APOE Genotype Modulates the Relationship of Stroke With Dementia Risk: Associations and Peripheral Clues for Biological Mechanisms

Abstract

Background

Stroke and APOE ε4 are established risk factors of dementia. However, it remains unclear whether stroke interacts with APOE genotypes to influence dementia occurrence. This study aims to investigate the associations of stroke and its interaction by APOE genotypes with incident risk of dementia, with a specific focus on peripheral clues for biological mechanisms.

Methods

This prospective cohort study included 336 903 participants (mean age: 56.3 years, stroke history: 1.3%, APOE ε4: 28.5%) from the UK Biobank, with a median follow‐up of 13 years. Cumulative incidence curves were constructed using Fine–Gray death competing risks analysis. The additive and multiplicative interaction of stroke (including ischemic and hemorrhagic stroke) with APOE genotypes on incident risk of all‐cause dementia (ACD) and Alzheimer disease were examined. Blood proteomics combined with bioinformatics analyses were used to explore the peripheral clues for biological mechanisms.

Results

Either ischemic or hemorrhagic stroke was significantly associated with elevated risk of ACD and Alzheimer disease (P<0.001). A significant multiplicative interaction was observed between stroke and APOE ε4 (P<0.001). The association of stroke with increased risk of dementia was stronger in APOE ε4 non‐carriers than carriers, for both ACD (hazard ratio [HR], 1.93 for carriers and 3.36 for non‐carriers, P<0.001) and AD (HR, 1.14 for carriers and 2.67 for non‐carriers, P<0.001). Inflammation‐related pathways could be mechanisms underpinning the association of stroke with ACD risk. We identified 191 functionally interconnected (P<1.0×10⁻¹⁶) proteins associated with both stroke and ACD only in APOE ε4 non‐carriers. CD4‐related and TGF‐beta (transforming growth factor beta) signaling pathway could mediate the strengthened relationship in APOE ε4 non‐carriers.

Conclusions

Stroke interacts with APOE ε4 to influence dementia, with the association being more pronounced in APOE ε4 non‐carriers. Future studies are needed to verify the underpinning mechanisms to guide precise prevention.

Nonstandard Abbreviations and Acronyms

ACD

all‐cause dementia

FDR

false discovery rate

GDF15

growth differentiation factor 15

KEGG

Kyoto Encyclopedia of Genes and Genomes

NEFL

neurofilament light

UKB

UK Biobank

Clinical Perspective.

What Is New?

• This study presents novel evidence demonstrating that stroke interacts with the APOE ε4 genotype to influence the risk of dementia, with a stronger association observed in noncarriers compared with carriers of the APOE ε4 allele.

• Peripheral proteomics suggest that inflammation‐related pathways may mediate the association between stroke history and dementia.

What Are the Clinical Implications?

• The interaction of stroke history with APOE ε4 status should be considered for risk stratification and prediction, which could also guide the precise prevention of dementia in the future.

Neurological disorders are the leading cause of disability‐adjusted life years worldwide, accounting for 10% of the global burden of disease. Thereinto, approximately 42% of these disability‐adjusted life years can be attributed to stroke and ~10% to dementia. ¹ As a cerebrovascular stress event, stroke serves as a significant and potentially modifiable contributor to dementia risk, although its mechanisms remain unclear. ² In addition, the risk of dementia is also influenced by genetic factors. ³ APOE ε4 allele is recognized as the strongest genetic risk factor for Alzheimer disease (AD) as well as all‐cause dementia (ACD). Individuals carrying the ε4 allele are estimated to have a 2‐ to 6‐fold increased risk of dementia, whereas those with the APOE ε2 allele have a reduced risk. ⁴ Recently, the importance of genetic‐environmental interactions has been increasingly recognized, especially in prediction, prevention, and development of precise management therapy for dementia. ⁵ However, whether and how stroke and APOE genotype interact to influence dementia risk is still scarcely investigated. A recent study indicated that, among patients who have experienced a transient ischemic attack or stroke, APOE ε4 homozygosity but not ε4/ε3 was associated with an elevated risk of both pre‐ and postevent dementia. ⁶ A longitudinal study found an additive effect of stroke and the APOE ε4 allele on dementia risk. ⁷ In contrast, another study reported no additive nor multiplicative effect of stroke with APOE ε4 on dementia risk. ⁸ This inconsistency may stem from methodological limitations, as previous studies often overlooked specific stroke types (ischemic or hemorrhagic) and APOE genotypes (ε2/ε3/ε4). ⁹ , ¹⁰ Evidence suggests that APOE ε2 carriers have an elevated risk of ischemic stroke compared with APOE ε3 individuals and are strongly linked to hemorrhagic stroke recurrence. ¹¹ , ¹² Additionally, an elevated risk of mortality associated with the APOE ε4 genotype was observed among patients with hemorrhagic stroke, whereas no such association was found in individuals with ischemic stroke. ¹³ However, previous studies barely considered the bias due to competing risks of death. Therefore, when studying the interaction of stroke with APOE genotypes, it is essential to consider the diversity of stroke types and APOE genotypes.

