Plasma proteomics of cerebrovascular disease, cognitive decline, and clinical outcomes

Abstract

INTRODUCTION

The plasma proteomic signatures underlying cerebrovascular disease (CeVD) remains poorly understood.

METHODS

A total of N = 2534 participants across two independent longitudinal Southeast‐Asian cohorts were included. We profiled 1441 baseline plasma proteins in a memory‐clinic cohort (N = 518), followed‐up for 4 years. Proteins associating with baseline and longitudinal CeVD lesions (i.e., white matter hyperintensity volume, lacunes, cerebral microbleeds, and cortical infarcts) were reported. The prognostic value of CeVD‐associated proteins was evaluated for incident major cardiovascular/cerebrovascular events (MACCE) and mortality. External validation of key proteins for mortality was performed in the plasma proteome of an independent cardiovascular cohort (N = 2016).

RESULTS

We report distinct and overlapping plasma proteins for baseline and longitudinal CeVD, representing diverse biological processes. Four proteins were prioritized as mediators of CeVD‐associated cognitive decline, and predictors of MACCE. These proteins were validated for incident mortality across both cohorts: neurofilament light chain (NEFL), latent‐transforming growth factor beta‐binding protein 2 (LTBP2), cysteine‐rich motor neuron 1 protein (CRIM1), and urokinase plasminogen activator surface receptor (PLAUR).

DISCUSSION

The prognostic proteins prioritized in our study provide robust signals in two cohorts, representing potential mechanistic targets for CeVD and health outcomes.

Highlights

• We report cross‐sectional and longitudinal proteomic signatures of cerebrovascular disease (CeVD).

• These proteins linked CeVD‐associated cognitive decline, major cardiovascular/cerebrovascular events (MACCE), and mortality.

• External validation was performed in the plasma proteome of an independent cohort.

• A concordant proteomic signature was identified across both cohorts.

• Further studies are needed to explore their mechanistic relevance for CeVD.

1. BACKGROUND

Neuroimaging markers of cerebrovascular disease (CeVD) include lacunes, cortical infarcts, cerebral microbleeds (CMB), and white matter hyperintensities (WMH). ¹ The neuroimaging phenotype of CeVD has been recognized as a prognostic marker for stroke, dementia, as well as acceleration of cognitive decline. ² , ³ , ⁴ , ⁵ CeVD lesions such as WMHs and lacunar infarcts are hypothesized to contribute independently and synergistically to dementia and adverse cognitive outcomes. ⁴ , ⁵

Epidemiological studies have highlighted vascular risk factors, cardiac dysfunction, and cardiometabolic health as potentially modifiable contributors to CeVD. ⁶ However, the underlying biological mechanisms driving CeVD pathogenesis and progression remain poorly understood. This knowledge gap poses an impediment to progress in developing novel CeVD therapeutics, and the discovery of validated blood‐based biomarkers for predicting and diagnosing CeVD. ⁶ , ⁷ , ⁸ Prior blood biomarker studies for CeVD have primarily focused upon single candidate platforms including biomarkers related to vascular integrity, endothelial dysfunction, neuronal injury, and subclinical cardiac dysfunction. ³ , ⁸ However, the multifactorial pathobiology of CeVD and its progression remains under‐investigated.

The advent of multiplexed affinity‐based proteomics has provided emerging insights into the diverse biological processes underpinning CeVD. Several studies have evaluated proteomic correlates of baseline CeVD burden; however, the majority lack longitudinal neuroimaging data and fail to elucidate the pathogenic mechanisms driving CeVD lesion‐specific progression. ⁹ , ¹⁰ , ¹¹ , ¹² , ¹³ Asian patients, who are known to be characterized by a significantly higher burden of CeVD compared to their Caucasian counterparts, are under‐represented in existing reports. ¹⁴ Taken together, this underscores the importance of longitudinal studies characterizing the proteomic landscape of CeVD in the Asian vascular cognitive endophenotype.

We profiled the plasma proteome of a well‐annotated Southeast‐Asian memory clinic cohort followed‐up for 4 years and investigated protein signatures related to CeVD burden. We first identified proteins associated with the burden of baseline and longitudinal CeVD lesion‐specific topography (lacunes, CMBs, cortical infarcts and WMH volume). Next, we elucidated how these proteins served as a mediating link between CeVD and future cognitive decline. We then ascertained their clinical prognostic value internally, by benchmarking these proteins as predictive markers of incident clinical events (i.e. major adverse cardiovascular and cerebrovascular events [MACCE] and mortality). Finally, we externally validated these key proteins as predictors of mortality within an independent Southeast‐Asian cardiovascular cohort.

RESEARCH IN CONTEXT

1. Systematic review: Cerebrovascular disease (CeVD) is significantly associated with accelerated cognitive decline and morbidity. However, current understanding of its underlying mechanisms remains poor. The advent of plasma proteomics allows for large‐scale screening for the identification of diverse biomarkers for CeVD, yielding insights into its underlying mechanisms.

2. Interpretation: We profiled 1441 plasma proteins in a longitudinally followed memory clinic cohort. We identified plasma proteomic signatures of cross‐sectional and longitudinal CeVD lesion‐specific topography—spanning lacunes, cortical infarcts, cerebral microbleeds and white matter hyperintensities. These plasma proteins implicated in CeVD represented diverse biological processes. A four‐protein signature was prioritized for CeVD, cognitive decline, major adverse cardiac/cerebrovascular events and mortality. External validation in the plasma proteome of an independent longitudinally followed cardiovascular cohort replicated these proteins as predictors of incident mortality: neurofilament light chain (NEFL), latent‐transforming growth factor beta‐binding protein 2 (LTBP2), cysteine‐rich motor neuron 1 protein (CRIM1), and urokinase plasminogen activator surface receptor (PLAUR).

