Association of Enlarged Perivascular Spaces With Early Serum and Neuroimaging Biomarkers of Alzheimer Disease Pathology

Abstract

Background and Objectives

Enlarged perivascular spaces (EPVS), recognized as a key feature of cerebral small vessel disease (CSVD), have emerged as a promising biomarker for vascular contribution to Alzheimer disease (AD) and other neurodegenerative diseases. Although previous studies have linked EPVS to cerebrovascular dysfunction, their relationship with AD pathology and cognitive decline remains underexplored, particularly in multiethnic cohorts. This study investigates the associations between basal ganglia EPVS burden, blood-based biomarkers (BBM), and cognitive outcomes in a Southeast Asian cohort.

Methods

This cross-sectional study drew from the Biomarkers and Cognition Study, Singapore, comprising participants recruited from the community, at Dementia Research Centre (Singapore) from 2022 to 2024. Participants underwent comprehensive neuropsychologic assessments and were classified into cognitively normal, subjective cognitive decline, and mild cognitive impairment (MCI) groups according to established diagnostic criteria. BBM including amyloid β oligomers, amyloid β42 (Aβ42) and β40 (Aβ40), phosphorylated tau 181 (p-tau181), neurofilament light chain (NfL), and glial fibrillary acidic protein (GFAP) were quantified. MRI Markers of CSVD (EPVS, white matter hyperintensities [WMH], lacunes, and microbleeds) were visually rated using validated scales. Associations between EPVS, biomarkers, and cognitive outcomes were analyzed using correlation or multivariable regression models adjusting for age, sex, education, cognitive diagnosis, and APOE ε4 carrier status.

Results

A total of 979 participants were included (mean age: 58.2 ± 10.7 years; mean education: 14.9 ± 3.5 years; 60.7% female). Elevated EPVS burden was positively correlated with higher GFAP (ρ = 0.166, 95% CI 0.104 to 0.228, p < 0.01), NfL (ρ = 0.169, 95% CI 0.112 to 0.242, p < 0.01), and p-tau181 (ρ = 0.087, 95% CI 0.022 to 0.152, p < 0.01) and inversely with the Aβ42/40 ratio (ρ = −0.077, 95% CI −0.144 to 0.006, p < 0.05)—suggesting links to neuroinflammation and amyloid pathology. Among CSVD markers, EPVS exhibited the strongest association with Aβ pathology in participants with MCI (odds ratio [OR] 1.877, 95% CI 1.045 to 3.370, p = 0.035). In addition, higher EPVS burden was linked to poorer visuospatial skills and executive function (Block Design Test, OR 0.182, 95% CI 0.037 to 0.890, p = 0.035).

Discussion

These findings suggest EPVS burden to be a potential marker of both CSVD and AD-related BBM pathology. The results support the potential of incorporating EPVS assessments into routine MRI evaluations to enhance early detection and risk stratification in AD. Longitudinal studies are needed to confirm the prognostic value of EPVS and to clarify mechanisms linking vascular dysfunction to amyloid and tau pathology.

Introduction

Dementia is a progressive neurodegenerative disorder marked by cognitive decline across multiple domains, including memory, executive function, and attention, ultimately disrupting brain networks essential for daily functioning. As the disease advances, it erodes personal autonomy and imposes substantial emotional and financial burdens on caregivers and health care systems worldwide.¹ With patients with dementia expected to rise to 78 million in 2030 and 139 million in 2050,² Asia is expected to bear a disproportionate share of new cases because of rapid demographic aging. This growing crisis underscores the urgent need for improved early detection and management strategies, with neuroimaging biomarkers emerging as promising, cost-effective tools for clinical diagnosis and risk stratification.

One key contributor to cognitive decline in dementia is cerebral small vessel disease (CSVD), a group of disorders resulting from damage to the brain's small vessels.³ CSVD has been consistently linked to Alzheimer disease (AD) pathology³ and cognitive decline,⁴ highlighting its relevance in understanding dementia progression. Neuroimaging plays a crucial role in detecting CSVD, with key markers including white matter hyperintensities (WMH), cerebral microbleeds (MB), lacunes, and enlarged perivascular spaces (EPVS).³ Although other CSVD markers have been extensively studied, EPVS remains relatively underexplored. Given its potential role in dementia pathophysiology, additional investigation into EPVS is essential to better understand its contribution to neurodegenerative processes.

When enlarged, perivascular spaces (PVS) become readily visible on MRI and are increasingly regarded as a hallmark of CSVD and compromised perivascular clearance.⁵ They are most conspicuous in the basal ganglia, an area with rich vascularization that not only governs motor control but also influences higher-order cognition through extensive connections with the prefrontal cortex and other cerebral regions.⁶ Unlike other markers such as lacunes and medial temporal lobe atrophy which indicate structural damage, EPVS dilatation is not classically regarded as a lesion because it typically does not involve substantial parenchymal loss. Instead, it represents an anatomical variation that may reflect impaired perivascular clearance and inflammatory processes rather than direct tissue injury.⁵

EPVS have garnered increasing attention because of their role in interstitial fluid drainage and the clearance of neurotoxic metabolic byproducts through the glymphatic system.⁷ Emerging evidence suggests that PVS burden is strongly associated with AD pathology, particularly because of its links with β-amyloid (Aβ) accumulation, tau pathology, and neurodegeneration.⁸ However, the mechanisms connecting EPVS to AD are complex. Vascular changes—most notably chronic hypertension, arteriosclerosis, and endothelial dysfunction—play a central role in driving PVS dilatation. Chronic hypertension induces stiffening of vessel walls,⁹ impairing vascular pulsatility essential for glymphatic flow.⁷ Vascular inflammation and endothelial dysfunction further exacerbate the issue by causing blood-brain barrier disruption, allowing plasma proteins and immune cells to infiltrate the perivascular space.¹⁰ This cascade promotes local inflammation, causing astrocytic endfoot swelling¹⁰ and mislocalization of aquaporin-4,⁵ further impairing glymphatic function. As clearance of interstitial fluid becomes inefficient, interstitial solutes accumulate, and PVS become pathologically enlarged to accommodate the retained fluid and metabolic waste.¹¹ Over time, this impaired clearance has to been shown to contribute to the extracellular deposition of Aβ and the intracellular aggregation of hyperphosphorylated tau.⁵ Thus, PVS dilatation may represent not only a downstream consequence of vascular disease but also a contributing factor that links vascular pathology to the hallmark proteinopathies of AD, establishing a self-reinforcing cycle of impaired clearance, protein and neurofibrillary accumulation, and neurodegeneration.

