A total of 982 participants were included in the analyses. Mean age-at-death was 90 years with 69% being women. Characteristics of participants are presented in Table 1. Among those with dementia, 425 participants were diagnosed with Alzheimer’s dementia and 40 were diagnosed with vascular dementia. Among 982 participants, 22% older persons had diabetes, 16% were taking diabetes insulin medications, and 6% taking insulin injections. 70% of participants had history of hypertension and 79% were taking any anti-hypertensive medication. Smoking history was present in 33% of persons, with 31% being former smokers and 2% being current smokers. The mean systolic blood pressure across all visits was 132.7 (SD = 14.0) and the mean diastolic blood pressure across all visits was 72.5 (SD = 7.7). History of claudication was present in 31%, history of stroke in 20%, and history of myocardial infarction in 21%. Watershed arteriolosclerosis was common in the brain (Fig. 3), with moderate-to-severe arteriolosclerosis pathology being more frequent in the AWS than the PWS region (45% vs. 35% of older persons). Persons with higher severity of AWS arteriolosclerosis pathology were also likely to have higher severity of PWS arteriolosclerosis pathology (X2 = 251.97, degrees of freedom = 9, p < 0.001). Both AWS and PWS arteriolosclerosis pathologies were associated with age [(F (3, 978) = 8.28, p < 0.001 for AWS] and [F (3, 978) = 12.80, p < 0.001 for PWS].

First, we examined the burden of β-amyloid and PHF-tau pathology across watershed vessel severity using Wilcoxon sum test. For persons with more severe arteriolosclerosis pathology in PWS region, PHF-tau-tangle burden was 40% higher than those with less severe vessel pathology (p < 0.001). For persons with more severe arteriolosclerosis pathology in AWS region, tau-tangle burden was 10% higher than those with less severe vessel pathology (p = 0.03) (Fig. 4). There was no significant difference between β-amyloid levels across AWS arteriolosclerosis severity (p = 0.30) or PWS arteriolosclerosis (p = 0.35). Histological sections showing arteriolosclerosis severity and tangle burden from age-matched participants are presented in Fig. 5.

Cortical β-amyloid and tau-tangle burden across vessel severity. Boxplots of β-amyloid and tau-tangle burden across severity levels of AWS/PWS arteriolosclerosis pathology

Illustrative cases. Histological sections from representative participants. Case 1 is a female participant, (age-at-death is 89.2 years), with a tau-tangle burden within the 75th–100th percentile and severe arteriolosclerosis pathology in watershed brain regions. Case 2 is a female participant (age-at-death is 89.6 years), with a tau-tangle burden within the 25th percentile and mild arteriosclerosis pathology in watershed brain regions. Images represent H&E-stained sections of the posterior watershed brain regions (i and ii) and AT8-stained sections for PHF-tau-tangle pathology in the CA1 subregion of the mid-hippocampus (iii and vi), midfrontal gyrus (iv), and inferior parietal cortex (v)

Next, we examined the association of watershed arteriosclerosis in relation to AD pathology. In linear regression models adjusted for age, sex, education, and common age-related pathologies (CAA, atherosclerosis, macroscopic and microscopic infarcts, TDP-43, and Lewy body pathology), more severe PWS arteriolosclerosis pathology was associated with a higher burden of PHF-tau-tangle pathology. There was no significant association between PWS arteriolosclerosis and cortical β-amyloid or between AWS arteriolosclerosis and cortical β-amyloid/PHF-tau-tangle pathology (Table 2). Because both β-amyloid and tau-tangle pathology have stereotypical progression patterns in the brain, subsequent analyses examined whether associations were driven by regional burden; specifically, with mesial temporal and neocortical burden. Adjusted for demographics and common age-related pathologies, AWS arteriolosclerosis was not associated with neither mesial temporal nor neocortical β-amyloid/tau-tangle burden, while PWS arteriolosclerosis was associated with only neocortical tau-tangle burden (Table 2).