3. Future directions: These proteins may serve as prognostic biomarkers and mechanistic targets for CeVD and its clinical sequelae.

2. METHODS

2.1. Primary study cohort: Memory Aging and Cognition Centre Harmonisation Cohort

A Southeast‐Asian memory clinic cohort (Memory Aging and Cognition Centre [MACC] Harmonisation Cohort) was followed up with serial neuroimaging and cognitive assessments for up to 4 years. The MACC Harmonisation Cohort comprised participants recruited from memory clinics from two Singaporean study sites (National University Hospital, and the St Luke's Hospital Singapore). Participants had to be ≥ 50 years, with adequate language skills to partake in neuropsychological assessments. Exclusion criteria included substance abuse disorders, major psychiatric illness, tumors, multiple sclerosis, traumatic brain injury resulting in cognitive impairment, and epilepsy, as described previously within the detailed study protocol. ¹⁵ Written informed consent was obtained prior to study recruitment. All research was conducted in accordance with the principles outlined in the Declaration of Helsinki. Ethics approval for this study was obtained from all participating institutions (National Healthcare Group Domain Specific Review Board (NHG DSRB reference numbers 2018/01098 and 2010/00017). An overview of the study design is presented in Figure 1.

FIGURE 1.

Overview of study. CeVD, cerebrovascular disease; MACCE, major adverse cardiovascular or cerebrovascular events.

2.2. Assessment of clinical characteristics

Clinical data was collected from all participants including age, gender, education level, and clinical risk factors such as hypertension, hyperlipidemia, diabetes, smoking status, anti‐hypertensive use, and lipid‐lowering medications using questionnaires administered in a standardized manner by trained study personnel. ¹⁵ Baseline systolic blood pressure was collected, as was apolipoprotein E4 (APOE4) status as determined using genotyping described previously. Subjects possessing one or more APOE4 allele were defined as APOE4 carriers. ¹⁶ Clinical characteristic data was available for all participants.

2.3. Neurocognitive assessments and follow‐up

Neurocognitive diagnoses for participants were ascertained at regular consensus meetings attended by study clinicians and neuropsychologists, where brain neuroimaging data and clinical cognitive assessments were reviewed. ¹⁵ Baseline Clinical Dementia Rating ‐ Global scores (CDR‐GS) were assigned at baseline. Yearly cognitive assessments were administered in the participant's native language (English, Mandarin, or Malay) by trained research psychologists, comprising a locally validated multi‐domain neuropsychological test battery. The domains assessed included: Attention (Digit span forward and backward test), Language (15‐item modified Boston Naming Test), Executive function (verbal fluency, and color trails test A and B), Visuomotor speed (symbol digit modalities test), Visuospatial function (Rey Complex Figure Test copy), Memory (Rey Complex Figure Test immediate/delayed test recall and recognition). ¹⁵ Raw scores for individual tests were converted to standardized Z‐scores using the standard deviations and mean values of a study reference group with no cognitive impairment (NCI). The Global Cognition Z‐score for each participant was obtained by averaging the Z‐scores of the individual tests of each domain, as we have previously published. ¹⁵ , ¹⁷ Dementia was diagnosed utilizing the Diagnostic and Statistical Manual of Mental Disorders Fourth edition. Participants exhibiting significant impairment in ≥ 1 cognitive domain without loss of independent daily function were classified as cognitive impairment no dementia (CIND) as previously described. ¹⁵ , ¹⁸ , ¹⁹

2.4. Plasma proteomic profiling

Non‐fasting blood samples were collected within 12 months of baseline neuroimaging assessment, drawn into ethylenediaminetetraacetic acid tubes, and centrifuged at 2000 × g for 10 minutes at 4 degrees Celsius. Plasma fractions were aliquoted and stored at ‐80 degrees Celsius until use and centrifuged at 3500 × g to remove particulate material prior to assay. Plasma proteomic profiling was performed using the Olink 1536 platform (Thermo Fisher Scientific, Waltham, MA, USA). A total of 1441 proteins representing major biological pathways were included for final analysis after rigorous quality control in accordance with the manufacturer's protocols and as we have previously published (the final proteins included for analysis are presented in Table S1). ¹⁷ , ²⁰ Protein levels were expressed as Normalized Protein eXpression (NPX) values for analysis. ¹⁷

2.5. Neuroimaging assessments of baseline and longitudinal CeVD lesions

Neuroimaging protocols were standardized and included three‐dimensional T1 and T2‐weighted fluid‐attenuated inversion recovery and susceptibility weighted images, performed on a 3T Siemens Magnetom Trio Tim Scanner, comprising a 32‐channel head receive coil. The full neuroimaging protocol is detailed previously. ¹⁵ , ¹⁷ , ²¹ , ²² Brain magnetic resonance imaging (MRI) scans were performed at baseline and every 2 years, up until 4 years of follow‐up. Markers of CeVD were graded at each timepoint including: WMH volume, lacunes, CMBs, and cortical infarcts. WMH volume was assessed for each participant using automated MRI brain volumetric analysis (AccuBrain IV2.0) following our previously described approach. ¹⁷ These were subject to log10 transformation and adjusted for total intracranial volume prior to analysis. The AccuBrain platform has been validated previously to have comparable accuracy to commonly utilized volumetric analytical platforms including FreeSurfer. ²³ Longitudinal WMH volume progression was defined as the yearly absolute increase in the volume of WMH volume for each participant, ascertained across serial MRI scans. The burden of CeVD structural lesions (lacunes, CMBs, and cortical infarcts) was graded by an expert clinical researcher using standardized approaches, as previously published. ¹⁵ , ¹⁷ , ²² In brief, CMBs were defined as round focal hypointense lesions with a blooming effect on susceptibility‐weighted images graded using the Brain Observer Microbleed Scale ²⁴ ; cortical infarcts defined as hypointense lesions interrupting the grey white cortical junction; and lacunes defined as 3‐ to 15‐mm round or ovoid lesions with a hyperintense rim within the subcortical region, with a high signal on T2‐weighted images coupled with a low signal on T1‐weighted images and fluid attenuation inversion recovery. ¹⁸ , ²² Longitudinal CeVD lesion burden was defined by the burden of new lacunes, CMBs, or cortical infarcts on the follow‐up scans which were absent at baseline. Significant CeVD at baseline was defined as the presence of two or more lacunes, and/or cortical infarcts, and/or confluent white matter hyperintensities in two or more regions of the brain defined by an age‐related white matter hyperintensities changes (ARWMC) score of ≥8, as we have published previously. ¹⁷ , ¹⁸