These mechanisms are supported by postmortem studies that have demonstrated a close spatial and pathologic overlap between PVS enlargement and AD proteinopathies. PVS enlargement and perivascular Aβ have been shown as common features in AD brains in autopsy studies,^12,13 implicating these changes as markers—and potentially mediators—of impaired clearance mechanisms central to AD progression. Notably, perivascular Aβ burden in these studies¹² showed a significant correlation with the Braak neurofibrillary tangle (NFT) staging,¹⁴ reinforcing its close association with tau pathology and overall disease severity. Although these findings are rooted in basic science, they provide a strong biological rationale for our investigation of EPVS in the Biomarkers and Cognition Study, Singapore (BIOCIS) cohort, where in vivo associations with cognition can be explored in a well-characterized clinical population.

Building on these pathophysiologic and histologic findings, it is important to consider the clinical utility of EPVS assessments with neuroimaging. Given the routine use of MRI in dementia evaluations, elucidating the relationship between EPVS burden and established Alzheimer biomarkers could streamline diagnostics. Specifically, if EPVS reliably reflects AD pathology or related cognitive deficits, MRI-based assessments may serve as a single, cost-effective tool for earlier or more accurate detection of cognitive decline. However, the relationship between EPVS burden and cognitive function remains contentious, with studies reporting either significant negative associations,^15,16 or no clear relationship.¹⁷ These inconsistencies, coupled with a relative paucity of data from Southeast Asian populations, highlight the necessity for region-specific research—particularly as Southeast Asian cohorts may exhibit distinct vascular and pathologic profiles compared to Western populations.^18,19,20

Current modern strategies to characterize AD and other neurodegenerative diseases include blood-based biomarkers (BBM) and advanced neuroimaging. BBM include amyloid β oligomers (OAβ), amyloid β40 (Aβ40) and β42 (Aβ42), phosphorylated tau 181 (p-tau181), neurofilament light chain (NfL) and glial fibrillary acidic protein (GFAP).^21–26 Meanwhile, advanced neuroimaging techniques such as volumetric analyses of gray/white matter and arterial spin labeling (ASL) for perfusion have shown promise in detecting early AD pathology.^27,28 However, these methods face financial and logistical barriers that limit widespread clinical adoption. BBM assays are not universally available, and advanced imaging demands extended scan times and specialized expertise, further constraining their routine use. This highlights the need for a simpler, MRI-based marker that is both widely accessible and clinically informative. Clinically, such insights could advance personalized approaches to diagnosing and managing prodromal and established dementia, thereby guiding more tailored investigations and interventions.

The primary aim of this study was to test the hypothesis that basal ganglia EPVS is associated with AD-related pathology, as reflected in BBM (OAβ, Aβ40, Aβ42, p-tau181, NfL and GFAP) and neuroimaging markers (gray matter volume, white matter volume, and gray matter perfusion). In addition, we evaluate whether EPVS provides useful clinical information, compared with other CSVD markers, in distinguishing individuals with AD-related plasma pathology. The secondary aim explores how a higher EPVS burden relates to cognitive performance in distinct domains. By focusing on the richly vascularized basal ganglia and its extensive cortical connections, we aim to illuminate the relationship between cerebrovascular integrity and cognitive function. Identifying a distinct biochemical and cognitive profile linked to elevated EPVS burden may enhance our understanding of the EPVS-AD relationship in a Southeast Asian cohort.

Methods

Participants

This cross-sectional study recruited community-dwelling participants in Singapore from the BIOCIS, conducted at the Dementia Research Centre (Singapore). Data were collected from February 2022 to June 2024. All included participants had completed brain MRIs, BBM evaluations, and neuropsychological assessments. Individuals were eligible if they were age 30–95 years, had intact mental capacity, and were literate in English or Mandarin. Exclusion criteria encompassed serious neurologic, psychiatric, or systemic illnesses, psychotic disorders, major depressive disorder, alcoholism, or drug dependency within the past 2 years. Full methodological details can be found in the published BIOCIS protocol.²⁹

Diagnostic Classification

Based on the National Institute on Aging-Alzheimer's Association criteria³⁰ and published criteria,³¹ participants were categorized into cognitively normal (CN), subjective cognitive decline (SCD), and mild cognitive impairment (MCI) groups. CN individuals had normal cognitive test performance with no subjective memory complaints. Individuals with SCD had normal cognitive test performance while reporting subjective memory impairment. Individuals with MCI had impaired cognitive test scores (>1.5 SDs below the CN mean) and subjective memory complaints without functional impairment.

Demographics and Clinical Data

Demographic information—including age, sex, years of education, medical history, and medication use—was collected, and blood pressure measurements were obtained. Diabetes mellitus was defined as glycated hemoglobin (HbA1C) ≥ 7.2% and/or current use of diabetes medications. Hypertension was defined as systolic blood pressure ≥140 mm Hg and/or current use of antihypertensive drugs. Hyperlipidemia was defined as total cholesterol ≥5.2 mmol/L and/or current use of lipid-lowering medication.