Moreover, despite growing evidence linking stroke to dementia, the precise mechanisms are yet to be determined. Although previous studies have suggested possible mechanisms such as inflammation that link a history of stroke to dementia, ¹⁴ these hypotheses lack support from large‐scale population‐based study. Proteomics can serve as an objective indicator linking stroke to dementia risk, with its expression levels providing insights into the underlying biological pathways and changes. Additionally, large‐scale proteogenomic studies have consistently identified the APOE gene as a hotspot for protein regulation, influencing the levels of hundreds of proteins in circulation. ¹⁵ , ¹⁶ However, it remains unclear whether the interaction between stroke and the APOE ε4 allele may alter the plasma protein expression profile and how these changes contribute to an increased risk of dementia risk in individuals with a history of stroke.

To address these uncertainties, this study used a comprehensive longitudinal data set from the UK Biobank (UKB). The primary objectives were (1) to elucidate the association between stroke and its subtypes with ACD and AD risk and to analyze the underlying mechanisms by proteomic analysis; (2) to explore the interaction between stroke and APOE genotypes on ACD and AD risk and to explore the possible mechanisms.

METHODS

All data are available upon reasonable request or can be obtained from the UKB (https://biobank.ctsu.ox.ac.uk/).

Study Population

Data are from a large population‐based cohort study based on UKB (https://www.ukbiobank.ac.uk/). Participants were recruited between 2006 and 2010 across 22 centers in the United Kingdom. Follow‐up was calculated from the date of recruitment to the earliest of the following events: dementia diagnosis, death, loss to follow‐up, or the most recent available data. Biological samples, including blood, urine, and feces, were collected and preserved for future testing following written informed consent. Ethical approval was granted by the North West‐Haydock Research Ethics Committee, and all participants provided written informed consent. In the present study, participants with baseline diagnoses of dementia or major psychiatric disorders were excluded. Furthermore, participants who lacked APOE genotyping, had missing covariate data, had a diagnosis of dementia within 1 year of follow‐up, were followed for <1 year, or had withdrawn during the follow‐up were excluded from the analyses.

History of Stroke

Stroke history was determined based on self‐reports. Participants were asked whether they had a prior history of stroke and if any the stroke types (ischemic or hemorrhagic type). Diagnoses recorded in hospital records were classified using International Classification of Diseases (ICD) codes from the Ninth Revision (ICD‐9) and Tenth Revision (ICD‐10). We defined stroke subtypes based on ICD codes, including ischemic stroke.

APOE Genotyping

The genotyping of APOE alleles in the UKB has been described in previous research. ¹⁷ APOE genotypes were determined based on rs429358 and rs7412. ¹⁸ , ¹⁹ APOE ε4 positive status was defined as carrying at least 1 ε4 allele (ε2/ε4, ε3/ε4, or ε4/ε4), and individuals with ε2/ε2, ε2/ε3, or ε3/ε3 genotypes were categorized as APOE ε4 negative. To assess ε2 and ε4 allele‐specific effects, individuals with the ε2/ε4 haplotype (2.52%) were excluded from subsequent analyses. Participants were finally categorized into 3 groups: APOE ε2 (ε2/ε2 and ε2/ε3), APOE ε3 (ε3/ε3), and APOE ε4 (ε3/ε4 or ε4/ε4).

Dementia Diagnosis

Dementia was ascertained using diagnoses obtained from primary care, hospital inpatient, and death registry records. These data were algorithmically integrated and classified using the ICD‐9 and ICD‐10. The diagnostic algorithm has been validated in previous studies. ²⁰ , ²¹ ACD covers all types of dementia including AD, vascular dementia, Lewy body dementia, frontotemporal dementia, Huntington’s disease dementia, Parkinson’s disease dementia, cortical basal ganglia degeneration dementia, and other neurodegenerative or any other dementia due to specific diseases. AD was ascertained based on ICD codes from linked hospital episode statistics as well as death certificates, primary care, self‐reports, and nurse interviews.