2.6. Follow‐up for incident MACCE and mortality

Patients were followed up for incident MACCE, defined as the earliest occurrence of new stroke, transient ischemic attack (N = 1), or cardiac events in alignment with the World Health Organization Criteria. ²⁵ Transient ischemic attack was included in our MACCE definition, given its recognition as a cerebrovascular disorder and its clinical relevance due to its association with future stroke risk. ²⁶ This was conducted via yearly study interviews, and review of medical records commencing within 12 months prior to blood sampling. Incident all‐cause mortality was defined as death from any cause during follow‐up. MACCE and mortality data for each participant underwent two‐level clinical adjudication by a board‐certified Cardiologist and Intensivist, who were blinded to the study conclusions (M.A.S. and E.S.J.T.).

2.7. External validation: Prognostic value of key proteins for mortality within a Southeast‐Asian cardiovascular cohort

Proteins commonly associating with CeVD (baseline and longitudinal), CeVD‐associated cognitive decline, MACCE, and mortality within the MACC Harmonisation Cohort were tested for their prognostic associations for all‐cause mortality within an independent Southeast‐Asian cardiovascular cohort. The Southeast‐Asian cardiovascular cohort comprised a prospective longitudinal cohort design, of N = 2016 participants followed longitudinally, recruited through the Singapore Heart Failure Outcomes and Phenotypes study and the Asian Network for Translational Research and Cardiovascular Trials, with the detailed study protocol described previously. ²⁷ , ²⁸ The prospective cohort comprised patients with heart failure recruited across six tertiary centers in Singapore (N = 1220), as well as age‐, sex‐, ethnicity‐matched participants recruited from the community without heart failure (N = 796) aged above 55 years of age, who were able to self‐ambulate and possessed sufficient cognitive capacity for participation. The exclusion criteria included heart failure secondary to severe valvular disease, heart failure secondary to acute coronary syndrome resulting in acute pulmonary edema, renal failure (estsimated glomerular filtration rate [eGFR] < 15 ml/min per 1.73 m²), and specific causes of heart failure such as constrictive pericarditis, complex adult congenital heart disease, hypertrophic cardiomyopathy, and isolated right heart failure. Participants with life‐threatening non‐cardiac comorbidities with an estimated life expectancy of less than 1 year were excluded. Institutional review board approval was obtained for all participating centers (ACTRN12610000374066, NCT02791009). All participants provided written informed consent.

Clinical covariates including age, gender, race, and medical comorbidities (such as hypertension, diabetes) were collected for each participant. New York Heart Association (NYHA) status was ascertained for all participants. ²⁷ Plasma samples were collected from each participant at baseline, with proteomic profiling performed using the Olink Explore 3072 platform. Data quality control processes followed a similar approach as the MACC Harmonisation Cohort, with relative protein abundances expressed as NPX values for downstream analysis. The outcome of interest was all‐cause mortality prospectively ascertained from medical record linkage or review of clinical records. ²⁷

2.8. Statistical analysis

Statistical analysis was performed using R version 4.3.2 (R Core Team. 2023. R Foundation for Statistical Computing, Vienna, Austria) and STATA (StataCorp. 2023. Release 18. College Station, TX: Statacorp LLC). Statistical significance was taken at p < 0.05, or q‐value < 0.05 where correction for false discovery rate (FDR) was performed, unless stated otherwise. ²⁹ . All related data is available upon reasonable request from the corresponding author.

2.8.1. Cohort demographics

Participant characteristics were expressed as means ± standard deviations (SD) for continuous variables, and as percentages for categorical variables. Independent t‐tests were used for continuously expressed variables, and Chi square tests for the analysis of categorical data.

2.8.2. Protein signatures of baseline and longitudinal CeVD lesion‐specific burden

The associations of circulating proteins with baseline WMH volume were evaluated using multivariable linear regression, with derived regression coefficients, 95% confidence intervals and p‐values presented. WMH volume values were subject to log‐10 transformation and adjusted for total intracranial volume for analysis. The associations of circulating proteins with the baseline burden of CeVD structural lesion subtypes were evaluated using multivariate Poisson regression (for lacunes, cortical infarcts) and zero‐inflated negative binomial regression (for CMBs). The fully adjusted models were adjusted for age, gender, years of education, hypertension, hyperlipidemia, diabetes, smoking status, the difference between timing of MRI and blood draw, anti‐hypertensive use, lipid‐lowering medication use, APOE4 status, baseline systolic blood pressure, and individual protein levels. Results derived from the basic model adjusting for clinical biomarkers: age, gender, years of education, hypertension, hyperlipidemia, and diabetes, are presented within the Supplemental. Derived incident risk ratio (IRR), 95% confidence intervals, and p‐values were reported.