Neuropsychologic Assessments

A comprehensive neuropsychologic battery of cognitive tests was conducted by trained research staff in either English or Mandarin. Each test assesses 1 or more cognitive domains, with the primary domains and their corresponding tests outlined: Global cognition—Montreal Cognitive Assessment, Visual Cognitive Assessment Test (VCAT); learning/memory—Rey Auditory Verbal Learning Test (RAVLT), Free and Cued Selective Reminding Test (FCSRT); working memory—Wechsler Adult Intelligence Scale, Fourth Edition (WAIS-IV) Digit Span (DS) backward; executive function—Test of Practical Judgment (TOPJ) 9 items, Color Trials (CT) 2, WAIS-IV Block Design (BDN); processing speed—CT-1, CT-2, WAIS-IV Coding; language—Semantic fluency (animals); visuospatial skills—BDN, Rey Complex Figure Test (RCFT); attention—WAIS-IV DS forward.

Neuroimaging Protocol

Participants underwent brain MRI scans on a 3T Siemens Prisma Fit (Siemens, Erlangen, Germany) MRI machine, where T1-weighted (T1), T2-weighted fluid-attenuated inversion recovery (T2-FLAIR), susceptibility-weighted imaging (SWI), and ASL sequences were obtained. The T1 sequence was acquired with a repetition time (TR) of 2,000 milliseconds, inversion time (TI) 800 milliseconds, echo time (TE) 2.26 milliseconds, flip angle 8°, 176 slices, 1 mm slice thickness, and a voxel size of 1 × 1 × 1 mm. The T2-FLAIR sequence had a TR of 7,000 milliseconds, TI 2,100 milliseconds, TE 394 milliseconds, flip angle 120°, 192 slices, a slice thickness of 1.56 mm, and a voxel size of 0.8 × 0.8 × 1 mm. The SWI sequence was acquired with a TR of 28 milliseconds, TE 20 milliseconds, flip angle 15°, 1 average, 2 mm slice thickness, an image resolution of 308 × 352 pixels, phase encoding direction set to ROW, voxel size 0.625 × 0.625 mm, 72 slices, and a pixel bandwidth of 120. The ASL sequence was obtained with a TR of 2,500 milliseconds, TE 11 milliseconds, TI 1,800 milliseconds, bolus duration 700 milliseconds, flow limit 100 cm/s, a field of view of 256 mm, a slice gap of 20.9 mm, a distance factor of 25%, flip angle 90°, and 91 measurements (1 calibration M0, 45 label, and 45 control). The ASL voxel size was 4.0 × 4.0 × 8.00 mm, with a matrix size of 64 × 64.

T1 sequences were processed using Computational Anatomy Toolbox in Statistical Parametric Mapping to segment gray matter, white matter, and CSF. ASL postprocessing was performed using FSL's Bayesian Inference for ASL MRI (BASIL) toolbox to get calibrated total gray matter perfusion.

MRI visual ratings were conducted on as per the Standards for Reporting Vascular Changes on Neuroimaging 2 criteria and previously published literature,^32,33 by 2 trained raters blinded to participants' cognitive diagnoses. EPVS, WMH, lacunes, and MB were quantified using this method and were further categorized into high and low using Staals criteria, where a Staals score of 0 indicates a low CSVD burden and a score of 1 indicates a high CSVD burden, in accordance with previously published criteria.³⁴

Blood Biomarkers

Venepuncture was performed by a certified phlebotomist at the baseline study visit. Blood samples were collected in EDTA vacutainers (Becton Dickinson, Franklin Lakes, NJ) and centrifuged at 2,000g for 10 minutes at 4°C. Plasma was aliquoted and stored at −80°C until analysis.

Genomic DNA was extracted from whole blood using the QIAamp DNA Blood Maxi Kit (Qiagen, Hilden, Germany). DNA concentration and purity were assessed using a Nanodrop One spectrophotometer (Thermo Fisher Scientific, Oxford, United Kingdom). APOE genotyping was performed using real-time PCR (StepOne Plus or QuantStudio 7 Pro, Applied Biosystems, Foster City, CA), targeting SNPs rs429358 and rs7412 (Life Technologies, Carlsbad, CA). PCR reactions were conducted in 96-well MicroAmp Fast/Optical plates (Life Technologies) with 10 ng of DNA in a 10 µL TaqPath ProAmp master mix reaction (Applied Biosystems). The results were analyzed using Design and Analysis 2.5.1 Real-Time PCR system software (Applied Biosystems), with independent raters verifying APOE genotypes.

Plasma biomarkers, including NfL, GFAP, Aβ40, Aβ42, p-tau181, and OAβ, were quantified using Quanterix's Single Molecule Array technology on the HD-X Analyzer (Quanterix, Billerica, MA). The Multi Detection System OAβ (PeopleBio Inc., Seongnam-si, South Korea) was used to quantify the oligomeric Aβ values using the ELISA method. The Neurology 4-Plex E (NfL, GFAP, Aβ40, Aβ42) and p-tau181 Advantage V2.1 kits were used according to manufacturer protocols.

Fasting glucose, HbA1C, and lipid profiles, including cholesterol and triglycerides, were analyzed by an external laboratory (Innoquest, Singapore).

Statistical Analysis

Demographic and Group Comparisons

Demographic comparisons across the CN, SCD, and MCI groups were performed using 1-way analysis of variance (ANOVA) for continuous variables and χ² tests for categorical variables. For comparisons between high and low EPVS burden groups, ANOVA and χ² tests were similarly applied.

