Associations of Circulating Platelet Endothelial Cell Adhesion Molecule‐1 Levels With Progression of Cerebral Small‐Vessel Disease, Cognitive Decline, and Incident Dementia

Abstract

Background

The association between platelet endothelial cell adhesion molecule‐1 (PECAM‐1) with cerebral small‐vessel disease and cognition in dementia‐free subjects remains uninvestigated.

Methods and Results

A prospective cohort of dementia‐free subjects was recruited from memory clinics and followed up for 5 years. Annual neurocognitive assessments and twice‐yearly brain magnetic resonance imaging scans were performed. Associations of baseline plasma PECAM‐1 levels with cerebral small‐vessel disease, cognitive decline (Montreal Cognitive Assessment scores and executive function Z scores), and incident dementia were evaluated. Of 213 subjects (aged 70.2±7.7 years, 51.2% men), median PECAM‐1 levels were 0.790 (interquartile range, 0.645–0.955 ng/mL). Compared with the highest tertile, subjects within the lowest PECAM‐1 tertile had greater cross‐sectional white matter hyperintensity volume (β=4.84 [95% CI, 0.67–9.01]; P=0.023), age‐related white matter change scores (β=1.39 [95% CI, 0.12–2.67]; P=0.033), and cerebral microbleeds (Adjusted risk ratio, 2.59 [95% CI, 1.19–5.62]; P=0.016). Of the 204 participants with follow‐up data (median, 60.0 [interquartile range, 60.0–60.0] months), 24 (11.8%) developed incident dementia. Compared with the highest tertile, subjects within the lower tertiles of PECAM‐1 had a higher risk of incident dementia (first tertile: adjusted hazard ratio [AHR], 4.52 [95% CI, 1.35–15.13]; P=0.024; second tertile: AHR, 3.28 [95% CI, 1.02–10.60]; P=0.047). The lowest PECAM‐1 tertile was associated with greater progression of white matter hyperintensity volume (β=4.15 [95% CI, 0.06–8.24]; P=0.047), cerebral microbleeds (incident relative risk [IRR], 2.21 [95% CI, 1.05–4.65]; P=0.036), and decline in executive function (β=−0.45 [95% CI, −0.76 to −0.14]; P=0.004), and Montreal Cognitive Assessment (β=−1.32 [95% CI, −2.30 to −0.35]; P=0.008) scores.

Conclusions

In dementia‐free subjects, lower circulating PECAM‐1 levels are associated with greater cerebral small‐vessel disease progression and cognitive decline, thus warranting future study as a potential therapeutic target.

Nonstandard Abbreviations and Acronyms

ARWMC

age‐related white matter changes

CMB

cerebral microbleed

CSVD

cerebral small‐vessel disease

MCI

mild cognitive impairment

NfL

neurofilament light chain

NMRC

National Medical Research Council

PECAM‐1

platelet endothelial cell adhesion molecule‐1

p‐tau181

phosphorylated tau181

WMH

white matter hyperintensity

Clinical Perspective.

What Is New?

• Our study demonstrated that in dementia‐free subjects, lower platelet endothelial cell adhesion molecule‐1 levels associated with greater cerebral small‐vessel disease burden and progression, as well as longitudinal cognitive decline.

What Are the Clinical Implications?

• Lower platelet endothelial cell adhesion molecule‐1 levels may help identify patients at risk of cerebral small‐vessel disease and worse cognitive outcomes.

• Larger, prospective studies are warranted to confirm the protective role of platelet endothelial cell adhesion molecule‐1 in early‐stage cognitive decline and cerebral small‐vessel disease, as well as its utility as a potential therapeutic target.

Dementia is associated with high socioeconomic costs and reduced life expectancy. There is increasing recognition of the diverse neuropathological processes which underpin age‐related cognitive decline. ¹ , ² Of these, cerebral small‐vessel disease (CSVD), manifested by the presence of lesions such as cerebral microbleeds (CMBs), lacunes, and white matter hyperintensities (WMHs) on brain imaging, ³ is a clinically significant disease entity, having been implicated in an increased risk of stroke, progression of Alzheimer disease and vascular dementias, and cognitive decline affecting preferentially executive function, with preservation of memory domains. ⁴ , ⁵ , ⁶ , ⁷ , ⁸ However, a pertinent knowledge gap that impedes the progress of therapeutics in this area is the lack of validated blood biomarkers for CSVD. ⁹ , ¹⁰ , ¹¹ Moreover, there remains a need for the elucidation of mechanistic pathways, beyond basic vascular risk factors, for the development and progression of CSVD. ¹⁰

Endothelial dysfunction is closely linked to CSVD pathophysiology, through its contributions to altered cerebral vasoreactivity, blood–brain barrier dysfunction, and impaired autoregulation of cerebral vessel blood flow. ⁴ , ⁵ , ⁸ Consequently, circulating markers of endothelial dysfunction hold promise as biomarkers for CSVD and cognitive decline. These markers may serve as potential indicators of underlying vascular disease pathology, thereby offering valuable insights into disease mechanisms.

Platelet endothelial cell adhesion molecule‐1 (PECAM‐1) is a glycosylated adhesion molecule expressed on endothelial cells. ¹² , ¹³ It is located strategically at cell–cell junctions and is integral to maintaining endothelial junctional integrity, vascular integrity, microvascular mechanosensory function, and endothelial barrier resistance to inflammatory challenges. ¹² , ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ , ¹⁸ Circulating PECAM‐1 levels may thus be a biomarker of functional endothelium, and an indicator of intact endothelial repair mechanisms. ¹⁷ Indeed, lower PECAM‐1 levels have been found to be associated with an increased risk of incident stroke, and slower poststroke neurological recovery. ¹⁹ , ²⁰ However, beyond stroke, no prior study has evaluated the role of PECAM‐1 as a prognostic biomarker for CSVD and cognitive decline within dementia‐free subjects. Therefore, investigating the utility of PECAM‐1 as a circulating biomarker for CSVD progression and cognitive decline fills crucial knowledge gaps in the risk stratification and mechanistic elucidation of therapeutic targets for these entities.