Covariates Measurements

The covariates included age, sex, Townsend deprivation index, educational level, alcohol dependency, body mass index (BMI), smoking status, anxiety, cancer, depression, diabetes, hyperlipidemia, hypertension, and APOE ε4 carrier status. Demographic and behavioral data were collected via touchscreen questionnaires. Socioeconomic status was assessed using the Townsend deprivation index. BMI was calculated from objectively measured height and weight. Comorbidity conditions were identified from self‐reports and medical records.

Blood Proteomics

Proteomics data were obtained from the Pharmaceutical Proteomics Project, a collaborative initiative involving the UKB and pharmaceutical industry partners. The project analyzed plasma samples from >54 000 participants at baseline, encompassing 2923 proteins. Olink’s Explore platform was used for data acquisition via next‐generation sequencing technology. Quality control measures included principal component analysis to identify and remove outliers, as well as the exclusion of flagged or unprocessed samples.

Statistical Analysis

The characteristics of participants enrolled in the study were summarized based on whether they developed dementia during the follow‐up. The normally distributed continuous variables were expressed as mean±SD, nonnormally distributed continuous variables were expressed as median and quartiles, and categorical variables were expressed as number of observations and percentages. Group comparisons were made using t test for normally distributed continuous variables, Mann–Whitney U test for nonnormally distributed continuous variables, and chi‐square test for categorical variables.

First, Cox proportional risk models were used to quantify the association of stroke and its subtypes with incident risk of ACD or AD. The results were expressed as hazard ratios (HRs) with 95% CIs. Model 1 was adjusted for age, sex, Townsend deprivation index, education level, alcohol dependency, BMI, smoking status, and APOE ε4 carrier status. Model 2 was further adjusted for anxiety, cancer, depression, diabetes, hyperlipidemia, and hypertension. The proportional hazards assumption was verified using the Schoenfeld residual method, and no violations were detected. To explore the multiplicative and additive interaction effects, the interaction terms between stroke history and APOE ε4 status were included in all analyses. Additive interactions were measured using the relative excess risk due to interaction, proportion attributable to interaction, and synergy index. Subgroup analyses stratified by APOE ε4 carrier status (ε4 carriers versus ε4 noncarriers; ε2, ε3, and ε4 group) were conducted. To address the potential competing risks of death, we also performed sensitivity analyses using Fine and Gray’s competing risks model.

Second, to investigate potential proteins and mechanisms mediating the relationship between stroke and incident dementia, plasma proteomic analyses were conducted to identify significant proteins for bioinformatic as well as mediation analyses. Linear regression analysis was employed to identify proteins significantly associated with stroke, using stroke history as the independent variable. Cox proportional‐hazards modeling was used to determine proteins significantly associated with dementia risk, with plasma protein levels serving as the independent variable. The intersection of these 2 sets of results, characterized by coefficients in the same direction, was considered to represent proteins significantly associated with both stroke and dementia events. These models were adjusted for age, sex, Townsend deprivation index, educational level, alcohol dependency, BMI, smoking status, and APOE ε4 carrier status. To adjust for multiple comparisons, we controlled false discovery rate (FDR) using the Benjamini–Hochberg method by fdrtool package (version 1.2.17).

Third, the selected significant proteins were imported into the STRING database (https://string‐db.org/) for functional analyses. Gene Ontology terms and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses were conducted using the STRING platform. Pathways with an FDR value <0.05 and strength >0.01 were selected, focusing on the top 10 pathways with the smallest FDR values. In addition, protein–protein interaction (PPI) networks were constructed and clustering analyses were performed using k‐means clustering. The cluster with the highest number of enriched proteins was visualized using Cytoscape software (v3.10.2). In addition, we used mediation analyses to identify proteins that may mediate the effect of stroke on ACD or AD risk. We performed mediation analyses using the mediation package (v 4.5.0) and adjusted the mediation models using age, sex, Townsend deprivation index, educational level, alcohol dependency, BMI, smoking status, and APOE ε4 carrier status. FDR were controlled using the Benjamini–Hochberg method and identified proteins with FDR <0.05 as mediating proteins. Finally, we explored the differentially expressed proteins associated with both stroke and incident dementia separately among APOE ε4 carriers and noncarriers. The same procedures were performed to provide clues that underpinned the interaction of stroke with APOE ε4 associated with incident dementia.