Correlation Analysis

To address data skewness, OAβ, NfL, GFAP, and p-tau181 values were log-transformed before analysis. Spearman correlation was used to assess the relationships between basal ganglia EPVS, other CSVD markers, BBM (Aβ42, Aβ40, Aβ42/Aβ40 ratio, OAβ, NfL, GFAP, p-tau181), and neuroimaging measures (gray matter/total intracranial volume [GM/TIV], white matter/total intracranial volume [WM/TIV], total gray matter perfusion). To estimate the 95% CIs for the correlation coefficients, a bias-corrected and accelerated bootstrapping approach with 1,000 resamples was used.

Logistic Regression Analysis

For the MCI subgroup, participants were classified as having AD-related pathology (AD-MCI) if their Aβ42/Aβ40 ratio was ≤0.05 and as non–AD-MCI if their ratio exceeded 0.05.³⁵ A forward stepwise logistic regression (likelihood ratio method) was conducted to determine which CSVD marker (classified via Staals scoring) was most strongly associated with AD pathology within the MCI group.

For neuropsychological outcomes, logistic regression was used to examine the association between EPVS burden (Staals classification) as the independent variable and cognitive performance on individual tests as the dependent variable. All regression models adjusted for age, sex, years of education, APOE ε4 status, and cognitive diagnosis, as appropriate.

Multiple Comparisons and Effect Size Reporting

Multiple comparison corrections were not applied, given the exploratory nature of the analyses. Instead, effect sizes, CIs, and a significance threshold of p < 0.05 were reported to provide an interpretable and robust assessment of findings, consistent with prior hypothesis-driven studies.³⁶

Software and Power Considerations

All statistical analyses were performed using IBM SPSS Statistics (version 29.0; IBM Corp., Armonk, NY). All figures and image visualizations were generated using R statistics (version 4.4.2; R Foundation for Statistical Computing, Vienna, Austria). The BIOCIS study was adequately powered for its primary end points, allowing for an exploratory analysis of EPVS, biomarkers, and cognition as secondary end points.

Standard Protocol Approvals, Registrations, and Patient Consents

Informed consent was obtained from all participants according to the Declaration of Helsinki and local clinical research regulations, and procedures used in the study were in accordance with ethical guidelines. This study has been approved by the Nanyang Technological University Institutional Review Board (NTU-IRB-2021-1036).

Data Availability

The deidentified data that support the findings of this study are available from the corresponding author on reasonable request.

Results

Demographic and Clinical Characteristics

Of the 1,500 participants with available demographic and neuropsychological assessment data, 1,481 had blood biomarker data. Among them, 979 participants had MRI data available and were included in the final analysis. Figure 1 illustrates the participant inclusion process.

Figure 1. Flowchart of Participants.

Table 1 summarizes participant demographics for the full cohort of 979 individuals, subdivided into CN (n = 347), SCD (n = 259), and MCI (n = 373) groups. Overall, 748 participants had low EPVS burden (Staals = 0), whereas 231 had high EPVS burden (Staals = 1). Compared with the cognitively normal group, participants in the SCD and MCI groups were older (p < 0.001), had fewer years of education (p < 0.001), and demonstrated a higher prevalence of hypertension (p < 0.001). EPVS visual rating scores (left p = 0.023; right p = 0.021; total p = 0.010) and total Staals score (p < 0.001) were higher in cognitively impaired (i.e., MCI) participants compared with those who were cognitively unimpaired (i.e., CN and SCD). The cognitively impaired group also had a higher percentage of EPVS Staals (high) (p = 0.047) and WMH Staals (high) (p < 0.001) participants compared with the cognitively unimpaired group.

Table 1.

Participant Demographics, Cardiovascular Risk Factors, and Cerebral Small Vessel Disease Markers, Grouped by Cognitive Diagnosis

Variable	CN (n = 347)	SCD (n = 259)	MCI (n = 373)	Total cohort (N = 979)	p Value
Demographics
Agea	54.33 ± 10.14	56.65 ± 9.37	62.89 ± 10.38	58.21 ± 10.72	<0.001c
Sex (female)b	218 (63.0)	158 (61.0)	217 (58.4)	593 (60.7)	0.455
Education (y)a	15.76 ± 3.23	15.14 ± 3.03	13.87 ± 3.88	14.88 ± 3.54	<0.001c
APOE ε4 carrierb	64 (20.6)	44 (18.4)	65 (18.0)	173 (18.9)	0.674
Diabetes mellitusb	27 (7.8)	19 (7.4)	44 (11.8)	90 (9.2)	0.087
Hypertensionb	71 (20.5)	69 (26.6)	154 (41.3)	294 (30.0)	<0.001c
Hyperlipidemiab	252 (72.6)	175 (67.6)	282 (76.0)	709 (72.6)	0.065
CSVD markers
EPVS lefta	1.11 ± 0.46	1.17 ± 0.47	1.21 ± 0.51	1.17 ± 0.49	0.023c
EPVS righta	1.01 ± 0.50	1.09 ± 0.52	1.11 ± 0.51	1.07 ± 0.51	0.021c
EPVS totala	2.21 ± 0.88	2.26 ± 0.91	2.32 ± 0.93	2.23 ± 0.91	0.010c
EPVS Staals (high)b	67 (19.3)	63 (24.3)	101 (27.1)	231 (23.6)	0.047c
WMH Staals (high)b	118 (34.0)	91 (35.1)	193 (51.7)	402 (41.1)	<0.001c
Lacunes Staals (high)b	81 (23.3)	60 (23.2)	100 (26.8)	241 (24.6)	0.458
MB Staals (high)b	44 (12.7)	33 (12.7)	56 (15.0)	133 (13.6)	0.592
Staals totala	0.89 ± 0.97	0.95 ± 1.01	1.21 ± 1.06	1.03 ± 1.03	<0.001c

Abbreviations: ANOVA = analysis of variance; CN = cognitively normal; CSVD = cerebral small vessel disease; MB = microbleeds; MCI = mild cognitive impairment; EPVS = enlarged perivascular spaces; SCD = subjective cognitive decline; WMH = white matter hyperintensity.