Associations of circulating proteins for the burden of longitudinal CeVD followed a similar approach. The associations of circulating proteins with WMH volume progression were evaluated using multivariable linear regression models of the total increase in WMH volume over the follow‐up duration. Derived regression coefficients, 95% confidence intervals, and p‐values were reported. Protein associations with the longitudinal increment in CeVD lesion‐specific burden as compared to the baseline, were evaluated using Poisson regression (lacunes and cortical infarcts) or zero‐inflated negative binomial regression (CMBs). The basic model (for presentation in the supplemental) adjusted for age, gender, hypertension, hyperlipidemia, diabetes, follow‐up duration, baseline CeVD (expressed as the presence of CMBs, cortical infarcts, lacunes, or significant WMH (ARWMC score ≥ 2) at baseline, respectively); while the fully adjusted model (presented in the main results) adjusted additionally for the difference between timing of MRI and blood draw, anti‐hypertensive use, lipid‐lowering medication use, APOE4 status, smoking status, baseline systolic blood pressure, and individual protein levels. IRRs, 95% confidence intervals, and p‐values were reported. All p‐values were subject to multiple testing correction, and statistical significance calls were made based on q‐values. ²⁹ The top 10 most statistically significant proteins (ranked by p‐value) for each CeVD subtype identified on the fully adjusted model were presented as volcano plots.

Over‐representation analysis of all significant proteins associating with one or more CeVD lesion subtype at baseline or longitudinally (p < 0.05, q < 0.05) was performed using the DAVID bioinformatics resource, with biological pathways mapped to KEGG, Reactome and Gene Ontology terms. ³⁰ Where applicable, p‐values were corrected for multiple testing using the Bonferroni–Hochberg procedure.

2.8.3. Cross‐cohort prognostic validation of key proteins for clinical and cognitive outcomes

Mediation analysis (MACC Harmonisation Cohort)

All proteins associated with CeVD lesion burden (at baseline and longitudinally) were evaluated as potential mediators of CeVD‐related cognitive decline using regmedint of R. ³¹ We reported significant protein mediators of the association between significant baseline CeVD and subsequent slope of cognitive decline. Significant baseline CeVD was defined as the presence of two or more lacunes, and/or cortical infarcts, and/or confluent WMH in two or more regions (ARWMC score of ≥8). ¹⁷ , ¹⁸ The longitudinal slope of cognitive decline was modelled as participant‐specific slopes estimated from linear mixed effects models with random intercepts and random slopes of global cognition Z‐score over time. Models were adjusted for age, gender, hypertension, hyperlipidemia, diabetes, and baseline cognitive impairment (defined as CDR‐GS ≥ 0.5). For each protein, we reported the regression coefficients, standard error, and p‐values, for their direct and indirect effects on baseline CeVD, and slope of cognitive decline.

Prognostic associations for MACCE and mortality (MACC Harmonisation Cohort)

Proteins significantly associated with baseline and/or longitudinal CeVD were further assessed as candidate prognostic markers of incident mortality and MACCE. This was evaluated using Cox proportional hazards regression models adjusting for age, gender, hypertension, hyperlipidemia, baseline CeVD, diabetes, and years of education. Time to event (where event refers to mortality or MACCE) was specified as the timepoint when a participant demised or developed MACCE respectively. Harrel's C‐index was assessed for each protein. The overlapping protein signature associating with CeVD, future cognitive decline, incident MACCE, and mortality, was reported.

External validation: Prognostic associations for mortality (independent Southeast‐Asian cardiovascular cohort)

Demographic analysis of the independent Southeast‐Asian cardiovascular cohort followed a similar approach as the MACC Harmonisation Cohort.

The prognostic performance of common proteins relating to CeVD‐associated cognitive decline, MACCE, and mortality in the MACC Harmonisation cohort, were validated for their associations with all‐cause mortality in the Southeast‐Asian cardiovascular cohort. Cox proportional hazards regression models were employed, adjusting for age, gender, race, hypertension, diabetes, and NYHA status. Time to event (where the event refers to mortality) was determined as the timepoint when a participant demised. Harrel's C‐index was reported for each protein. Kaplan–Meier curves for each protein (stratified by quartiles) were visualized, with p‐values from log‐rank tests reported.

3. RESULTS

3.1. Clinical characteristics: MACC Harmonisation Cohort

Among 518 memory clinic participants included (mean age 72.9 ± 7.8 years), 56% were female, 71.6% had hypertension, and 33.8% had diabetes. Figure S1 shows a flowchart of study participant recruitment. Cohort demographics are presented in Table 1. In total, 300 (57.9%) had significant CeVD at baseline, of which 13.1% had cortical infarcts, 26.8% had lacunes, and 44.3% had CMBs. The average volume of WMH at baseline was 14.01 ± 16.22 mL. Table S2 shows the baseline demographics stratified according to baseline CeVD status.

TABLE 1.

Primary study cohort demographics (MACC Harmonisation Cohort).