Given its close association with vascular integrity and function, we hypothesize that lower PECAM‐1 levels are associated with CSVD progression, worse cognitive performance affecting, in particular, executive function (as a hallmark of CSVD‐related cognitive impairment) and incident dementia. ⁸ , ²¹ , ²² , ²³ Accordingly, we investigated the associations of circulating PECAM‐1 levels with (1) cognitive decline and incident dementia, (2) cross‐sectional and longitudinal neuroimaging markers of CSVD, and (3) plasma and neuroimaging biomarkers of amyloid and neurodegeneration. Our study design, undertaken within a memory clinic cohort involving longitudinal assessments of neurocognitive function and neuroimaging, is well positioned to ascertain the temporal relationship between circulating PECAM‐1 levels and cognitive trajectory as well as CSVD progression.

METHODS

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

Study Cohort

A longitudinal, prospectively followed up cohort involving subjects recruited from memory clinics from 2 Singaporean study sites (National University Hospital Singapore and St Luke's Hospital) was used (Figure 1). Subjects were followed‐up for up to 5 years. The inclusion criteria were subjects aged ≥50 years with sufficient language skills to participate in neuropsychological assessments. Subjects with a diagnosis of major psychiatric illness, substance abuse disorder, traumatic brain injury resulting in cognitive impairment, tumors, multiple sclerosis, and epilepsy were excluded. To study the role of PECAM‐1 in early cognitive impairment, subjects with baseline dementia were excluded. The detailed study protocol has been previously described. ²⁴ , ²⁵ , ²⁶

Figure 1. Flowchart of subject recruitment and follow‐up.

Created with biorender.com. ARWMC indicates age‐related white matter changes; CSVD, cerebral small‐vessel disease; MoCA, Montreal Cognitive Assessment; and PECAM‐1, platelet endothelial cell adhesion molecule‐1.

Informed written consent was obtained before study recruitment. Ethics approval was obtained from the National Healthcare Group Domain Specific Review Board (NHG DSRB reference numbers 2018/01098 and 2010/00017).

Data from all subjects regarding age, sex, and education levels were collected via detailed questionnaires administered in a standardized fashion. Medical comorbidities including hypertension, hyperlipidemia, diabetes, body mass index, cardiovascular diseases (inclusive of previous myocardial infarction, stroke, heart failure, or atrial fibrillation) and smoking status were collected and verified using medical records. Apolipoprotein E4 status was determined using genotyping as previously described. ²⁵ Subjects with at least 1 apolipoprotein E4 allele were defined as apolipoprotein E4 carriers.

Among the 410 subjects recruited from memory clinics from August 2010 to August 2014, 79 subjects without available plasma samples and 118 subjects with dementia at baseline were excluded (Figure 1). Of 213 dementia‐free subjects included for analysis (aged 70.2±7.7 years, 51.2% men), 122 (57.3%) had no cognitive impairment, and 91 had (42.7%) mild cognitive impairment (MCI). Baseline subject demographics are presented in Table 1.

Table 1.

Baseline Characteristics of Study Subjects

Demographics, mean±SD/n (%)	All subjects (N=213)	PECAM‐1 tertile
		Tertile 1 (N=71)	Tertile 2 (N=73)	Tertile 3 (N=69)	P value
Age, y	70.2±7.7	68.4±7.5*	69.9±7.7	72.4±7.4*	0.007
Male, %	109 (51.2)	36 (50.7)	33 (45.2)	40 (58.0)	0.312
Body mass index, kg/m2	24.1±3.8	24.1±4.1	24.0±3.5	24.2±3.8	0.962
Education, y	8.4±5.0	8.7±4.9	8.7±5.0	8.0±5.1	0.614
Apolipoprotein E4 carrier	55 (25.8)	19 (26.8)	17 (23.3)	19 (27.5)	0.826
Ethnicity
Chinese	170 (79.8)	62 (87.3)	57 (78.0)	51 (74.0)	0.137
Malay	17 (8.0)	4 (5.6)	4 (5.5)	9 (13.0)
Indian	22 (10.3)	4 (5.6)	9 (12.3)	9 (13.0)
Others †	4 (1.9)	1 (1.4)	3 (4.1)	0 (0.0)
Hypertension	134 (62.9)	45 (63.3)	45 (61.6)	44 (63.8)	0.961
Hyperlipidemia	153 (71.8)	50 (70.4)	50 (68.5)	53 (76.8)	0.518
Diabetes	68 (31.9)	23 (32.4)	17 (23.2)	28 (40.6)	0.087
Smoking	58 (27.2)	18 (25.3)	21 (28.8)	19 (27.5)
Cardiovascular disease	37 (17.4)	10 (14.1)	8 (11.0)*	19 (27.5)*	0.022
Cognitive performance
MoCA score	22.0±4.6	22.1±4.4	22.4±5.0	21.4±4.3	0.397
Executive function Z score	−1.42±1.78	−1.29±1.79	−1.38±1.92	−1.60±1.62	0.565
Cerebral small‐vessel disease markers
Presence of lacunes	46 (21.7)	18 (25.7)	13 (17.8)	15 (21.7)	0.518
Presence of CMBs	69 (33.0)	29 (42.0)	23 (31.9)	17 (25.0)	0.103
Presence of cortical infarcts	15 (7.1)	4 (5.7)	5 (6.8)	6 (8.7)	0.787
ARWMC scores	5.95±3.87	6.58±4.39	5.63±3.38	5.65±3.77	0.255
Cognitive diagnoses at baseline
No cognitive impairment	122 (57.3)	40 (56.3)	43 (58.9)	39 (56.5)	0.941
MCI	91 (42.7)	31 (43.7)	30 (41.1)	30 (43.5)

ARWMC indicates age‐related white matter changes; CMB, cerebral microbleed; MCI, mild cognitive impairment; MoCA, Montreal Cognitive Assessment; and PECAM‐1, platelet endothelial cell adhesion molecule‐1.

^* P<0.05.

^† Includes Burmese and mixed ethnicity.

Blood Biomarker Assessment

Nonfasting blood samples were obtained from subjects at baseline. Blood samples were drawn into EDTA tubes and centrifuged at 2000g for 10 minutes at 4 °C. Plasma fractions were then extracted, aliquoted, and stored at −80 °C until use. Blood‐based biomarkers were analyzed in duplicates.