^aMean ± SD.

^bFrequency (percentages).

^cSignificance level set at p < 0.05. Comparing the 3 groups using χ² for categorical variables or 1 way ANOVA for continuous variables.

Correlation of CSVD Markers and BBM

Table 2 depicts the Spearman correlation matrix of 4 CSVD markers against BBM. EPVS—as well as all other CSVD markers—positively correlated with NfL (ρ = 0.169, 95% CI 0.112 to 0.242, p < 0.01), GFAP (ρ = 0.166, 95% CI 0.104 to 0.228, p < 0.05), and p-tau181 (ρ = 0.087, 95% CI 0.022 to 0.152, p < 0.01), but negatively correlated with GM/TIV (ρ = −0.100, 95% CI −0.156 to −0.031, p < 0.01), WM/TIV (ρ = −0.161, 95% CI −0.220 to −0.096, p < 0.01), and the Aβ42/Aβ40 ratio (ρ = −0.077, 95% CI −0.144 to −0.006, p < 0.05). Notably, MB were also positively correlated with Aβ40 (ρ = 0.097, 95% CI 0.032 to 0.160, p < 0.01). Lacunes correlated negatively with total gray matter perfusion (ρ = −0.140, 95% CI −0.207 to −0.065, p < 0.01) but also positively with Aβ40 (ρ = 0.067, 95% CI −0.002 to 0.130, p < 0.05). The WMH Staals score showed positive correlations with Aβ40 (ρ = 0.165, 95% CI 0.100 to 0.229, p < 0.01) and Aβ42 (ρ = 0.072, 95% CI 0.015 to 0.142, p < 0.05), whereas negatively correlating with total gray matter perfusion (ρ = −0.116, 95% CI −0.175 to −0.044, p < 0.01).

Table 2.

Correlation Matrix of EPVS, WMHs, Lacunes, and Cerebral MBs With Neuroimaging and Blood-Based Biomarkers

Variable	EPVS (Staals)	WMH (Staals)	Lacunes (Staals)	MB (Staals)
MRI markers
GM/TIV	−0.100** (−0.156 to −0.031)	−0.288** (−0.344 to −0.233)	−0.192** (−0.249 to −0.126)	−0.174** (−0.229 to −0.107)
WM/TIV	−0.161** (−0.220 to −0.096)	−0.188** (−0.252 to −0.124)	−0.117** (−0.174 to −0.046)	−0.144** (−0.201 to −0.075)
Total GM perfusion (mL/min/100 g)	−0.040 (−0.104 to 0.024)	−0.116** (−0.175 to −0.044)	−0.140** (−0.207 to −0.065)	0.003 (−0.074 to 0.069)
BBM
Aβ42 (pg/mL)	−0.008 (−0.071 to 0.057)	0.072* (0.015 to 0.142)	−0.007 (−0.071 to 0.057)	0.027 (−0.028 to 0.090)
Aβ40 (pg/mL)	0.056 (−0.011 to 0.120)	0.165** (0.100 to 0.229)	0.067* (−0.002 to 0.130)	0.097** (0.032 to 0.160)
Aβ42/Aβ40 (ratio)	−0.077* (−0.144 to −0.006)	−0.084* (−0.149 to −0.017)	−0.086* (−0.147 to −0.023)	−0.069* (−0.135 to −0.002)
OAβ (ratio)	0.012 (−0.077 to 0.092)	0.062 (−0.026 to 0.151)	0.022 (−0.071 to 0.111)	0.034 (−0.059 to 0.134)
NfL (pg/mL)	0.169** (0.112 to 0.242)	0.329** (0.267 to 0.390)	0.166** (0.109 to 0.227)	0.134** (0.074 to 0.190)
GFAP (pg/mL)	0.166** (0.104 to 0.228)	0.266** (0.201 to 0.328)	0.073* (0.007 to 0.140)	0.109** (0.049 to 0.172)
p-tau181 (pg/mL)	0.087** (0.022 to 0.152)	0.193** (0.132 to 0.256)	0.142** (0.080 to 0.205)	0.095** (0.025 to 0.161)

Abbreviations: Aβ = β-amyloid; BBM = blood biomarkers; GFAP = glial fibrillary acidic protein; GM = grey matter; MB = microbleeds; NfL = neurofilament light chain; OAβ = amyloid β oligomers; EPVS = enlarged perivascular spaces; p-tau181 = phosphorylated tau 181; TIV = total intracranial volume; WM = white matter; WMH = white matter hyperintensities.

Correlation coefficients (95% CI), *p < 0.05, **p < 0.005. OAβ, NfL, GFAP, and pTau181 log transformed.

Validation of EPVS Association With BBM in Participants With MCI

Table 3 presents the results of a forward stepwise logistic regression in participants with MCI, identifying demographic, blood, and CSVD variables most strongly associated with AD-related BBM. After 3 steps, age (odds ratio [OR] 1.041, 95% CI 1.010 to 1.074, p = 0.009), APOE ε4 carrier status (OR 2.456, 95% CI 1.305 to 4.623, p = 0.005), and EPVS Staals score (OR 1.877, 95% CI 1.045 to 3.370, p = 0.035) were retained in the final model, whereas other CSVD markers were not retained.

Table 3.

Logistic Regression of Incidence of Participants With Alzheimer Disease-Related Pathology Against MRI Cerebral Small Vessel Disease Markers (EPVS, White Matter Hyperintensities, Lacunes, Microbleeds) in the Mild Cognitive Impairment Cohort

Variable	OR	95% CI	p Value
Age	1.041	1.010–1.074	0.009a
APOE ε4 carrier status	2.456	1.305–4.623	0.005a
EPVS Staals score	1.877	1.045–3.370	0.035a

Abbreviations: OR = odds ratio; EPVS = enlarged perivascular spaces.