RESULTS

Population Characteristics

A total of 336 903 participants with a mean age of 56.3 years (SD=8.02) were included, of whom 45.1% were female and 28.5% APOE ε4 carriers. The flow chart of the study design was shown in Figure 1. The baseline characteristics of the participants were given in Table1. During a mean follow‐up of 13 years, 7557 participants were diagnosed with ACD, including 2804 cases of AD and 1410 cases of vascular dementia. Participants who have incident ACD were more likely to be older and female, and to have poorer economic status, lower educational attainment, higher BMI, alcohol dependence, and current smoking habits. Participants with incident AD tended to be older and to have poorer economic status, lower education levels, and current smoking habits. Participants with incident vascular dementia tend to be older and female, lower educational level, reduced APOE ε4 allele prevalence, and an increased risk of comorbid diabetes, hypertension, hyperlipidemia, and depression (details can be found in Table S1). These baseline differences remained statistically robust following adjustment for age and sex. Furthermore, significant differences in alcohol dependence and anxiety levels were observed between groups with and without AD, after controlling for age and sex.

Figure 1. Selection of study participants in the UK Biobank.

ACD indicates all‐cause dementia; and AD, Alzheimer disease.

Table 1.

Baseline Characteristics of Participants

	Total (N=336 903)	ACD		P value	P value *	Total (N=319 516)	AD		P value	P value *
		No (N=329 346)	Yes (N=7557)				No (N=316 712)	Yes (N=2804)
Age, y	56.3 (8.02)	56.1 (7.99)	64.3 (4.68)	<0.001		56.3 (8.00)	56.2 (7.99)	64.6 (4.18)	<0.001
Sex, %				<0.001					0.087
Female	151 869 (45.1%)	147 975 (44.9%)	3894 (51.5%)			143 768 (45.0%)	142 461 (45.0%)	1307 (46.6%)
Male	185 034 (54.9%)	181 371 (55.1%)	3663 (48.5%)			175 748 (55.0%)	174 251 (55.0%)	1497 (53.4%)
Thompson deprivation index	−1.49 (2.98)	−1.50 (2.97)	−1.10 (3.22)	<0.001	<0.001	−1.49 (2.98)	−1.49 (2.98)	−1.27 (3.16)	<0.001	<0.001
APOE ɛ4, %				<0.001	<0.001				<0.001	<0.001
Noncarriers	240 747 (71.5%)	237 263 (72.0%)	3484 (46.1%)			229 119 (71.7%)	228 085 (72.0%)	1034 (36.9%)
Carriers	96 156 (28.5%)	92 083 (28.0%)	4073 (53.9%)			90 397 (28.3%)	88 627 (28.0%)	1770 (63.1%)
APOE status, %				<0.001	<0.001				<0.001	<0.001
ɛ2ɛ2	1914 (0.57%)	1888 (0.57%)	26 (0.34%)			1800 (0.56%)	1794 (0.57%)	6 (0.21%)
ɛ2ɛ3	41 119 (12.2%)	40 615 (12.3%)	504 (6.67%)			39 181 (12.3%)	39 041 (12.3%)	140 (4.99%)
ɛ2ɛ4	8477 (2.52%)	8257 (2.51%)	220 (2.91%)			8023 (2.51%)	7951 (2.51%)	72 (2.57%)
ɛ3ɛ3	197 714 (58.7%)	194 760 (59.1%)	2954 (39.1%)			188 138 (58.9%)	187 250 (59.1%)	888 (31.7%)
ɛ3ɛ4	79 705 (23.7%)	76 666 (23.3%)	3039 (40.2%)			75 077 (23.5%)	73 772 (23.3%)	1305 (46.5%)
ɛ4ɛ4	7974 (2.37%)	7160 (2.17%)	814 (10.8%)			7297 (2.28%)	6904 (2.18%)	393 (14.0%)
Educational level, y	15.5 (5.24)	15.5 (5.24)	13.5 (4.93)	<0.001	<0.001	15.5 (5.23)	15.5 (5.23)	13.4 (4.85)	<0.001	<0.001
Alcohol dependency, %	370 (0.11%)	350 (0.11%)	20 (0.26%)	<0.001	<0.001	341 (0.11%)	336 (0.11%)	5 (0.18%)	0.232	0.0163
Body mass index, kg/m2	27.2 (4.68)	27.2 (4.68)	27.7 (4.78)	<0.001	<0.001	27.2 (4.68)	27.2 (4.68)	27.3 (4.60)	0.284	<0.001
Smoking status, %	148 929 (44.2%)	144 926 (44.0%)	4003 (53.0%)	<0.001	<0.001	141 128 (44.2%)	139 706 (44.1%)	1422 (50.7%)	<0.001	<0.001
Anxiety, %	4511 (1.34%)	4415 (1.34%)	96 (1.27%)	0.635	0.645	4289 (1.34%)	4239 (1.34%)	50 (1.78%)	0.051	0.0352
Cancer, %	1854 (0.55%)	1830 (0.56%)	24 (0.32%)	0.007	<0.001	1679 (0.53%)	1671 (0.53%)	8 (0.29%)	0.102	<0.001
Depression, %	18 094 (5.37%)	17 498 (5.31%)	596 (7.89%)	<0.001	<0.001	17 047 (5.34%)	16 842 (5.32%)	205 (7.31%)	<0.001	<0.001
Diabetes, %	14 788 (4.39%)	13 844 (4.20%)	944 (12.5%)	<0.001	<0.001	13 830 (4.33%)	13 540 (4.28%)	290 (10.3%)	<0.001	<0.001
Hyperlipidemia, %	38 986 (11.6%)	37 247 (11.3%)	1739 (23.0%)	<0.001	<0.001	36 944 (11.6%)	36 299 (11.5%)	645 (23.0%)	<0.001	<0.001
Hypertension, %	84 822 (25.2%)	81 611 (24.8%)	3211 (42.5%)	<0.001	<0.001	80 298 (25.1%)	79 186 (25.0%)	1112 (39.7%)	<0.001	<0.001
Stroke group, %				<0.001	<0.001				<0.001	<0.001
Nonstroke	332 652 (98.7%)	325 444 (98.8%)	7208 (95.4%)			315 620 (98.8%)	312 895 (98.8%)	2725 (97.2%)
Hemorrhagic stroke	515 (0.15%)	491 (0.15%)	24 (0.32%)			482 (0.15%)	478 (0.15%)	4 (0.14%)
Ischemic stroke	3736 (1.11%)	3411 (1.04%)	325 (4.30%)			3414 (1.07%)	3339 (1.05%)	75 (2.67%)