Plasma PECAM‐1 concentrations were analyzed by multiplex xMAP‐based Luminex immunoassays (MILLIPLEX, Merck Millipore, Billerica, MA; catalog number HCVD4MAG‐67K), in accordance with the manufacturer's protocols. Plasma samples were incubated overnight with a mixture of fluorescent‐coded magnetic beads coated with specific capture antibodies. Biotinylated antibody and streptavidin–phycoerythrin conjugates were then added. Median fluorescent intensities were analyzed on a Luminex 200 platform using xPONENT software (Thermo Fisher Scientific, Waltham, MA). The standard curves were fitted to a 5‐parameter logistic model, ranging from 0.27 ng/mL to 200 ng/mL.

To study associations of PECAM‐1 levels with blood‐based biomarkers of amyloid burden and neurodegeneration, plasma phosphorylated tau181(p‐tau181) and neurofilament light chain (NfL) levels were measured, respectively, using the SIMOA platform (Quanterix, Billerica, MA) as previously described. ²⁷ , ²⁸

Neurocognitive Assessments and Cognitive Diagnoses

Neurocognitive diagnoses of study subjects were discussed at regular consensus meetings attended by study clinicians and neuropsychologists, during which detailed clinical neurocognitive and brain MRI data were reviewed. ²⁴ , ²⁵ , ²⁶ Given its close association with vascular integrity and function, we hypothesized a priori that PECAM‐1 is associated with neurocognitive tests sensitive to CSVD, and with incident dementia. This is aligned with current recommendations for the use of Montreal Cognitive Assessment (MoCA) scores as a screening tool, and the known predominance of executive function deficits in vascular cognitive impairment. ⁸ , ²¹ , ²² , ²⁹ , ³⁰ , ³¹

Yearly cognitive assessments including MoCA, and a locally validated neuropsychological test battery was administered, by trained research psychologists, in the participants' native language (English, Mandarin, or Malay). ²⁶ Executive function was assessed using Verbal Fluency (naming), and the Color Trails tests 1 and 2 as previously described. ²⁶ , ³² , ³³ Raw scores for individual tests were then converted to standardized Z scores using the mean±SD values of the study reference group of subjects with no cognitive impairment. The executive function Z score was then obtained via averaging the Z scores of individual tests, and subsequently standardized using the mean±SD values of the reference group. ²⁶ To supplement our investigation of PECAM‐1 as a biomarker of CSVD‐related cognitive decline, we also sought to evaluate whether PECAM‐1 levels were associated with memory domains. Tests of memory were assessed yearly using the Rey Complex Figure Test–immediate/delayed recall and recognition and the Hopkins Verbal Learning Test–immediate/delayed test recall and recognition. ³⁴ , ³⁵ The memory Z score was subsequently obtained by averaging the Z scores of the individual tests, which were first converted to standardized Z scores using the mean±SD values values of the study reference group of subjects with no cognitive impairment, in accordance with our previously described approach. ²⁶ On longitudinal follow‐up, executive function and memory Z scores were calculated in a similar fashion, using the mean±SD values of the control group evaluated at baseline. ²⁶ Subjects without objective cognitive impairment were categorized as no cognitive impairment. In accordance with Petersen's criterion, subjects with subjective cognitive impairment together with objective cognitive impairment in ≥1 cognitive domains (defined by a score of at least 1.5 SDs below established education‐adjusted cutoff values on any test) were classified as MCI. ²⁴ , ²⁵ , ³⁶

Clinical Dementia Rating scores were assigned to each patient at baseline, and at every follow‐up visit to stage the severity of cognitive impairment and dementia. ³⁷ A global score (Clinical Dementia Rating –Global) was obtained by evaluating 6 domains of function (a score of 0 or 0.5 indicates no dementia, while ≥1 indicates dementia, with scores >1 indicating increasing dementia severity). ³⁷ Incident dementia was defined by a change in Clinical Dementia Rating –Global score from ≤0.5 to ≥1 at any annual follow‐up, compared with the baseline. ³⁷ , ³⁸

Neuroimaging Assessments of CSVD and Neurodegeneration Markers

Serial brain magnetic resonance imaging (MRI) scans were performed at baseline, at 2 years and at the fourth (or fifth) year of follow‐up. A 3T Magnetom Trio Tim Scanner (Siemens, Munich, Germany) was employed, consisting of a 32‐channel head receive coil. Neuroimaging protocols were standardized, and included 3‐dimensional T1‐ and T2‐weighted fluid‐attenuated inversion recovery and susceptibility‐weighted images as previously described. ²⁴ , ³⁹ Automated MRI brain volumetric analysis was performed for all subjects using AccuBrain (IV2.0; Shenzhen, China), to obtain serial quantitative MRI volumetric markers of neurodegeneration (hippocampal volume, gray matter volume) andCSVD (WMH volume). These were adjusted for total intracranial volume for subsequent analysis. The AccuBrain platform has been previously validated, demonstrating comparable accuracy with commonly used volumetric analytical platforms such as FreeSurfer. ⁴⁰ Additionally, markers of CSVD including lacunes, cortical infarcts, and CMBs at baseline, 2 years, and 4 years were graded by an expert clinical researcher (S.H.), in accordance with the Standards for Reporting Vascular Changes on Neuroimaging criteria as previously described ²⁴ , ³⁹ , ⁴¹ , ⁴² , ⁴³ :

1. Cerebral microbleeds: Focal, round, hypointense lesions with a blooming effect observed on susceptibility‐weighted images, graded using the Brain Observer Microbleed Scale. ⁴³

2. White matter hyperintensity burden: Hyperintense lesions on fluid‐attenuated T2‐weighted inversion recovery sequences and as hypointense lesions on T1‐weighted images. These were graded using the age‐related white matter changes (ARWMC) scale. ⁴² The presence of significant white matter hyperintensity burden at baseline was defined by an ARWMC score of ≥8 in accordance with our previously published approach. ⁴⁴

3. Cortical infarcts: Hypodense lesions interrupting the cortical gray–white junction.

4. Lacunes: Ovoid or round lesions (3–15 mm) with a hyperintense rim within the subcortical region, with low signal observed on T1‐weighted images and fluid‐attenuated inversion recovery, coupled with a high signal on T2‐weighted images. ²⁴ , ⁴¹

Neuroimaging findings graded by the same researcher (S.H.) within this cohort previously demonstrated good to excellent intrarater agreement of 0.79 to 0.88. ⁴⁵

Statistical Analysis

Statistical analysis was performed using STATA (Release 18; StataCorp, College Station, TX), by study authors M.A.S and S.P.C. All statistical tests were conducted at a 5% level of significance.

