Alzheimer’s disease (AD) is the most common form of dementia, accounting for 60–70% of all dementia cases. AD is a progressive neurodegenerative disorder causing irreversible deterioration in cognitive function secondary to neuronal cell death and brain atrophy [1]. The main pathological features are the accumulation of two proteins: amyloid-β (Aβ) peptide, generated from its parent molecule, the amyloid precursor protein (APP), which aggregates into extracellular plaques, and hyper-phosphorylated tau, which forms intracellular neurofibrillary tangles [2]. Evidence now implies that neuropathological changes involving the accumulation of Aβ occur up to 20 years prior to the emergence of clinical symptoms, emphasising the importance of developing methods for early diagnosis [3].

Currently, diagnosis is based primarily upon cognitive assessments of patients presenting with symptomatic features of cognitive and behavioural change [4, 5]. On the basis of these assessments, an individual may fall into one of three categories: (1) probable AD dementia, (2) possible AD dementia or (3) probable or possible AD dementia with evidence of the AD pathophysiological process [5]. Computed tomography and/or magnetic resonance imaging (MRI) are frequently used clinically in the initial diagnosis of cases where probable AD is likely, whereas positron emission tomographic (PET) amyloid imaging is less frequently included where confirmation is needed; however, this is not universal [6]. Johnson and colleagues [7] devised a set of criteria that an individual must meet before PET amyloid imaging can be considered ‘appropriate’. These criteria include (1) a cognitive complaint, (2) possible AD diagnosis and (3) whether the knowledge of the presence or absence of Aβ will increase diagnostic accuracy or certainty. However, PET is not applicable for use in population-wide screening, owing to the relatively high cost of this test. Besides PET amyloid imaging, cerebrospinal fluid (CSF) pathophysiological markers, including Aβ_1–42, total tau and phosphorylated Tau, have also shown high specificity in confirming AD pathophysiology [4].

Because the retina is considered an extension of the central nervous system (CNS) via the optic nerve, researchers have investigated it to determine whether it can serve as a marker for AD change [8]. The advantages of including the retina in investigations for diagnosing probable AD include its low cost, easy accessibility and the non-invasive nature of the tests. Recently, researchers have identified Aβ deposits in eyes from both human AD and transgenic mouse models. Retinal imaging techniques for Aβ are being investigated for detection of potential biomarkers for preclinical AD [9]. Although human studies have identified retinal abnormalities in patients with AD, such as reduced retinal nerve fibre layer (RNFL) thickness [10] and degeneration of retinal ganglion cells [11], these deficits are not AD-specific and are seen in patients with other degenerative diseases, such as glaucoma [12]. Animal models including APP/PS1-transgenic mice show Aβ plaques depositing in the retina, and these plaques have been correlated with plaque load in the brain [1]. The investigators in the latter study also reported detecting the amyloid burden in the retina at 2.5 months of age, which preceded the brain deposition occurring from 5 months of age.

Researchers have also identified a potential association between changes in vascular parameters, both static and dynamic, and AD changes. These include retinal vessel diameter [13, 14] as well as systemic cardiovascular measurements of blood pressure and pulse wave velocity (PWV) [15, 16]. Amyloid plaques have been reported to be closely located to blood vessels in the brain [17]. Systemic changes in parameters such as increased blood pressure [15] and arterial stiffness [16] have been correlated with declining cognition. In APP/PS1-transgenic mice, increased blood pressure was associated with higher concentrations of Aβ plaques in the brain. In addition, endothelial cell function includes regulation of APP, which could support the concept of increased mechanical stress as a contributing factor in the overproduction of Aβ in the vasculature.

Therefore, in this study, we aimed to determine if there was an association between retinal vascular parameters and neocortical Aβ concentrations. Vascular changes and PET amyloid scan results were documented in an elderly group of participants who had subjectively reported some memory issues but had no clinical signs of AD confirmed in testing. Vascular changes identified were then also investigated in a known group with clinically diagnosed AD dementia.

