebral perivascular spaces (a postulated indirect neuroimaging biomarker of impaired glymphatic function) in AD patients as shown by magnetic resonance imaging (MRI) [
50‐
52], indicating possible changes in the perivascular spaces in the retina (mid-peripheral CFZs) of these patients. APOE e4 genotype has been associated with vascular impairment in AD, and it is an important risk marker for abnormal Aβ accumulation and impaired clearance within the brain vasculature [
53‐
56], positing similar changes in the retinal vasculature of these patients. Building on these prior studies and in support of our hypothesis, we found evidence that the mid-peripheral CFZs were significantly enlarged in CU older participants at high risk for AD compared to age-matched low-risk CU older participants, which posits similar perturbation in the retinal vasculature as has been reported in the brain [
50‐
56]. Larger retinal mid-peripheral CFZs in the high-risk group indicates large spaces or passageways around the retinal arterioles and venules in these patients for diffusion of oxygen, nutrients, and other waste products of metabolism between the retinal vascular and neural system.
Perivascular spaces in the brain basically include gaps or passageways around arterioles, capillaries, and venules along which a range of substances can move [
50‐
52] similar to the structure and function of the retinal mid-peripheral CFZs (retinal perivascular spaces) [
22,
23]. However, it is currently under debate whether MRI-visible perivascular spaces surround both arterioles and venules [
57‐
59]. Most MRI at conventional strengths cannot easily distinguish between perforating arterioles and venules [
50]. The use of 7-T MRI has demonstrated that MRI-visible perivascular spaces are spatially more correlated with arterioles but not venules [
60]. The use of lower field T2 sequence (if images are of good enough quality) to visualize perivascular spaces and venules in the centrum semiovale [
50,
61] suggests that perivascular spaces are distinct from venules [
50,
62]. Thus, most evidence in the literature suggests that MRI-visible perivascular spaces are periarteriolar rather than perivenular [
60‐
62]. Since the human retina is an extension of the brain, and perivascular spaces in the brain are more periarteriolar than perivenular [
60‐
62], this may explain the moderate effect size for the periarteriole CFZ compared to the small effect size for the perivenule CFZ.
We did not find significant differences between the two AD risk groups with respect to FAZ size, FAZ effective diameter, and vessel density. The lack of significant differences in the FAZ metrics in our study is similar to that reported in previous studies that compared the FAZ metrics between participants with MCI and controls [
29,
31], as well as preclinical AD and controls [
30,
32]. The lack of significant differences in the FAZ metrics between the two groups in our study could be explained by the large individual variability in the FAZ metrics even in the CU older adult population [
23,
25‐
28] leading to the overlaps between the two groups. Interestingly, a paradoxical smaller FAZ size has been reported in participants with genetic risk for AD (APOE e4) compared to those without in a previous study [
15]. The lack of significant differences in vessel density between the two groups is also similar to that reported in previous studies that compared vessel density metrics between controls and preclinical AD [
32], as well as MCI and controls [
29,
31]. Even in studies that found significant differences, these differences were found in the later stages of the disease (AD vs. MCI, and AD vs. controls) rather than in the early stages [
31,
33]. A longitudinal study reported reduced baseline retinal vessel density metrics in APOE e4 carriers compared to non-carriers, but these metrics were not significantly different between the two groups after a 2-year follow-up [
16]. Interestingly, one study found a paradoxical large vessel density in preclinical AD patients compared to controls [
30]. The lack of significant differences in vessel density between the two groups could be explained by the fact that OCTA vessel density computations are influenced by noise in the image along with variable anatomic features such as vessel diameter [
20,
21,
23].
A single retinal biomarker related to either neural (retinal nerve fiber layer; RNFL thickness), vascular (vessel density and FAZ metrics), or proteinopathy changes (retinal amyloid/inclusion bodies) may not be sensitive and specific to AD. For example, RNFL thickness that has been shown to be thinner in AD [
63‐
65] is also implicated in glaucoma [
66‐
68]. Thus, it becomes possible that a retinal biomarker study for AD could include participants with early glaucoma who have not yet been formally diagnosed with glaucoma. Retinal vascular metrics, such as vessel density and FAZ size, that have been previously studied in AD [
29‐
31,
33,
34] are also implicated in diabetic retinopathy [
69‐
71]. Changes in these retinal vascular metrics precede clinically detected diabetic retinopathy (diagnosed using dilated fundus examination or color fundus images) [
69‐
71]. Thus, an AD retinal biomarker study that includes diabetics but has excluded clinical diabetic retinopathy may have participants with changes in retinal vessel density and FAZ size that are unrelated to AD. Also, retinal amyloid/inclusion bodies indicated in AD [
72,
73] are also found in retinal drusenoid structures in age-related macular degeneration (AMD) [
74]. It must however be noted that other studies have found no association between a family history of AD/APOE e4 genotype and the presence of drusen, and that amyloid deposits are distinct from drusenoid structures [
72,
75]. Without a multimodal imaging model of blue autofluorescence imaging, color fundus, and spectral domain OCT, drusenoid structures in AMD may be falsely counted as retinal amyloid or inclusion bodies. Thus, several possible retinal biomarkers that have been investigated for AD are also affected by other retinal disease processes. While a retinal vascular metric, e.g., the mid-peripheral CFZs may have advantages over other known vascular metrics (FAZ and vessel density metrics), it currently appears that the way forward as a field is to utilize a multimodal approach that combines all the vascular metrics to improve sensitivity and specificity of these metrics for early AD risk detection. This proposal is supported by our finding of a multimodal ROC model that combined the mid-peripheral CFZs with other retinal vascular metrics to yield a better ROC model to distinguish between two CU groups with different risks of AD, as well as by a previous study that investigated a multimodal model of different types of fractal and lacunarity analysis to distinguish between cognitively impaired and CU older adults [
76].