Patient characteristics were expressed as percentages for categorical variables and mean±SD for continuous variables. Plasma PECAM‐1 levels were stratified by tertiles for analyses. χ² tests were used in the analysis of categorical data. Independent t tests or 1‐way ANOVA tests were used in the analysis of normally distributed continuous variables. Mann–Whitney U tests or Kruskal–Wallis tests were used for the analysis of nonparametric data.

In view of its skewed distribution, PECAM‐1 levels were included as categorical variables, and expressed as tertiles for the subsequent models. To study dose‐dependent relationships of PECAM‐1 expressed as tertiles, we employed tests of linear trends (P‐trend) to assess for significant upward or downward trends across PECAM‐1 tertiles with the outcomes studied, in accordance with previously published approaches. ⁴⁶ , ⁴⁷ Tests for linear trends (P‐trend) were evaluated by modeling PECAM‐1 tertiles as numeric variables, as per previously published approaches. ⁴⁶

Considering its skewed distribution, PECAM‐1 levels were next subject to transformation to Z scores to approximate a normal distribution for subsequent analysis as a continuous variable and the results reported within Tables S1 through S5.

Multivariable regression models were used to study associations of baseline PECAM‐1 levels with cross‐sectional and longitudinal outcomes as described below. In all multivariable regression models, variables were entered simultaneously, to ascertain the effect of PECAM‐1 on the outcomes of interest, while still considering the effect of a clinically relevant subset of covariates (as described within each model below) for the outcomes of interest. This allowed for the demonstration of PECAM‐1's independent effect on the outcomes of interest while still considering the effect of these covariates within each model.

Cross‐sectional analysis to identify associations between PECAM‐1 levels with MoCA scores, memory Z scores, and executive function Z scores were performed with linear regression models adjusted for age, sex, education, apolipoprotein E4 status, hypertension, hyperlipidemia, smoking, diabetes, and the presence of cardiovascular diseases.

Cross‐sectional analysis of neuroimaging parameters was performed with multivariable linear regression models for WMH volume and ARWMC scores, and Poisson regression for the burden of CMBs, cortical infarcts, and lacunes by counts. The models adjusted for relevant covariates including age, sex, hypertension, hyperlipidemia, diabetes, smoking, body mass index, antiplatelet use, the presence of cardiovascular disease, and total intracranial volume (for WMH volume).

Linear mixed‐effect regression models were used to examine associations of baseline PECAM‐1 levels with cognitive decline in MoCA scores, memory Z scores and executive function Z scores. The linear mixed‐effect model constructed was a 2‐level random‐intercept model estimated with restricted maximum likelihood technique. The longitudinal visit data (level 1) were nested within subjects (level 2), and the intercept was allowed to vary. The model was applied for the continuous cognitive outcomes studied including executive function Z scores, memory Z scores, and MoCA scores. The model adjusted for age, sex, education, apolipoprotein E4 carrier status, cardiovascular disease, hypertension, hyperlipidemia, diabetes, smoking status, and all relevant baseline cognition (expressed as MoCA scores, memory, or executive function Z scores, respectively). The models adjusted for the timing of assessment of cognitive performance across longitudinal follow‐up. Marginal plot effects were used to visualize the results derived from mixed‐effect models for executive function Z scores and MoCA scores, using post hoc average marginal effects, for presentation in Figures S1 and S2.

Cox proportional hazards regression models were used to evaluate associations between baseline PECAM‐1 levels and incident dementia. The model adjusted for age, education, sex, and baseline MoCA scores. Nelson–Aalen cumulative hazard estimate graphs were constructed for the association of PECAM‐1 (stratified by tertiles) with incident dementia.

Longitudinal analysis of associations between PECAM‐1 with CSVD markers was performed using linear mixed‐effect models (for WMH volume and ARWMC scores), and multilevel mixed‐effect Poisson models (for CMBs, cortical infarct, and lacune burden by counts). The multilevel mixed‐effect Poisson model for count data was a 2‐level model with data nested within subjects. The reported coefficients (converted to incidence rate ratios) were the fixed effects, while the intercept was allowed to vary (random effect). The models were estimated with restricted maximum likelihood technique. The models adjusted for age, sex, hypertension, hyperlipidemia, diabetes, smoking, body mass index, antiplatelet use, cardiovascular diseases, timing of neuroimaging assessment, and all relevant CSVD markers at baseline (expressed as CMBs, cortical infarcts, lacunes, or significant WMH burden at baseline defined by an ARWMC score of ≥8, respectively).

Finally, cross‐sectional associations between PECAM‐1 levels and biomarkers of neurodegeneration and amyloid burden were evaluated. Cross‐sectional analysis of PECAM‐1 levels with MRI markers of neurodegeneration (gray matter volume and hippocampal volume) was evaluated using linear regression, adjusted for age, sex, and total intracranial volume. Associations between PECAM‐1 levels with log₁₀‐transformed NfL levels, and log₁₀‐transformed p‐tau181 levels were evaluated using linear regression adjusted for age and sex. To investigate if associations between PECAM‐1 and cognitive change remained independent of the amyloid‐related and neurodegenerative processes (represented by p‐tau181 and NfL), we conducted several sensitivity analyses. We first examined associations of PECAM‐1 levels with decline in MoCA scores, executive function Z scores, and memory Z scores using the aforementioned linear mixed‐effect models, additionally adjusted for baseline p‐tau181 and NfL levels. Subsequently, we assessed the independent effect of baseline PECAM‐1 levels on incident dementia, using Cox proportional hazards regression, additionally adjusted for baseline p‐tau181 and NfL levels. These sensitivity analyses are detailed in Tables S7 and S8.