Prior studies have shown vascular risk factors to be associated with β-amyloid and tau burden. In sensitivity analyses, we further adjusted models for overall vascular risk burden (estimate = 0.11, SE = 0.04, p = 0.01), and in separate models, we adjusted for demographics and average systolic (estimate = 0.15, SE = 0.05, p = 0.001) and diastolic blood pressure (estimate = 0.15, SE = 0.05, p = 0.001). Notably, the association between PWS arteriolosclerosis and tangle burden in all sensitivity analyses remained unchanged. Using Spearman correlation, we further examined regional association between local arteriolosclerosis and local tau-tangle burden. We find that AWS arteriolosclerosis was not correlated with tangle burden in the midfrontal gyrus (rs = 0.05; p = 0.11), while PWS vessel disease was weakly correlated with cortical tau burden in both PWS region (rs = 0.10; p = 0.03) and angular gyrus (rs = 0.07; p = 0.02).

Because our primary findings revealed that watershed arteriolosclerosis pathology is associated with tau-tangle pathology, secondary analyses leveraged available tau proteomic data from the dorsolateral prefrontal cortex (N = 654) to examine the association between watershed arteriosclerosis and tau peptides. First, we examined Spearman correlations between each phosphorylated tau (ptau) epitope (S202, T217, S262, S305, S404) and tau-tangle pathology burden. The strongest correlations were the positive correlation between ptau S262 with tangle pathology (rs = 0.43; p < 0.001), as well as between ptau T217 and tangle pathology (rs = 0.35; p < 0.001). Next, in separate linear regression models adjusting for demographics and post-mortem interval (PMI), we examined whether AWS/PWS arteriolosclerosis was associated with each phosphorylated peptide. More severe PWS arteriolosclerosis pathology was associated with a higher abundance of all five phosphorylated tau peptides. Interestingly, we find AWS arteriolosclerosis pathology is associated with higher abundance of ptau S262 (Table 3).

To translate our pathology findings to neuroimaging, we leveraged available whole brain measures of ex-vivo WMH data (N = 389). WMH lesions in aging have often been used as a surrogate marker for SVD in clinical studies and shown to be associated with arteriolosclerosis pathology in imaging-pathologic studies [2, 4]. Out of 389 participants, 19% had WMH burden = 1 (mild), 37% had WMH burden = 2 (moderate), and 44% had WMH burden = 3 (severe). In linear regression models adjusted for demographics, PMI, and common age-related pathologies, WMH burden was associated with greater tau-tangle pathology (estimate = 0.29, SE = 0.08, p = 0.001) but not with β-amyloid (estimate = 0.11, SE = 0.08, p = 0.19).

In the present study, we leveraged neuropathologic, neuroimaging, and proteomic data to examine the association between small vessel disease and AD pathological changes. Our main findings show that watershed arteriolosclerosis is associated with tau-tangle pathology and not β-amyloid. We extend these findings, indicating that watershed arteriosclerosis pathology is associated with proteomic tau phosphopeptides. Further, supporting these findings, we also show that whole brain measures of WMH burden, that are commonly used as a surrogate marker for clinical SVD, are associated with tau-tangle pathology. Together, these findings provide compelling evidence that SVD-related pathologic changes and tau are interconnected.

Watershed brain regions can provide a novel window into understanding the mechanisms associated with SVD. These brain regions have an idiosyncratic blood supply and can be considered as ‘weak points’ in the cerebral blood supply. We and others have shown watershed brain regions to be vulnerable to both hypoxic–ischemic-tissue injuries and arteriosclerotic vessel changes [1, 29]. Arteriolosclerosis is one of the most predominant SVD lesions co-existing with AD pathology in post-mortem brains [11]. Our current study extends these findings by showing that watershed arteriolosclerosis, specifically posterior watershed, is associated with greater tau-tangle pathology burden and higher abundance of tau phosphorylation epitopes. These findings are consistent with recent studies showing markers of SVD (both pathological and MRI) are associated with expression of tau pathological changes [31, 32, 34, 62]. To translate our findings to neuroimaging, we used WMH burden as a surrogate neuroimaging marker for SVD-related tissue injury. However, we recognize that while WMH are often presumed to be of vascular etiology, they are not specific to vascular disease, with their etiology most likely driven by a combination of both vascular and neurodegenerative factors. Notably, our findings between WMH and elevated tau burden corroborate other neuroimaging-pathological studies that report similar findings [24, 39]. Interestingly, though arteriosclerosis was more severe in the AWS compared to the PWS, it was largely the PWS arteriolosclerosis that we found an association with tau-tangle pathology, supporting the idea of brain vascular heterogeneity in the aging brain. The posterior watershed region presents as an interesting region. It is irrigated by terminal branches of all three anterior, middle, and posterior cerebral arteries, and overlaps specific brain regions (precuneus and posterior cingulate) susceptible to AD pathologic change [20, 21, 30, 49, 53]. We find an association between arteriosclerosis and neocortical tangle burden, suggesting that it is those persons with advanced tau pathologic changes that is driving the association, and suspect that this relationship is most prominent in symptomatic individuals. We did not find strong evidence of local SVD being associated with local tau pathology; however, further studies are needed to address whether watershed arteriolosclerosis is associated with tau burden among regionally or functionally connected brain regions. In contrast to other studies [19, 65, 66], we did not find an association of SVD (arteriolosclerosis or WMH) with β-amyloid burden. This may be due to study design as most of these studies have focused on neuroimaging markers of SVD, as well as PET markers for β-amyloid. While, PET ligands demonstrate excellent sensitivity and specificity to Aβ detection and have correlated well with the pathologic diagnosis of AD, these ligands are often used with a positive/negative threshold and cut-offs, which may not reflect diffuse Aβ positivity. Studies have shown that PET ligands have high specificity once Aβ adopts a beta-sheet fibrillar structure; therefore, a large proportion of PET amyloid is reflected by aggregated Aβ42, and therefore, PET binding to amyloid may be more prominent in neuritic plaques compared to diffuse plaques/vascular deposits [23].