^aSignificance level set at p < 0.05. Age, APOE ε4 status and PVS Staals score were included in the model; all other factors were not retained.

Demographic and Cognitive Profiles of High and Low EPVS Burden Groups

In the full cohort, univariate analyses revealed that individuals with high EPVS burden were generally older (p < 0.001), had fewer years of education (p = 0.007), and exhibited higher incidence of hypertension (p < 0.001) compared with those with low EPVS burden. Table 4 summarizes these demographic differences and cognitive test results across high vs low EPVS groups. Significant group differences emerged for VCAT (p < 0.001), RAVLT (p = 0.041), FCSRT (p = 0.009), CT-1 (p = 0.004), CT-2 (p = 0.001), coding (p = 0.003), and semantic fluency (p = 0.010).

Table 4.

Participant Demographics and Neuropsychologic Assessment Scores, Grouped by EPVS Staals Score

Variable	EPVS Staals (low) (n = 748)	EPVS Staals (high) (n = 231)	Total cohort (N = 979)	p Value
Demographics
Agea	56.71 ± 10.56	63.06 ± 9.77	58.21 ± 10.72	<0.001c
Sex (female)b	445 (59.5)	149 (64.8)	593 (60.7%)	0.151
Education (y)a	15.05 ± 3.58	14.33 ± 3.36	14.88 ± 3.54	0.007c
APOE ε4 carrierb	133 (18.9)	40 (19.2)	173 (18.9%)	0.920
Diabetes mellitusb	61 (8.2)	29 (12.6)	90 (9.2%)	0.460
Hypertensionb	199 (26.6)	95 (41.1)	294 (30.0%)	<0.001c
Hyperlipidemiab	537 (72.0)	172 (74.5)	709 (72.6%)	0.461
Neuropsychologic assessments
Global cognition
MoCAa	26.14 ± 2.61	25.9 ± 2.69	26.08 ± 2.63	0.241
VCATa	27.19 ± 2.62	26.29 ± 3.25	26.97 ± 2.80	<0.001c
Learning/memory
RAVLTa	50.21 ± 10.41	48.55 ± 10.84	49.82 ± 10.53	0.041c
FCSRTa	12.63 ± 2.34	12.16 ± 2.33	12.52 ± 2.34	0.009c
Working memory
DS (backward)a	9.54 ± 2.48	9.27 ± 2.40	9.48 ± 2.46	0.140
Executive function
TOPJ 9a	16.64 ± 4.06	16.33 ± 3.94	16.56 ± 4.03	0.330
CT-2a	88.15 ± 31.90	96.51 ± 34.94	90.12 ± 32.81	0.001c
Processing speed
CT-1a	44.50 ± 17.99	48.76 ± 21.78	45.50 ± 19.02	0.004c
Codinga	72.16 ± 16.79	68.4 ± 16.98	71.28 ± 16.90	0.003c
Language
Semantic fluency (animals)a	19.37 ± 4.74	18.47 ± 4.52	19.16 ± 4.70	0.010c
Visuospatial skills
BDNa	42.28 ± 11.15	41.31 ± 11.44	42.06 ± 11.22	0.259
RCFTa	32.69 ± 3.47	32.72 ± 3.11	32.69 ± 3.39	0.902
Attention
DS (forward)a	11.15 ± 2.40	10.84 ± 2.46	11.08 ± 2.42	0.096

Abbreviations: ANOVA = analysis of variance; BDN = Block Design; CT = Color Trial; DS = Digit Span; FCRST = Free and Cued Selective Reminding Test; MoCA = Montreal Cognitive Assessment; EPVS = enlarged perivascular spaces; RAVLT = Rey Auditory Verbal Learning Test; RCFT = Rey Complex Figure Test; TOPJ = Test of Practical Judgment; VCAT = Visual Cognitive Assessment Test.

^aMean ± SD.

^bFrequency (percentages).

^cSignificance level set at p < 0.05. Comparing the 2 groups using χ² for categorical variables or 1 way ANOVA for continuous variables.

Table 5 presents a logistic regression model adjusting for potential confounders, in which BDN emerged as the sole neuropsychological test variable significantly associated with EPVS burden (OR 0.182, 95% CI 0.037 to 0.890, p = 0.035). A forest plot to summarize multiple regression results can be found in Figure 2.

Table 5.

Logistic Regression of Neuropsychological Assessment Scores Against Enlarged Perivascular Spaces Staals Score

Variable	OR	95% CI	p Value
Global cognition
MoCAa	0.821	0.559–1.207	0.316
VCATa	1.246	0.832–1.866	0.285
Learning/memory
RAVLTa	1.690	0.353–8.095	0.511
FCSRTa	1.342	0.921–1.953	0.125
Working memory
DS (backward)a	0.931	0.650–1.334	0.698
Executive function
TOPJ 9a	1.313	0.681–2.531	0.417
CT 2a	0.475	0.004–60.908	0.764
Processing speed
CT 1a	0.322	0.018–5.841	0.443
Codinga	0.345	0.034–3.547	0.371
Language
Semantic fluency (animals)a	0.958	0.487–1.886	0.901
Visuospatial skills
BDNa	0.182	0.037–0.890	0.035a
RCFTa	0.877	0.524–1.466	0.616
Attention
DS (forward)a	1.166	0.802–1.697	0.421

Abbreviations: BDN = Block Design; CT = Color Trials; DS = Digit Span; FCRST = Free and Cued Selective Reminding Test; MoCA = Montreal Cognitive Assessment; OR = odds ratio; RAVLT = Rey Auditory Verbal Learning Test; RCFT = Rey Complex Figure Test; TOPJ = Test of Practical Judgment; VCAT = Visual Cognitive Assessment Test.

^aSignificance level set at p < 0.05.