RESULTS

Baseline demographics of the 213 subjects stratified by PECAM‐1 tertiles are presented in Table 1. Median PECAM‐1 levels were 0.790 (interquartile range, 0.645–0.955) ng/mL. PECAM‐1 levels were stratified by tertiles (first tertile: 0.31–0.692 ng/mL; second tertile: 0.693–0.900 ng/mL; and third tertile: 0.901–2.02 ng/mL) for analysis. Compared with those in the highest tertile, subjects within the lowest tertile of PECAM‐1 were younger (P=0.007) and had a lower prevalence of cardiovascular diseases (P=0.022; Table 1).

Association of PECAM‐1 Levels With Cross‐Sectional Outcomes

Cross‐Sectional Association of PECAM‐1 and Baseline Neurocognitive Assessment

PECAM‐1 levels by tertiles were similar among subjects with no cognitive impairment and MCI at baseline (P=0.941). Compared with the highest tertile, subjects within the lower tertiles of PECAM‐1 levels did not have significantly different MoCA scores (first tertile: β=−0.18 [95% CI, −1.50 to 1.14]; P=0.78; second tertile: β=0.28 [95% CI, −1.05 to 1.60]; P=0.68, P‐trend=0.757), memory Z scores (first tertile: β=0.02 [95% CI, −0.37 to 0.41]; P=0.919; second tertile: β=0.01 [95% CI, −0.38 to 0.40]; P=0.958, P‐trend=0.919), and executive function Z scores (first tertile: β=−0.10 [95% CI, −0.59 to 0.40]; P=0.70; second tertile: β=−0.12 [95% CI, −0.61 to 0.38; P=0.64, P‐trend=0.715]; Table S1). Similarly, there were no significant associations found when PECAM‐1 was analyzed as a continuous variable (Table S1).

Cross‐Sectional Association of PECAM‐1 With Neuroimaging Markers of CSVD

Cross‐sectional associations between circulating PECAM‐1 levels and CSVD markers on MRI are presented in Table 2. Compared with the highest tertile, subjects within the lowest tertile of PECAM‐1 had higher WMH volume (β=4.84 [95% CI, 0.67–9.01]; P=0.023, P‐trend=0.022), ARWMC scores (β=1.39 [95% CI, 0.12–2.67]; P=0.033, P‐trend=0.032), and burden of CMBs (adjusted risk ratio, 2.59 [95% CI, 1.19–5.62]; P=0.016, P‐trend=0.040). Consistent with analysis by tertiles, when expressed as a continuous variable, lower levels of PECAM‐1 were associated with higher cross‐sectional WMH volume, ARWMC scores, and burden of CMBs (Table S2).

Table 2.

Association of PECAM‐1 Levels With Cross‐Sectional Markers of Cerebral Small‐Vessel Disease

Cross‐sectional cerebral small‐vessel disease markers (max N=212)
PECAM‐1 (ng/mlL)	WMH volume†			ARWMC score†			Cerebral microbleed count‡			Lacune count‡			Cortical infarct count‡
	β	95% CI	P value	β	95% CI	P value	ARR	95% CI	P value	ARR	95% CI	P value	ARR	95% CI	P value
Tertile 3	Reference	Reference	Reference	Reference	Reference	Reference	Reference	Refeference	Reference	Reference	Reference	Reference	Reference	Reference	Reference
Tertile 2*	−0.71	−4.91 to 3.48	0.737	0.22	−1.05 to 1.49	0.733	0.75	0.24 to 2.32	0.617	0.79	0.30 to 2.06	0.627	0.43	0.13 to 1.40	0.161
Tertile 1*	4.84	0.67 to 9.01	0.023	1.39	0.12 to 2.67	0.033	2.59	1.19 to 5.62	0.016	0.97	0.49 to 1.92	0.941	1.45	0.40 to 5.18	0.571
P‐trend	0.022			0.032			0.040			0.934			0.873

ARR indicates adjusted risk ratio; ARWMC, age‐related white matter changes; PECAM‐1, platelet endothelial cell adhesion molecule‐1; and WMH, white matter hyperintensity.

^*With reference to the third (highest) tertile of PECAM‐1.

^† Linear regression model, adjusted for age, sex, education, intracranial volume, hypertension, hyperlipidemia, diabetes, smoking status, body mass index, antiplatelet use, cardiovascular disease. β (regression coefficient) represents adjusted associations with reference to the third (highest) PECAM‐1 tertile. Tests for linear trends (P‐trend) were obtained by modeling PECAM‐1 tertiles as numeric variables.

^‡ Poisson regression model, adjusted for age, sex, education, hypertension, hyperlipidemia, diabetes, smoking status, body mass index, antiplatelet use, cardiovascular disease. ARR represent adjusted risk ratio with reference to the third (highest) PECAM‐1 tertile. Tests for linear trends (P‐trend) were obtained by modeling PECAM‐1 tertiles as numeric variables.

Association of PECAM‐1 Levels With Longitudinal Outcomes

Association of PECAM‐1 With Longitudinal Cognitive Function

Of 204 subjects with at least 1 year of follow‐up data (median, 60.0 [interquartile range, 60.0–60.0] months), conversion to dementia occurred in 24 (11.8%) subjects. Of 82 subjects with baseline MCI, 22 (26.8%) of subjects converted to dementia. Of 122 subjects without baseline cognitive impairment, conversion to dementia occurred in 2 (1.6%) of subjects. Multivariable Cox regression analysis of associations between circulating PECAM‐1 levels and conversion to dementia are presented in Table 3. When compared with the highest tertile, subjects within the lower 2 tertiles were independently associated with 3 to 4 times higher hazards of dementia conversion (first tertile: adjusted hazard ratio [AHR], 4.52 [95% CI, 1.35–15.13]; P=0.024; second tertile: AHR, 3.28 [95% CI, 1.02–10.60]; P=0.047, P‐trend=0.012). Nelson–Aalen cumulative hazard estimates of circulating PECAM‐1 levels stratified by tertiles with conversion to dementia are shown in Figure 2.

Table 3.