Furthermore, various methodologies of quantifying amyloid (via PET/CSF/plasma) introduce differences across study design. For example, PET amyloid studies do not differentiate between parenchymal vs. vascular Abeta deposition, whereas our study included measures for meningeal and parenchymal CAA as confounding factors.

Typically, tau tangles accumulate in the entorhinal cortex and hippocampus followed by the neocortex. However, atypical AD pathologic subtypes have been described, where tau pathology spares the hippocampus [44]. Examining vascular changes in relation to neocortical tau in these atypical AD subtypes would be of interest. In addition, phosphorylated tau can also be detected in astrocyte glial cells and around the peri-vascular space, lesions often found in post-mortem brains with AD and other tauopathies [14, 33], raising the question of whether SVD differentially impacts tau pathognomonic lesions. Furthermore, supporting the hypothesis that SVD and tau are inter-related, neuropathologic studies have showed an association between SVD, white matter demyelination and Pick’s disease, and FTLD-tauopathy [62]. We and others have shown SVD pathology to be related to TDP-43 [1, 48], a proteinopathy frequently co-existing with AD pathology and associated with rapid cognitive decline and Alzheimer’s dementia in older persons [47]. While our data suggest that watershed arteriolosclerosis is associated with tau pathology independent of TDP-43, future work exploring key molecular vascular markers linked to TDP-43 encephalopathy, and more broadly to protein aggregation and proteotoxicity, will be important.

The exact mechanisms linking SVD and tau pathology are not well established. We conceptualize that SVD may be associated with tau phosphorylation through several downstream biological processes, including but not limited to neuroinflammation and oxidative stress associated with blood–brain barrier (BBB) dysfunction, which could impact key pathways that regulate intrinsic neuronal/cellular activity and function, thus leading to abnormal protein aggregation and neurodegeneration [52, 54]. Conversely, accumulation of dysfunctional tau protein may also lead to SVD-related lesions by directly impacting processes associated with vessel wall remodeling [42], as well as by altering white matter homeostasis [62]. Thus, resulting in a detrimental feedback loop mechanism between SVD and tau protein. One emerging topic in the field suggests that the glymphatic system represents a dynamic network to remove waste from the brain, including Aβ and tau [58, 69]. A myriad of biological processes including peri-vascular integrity, aquaporin-4 polarization, pericyte and endothelial cell organization, and inflammation within the vessel wall and within the peri-vascular space are important for facilitating glymphatic clearance. It is possible that morphological changes within the vessel due to the presence of arteriosclerosis can affect one/multiple biological processes necessary for conducting glymphatic clearance of Aβ and tau. In addition, studies have shown that neuroinflammation may lead to impaired glymphatic clearance [16]; therefore, it is possible that in the presence of heightened neuroinflammation-associated with AD pathology may impact the functioning of the glymphatic system and possibly exacerbate the frailty of this clearance system in the presence of severe vessel disease. Another intriguing topic is the tau propagation hypothesis which proposes a mechanism by which the pathological form of tau transfers between communicating neurons. The mechanisms associated with tau uptake in the extracellular space and subsequent intracellular uptake/aggregation are complex and many. It is possible that mechanisms of associated with SVD and tau propagation are linked. For example, in pathological conditions where morphological changes within the brain vasculature result in poor glymphatic clearance of tau from the extracellular space, it is likely that this would lead to intensifying tau propagation downstream. Furthermore, microglia may promote tau propagation via their ability to phagocytose and exocytose tau protein; thus, a heightened neuroinflammatory environment would also affect the vasculature and BBB function [14]. Altogether, the relationship between tau toxicity, SVD/vasculature, and neurodegeneration remains an interesting area and calls for the need of mechanistic vasculature studies.