ACD indicates all‐cause dementia; and AD, Alzheimer disease.

^* P value adjusts for age and sex.

Stroke Interacted With APOE ε4 to Influence Dementia Risk

A history of stroke and both of its subtypes were significantly associated with incident risk of ACD (Figure 2A) and AD (Figure 2B). In model 1, participants with stroke history were more likely to develop ACD compared with those without stroke history (HR, 2.52 [95% CI, 2.26– 2.81], P<0.001). The associations were observed for both ischemic (HR, 2.61 [95% CI, 2.33– 2.92], P<0.001) and hemorrhagic stroke (HR, 1.75 [95% CI, 1.17– 2.61], P<0.01). Similarly, participants with stroke history had a 67% increased risk of AD compared with their counterparts (HR, 1.67[ 95% CI, 1.33– 2.08], P<0.001). The association remained significant for ischemic stroke (HR, 1.76 [95% CI, 1.40– 2.21], P<0.001) but not hemorrhagic stroke (HR, 0.85 [95% CI, 0.32– 2.26]). Additionally, as presented in Table S2, participants with a history of stroke exhibited a higher risk of vascular dementia compared with those without such a history (HR, 5.13 [95% CI, 4.26– 6.17], P<0.001). The aforementioned significant findings remained robust after including more confounders and in Fine and Gray’s death‐competing models (P<2.0×10⁻¹⁶, Figures S1 and S2). We additionally excluded participants diagnosed with dementia within 3 years of follow‐up to prevent prestroke dementia from confounding our results (Tables S3 and S4). In addition, to further verify the reliability of the conclusion, we also excluded the participants who experienced stroke during the follow‐up period. The results are recorded in Tables S5 and S6.

Figure 2. Association of stroke and its subtypes with ACD and AD.