Longitudinal Associations of PECAM‐1 Levels (By Tertiles) and Conversion to Dementia

PECAM‐1 (ng/mL)	Conversion to dementia (N=204)		Adjusted hazard ratio†	95% CI	P value
	Nonconverters, n (%)	Converters to dementia, n (%)
Tertile 3	60 (93.7)	4 (6.3)	Reference	Reference	Reference
Tertile 2*	60 (85.7)	10 (14.3)	3.28	1.02–10.60	0.047
Tertile 1*	60 (85.7)	10 (14.3)	4.52	1.35–15.13	0.014
P‐trend	0.012

PECAM‐1 indicates: Platelet endothelial cell adhesion molecule‐1.

^*With reference to the third (highest) tertile of PECAM‐1.

^† Adjusted hazard ratios and 95% CIs were derived from Cox regression models adjusted for age, sex, education, and baseline cognition. Tests for linear trends (P‐trend) were obtained by modeling PECAM‐1 tertiles as numeric variables.

Figure 2. Nelson–Aalen cumulative hazard estimates for conversion to dementia stratified by PECAM‐1 levels (tertiles).

PECAM‐1 indicates platelet endothelial cell adhesion molecule‐1.

Decline in cognitive function was significantly worse in the lowest tertile of PECAM‐1, with greater decline observed in executive function Z scores (β=−0.45 [95% CI, −0.76 to −0.14; P=0.004], P‐trend=0.004), and MoCA scores (β=−1.32 [95% CI, −2.30 to −0.35; P=0.008], P‐trend=0.008; Table 4). Marginal plot effects derived from post hoc average marginal effect plots (Figures S1 and S2) demonstrated that the highest tertile of PECAM‐1 had higher longitudinal MoCA and executive function Z scores, compared with the lower 2 tertiles over time. There was no significant association of PECAM‐1 levels with decline in memory Z scores (first tertile: β=−0.28 [95% CI, −0.57 to 0.02; P=0.064]; second tertile: β=−0.16 [95% CI, −0.045 to 0.14]; P=0.294; Table 4).

Table 4.

Associations of Longitudinal Cognitive Performance (MoCA Scores and Tests of Executive Function and Memory) With PECAM‐1 Levels Stratified by Tertiles

PECAM‐1 (ng/mL)	Adjusted coefficient†	95% CI	P value
MoCA scores
PECAM‐1 tertile 3	Reference	Reference	Reference
PECAM‐1 tertile 2*	−0.87	−1.86 to 0.11	0.08
PECAM‐1 tertile 1*	−1.32	−2.30 to −0.35	0.008
P‐trend	P=0.008
Executive function Z score
PECAM‐1 tertile 3	Reference	Reference	Reference
PECAM‐1 tertile 2*	−0.26	−0.57 to 0.05	0.105
PECAM‐1 tertile 1*	−0.45	−0.76 to −0.14	0.004
P‐trend	0.004
Memory Z score
PECAM‐1 tertile 3	Reference	Reference	Reference
PECAM‐1 tertile 2*	−0.16	−0.045 to 0.14	0.294
PECAM‐1 tertile 1*	−0.28	−0.57 to 0.02	0.064
P‐trend	0.064

MoCA indicates Montreal Cognitive Assessment; and PECAM‐1, Platelet endothelial cell adhesion molecule‐1.

^*With reference to the third (highest) tertile of PECAM‐1.

^† Linear mixed‐effect models adjusted for age, sex, apolipoprotein E4 status, education, time of cognitive assessment, baseline cognitive performance (expressed as MoCA, memory or executive function Z scores, respectively), hypertension, hyperlipidemia, smoking, diabetes, and other cardiovascular diseases. Adjusted regression coefficients represent mean decrease in Z scores or MoCA scores with reference to the third (highest) PECAM‐1 tertile. Tests for linear trends (P‐trend) were obtained by modeling PECAM‐1 tertiles as numeric variables.

Analyses of PECAM‐1 expressed as a continuous variable yielded largely similar results. Lower PECAM‐1 levels were associated with significantly increased hazard of incident dementia and greater decline in MoCA scores (Tables S3 and S4). Lower PECAM‐1 levels displayed a trend toward a greater decline in executive function Z scores, although this did not reach statistical significance (Table S4).

Longitudinal Associations of PECAM‐1 With Progression of Neuroimaging Markers of CSVD

Longitudinal associations between PECAM‐1 levels with progression of CSVD markers over 4 years are presented in Table 5. Compared with the highest tertile, subjects within the lowest tertile of PECAM‐1 had significantly greater progression in WMH volume (β=4.15 [95% CI, 0.06–8.24]; P=0.047, P‐trend=0.044) and CMB burden (incident relative risk, 2.21 [95% CI, 1.05–4.65]; P=0.036, P‐trend=0.034) over 4 years. However, the association between the lowest PECAM‐1 tertile and ARWMC scores was attenuated (β=0.42 [95% CI, −0.56 to 1.40]; P=0.402, P‐trend=0.398). When PECAM‐1 was analyzed as a continuous variable, results remained consistent with the primary analysis of PECAM‐1 expressed as tertiles (Table S5).

Table 5.

Association of PECAM‐1 Levels With Longitudinal MRI Markers of Cerebral Small‐Vessel Disease

Progression of cerebral small‐vessel disease markers over 4 years (max N=180)
PECAM‐1 (ng/mL)	Progression of WMH volume†			Progression of ARWMC score†			Progression of CMB count‡			Progression of lacune count‡			Progression of cortical infarct count‡
	β	95% CI	P value	β	95% CI	P‐value	IRR	95% CI	P value	IRR	95% CI	P value	IRR	95% CI	P value
Tertile 3	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref
Tertile 2*	1.12	−0.29 to 5.16	0.588	0.16	−0.81 to 1.14	0.746	1.64	0.76 to 3.52	0.208	1.33	0.70 to 2.52	0.388	1.01	0.47 to 2.17	0.987
Tertile 1*	4.15	0.06 to 8.24	0.047	0.42	−0.56 to 1.40	0.402	2.21	1.05 to 4.65	0.036	1.14	0.63 to 2.05	0.666	0.57	0.23 to 1.43	0.231
P‐trend	0.044			0.398			0.034			0.673			0.227

ARWMC indicates age‐related white matter changes; CMB, cerebral microbleed; IRR, incident relative risk; PECAM‐1, platelet endothelial cell adhesion molecule‐1; and WMH, white matter hyperintensity.