In this study, we investigated the association between retinal haemodynamics and cerebral Aβ load in a cohort of elderly participants with no clinical diagnosis of AD but who had subjective memory complaints, and we subdivided these for further analysis according to PET scan results (elderly control subjects [NAL <1.4], preclinical AD [NAL ≥1.4]). Participants with confirmed clinical signs of AD were also included to replicate the trend observed in our preclinical AD cohort. We observed a negative correlation between amplitude of RVP and Aβ SUVR in the preclinical cohort. The correlation was positive when we compared RAP amplitude and Aβ SUVR. When data from our clinical AD cohort were added for comparison, with the assumption that these patients would have an Aβ SUVR greater than 2.0 [20], we observed a continuing trend in our control subjects and preclinical AD cohort, further supporting the association between retinal haemodynamics and cerebral Aβ load.

Recent studies have shown a significant association between measures of vascular function (PP, arterial stiffness, endothelial function, carotid intima-media thickness, endothelial cell response to stress) and cognitive function [16, 21, 22]. Mechanical forces on endothelial cells result in activation of a range of cell-signalling pathways as well as alterations in cell morphology, cell function and gene expression [23]. In contrast, effects of dynamic forces due to pulsatile pressure and flow result in cyclic strain on the cell membrane. Cyclic stretch has also been shown to induce retinal expression of vascular endothelial growth factor and has been implicated in the exacerbation of diabetic retinopathy by high blood pressure [24].

Recent studies have highlighted the link between endothelial nitric oxide (NO) and cerebrovascular function, in modulation of the processing of APP, in affecting the functional status of microglia, and on cognitive function [25]. Loss of NO in cultured human cerebrovascular endothelium has been shown to increase the expression of APP and β-site amyloid precursor protein-cleaving enzyme 1 (BACE1), thereby resulting in increased secretion of Aβ peptides (Aβ_1–40 and Aβ_1–42) [26]. Increased expression of APP and BACE1, as well as increased production of Aβ peptides, was also detected in the cerebral microvasculature and brain tissue of endothelial nitric oxide synthase-deficient mice.

Collectively, the higher retinal arterial pulse amplitude observed may be due to mechanical cyclic stretch (to simulate the mechanical pulsatile effect of arterial pressure) in response to upregulation of APP expression as well as deposition of Aβ plaques. Further studies are required to elucidate underlying mechanisms involved in regulating retinal arterial pulse amplitude in AD. It may be a primary risk factor, or it may occur secondary to local endothelially induced changes.

With respect to static retinal imaging, many researchers have investigated the correlation between retinal vascular calibre and cognitive decline [27–29]. However, the limitation with static measures is that often the biomarkers observed are not specific to the disease and can be seen in other diseases that may affect retinal pathology (e.g., hypertension, diabetes). In the Rotterdam study, researchers reported that larger retinal venular calibres were associated with an increased risk of dementia [30]. These findings are contradictory to those of another study in which investigators reported that narrower venular calibre is associated with likeliness of AD development [28]. In our cohort, we did not observe any significant association between retinal venous and arterial diameters with Aβ SUVR. However, studying dynamic indices of retinal vessels, including retinal pulsatility and local arterial PWV, within retinal vessels (a measure of arterial stiffness) is more likely to be applicable in identifying those at risk.

In this study, we also investigated endothelial dysfunction in retinal vasculature and its association with neocortical Aβ scores. Endothelial dysfunction plays a significant role in the pathogenesis of vascular abnormalities, and various feedback mechanisms are involved in retinal autoregulation. Previous studies have demonstrated that NO is involved in retinal vasodilation during flicker light stimulation [31, 32]. Although we observed significantly lower vasodilation in retinal arteries and veins in the clinical AD cohort in comparison with the participants having an Aβ SUVR less than 1.4, we did not see any significant association between the percentage of retinal vasodilation during flicker stimulation and Aβ SUVR.