Demographic	Total N = 518 (N (%), mean ± SD)
Age, years	72.9 ± 7.8
Education, years	7.1 ± 5.0
Gender, female (%)	290 (56.0)
Hypertension (%)	371 (71.6)
Hyperlipidemia (%)	392 (75.7)
Diabetes (%)	175 (33.8)
Baseline CDR‐GS	0.67 ± 0.69
Baseline cognitive diagnosis
NCI (%)	117 (22.6)
CIND (%)	212 (40.9)
Dementia (%)	189 (36.5)
Baseline significant CeVD (%)	300 (57.9)
APOE4 (%)	152 (29.3)
Anti‐hypertensive medications (%)	267 (51.5)
Lipid lowering medications (%)	336 (64.9)
Smoker (%)	130 (25.1)
Baseline systolic blood pressure, mmHg	141.17 ± 16.67
Presence of CeVD lesion subtype at baseline
WMH volume, ml	14.01 ± 16.22
Cortical infarcts (%)	68 (13.1)
Lacune (%)	139 (26.8)
Cerebral microbleeds (%)	225 (44.3)
Longitudinal CeVD lesion burden*
Progression in WMH volume, ml	5.89 ± 7.82
Cortical infarcts (%)	27 (5.4)
Cerebral microbleeds (%)	147 (30.4)
Lacunes (%)	66 (13.2)
Mortality (%)	47 (9.1)
Incident MACCE (%)	30 (5.8)

Abbreviations: APOE4, apolipoprotein E4; CDR‐GS, Clinical Dementia Rating ‐ Global Score; CeVD, cerebrovascular disease; CIND, cognitive impairment no dementia; MACCE, major adverse cardiovascular and cerebrovascular events; NCI, no cognitive impairment; WMH, white matter hyperintensity.

*max N = 506.

3.2. Protein signatures for baseline burden of CeVD lesion subtypes

Distinct and overlapping protein signatures were identified for each CeVD lesion subtype. Of the 1441 proteins profiled, 703 proteins associated with the baseline burden of one or more CeVD lesion subtype, on the fully adjusted model adjusting for age, gender, education, smoking, hypertension, hyperlipidemia, diabetes, the difference between timing of MRI and blood draw, anti‐hypertensive use, lipid‐lowering medication use, APOE4 status, and baseline systolic blood pressure. The 177 proteins associated with the burden of lacunes are presented in Table S3. The proteins with strongest statistical association with lacune burden included markers of axonal injury (neurofilament light chain [NEFL]), inflammation (tumor necrosis factor [TNF]), and cell adhesion (FES). ³² , ³³ Table S4 presents the 270 proteins associated with CMB burden, with the top three proteins being markers of neurogenesis (CHRDL1) and cell adhesion (GPNMB, CX3CL1). ³⁴ , ³⁵ For cortical infarct burden, 274 proteins showed statistically significant association (Table S5), with the top proteins including markers of myocardial strain (N‐terminal pro‐B‐type natriuretic peptide [NT‐proBNP]), neuroinflammation (glial fibrillary acidic protein [GFAP]), cytokine signaling (GDF15, IL4R), and cardiac fibrosis (SDC4). ³⁶ , ³⁷ , ³⁸ , ³⁹ A total of 309 proteins associated with WMH volume (Table S6), with the most significant proteins relating to axonal injury (NEFL), inflammatory response (CCL27), and cytokine response (TIMP4). ³² , ⁴⁰ Results derived from the basic analysis models adjusting for age, gender, education, hypertension, hyperlipidemia, diabetes are presented within Tables S3‐S6.

The top 10 proteins showing statistically significant association for baseline CeVD lesions are indicated on volcano plots in Figure 2A‐D. Selected proteins for the baseline burden of all CeVD lesion subtypes are summarized by a Venn diagram in Figure 2E (Table S7). A seven‐protein signature was commonly dysregulated across CeVD lesion subtypes: GFAP, interleukin (IL) 19, NEFL, PTGDS, TIMP4, TNC, and VCAM1. Overrepresentation analysis of all proteins associating with the baseline burden of one or more CeVD lesion subtype implicated diverse biological pathways including cell adhesion, immune response, and extracellular matrix organization (Figure 2F, Table S8).

FIGURE 2.

Volcano plots of the top 10 most significant proteins for baseline CeVD lesions: (A) CMB burden; (B) lacune burden; (C) cortical infarct burden; (D) WMH volume. (E) Venn diagram of shared and distinct proteins associated with baseline CeVD lesion subtypes. (F) Top five most significant biological processes overrepresented within all significant proteins associative of one or more CeVD subtype. CeVD, cerebrovascular disease; CMB, cerebral microbleeds; GO, gene ontology; IRR, incident risk ratio; KEGG, Kyoto Encyclopedia of Genes and Genomes; WMHv, white matter hyperintensity volume.

3.3. Protein signatures of longitudinal CeVD progression

A total of 506 participants had available longitudinal MRI data over a mean follow‐up period of 39.7 ± 11.4 months. The average absolute increment in WMH volume over the follow‐up duration was 5.89 ± 7.82 mL. The incidence of new cortical infarcts was 5.4%, CMBs 30.4%, and lacunes 13.2%.

All proteins significantly predictive of longitudinal increments in the burden of CeVD lesion subtypes on the fully adjusted analysis model, are presented in Tables S9‐S12. The proteins most predictive of longitudinal CMB burden implicated Akt/AKT1 activity regulation (FKBP5) ⁴¹ and cellular response to oxidative stress (MSRA). ⁴² For the longitudinal burden of cortical infarcts, the most predictive proteins related to platelet activation (ENTPD2) ⁴³ and cell adhesion (SIGLEC7). ⁴⁴ For lacune progression, the most significant proteins included microtubule bundle formation (PSRC1) ⁴⁵ and positive regulation of autophagy (RAB37). ⁴⁶ The predictive proteins with the strongest statistical significance for the longitudinal increase in WMH volume progression related to axonal injury (NEFL), ³² protease inhibition (SERPINA11), and cardiovascular inflammatory processes (TIMP4). ⁴⁰ , ⁴⁷ Results derived from the basic analysis models are presented in Tables S9‐S12. The top 10 predictive proteins for longitudinal CeVD subtype burden (in terms of p‐value, with p < 0.05, q < 0.05) are presented as volcano plots in Figure 3A–D.