Given the universality of blood collection in medical settings, blood-based biomarkers have the potential to improve widespread access to AD screening and diagnosis in both high- and low-resource areas. These blood-based biomarkers for AD include Aβ42/Aβ40 ratio [
10], p-tau 181 [
11], p-tau 217 [
77], p-tau 231 [
78], neurofilament light protein (NfL) [
12], and glial fibrillary acidic protein (GFAP) [
13]. p-tau 217 shows better results at detecting AD pathology (including preclinical AD) and for monitoring of disease progression [
77,
79,
80]. However, the robustness of the measured Aβ42/Aβ40 ratio in blood plasma is only 0.9-fold times lower in patients with brain amyloidosis compared to controls and therefore the challenge with implementing blood-based Aβ42/Aβ40 ratio for screening purposes is the smaller effect size when compared to its CSF counterparts [
81]. This could be explained by the contamination of results based on the fact that plasma Aβ is derived from peripheral sources (downstream effect) [
82,
83] unlike the retina, which is a direct extension of the brain, and can also be influenced by genetic factors and renal function [
82]. With respect to p-tau biomarkers, the robustness ranges from low to high effect size from preclinical AD to prodromal AD, respectively, with the highest levels in AD dementia with p-tau 181 and also p-tau 231 both not performing as well as p-tau 217 [
79,
80], positing that blood-based biomarkers may do better in later rather than early disease. This could be explained by the fact that unlike the retina which is a direct extension of the brain and can detect subtle early changes in AD, blood-based biomarkers are derived via a downstream effect from the CNS. While retinal biomarkers may have some advantages over blood-based biomarkers and vice-versa, the argument today cannot be choosing one over the other, but rather investigating how both groups of biomarkers are related to each other, as well as developing a model that incorporates both biomarkers to improve the sensitivity, specificity, and AUC for early detection of AD. The next phase of our research will investigate the relationship between the mid-peripheral CFZs (especially the periarteriole CFZ) and the abovementioned blood-based biomarkers as well as a model that incorporates both biomarkers.
Our other future research endeavors will investigate the associations between the mid-peripheral CFZ width (retinal perivascular spaces; especially the periarteriole CFZ width) [
22,
23] and quantitative features (length, width, volume, etc.) of MRI-defined perivascular spaces in the centrum semiovale in AD [
50‐
52]. Our results show that the mid-peripheral CFZs are enlarged in CU older adults at high risk for AD compared to CU low-risk participants. Also, in AD, there is impairment in the drainage of fluid in the brain glymphatic system which leads to the accumulation of Aβ and dilatation of the perivascular spaces [
50‐
52]. From a mechanistic perspective, the goal will be to use mid-peripheral measures of variability in retinal vasculature (mid-peripheral CFZs) as an intriguing approach to test hypotheses about potential vascular contributions to AD.
An inherent limitation of our current study is the limits of lateral resolution of OCTA technology. However, OCTA has better axial resolution than other superior lateral resolution devices, such as AOSLO. There are new developments in the technology to improve the field of view, and speed of image acquisition, which may be valuable to improve image quality and analysis in future studies. The cross-sectional nature of this proof-of-concept study provides support for a future longitudinal study to investigate the within and between subject changes over time (especially for the periarteriole CFZs) and as well as the association with blood-based and brain AD biomarkers (PET Aβ and tau SUVR). To ensure clinical applicability of the mid-peripheral CFZs (especially the periarteriole CFZs) for early AD risk detection, the MATLAB scripts used for their computations will need to be commercialized and incorporated into the current OCTA imaging modalities in the clinic in future research. At the present time, there is little broad agreement on how best to compute vessel density from OCTA images, as processing technology is still relatively new [
48]. There is a rapidly growing number of published methods for OCTA signal analyses and data reporting, and currently little agreement on standard metrics [
48]. In our current study, we used our previously reported methodology for vessel density computation for our OCTA images [
48].
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.