Hazard ratios and 95% CIs of incident ACD (A) and AD (B) according to different stroke subtypes assessed by Cox proportional hazards models. Model 1: adjusted for the following potential confounders: age, sex, Thompson deprivation index, educational level, alcohol dependency, body mass index, and smoking status. Model 2: based on Model 1 and additionally adjusted for anxiety, cancer, depression, diabetes, hyperlipidemia, and hypertension. ACD indicates all‐cause dementia; AD, Alzheimer disease; and HR hazard ratio. **P<0.01; ***P<0.001.

Statistical significance for the multiplicative interaction effect of stroke by APOE ε4 on the risk of both ACD (Table S7) and AD (Table S8) was observed (P<0.001). However, no multiplicative interaction between stroke and APOE ε4 on vascular dementia was observed (Table S9). When different stroke subtypes were analyzed, the multiplicative interaction effect of either ischemic or hemorrhagic stroke with APOE ε4 on ACD risk remained significant (P<0.05). The risk estimate for ACD in APOE ε4 carriers (HR, 1.93 [95% CI, 1.64– 2.26]) is much lower than that in APOE ε4 noncarriers (HR, 3.36 [95% CI, 2.90– 3.88], P<0.001). As for AD, the association was significant only in APOE ε4 noncarriers (HR, 2.67 [95% CI, 1.97– 3.63], P<0.001). No additive interactions were found between stroke and APOE ε4 on ACD (details in Table S10) or AD (details in Table S11).

To further evaluate the independent effects of the APOE ε2, ε3, and ε4 alleles, we examined the association between stroke and the risk of incident ACD (Figure S3A) and AD (Figure S3B) across populations stratified by APOE genotypes. Stroke was significantly associated with an increased risk of ACD in APOE ε2 carriers (HR, 3.21 [95% CI, 2.19– 4.71]; P<0.001) and APOE ε3 carriers (HR, 3.38 [95% CI, 2.88– 3.96]; P<0.001). The risk sizes of these estimates were higher than that in ε4 carriers (HR, 1.96 [95% CI, 1.66– 2.31]).

Proteomic Profile and Enriched Pathways Mediating the Association of Stroke With Incident Dementia

After correction for multiple testing, a total of 454 proteins were associated with both stroke and incident risk of ACD, with consistent direction of coefficient (Figure 3). Among 454 proteins, 80 proteins were downregulated and 374 proteins were upregulated in participants who had ever experienced stroke (details in Table S12). Gene Ontology/KEGG enrichment analyses revealed that these proteins were primarily involved in the cell surface receptor signaling pathway, regulation of multicellular organismal processes, response to external stimuli, and cytokine–cytokine receptor interaction pathway. The produced PPI network consisted of 450 nodes and 313 edges, with significant enrichment (P<1.0×10⁻¹⁶). K‐means clustering identified distinct clusters based on protein centroids, with cluster 1 containing the largest number of proteins. This cluster was predominantly associated with cytokine–cytokine receptor interactions (Figure S4) and IL‐6 (interleukin 6) was the core protein of the cluster. In the mediation analysis, we identified 408 proteins mediating the association between stroke and ACD risk, with GDF15 (growth differentiation factor 15) having the largest proportion of mediators (22.5%).

Figure 3. Bioinformatics analysis of proteins associated with both stroke and ACD.

A, Venn diagram showing the number of proteins associated with stroke (Group A), ACD (Group B), and both stroke and ACD (Group C). B, Volcano plot of protein (Group C) levels in participants with stroke history compared with those without stroke history. Blue represents downregulated proteins, and red represents upregulated proteins. C, GO analysis showing enrichment of differential proteins in biological processes, cellular components, and molecular function processes. D, Sankey plot of KEGG analysis showing enriched pathways. E, Differential proteins are mapped on a PPI network. Different colors indicate distinct clusters. F, Differential proteins in cluster 1, sorted by degree. The size and color of protein circles represent their degree values. G, Mediation of proteins on the association between stroke and ACD in the total population. ACD indicates all‐cause dementia; FDR, false discovery rate; GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes; and PPI, protein–protein interaction.