FIGURE 3.

Predictive proteins for the longitudinal burden of CeVD lesion progression. Volcano plots of the top 10 most significant proteins for longitudinal (A) CMB burden; (B) lacune burden; (C) cortical infarct burden; (D) longitudinal increase in WMH volume. (E) Venn diagram of shared and distinct proteins associating with longitudinal burden of CeVD lesion subtypes; (F) Top five most significant biological processes overrepresented within all proteins predictive of the progression burden of one or more CeVD subtype. CeVD, cerebrovascular disease; CMB, cerebral microbleeds; IRR, incident risk ratio; GO, gene ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes; WMHv, white matter hyperintensity volume.

The 372 proteins predictive of the progression of one or more CeVD lesion subtype are visualized as a Venn diagram in Figure 3E. While six proteins commonly predicted the progression in the burden of lacunes, CMBs, and WMH volume (EGF, GYS1, ITGB1BP2, PPIB, USP8, VSIR), no proteins were common to the progression of cortical infarcts and the aforementioned three CeVD lesions (Table S13). Overrepresentation analysis of the 372 predictive proteins for the longitudinal progression of one or more CeVD subtype, highlighted biological pathways including hemostasis, signal transduction, PI3K‐Akt, and MAPK signaling (Figure 3F, Table S14).

3.4. Validation across two independent cohorts: Prognostic value of key proteins for cognitive decline, MACCE, and mortality

Of the 1441 proteins profiled, 213 proteins were significantly associated with the baseline and longitudinal progression of one or more CeVD subtype on fully adjusted analysis models (p < 0.05, q < 0.05, Table S15). These proteins were benchmarked across the MACC Harmonisation cohort and subsequently, an independent Southeast‐Asian cardiovascular cohort, for their prognostic value for cognitive and clinical outcomes.

3.4.1. Internal validation for CeVD‐associated cognitive decline, mortality and MACCE (MACC Harmonisation Cohort)

To this end, we first leveraged cognitive follow‐up data from 518 participants over 39.8 ± 12.0 months. Participant‐level cognitive decline slopes were estimated using linear mixed effects models of the global cognitive Z‐score over time. Modelling these as continuous outcomes, we utilized linear regression (adjusted for age, education in years, gender, hypertension, hyperlipidemia, baseline cognitive impairment) and demonstrated that significant baseline CeVD associated with a steeper slope of cognitive decline over 4 years (β ‐0.16, 95% confidence interval [CI], ‐0.09, ‐0.24, p = 2.95E‐10). Mediation analysis (adjusting for age, education in years, gender, hypertension, hyperlipidemia, and baseline cognitive impairment) prioritized 31 proteins as significant mediators of the observed associations of CeVD with the longitudinal slope of cognitive decline (Table S16). Among those, NEFL mediated the observed association most strongly. The mediation pathway diagram for NEFL is presented in Figure 4A.

FIGURE 4.

Prognostic value of key proteins for CeVD‐associated cognitive decline and clinical events: (A) mediation pathway diagram of NEFL, significant baseline CeVD, and slope of 4‐year cognitive decline; (B) proportion of CeVD‐related cognitive decline mediated by the four overlapping proteins; (C) volcano plots of the four overlapping proteins for incident MACCE; and (D) all‐cause mortality (MACC Harmonisation Cohort); (E) four overlapping proteins (NEFL, LTBP2, CRIM1, PLAUR) commonly mediating CeVD‐associated cognitive decline, and predictive of incident MACCE and mortality in the memory clinic cohort. CeVD, cerebrovascular disease; MACC, Memory aging and Cognition center; MACCE, major adverse cardiovascular and cerebrovascular events; SE, standard error. * p < 0.05.

Of the 518 participants included, 30 (5.8%) developed incident MACCE, and 47 (9.1%) died. Using cox proportional hazards models adjusting for age, gender, education, hypertension, hyperlipidemia, baseline CeVD, and diabetes, we identified 13 significantly predictive proteins for incident MACCE, and 74 proteins predictive of all‐cause mortality (Table S17‐S18). Clinical biomarkers (age, gender, years of education, hypertension, hyperlipidemia, and diabetes) yielded a C‐index of 0.62 for mortality, and 0.65 for incident MACCE respectively. The most predictive protein for incident MACCE was CRIM1 (C‐index 0.76), while NEFL most predicted incident mortality (C‐index 0.74).

An overlapping four‐protein prognostic signature was validated as mediators of (1) CeVD‐associated cognitive decline, and predictive biomarkers of (2) MACCE and (3) mortality within the memory clinic cohort: NEFL, latent‐transforming growth factor beta‐binding protein 2 (LTBP2), cysteine‐rich motor neuron 1 protein (CRIM1), and urokinase plasminogen activator surface receptor (PLAUR) (Figure 4E). The proportion of CeVD‐associated cognitive decline mediated by the proteins are presented in Figure 4B. Prognostic associations of the 4 proteins for incident MACCE and mortality in the MACC Harmonisation cohort are visualized in Figure 4C–D, respectively. Inter‐protein correlations derived from pairwise Pearson's correlations of the four proteins are visualized as a heatmap in Figure S2. Overlapping and distinct prognostic proteins across outcomes in the memory clinic cohort are visualized in a Venn diagram in Figure 4E. When combined with clinical biomarkers these prioritized four proteins (NEFL, LTBP2, PLAUR, CRIM1) yielded a combined C‐index of 0.77 (delta‐C‐index 0.12) for MACCE, and 0.76 (delta C‐index 0.14) for mortality. The prioritized four‐protein signature related to biological processes including neuronal injury (NEFL), ³² heparin binding (LTBP2), ⁴⁸ regulation of cell adhesion (PLAUR), ⁴⁹ and vascular homeostasis (CRIM1). ⁵⁰

3.4.2. External validation: Prognostic associations for mortality (independent Southeast‐Asian cardiovascular cohort)

Baseline characteristics of the N = 2016 included participants from the Southeast‐Asian cardiovascular cohort are presented in Table S19. The mean age of participants was 58.9 ± 11.3 years, 68.4% were male, and 52.9% had hypertension. Participants were prospectively followed for a mean of 699 ± 288 days, with death occurring in 171 (8.5%) participants.