Figure 2. Forest Plot of Logistic Regression of Neuropsychological Assessment Scores Against Enlarged Perivascular Spaces Staals Score.

BDN = Block Design; CT = Color Trials; DS = Digit Span; FCRST = Free and Cued Selective Reminding Test; MoCA = Montreal Cognitive Assessment; OR = odds ratio; RAVLT = Rey Auditory Verbal Learning Test; RCFT = Rey Complex Figure Test; TOPJ = Test of Practical Judgment; VCAT = Visual Cognitive Assessment Test.

Discussion

This study demonstrates that higher basal ganglia EPVS burden correlates significantly with NfL, GFAP, p-tau181, and the Aβ42/Aβ40 ratio, with an inverse relationship to the latter suggesting that elevated EPVS reflects increased amyloid pathology consistent with AD pathology.²³ The observed trends for NfL and GFAP are consistent with previously published studies.^16,37 In line with established mechanisms,^5,9,10 elevated GFAP and NfL levels likely reflect astrocytic activation and axonal injury secondary to vascular dysfunction, local inflammation, and accumulation of toxic metabolites associated with PVS enlargement. Given that these plasma markers are well-established indicators of vascular inflammation^25,26—and with our study demonstrating a higher prevalence of hypertension in individuals with greater EPVS burden—these associations support the potential role of EPVS as a marker of both vascular inflammation and glymphatic dysfunction.

By contrast, relationships between EPVS and p-tau181 or the Aβ42/Aβ40 ratio have been less frequently reported, particularly in plasma-based biomarker studies. Previous research linking EPVS to p-tau181 has relied on CSF biomarkers, whereas studies which analyzed plasma p-tau181 did not report significant associations.³⁸ In addition, a previous study reported no significant relationships between EPVS in the basal ganglia and the plasma Aβ42/Aβ40 ratio.³⁹ In this context, our findings—demonstrating significant associations between elevated EPVS burden and both decreased plasma Aβ42/40 ratio and increased p-tau181—suggest a closer link between EPVS and amyloid and tau pathology than previously recognized. These results are consistent with proposed mechanisms whereby impaired glymphatic clearance contributes to the extracellular accumulation of Aβ and the intracellular aggregation of hyperphosphorylated tau,⁵ both key processes in AD pathophysiology.^22,23 Given this context, our findings raise the possibility that EPVS burden could serve as a surrogate imaging marker for early AD pathology. Nonetheless, discrepancies with previous studies may reflect differences in population characteristics or methodological approaches, reinforcing the need for broader validation efforts. These findings also align with autopsy studies,^12,13 which similarly demonstrated strong associations between EPVS enlargement, Aβ deposition, Braak NFT staging, and other CSVD and white matter histopathologic changes. Together, these results further underscore the strong association between EPVS, neurofibrillary pathology, and overall severity of neurodegenerative changes.

A key finding is that EPVS emerged as the strongest CSVD marker associated with AD-related pathology in participants with MCI, whereas WMH and other CSVD markers were not retained in the final regression model. This finding carries substantial clinical implications—although WMH is more widely used in clinical practice because of its recognizability and substantial literature base,^40,41 our results suggest that EPVS may hold unique prognostic value in detecting AD pathology at the prodromal stage.

Although these findings are promising, validation against neuropathologic evidence remains critical. Although postmortem studies provide valuable context, they only partially align with our in vivo observations. Large autopsy series consistently highlight white matter thinning as a core Alzheimer anomaly that correlates more strongly with cognitive decline than EPVS burden, and further show that WMH volume rises up to 2 decades before symptom onset in autosomal-dominant AD.^42,43 By contrast, other neuropathologic studies mirror our results, demonstrating that basal-ganglia EPVS colocalize with cerebral amyloid angiopathy and predict higher cortical Aβ and tau loads, thereby linking deep-perforator clearance failure to Alzheimer-specific proteinopathies.^44,45 Such heterogeneity may reflect variations in vascular architecture, perivascular clearance efficiency, and regional vulnerability to amyloid-related injury. Moreover, autopsy studies involving individuals at the stage of MCI remain scarce, precluding definitive conclusions about the sensitivity of basal-ganglia EPVS as an early biomarker of AD. Our findings therefore underscore the need for prospective validation of MRI-derived EPVS metrics in prodromal cohorts.

In our cohort, biomarker findings support a growing body of evidence that EPVS burden, as visualized on MRI, may serve as a surrogate marker of impaired glymphatic function and early AD-related proteinopathy. These findings extend evidence implicating EPVS as a marker of microvascular and clearance dysfunction in Alzheimer pathogenesis. Confirmation in longitudinal and neuropathologic studies is still needed, but routine EPVS quantification could refine dementia imaging workups and help stratify risk among patients with MCI.

Higher EPVS burden was selectively associated with impaired performance on the BDN test, a multidomain measure encompassing visual organization, visuospatial processing, and executive function.⁴⁶ This finding parallels previous research^47,48 which established links between basal ganglia EPVS and executive dysfunction. A large community-based autopsy-MRI study further found that higher EPVS burden was associated with faster decline in visuospatial ability and lower semantic memory at death, even after adjustment for coexisting neuropathologies,⁴⁹ suggesting a domain-specific cognitive signature of EPVS burden. Our study extends these observations by highlighting a visuospatial component of EPVS-related cognitive deficits, suggesting that EPVS burden may affect multiple cognitive domains. Notably, EPVS burden was not associated with global cognition, memory, or processing speed, in contrast to WMH.⁵⁰ These findings suggest that EPVS-related cognitive effects may be more domain-specific and subtle than those associated with WMH.