As for AD, 84 proteins were identified, among which 13 proteins were downregulated and 71 proteins were upregulated in participants with stroke history compared with their counterparts (Table S13). As shown in Figure S5, these proteins were functionally interconnected (PPI P<1.0×10–16) and involved in response to stimuli, cell communication, and signal transduction, KEGG analysis revealed that these proteins were significantly enriched in the cytokine–cytokine receptor interaction pathway. K‐means clustering identified a major cluster containing 66 proteins. EGFR (epidermal growth factor receptor) was found as the most connected protein in this cluster and EGFR tyrosine kinase inhibitor resistance as one of the key pathways (Figure S6). No proteins mediating the association between stroke and AD were identified in the mediation analyses.

Proteins and Pathways Potentially Underpinning the Interaction Role of Stroke by APOE ε4 in Influencing Dementia Risk

The aforementioned analyses showed a more pronounced correlation between stroke and dementia risk in APOE ε4 noncarriers, compared with APOE ε4 carriers. We hypothesized that the plasma protein profile and the involved pathways linking stroke to dementia incidence differed in the presence or absence of APOE ε4. Accordingly, we targeted those significant proteins specifically in APOE ε4 noncarriers to explain the aforementioned interaction relationship.

As shown in Figure 4, 191 proteins (32 downregulated and 159 upregulated) were associated with both stroke and ACD only in APOE ε4 noncarriers (group D, details in Table S14). These proteins are functionally interconnected (PPI P<1.0×10⁻¹⁶) and are primarily involved in stimulus response, cellular communication, and signal transduction. KEGG analyses showed that TGF‐β (transforming growth factor beta) signaling pathway was a specific pathway in APOE ε4 noncarriers, in contrast to the pathways enriched in the total population. In addition, we also investigated the differences in pathways enriched in various populations (APOE ɛ4 carriers versus noncarriers), which are detailed in Table S15. K‐means clustering analysis showed that most of the proteins are associated with the cell surface (clusters labeled in red). Visualization using Cytoscape revealed that CD4 was the most connected protein in this cluster. Detailed Gene Ontology/KEGG results for cluster 1 are presented in Figure S7. In addition, we identified 162 proteins mediating the association between stroke and ACD in APOE ε4 noncarriers. Among these proteins, NEFL (neurofilament light) played a key mediating role (mediation proportion: 13.1%).

Figure 4. Bioinformatics analysis to reveal proteins which were associated with both stroke and ACD only in the APOE ε4 noncarriers.

A, Venn diagram showing the number of proteins associated with both stroke and ACD in APOE ε4 carriers (Group A) and noncarriers (Group B). Group C represents proteins shared between carriers and noncarriers, and Group D represents proteins unique to APOE ε4 noncarriers. B, Volcano plot of protein (Group D) levels in APOE ε4 noncarriers with stroke history compared with those without stroke history. Blue represents downregulated proteins, and red represents upregulated proteins. C, GO analysis showing enrichment of different proteins in biological processes, cellular components, and molecular function processes. D, Sankey diagram of KEGG analysis illustrating enriched pathways. E, Mapping differential proteins on a PPI network. Distinct colors denote different clusters. F, Differential proteins in cluster 1, sorted by degree. Protein circle size and color indicate degree values. G, Mediation of proteins on the association between stroke and ACD in APOE ε4 noncarriers. ACD indicates all‐cause dementia; FDR, false discovery rate; GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes; and PPI, protein–protein interaction.

As for AD, 53 significant proteins (Group D, 3 downregulated and 50 upregulated, Table S16) were identified only in APOE ε4 noncarriers (Figure S8). These proteins were functionally interconnected (P<1.0×10⁻¹⁶) and primarily involved in response to stimuli, multicellular organismal processes, and complement and coagulation cascades. EGFR was identified as the most connected protein for the major cluster (Figure S9).

DISCUSSION

Using data from 336 903 participants in the population‐based UKB study, our analyses revealed an elevated risk of ACD and AD in individuals with stroke history. Dementia risk in individuals with stroke history was found to be associated with multiple neuroinflammation‐related signaling pathways such as cell surface receptor, regulation of multicellular organismal processes, response to external stimuli and cytokine–cytokine receptor interaction pathways. Notably, a significant interaction was observed between stroke and APOE ε4 in influencing dementia risk. The association between stroke and dementia is more pronounced in APOE ε4 noncarriers. Neuroinflammation due to an imbalance in the TGF‐β signaling pathway may be associated with a higher risk of dementia in APOE ε4 noncarriers with stroke history.