The four‐protein prognostic signature identified in the MACC cohort was externally benchmarked as prognostic biomarkers of mortality in the Southeast Asian cardiovascular cohort. Individual protein prognostic associations for mortality were evaluated using Cox proportional hazards regression adjusting for age, gender, diabetes, hypertension, race, and NYHA status. All four proteins were concordantly replicated as significant predictors of all‐cause mortality in the independent cardiovascular cohort. Kaplan–Meier survival curves and corresponding p‐values from log‐rank tests for each of the five proteins stratified by quartiles are presented in Figure 5. The most predictive protein for mortality was LTBP2 (C‐index 0.84, Table S20).

FIGURE 5.

External prognostic validation: Kaplan–Meier curves of the four overlapping protein signatures (NEFL, LTBP2, CRIM1, PLAUR) stratified by quartiles, for incident mortality in the independent cardiovascular cohort. p‐Values presented were derived from log‐rank tests.

4. DISCUSSION

We report novel plasma protein signatures of baseline and longitudinal progression of CeVD lesion burden. Diverse biological processes underlie these protein signatures including extracellular matrix organization, cell adhesion, inflammation, and immune response. Leveraging insights from 2534 participants across two independent cohorts, we identified four prognostic proteins strongly linked to CeVD‐associated cognitive decline, incident MACCE, and mortality: NEFL, LTBP2, PLAUR, and CRIM1. Apart from NEFL, these prioritized proteins are newly identified and may serve as novel prognostic biomarkers for cognition, cerebrovascular health and mortality, paving the way for future mechanistic studies.

We identified several associative biomarkers of baseline CeVD, which have been reported in previous cross‐sectional studies. A study conducted within the UK Biobank and Iceland 36K cohorts reported 12 inflammatory‐endothelial proteins associated with baseline CeVD burden. ⁹ Associations for five of the 12 proteins (METAP1D, EPHA2, MERTK, PEAR1, CD46) for one or more CeVD lesion subtype were replicated within our study. By contrast, the STRADL cohort of cognitively normal Scottish participants identified 2 proteins (APOB and TIM‐1) with the Somalogic platform associating with cross‐sectional WMH. ¹⁰ Neither protein was quantified in our study. The disparity in protein signatures across studies may also reflect inter‐platform differences in the coverage of protein panels utilised. ⁵¹ Nonetheless, we reaffirm the associations of cross‐sectional CeVD with SPON1, TNFRSF1A, EFEMP1, SPINK1 reported in the Atherosclerosis Risk in Communities (ARIC) study. ¹¹ We also replicated 16 of the 19 significant proteins for cross‐sectional WMH volume identified in the Heart Brain Connection Study in the Netherlands which profiled 92 inflammatory proteins using the Olink platform. ¹³ Notably, the protein NEFL which featured prominently in our cohort, emerged as a significant predictor of incident vascular dementia and stroke in two studies leveraging Olink data from the UK Biobank, thus affirming its critical role in the pathobiology of vascular cognitive health. ¹² , ⁵²

Our study is novel in several aspects. To our knowledge, we are one of few studies to report the proteomic profiles underpinning longitudinal CeVD progression, and the first to study this within a Southeast‐Asian longitudinal cohort. Previous studies have predominantly focused on Caucasian populations, resulting in a significant knowledge gap of the Asian vascular cognitive phenotype, known to exhibit an elevated prevalence of cerebrovascular disease. ⁹ , ¹⁰ , ¹¹ , ¹² , ¹³ , ¹⁴ , ⁵² Our work addressed this disparity, which has potential clinical implications, given the elevated prevalence of CeVD affecting this population. ⁹ , ¹⁰ , ¹¹ , ¹² , ¹³ , ¹⁴ , ⁵² Additionally, we extend current knowledge of the protein signatures of CeVD, previously characterized mainly in relation to WMH, by elucidating distinct protein signatures for cortical infarcts, lacunes, and CMBs. ¹⁰ , ¹³ These findings offer novel insights into the pathobiological underpinnings of the structural markers of CeVD pathophysiology (CMB, lacunes, and cortical infarcts) which are thought to be more closely related to vascular disease, especially considering that WMH volume may also have overlapping links to cerebral amyloid angiopathy. ⁵³ , ⁵⁴ , ⁵⁵ Interestingly, we found no significant overlap among the proteins predictive of incident cortical infarcts, compared with the rest of the CeVD subtypes; this is unsurprising, given that CMBs, lacunes, and WMH volume may correlate with small vessel etiology, while by contrast, cortical infarcts may relate additionally to larger vessel embolic phenomena. ⁵³ , ⁵⁴ Among the biological processes highlighted within the protein profiles underlying CeVD, the top pathways identified included extracellular matrix organization and cellular adhesion. The implication of cellular adhesion is unsurprising and accords with the known importance of neurovascular unit integrity and endothelial health in CeVD pathogenesis. ³ , ⁸

The four‐protein prognostic signature for CeVD‐related cognitive decline and clinical events replicated across both cohorts related to diverse pathobiological processes. These included: neuronal injury (NEFL), ³² heparin binding (LTBP2), ⁴⁸ regulation of cell adhesion (PLAUR), ⁴⁹ and vascular homeostasis (CRIM1). ⁵⁰ Of these, the protein which most strongly mediated the observed associations of CeVD with cognitive decline was NEFL. Amongst the cross‐validated proteins, LTBP2 was the most predictive of mortality within the independent cardiovascular cohort. The existing literature of NEFL and LTBP2 in the context of cerebrovascular health and cognition are hence discussed below.