^*With reference to the third (highest) tertile of PECAM‐1.

^† Linear mixed‐effect model, adjusted for age, sex, education, hypertension, hyperlipidemia, diabetes, smoking status, body mass index, antiplatelet use, cardiovascular disease, timing of assessment, and the baseline presence of significant white matter hyperintensity burden defined by ARWMC score of ≥8. β (regression coefficient) represents adjusted associations with reference to the third (highest) PECAM‐1 tertile. Tests for linear trends (P‐trend) were obtained by modeling PECAM‐1 tertiles as numeric variables.

^‡ Multilevel mixed‐effect Poisson model, adjusted for age, sex, education, hypertension, hyperlipidemia, diabetes, smoking status, body mass index, antiplatelet use, cardiovascular disease, timing of assessment, and all relevant baseline CSVD markers (cerebral microbleeds, cortical infarcts, or lacunes, respectively). IRR represents adjusted incident relative risk with reference to the third (highest) PECAM‐1 tertile. Tests for linear trends (P‐trend) were obtained by modeling PECAM‐1 tertiles as numeric variables.

Association of PECAM‐1 With Markers of Neurodegeneration and Amyloid Burden

Associations of circulating PECAM‐1 levels with blood and MRI markers of neurodegeneration are presented as regression coefficient plots in Figure 3. Plasma p‐tau181 levels were available for 184 of 213 subjects, while plasma NfL levels were available for 183 subjects. Adjusted for age and sex, no significant association was found between PECAM‐1 levels with p‐tau181 and NfL levels (Figure 3, Table S6). Gray matter and hippocampal volume data were available for 205 patients at baseline. PECAM‐1 levels were not significantly associated with gray matter volume or hippocampal volume using linear regression adjusted for age, sex, and total intracranial volume (Figure 3, Table S6).

Figure 3. Cross‐sectional associations of PECAM‐1 levels (stratified by tertiles), with MRI and plasma biomarkers of amyloid burden and neurodegeneration.

Regression coefficient plots present regression coefficients, 95% CIs, and P values derived from linear regressions adjusted for age and sex, with reference to the third (highest) PECAM‐1 tertile. Regression models for gray matter and hippocampal volume were additionally adjusted for intracranial volume. MRI indicates magnetic resonance imaging; NfL, neurofilament light; PECAM‐1, platelet endothelial cell adhesion molecule‐1; and p‐tau181, phosphorylated tau181.

Sensitivity analyses showed that lower PECAM‐1 levels remained significantly associated with greater hazard of incident dementia and cognitive decline (in MoCA scores and executive function Z scores, but not memory Z scores), even after adjustment for baseline circulating biomarkers of amyloid‐related/neurodegenerative processes (p‐tau181, and NfL; Tables S7 and S8).

DISCUSSION

In a cohort of dementia‐free subjects, we found novel associations between lower PECAM‐1 levels with increased CSVD progression and poorer cognitive outcomes. Lower PECAM‐1 levels were associated with increased cross‐sectional burden and longitudinal progression of CSVD (WMHs and CMBs). Longitudinally, PECAM‐1 was associated with increased incident dementia, and greater cognitive decline in MoCA scores and executive function with preservation of memory domains (hallmarks of CSVD‐related cognitive impairment). ²³ There were no significant associations of PECAM‐1 levels with biomarkers of neurodegeneration or amyloid burden. Taken together, our findings suggest the utility of PECAM‐1 as a prognostic biomarker of cognitive decline and incident dementia, relating primarily to its associations with CSVD.

Previous studies investigating the role of PECAM‐1 in cognition are scarce, conflicting, and largely cross‐sectional or post‐mortem in nature. Nielsen et al demonstrated higher levels of plasma PECAM‐1 levels in dementia cases, compared with cognitively normal controls. ⁴⁸ By contrast, Hochstrasser et al reported that monocytic PECAM‐1 levels were similar between subjects with cognitive impairment and healthy controls. ⁴⁹ However, clinically relevant covariates known to be associated with PECAM‐1 expression such as vascular risk factors were unaccounted for. ⁴⁸ , ⁴⁹ In 2 post‐mortem studies, PECAM‐1 expression was found to be elevated in the brains of patients with Alzheimer disease compared with healthy controls. ⁵⁰ , ⁵¹ The increase in PECAM‐1 expression in patients with dementia was postulated to be reflective of ongoing blood–brain barrier dysfunction and neuronal death, processes previously associated with dementia. ⁵⁰ , ⁵¹ , ⁵² The resultant endothelial damage and upregulation of endothelial repair responses may hence contribute to increased circulating PECAM‐1 levels. ¹³ , ¹⁷ , ²⁰ , ⁵³ However, the cross‐sectional nature of these studies raises uncertainties as to whether the observed increase in PECAM‐1 in dementia, was a consequence of, or rather, an underlying driver of the pathophysiological processes underpinning neurocognitive decline. Indeed, it has been increasingly recognized that biomarkers for cognitive decline may be disease‐stage specific, but it remains uncertain if these findings are applicable in dementia‐free subjects who might not have the extensive pathophysiological processes of late‐stage brain aging compared with their counterparts with dementia. ⁵⁰ , ⁵² , ⁵⁴ , ⁵⁵ , ⁵⁶ Our longitudinal study with serial cognitive and neuroimaging evaluations is well‐positioned to address this knowledge gap by establishing the temporality of these associations, demonstrating that lower PECAM‐1 levels precede and may thus be an early‐stage prognostic biomarker for cognitive decline and CSVD in dementia‐free subjects.