A possible explanation for the effect of EPVS on these domains is that a high EPVS burden may compromise cerebrovascular integrity and glymphatic waste clearance in the highly vascularized basal ganglia, disrupting connections involved in visuospatial and executive function.⁷ The basal ganglia's extensive connections with the prefrontal cortex may help explain EPVS-related impairments in executive function.⁶ In addition, the caudate nucleus, a key basal ganglia structure, has been implicated in visuospatial processing through its interactions with the parietal cortex.⁵¹ Additional mechanistic studies are needed to delineate the precise pathophysiologic pathways linking EPVS burden to cognitive dysfunction.

Given that MRI is a routine component of the diagnostic workup for cognitive impairment, our findings emphasize the potential utility of EPVS as an early and cost-effective imaging biomarker for AD. Although WMH remains the most studied CSVD marker in AD diagnostics,^40,41,50 EPVS may be more advantageous because it appears earlier in the course of CSVD⁵ and exhibits domain-specific effects on cognition,^47,48 as supported by our findings. Identifying high EPVS load, whether incidentally or during a dementia workup, could enable health care providers to implement regular monitoring and encourage patients to be proactive about their cognitive health by taking preventative steps before marked symptom deterioration. Moreover, integrating both WMH and EPVS ratings into standard MRI protocols would enable the simultaneous review of complementary neuroimaging data, thereby streamlining the diagnostic process for prodromal and clinical AD. This could help triage further targeted cognitive and biochemical testing. This cost-effective approach reduces the need for expensive biomarker testing and extensive neuropsychological assessments, making it particularly beneficial in hospital settings with limited resources.

A major strength of this study is its large sample size and focus on a multiethnic Southeast Asian population, a demographic that remains underrepresented in dementia research. With differences in dementia phenotypes and CSVD risks between ethnic and regional groups,⁵² our findings provide important insights into regional variations in AD pathophysiology. In addition, the detailed cognitive assessments used in this study allowed for a granular analysis of EPVS-related cognitive deficits, particularly in visuospatial and executive domains.

Several limitations should be acknowledged. The cross-sectional design precludes causal inferences, leaving it unclear whether EPVS burden drives cognitive decline or reflects underlying neurodegeneration. Longitudinal studies within the BIOCIS cohort will clarify its prognostic value. In addition, the absence of amyloid and tau PET imaging limits direct confirmation of AD pathology because plasma biomarkers, although informative, are not definitive diagnostic tools. Multiple comparison corrections were not applied, warranting cautious interpretation and independent replication to minimize false-positive results. Despite adjustment for key confounders, residual confounding—such as differentiating vascular-driven from amyloid-driven mechanisms—cannot be ruled out. Relying on visual EPVS ratings, although by blinded raters, remains subjective, and the adoption of automated MRI pipelines could enhance both accuracy and reproducibility. Finally, because this study was conducted in a Southeast Asian cohort, findings may not fully generalize to other populations with different cerebrovascular risk profiles.

This study demonstrates that a high basal ganglia EPVS burden correlates strongly with AD biomarkers and is selectively associated with visuospatial and executive dysfunction. Notably, EPVS burden emerged as the strongest CSVD marker of AD pathology in participants with MCI, suggesting that it may offer greater predictive value than WMH and other CSVD markers in early-stage AD. Although WMH remains fundamental in AD neuroimaging, our findings suggest that EPVS quantification could provide complementary information for risk stratification and clinical decision making.

Future research should focus on longitudinally examining causal relationships between EPVS, AD biomarkers, and cognitive decline. Further investigation into the mechanistic pathways linking EPVS burden to glymphatic dysfunction and neurodegeneration may clarify EPVS's role as an imaging biomarker for prodromal AD.

Glossary

Aβ

β-amyloid

AD

Alzheimer disease

ANOVA

analysis of variance

ASL

arterial spin labeling

BBM

blood biomarkers

BDN

Block Design

BIOCIS

Biomarkers and Cognition Study, Singapore

CN

cognitively normal

CSVD

cerebral small vessel disease

CT

Color Trial

DS

Digit Span

EPVS

enlarged perivascular spaces

FCRST

Free and Cued Selective Reminding Test

GFAP

glial fibrillary acidic protein

GM

gray matter

HbA1C

glycated hemoglobin

MB

microbleeds

MCI

mild cognitive impairment

NfL

neurofilament light chain

NFT

neurofibrillary tangle

OAβ

amyloid β oligomers

OR

odds ratio

p-tau181

phosphorylated tau 181

PVS

perivascular spaces

RAVLT

Rey Auditory Verbal Learning Test

RCFT

Rey Complex Figure Test

SCD

subjective cognitive decline

SWI

susceptibility-weighted imaging

T1

T1-weighted

T2-FLAIR

T2-weighted fluid-attenuated inversion recovery

TE

echo time

TI

inversion time

TIV

total intracranial volume

TOPJ

Test of Practical Judgment

TR

repetition time

VCAT

Visual Cognitive Assessment Test

WM

white matter

WMH

white matter hyperintensities

Author Contributions

J.J.H. Ong: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data. Y.J. Leow: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data. B. Qiu: drafting/revision of the manuscript for content, including medical writing for content. P. Tanoto: drafting/revision of the manuscript for content, including medical writing for content. F.Z. Zailan: drafting/revision of the manuscript for content, including medical writing for content. G.K. Sandhu: drafting/revision of the manuscript for content, including medical writing for content. N. Kandiah: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data.

Study Funding

This study received funding support from the Strategic Academic Initiative grant from the Lee Kong Chian School of Medicine, Nanyang Technological University, Singapore, National Medical Research Council, Singapore under its Clinician Scientist Award (MOH-CSAINV18nov-0007), Ministry of Education Start-up Grant, Ministry of Education Academic Research Fund Tier 1 (RT02/21), and Ministry of Education Science of Learning Grant (MOESOL2022-0002).

Disclosure

The authors report no relevant disclosures. Go to Neurology.org/N for full disclosures.