NEFL is a primary component of the axonal cytoskeleton which is released in response to axonal injury. It has been linked to incident stroke in atrial fibrillation, cognitive decline, as well as the progression of CeVD (WMHs, lacunar infarcts, and CMBs). ⁵⁶ , ⁵⁷ , ⁵⁸ Our findings are consistent with these reports, reinforcing consistency of our results with existing bodies of work. ⁵⁶ , ⁵⁷ , ⁵⁸ , ⁵⁹ The prognostic utility of NEFL for CeVD demonstrated in our study may reflect the extent of neuronal injury sustained due to the disruption of cerebral microvasculature and neurovascular unit, underlying CeVD pathogenesis. ⁶⁰ It is thus unsurprising that NEFL levels provided strong mediation between baseline CeVD and future cognitive outcomes, as well as predicted incident clinical events in our study. Our findings highlight the role of NEFL as a prognostic marker of CeVD and overall health outcomes across memory clinic and cardiovascular cohorts, possibly related to ongoing neurovascular insult.

LTBP2 is an extracellular matrix protein belonging to the latent transforming growth factor‐beta (TGF‐beta) ‐binding protein family. It serves as a critical regulator of TGF‐beta signaling which in turn, been linked to cell differentiation, proliferation, and tissue repair. ⁶¹ It is postulated to be an important structural component of microfibrils, in addition to maintaining cell adhesion. In cardiovascular cohorts, upregulated LTBP2 has been linked to adverse clinical outcomes related to right ventricular dysfunction and dilated cardiomyopathy. ⁶² , ⁶³ In the area of cognitive health, one study conducted within the UK Biobank identified LTBP2 as one of the most prognostic proteomic markers of all‐cause incident dementia. ⁵² Although the associations of LTBP2 and CeVD are relatively under‐investigated, our findings align with conclusions drawn from previous studies, highlighting the importance of LTBP2 as a marker of cognitive and vascular health. This study extends current knowledge by demonstrating the clinical relevance of LTBP2 in mediating CeVD‐associated cognitive decline, in addition to predicting MACCE and mortality across cardiovascular and memory clinic cohorts.

4.1. Limitations

Our study has several limitations. Firstly, the proteins assayed were obtained from commercially available panels, which carries the inherent biases of pre‐selected protein panels. The variation in the coverage of different proteomic platforms is further affected by inter‐antigen variability in effective assay signaling. ⁵¹ Secondly, as one of the first studies to evaluate the protein signatures of CeVD in a Southeast‐Asian cohort, our early findings should be interpreted with awareness of the risk of type I error, and these findings need to be validated in external CeVD cohorts of sufficient sample size. Nonetheless, we have shown strong correlations of proteomic measurements across quantitative immunoassay platforms including the single‐molecule array (Simoa) in our previous work. ¹⁷ We were unable to adjust for low‐density lipoprotein (LDL) levels and renal function in this study. As a result, it is plausible that residual confounding related to circulating lipid levels may persist despite the adjustment for hyperlipidemia and use of lipid lowering medications. Additionally, the levels of renally filtered proteins, such as cystatin C, may have been affected by differences in renal clearance. ⁶⁴ The findings reported here therefore require further external and mechanistic validation within other independent cerebrovascular disease cohorts. Although the N = 12 participants without longitudinal neuroimaging data were generally comparable to those retained in terms of patient characteristics, those lost to neuroimaging follow‐up were marginally older in age (Table S21). This age effect could have introduced a risk of attrition bias. Lastly, although we were unable to perform proteomic benchmarking for CeVD in an external cohort specifically designed for the same outcome, we have reinforced the clinical importance of four highly prioritized proteins, by demonstrating their concordant signals for clinical outcomes in two independent cohorts. It was also noteworthy that this prioritized four‐protein signature exhibited significant inter‐protein correlations, potentially reflecting their roles in complementary biological pathways, linked to cerebrovascular health outcomes. Future mechanistic studies are required to disentangle the individual and potentially interacting roles of these proteins in cerebrovascular health pathogenesis. Nonetheless, this cross‐validation of protein signals across dimensions of cohorts and longitudinal outcomes (cognitive decline, MACCE and mortality), highlights their potential as mechanistic targets for the optimization of cerebrovascular health.

5. CONCLUSION

Our study identifies novel plasma protein signatures underpinning CeVD and its longitudinal progression, in a Southeast‐Asian memory clinic cohort. We further prioritize a prognostic four‐protein signature for CeVD‐related cognitive decline and incident clinical outcomes in the memory clinic cohort, which was externally validated for mortality in an independent cardiovascular cohort: NEFL, LTBP2, CRIM1, and PLAUR. Combining insights from more than 2500 participants across memory clinic and cardiovascular cohorts, we highlight the pathophysiological relevance of proteins related to biological processes including neuronal injury, immune response, and cell adhesion in cerebrovascular health. Future mechanistic studies are required to elucidate their roles as predictive biomarkers for cerebrovascular health outcomes.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest. Author disclosures are available in the Supporting Information.

CONSENT STATEMENT

Informed consent was obtained from all human subjects.

Supporting information

Supporting Information

Supporting Information