The origin of spontaneous RVP in the eye is debated, with some researchers suggesting abnormalities in the ocular perfusion pressure in combination with high central retinal vein pressures as the main factor [33–35], whereas others argue for an imbalance in translaminar pressure gradient across the lamina cribrosa (IOP-CSF pressure) [36, 37]. Lower venous pulsations are also observed in primary open angle glaucoma [38]. Glaucoma and AD share common features of age-related risk, chronic decline in function and focal degeneration of neurons. Both diseases are characterised as degenerative disorders involving the CNS, and even though glaucoma is primarily an optic nerve disorder, secondary changes have been observed in the radiations and visual cortex. Glaucoma has been reported to be associated with AD pathophysiology [39, 40]. There are also reports suggesting that glaucoma prevalence is higher in AD (2.6 to 5.2 times) [41]; both diseases are associated with microbleeding [42, 43]. In our cohort, we did find a higher-than-average incidence of subjects with glaucoma (8 of 73 [about 11%]), but subanalysis did not identify this group of participants as showing any significant difference in parameters other than thinner RGCL/RNFL, as expected.

Retinal structural changes, including RNFL and RGCL thickness, have been reported in AD [44–46]. Although most reports of thinning of RNFL and GCL are associated with AD severity, some suggest a non-significant correlation between retinal structures and AD severity [45, 47]. Our findings show a non-significant correlation between RNFL thickness and Aβ SUVR. When assessing macular RGCL, we observed only a slightly significant reduction in the thickness when we compared patients with clinical AD with the elderly control subjects and participants with preclinical AD. The results were not changed when the subjects with glaucoma and macular degeneration were excluded.

The vascular contribution to the broad range of cognitive impairment leading to AD has been recognised [48, 49], including by professional societies providing statements based on epidemiological evidence linking loss of vascular function and development of dementia [50]. Studies of associations of vascular function and cognitive impairment are converging on the question whether AD can be considered part of the spectrum of vascular disease [51]. There is accumulating evidence suggesting an association of arterial blood pressure, arterial stiffness and reduced endothelial function with AD [22, 52], but there have been conflicting findings in the literature, and it is clearly not a simple relationship with arterial blood pressure per se. Infusing hypertensive doses of angiotensin II in APP/PS1-transgenic mice produced a 100% increase in soluble Aβ in the brain and a 25% reduction in cerebral microvessel density [53]. We assessed central PP (systolic − diastolic) and PWV and their association with Aβ SUVR. No significant association or correlation was observed between these cardiovascular parameters and neocortical Aβ scores.

Limitations of our study include that we did not have neuroimaging data from our clinically confirmed AD group, because we could not ethically justify scanning these patients. It was anticipated that 80% of participants with clinical AD would have high PET scores (>2). Additionally, there might be a false-positive rate within our cohort (i.e., healthy elderly subjects with SUVR >1.4) leading to classification as preclinical AD. Whereas this would not affect the association between SUVR and retinal vascular changes, future scans and recording will elucidate the number of false-positives within the cohort. We also acknowledge that a large proportion of our cognitively healthy participants were receiving treatment for systemic hypertension. In this age group, this is extremely common, but if our hypothesis is correct that there is a link with arterial pulse waves and chronic brain damage, then this treatment may be most likely masking the effect.

This study demonstrates a significant correlation between amplitude of retinal vascular pulsatility and neocortical Aβ scores, independent of other risk factors. Arterial pulsation was increased while venous pulsation was reduced. Future studies will further elucidate whether these biomarkers have a correlation not only with cerebral Aβ plaques but also with the development of subsequent clinical dementia, because these participants are all being followed longitudinally. If a clinical correlation is confirmed, although direct cause and effect cannot be certain, screening of eyes in those considered at risk of AD may aid in the development of an additional inexpensive approach for identifying increased risk as well as serving to potentially monitor vascular function in these individuals.