It is well recognized that perfusion changes in persons with mid-cognitive impairment and Alzheimer’s dementia involve both temporal and parietal regions, with evidence to suggest that the earliest perfusion changes occur in the medial parietal cortex in persons with AD pathology [21, 43]. Interestingly, Neltner et al., showed that several medial temporal lobe regions can be vulnerable to arteriolosclerosis pathology, with the presence of arteriolosclerosis in temporal brain regions being greater in subjects with comorbid neurodegenerative pathologies [48]. These data directly align with our findings. We conceptualize that persons with higher levels of neurodegenerative proteinopathies (e.g., tau burden) may have more severe arteriosclerosis pathology in parietal and temporal regions, which may contribute to the hypoperfusion seen on neuroimaging studies in patients with Alzheimer’s dementia. However, further studies are needed to test whether SVD burden is associated with different aspects of neurodegeneration (e.g., neuronal death, inflammation, and brain atrophy).

Tau is a microtubule-associated protein, abundant in axons, and stabilizes microtubule bundles. For optimal function of tau, a normal level of phosphorylation is required. However, in pathological condition, an imbalance of phosphorylation events initiates abnormal metabolism and toxicity of tau. We find association of PWS arteriolosclerosis with dorsolateral prefrontal cortex expression of ptau Ser202, Thr217, Ser262, and Ser404, all of which are common tau epitopes phosphorylated in the AD brain [59]. Overall, phosphorylation of these tau epitopes changes the shape of tau molecule and modifies tau’s biological activity with microtubule bundles. Interestingly, arteriolosclerosis pathology in both AWS and PWS regions was associated with higher abundance of tau phosphorylation site Ser262, highlighting a more widespread role of arteriolosclerosis pathology with this specific tau epitope. Consistent with these findings, multiple animal studies also show an accumulation of Ser262 tau phosphopeptides after modeling ischemic stroke [7, 35, 36]. To enhance molecular understanding of AD pathophysiology, future studies examining AD-related proteomic changes in brain regions vulnerable to vascular changes, as well as angiogenesis/ischemic proteomic changes in regions vulnerable to early AD, are warranted.

Our findings raise the possibility that SVD plays a propagating role in tau pathologic change. It is well documented that β-amyloid drives tau accumulation leading to Alzheimer’s dementia [12, 64]; however, we speculate that tau represents a common pathway to dementia triggered by multiple factors, with one factor being SVD burden. Because mixed AD and cerebrovascular pathology account for over 80% of Alzheimer’s dementia diagnoses. Understanding these complex relationships between AD and vascular changes will be pivotal toward understanding AD pathophysiology and finding disease-modifying therapies.

There are multiple strengths to this study. By leveraging neuropathologic, ex-vivo neuroimaging, and proteomic, our study systematically identifies a relationship between SVD and tau pathologic change. We assessed pathology from watershed regions, providing important information on regional vascular vulnerability. However, we note several limitations. Although we examined two markers of SVD pathology (arteriolosclerosis and WMH), future work examining additional SVD markers (CAA, microbleeds, enlarged peri-vascular spaces) with AD and neurodegenerative proteinopathies in general is needed. There are no standard pathological consensus criteria assessing arteriolosclerosis pathology; future studies implementing a quantitative approach in evaluating brain arteriolosclerosis are needed. In addition, despite following standard protocols, we recognize that subtle differences with tissue sampling may occur which were not addressed in this study. Furthermore, we recognize that in the current study, tau phosphopeptides’ data were derived from the dorsolateral prefrontal cortex, a region involved later as per Braak staging, and we may be missing important molecular information if we were to examine proteomic changes in other AD vulnerable regions (e.g., hippocampus and precuneus). Finally, due to the cross-sectional nature of this study, we were unable to depict the underlying mechanisms of association of SVD and tau pathologic change.