In this study, we present significant associations between lower PECAM‐1 levels with greater CSVD progression, incident dementia, and cognitive decline among dementia‐free subjects. This is likely related to a primarily vascular pathogenesis, as PECAM‐1 was only associated with MRI markers of CSVD, but not with biomarkers of neurodegeneration or amyloid burden (p‐tau181 levels, NfL levels, hippocampal volume, or gray matter volume). Indeed, PECAM‐1 remained significantly associated with longitudinal cognitive change even after the adjustment for baseline blood biomarkers of amyloid‐related and neurodegenerative processes. This is corroborated by the observed association of lower PECAM‐1 levels with a greater decline in executive function, a hallmark of CSVD‐related cognitive impairment. ⁸ , ²¹ , ²² , ²³ We also found a corresponding longitudinal decline in MoCA scores, which is known to comprise components evaluating executive function as compared with other cognitive screening tests such as the Mini‐Mental State Examination. ²² These findings are aligned with the cross‐sectional and longitudinal associations observed between lower PECAM‐1 levels with WMHs and CMBs. Indeed, these CSVD markers are known to relate to endothelial dysfunction, and hence remain consistent with PECAM‐1's postulated role as an integral regulator of endothelial junctional integrity. ³ , ¹³ , ¹⁶ , ⁵⁷ , ⁵⁸ While lower PECAM‐1 levels were associated with ARWMC scores cross‐sectionally, no associations were found with regard to ARWMC progression. This may have been contributed by the decreased sensitivity of the ARWMC in detecting temporal changes in WMHs. ⁵⁹ We also did not find significant associations of PECAM‐1 with lacunes or cortical infarcts. This may be contributed in part by their typically lower incidence rate and diverse competing causes apart from CSVD, such as embolic phenomena or parent artery occlusions. ⁶⁰ , ⁶¹ Nonetheless, the protective role found between PECAM‐1 levels for cognitive decline and CSVD progression is consistent with an earlier study demonstrating increased PECAM‐1 expression in response to in vivo interleukin‐6 administration, reinforcing its function as a marker of preserved endothelial homeostasis in response to physiological challenge. ²⁰ Another longitudinal study also reported associations of PECAM‐1 upregulation with improved poststroke neurological outcomes. ¹⁹ Our results may thus be explained by the stage‐specific nature of cognitive biomarkers, with PECAM‐1 being a prognostic biomarker in early‐stage disease, as a marker of endothelial reserve. ¹³ , ²⁰ , ⁵⁶ , ⁶²

The association of lower circulating PECAM‐1 levels with CSVD progression and cognitive decline may be explained by several factors. PECAM‐1 is an integral regulator of endothelial integrity, conferring a protective role in cardiac, vascular, and inflammatory disease phenotypes. ¹³ , ¹⁶ , ⁵⁷ , ⁵⁸ This is postulated to be due to its promotion of cell–cell junctional integrity, vascular mechanosensory function, endothelial homeostasis, and mediation of leukocyte transmigration in inflammation. ¹³ , ¹⁵ , ⁶² , ⁶³ PECAM‐1 deficiency is also known to adversely modulate neuroinflammation and blood–brain barrier instability. ¹² , ¹⁷ , ⁶⁴ While the role of PECAM‐1 has not been previously studied in the context of CSVD; endothelial dysfunction is known to precede the development of CSVD via hypoperfusion, blood–brain barrier disruption and altered cerebral vasoreactivity. ⁴ , ⁵ , ⁸ PECAM‐1 may thus be protective in mitigating endothelial dysfunction, neurovascular dysfunction, and blood–brain barrier disruption, which collectively contribute to the pathogenesis of CSVD and cognitive dysfunction. ¹² , ¹³ , ¹⁷ , ⁶⁵ Taken together, our findings suggest a role for PECAM‐1 as an early prognostic marker of CSVD and cognitive impairment, particularly among patients within the early stages of cognitive decline. Its role as a therapeutic target remains unknown and requires further study.

Study Limitations

The present study has several limitations. First, due to the limited number of dementia cases, our findings may be considered exploratory and require validation in larger cohorts with longer‐term follow‐up. However, when stratified by baseline cognitive status, our study maintains a 27% progression rate from MCI to dementia, and remains consistent with conversion rates reported within other memory clinic cohorts. ⁶⁶ Additionally, we have demonstrated significant prognostic associations of PECAM‐1 levels, with decline in executive function (a hallmark of CSVD‐related cognitive impairment) and progression of CSVD parameters on neuroimaging. ²³ These findings are aligned with the proposed role of PECAM‐1 as an indicator of endothelial reserve, a key underlying factor in CSVD pathogenesis. ³ Second, PECAM‐1 concentrations were exclusively blood based, and levels within the central nervous system were not studied. However, plasma concentrations of PECAM‐1 levels have been reported to correlate with cerebrospinal fluid levels. ⁵³ Thus, it may be reasonable to hypothesize that the circulating levels of PECAM‐1 in our study are reflective of PECAM‐1 expression within the central nervous system. We also acknowledge that data on longitudinal changes in hippocampal atrophy or cortical thickness were not available within this study. However, we have demonstrated that lower baseline PECAM‐1 levels associated with longitudinal cognitive change independent of baseline circulating biomarkers of amyloid burden and neurodegeneration. Furthermore, there were no significant cross‐sectional associations between PECAM‐1 levels with neuroimaging and circulating biomarkers of amyloid and neurodegenerative processes. Taken together, our results suggest that the contributions of PECAM‐1 to cognitive change remain independent of that contributed by p‐tau181 (amyloid burden) and NfL (neurodegeneration). Finally, as circulating PECAM‐1 levels were quantified only at baseline, further studies assessing longitudinal changes in PECAM‐1 levels are required to deepen our understanding of the pathophysiological role of PECAM‐1 in CSVD and neurocognitive decline.

In conclusion, in dementia‐free subjects, lower circulating PECAM‐1 levels are associated with greater CSVD progression, worse neurocognitive decline, and increased hazards of incident dementia. Larger prospective studies are required to confirm the protective role of PECAM‐1 in early‐stage cognitive decline and CSVD, as well as its utility as a therapeutic target in cognitive impairment.

Sources of Funding

The work is supported by funding from the National Medical Research Council (NMRC) Singapore (NMRC/CG/NUHS/2010, NMRC/CG/013/2013, NMRC/CSA‐SI/007/2016, NMRC/CIRG/1485/2018, NMRC/OFLCG/2019, NMRC/CG/M006/2017), awarded to Prof Chen. The work is supported by funding from the National University Health System Clinician Scientist Academy (NCSP2.0/2023/NUHS/SMA), National University of Singapore Clinician Scientist Development Unit (KCG/2023/NUSMED), and NMRC (MH 095:003/008‐340), awarded to Dr Sim.

Disclosures

None.

Supporting information

Tables S1–S8

Figures S1–S2