Consistent with the findings of several previous studies, our analyses confirmed that individuals with a history of stroke have a significantly higher risk of incident dementia compared with those without stroke history. ²² , ²³ Furthermore, our proteomic analyses indicated that stroke may contribute to dementia through neuroinflammation‐related signaling pathways. This is in line with the hypothesis that the onset of dementia following a stroke is influenced by both the acute inflammatory response to lipid‐rich debris generated by the lesion and the long‐term autoimmune response. ²⁴ As a pivotal factor in the PPI network composed of proteins linked to both stroke and dementia, IL‐6 indicates the role of proinflammatory status in the progression of dementia in stroke survivors. ²⁵ Notably, serum IL‐6 levels are significantly higher in patients with acute ischemic stroke accompanied by dementia compared with those with acute ischemic stroke alone. ²⁶ These findings suggest that targeting IL‐6 may be a promising strategy to mitigate cognitive impairment in populations with stroke history. ²⁷ In the mediation analysis, we found that GDF15 had the largest proportion of mediation in the association between stroke and ACD. GDF15 is a cellular stress induced factor that is typically activated in response to cellular injury, inflammation, or stress. ²⁸ A longitudinal study based on the UKB showed that GDF15 is consistently associated with ACD and ranked high in terms of protein importance. ²⁹ An experiment in a mouse model of septic encephalopathy showed that GDF15 exacerbated cognitive deficits by promoting an inflammatory response in microglia. ³⁰ These findings further support that GDF15 may mediate the association between stroke and ACD in an inflammation‐dependent manner.

Several studies have evaluated the potential interaction between APOE genotype and stroke history on the risk of dementia, but the results of the available population‐based studies are generally inconsistent. A population‐based longitudinal study found an additive effect of stroke and APOE ε4 on dementia risk but reported no multiplicative interaction between these factors. ⁷ In line with this, the Canadian Health and Aging Study demonstrated that the coexistence of stroke and APOE ε4 increased dementia risk compared with their absence, yet the effect of stroke on dementia appeared to be independent of APOE ε4. ⁸ In contrast, a study of familial AD in Latino subjects found that APOE ε4 and stroke independently raised the risk of familial AD and might interact to further amplify this risk. ³¹ Our study aims to provide evidence that stroke interacts with APOE ε4 to increase the risk of ACD and AD. Notably, this interaction remained significant in increasing the risk of ACD even when ischemic and hemorrhagic strokes were analyzed separately. However, the effect of the interaction between hemorrhagic stroke and APOE ε4 on AD risk was no longer significant, which may be due to reduced statistical power resulting from fewer cases of hemorrhagic stroke.

Our study demonstrated that APOE ε4 noncarriers with stroke history are at higher risk of incident dementia compared with APOE ε4 carriers. A possible mechanism for this phenomenon is that APOE ε2 triggers neuroinflammation by mediating vascular fibrinoid necrosis. ³² , ³³ Furthermore, studies have shown that APOE ε2 carriers, unlike APOE ε4 carriers, exhibit a higher incidence of moderate to severe small‐vessel atherosclerosis compared with APOE ε3/ε3 carriers. ³⁴ However, APOE ε2 carriers constituted only a minor fraction of the participants included in our analysis. Consequently, there are certain limitations in leveraging the association between APOE ε2 status and small vessel disease to account for the elevated risk of dementia among APOE ε4 noncarriers. We are puzzled by the fact that APOE ε3/ ε3 carriers with stroke history also have a higher risk of dementia than APOE ε4 carriers, and this unexpected finding warrants further investigation to elucidate the underlying mechanisms. Our bioinformatics analyses of 191 APOE ε4‐independent proteins provided insights into the differences in the association between stroke and incident dementia in APOE ε4 carriers and noncarriers. The TGF‐β signaling pathway was uniquely enriched in APOE ε4 independent proteins, suggesting it may play a central role in the stronger association observed between stroke and dementia in APOE ε4 noncarriers compared with carriers. A study integrating 2 independent cohorts with AD demonstrates that the TGF‐β signaling pathway is enriched during AD progression in participants independent of stroke. ³⁵ Moreover, the TGF‐β signaling pathway has been extensively investigated as a potential therapeutic target for AD in preclinical studies. ³⁶ , ³⁷ TGF‐β is a multifunctional cytokine that protects damaged brain tissue after stroke by inhibiting inflammatory responses, modulating microglial activation, and attenuating oxidative stress associated with stroke. ³⁸ , ³⁹ , ⁴⁰ Additionally, TGF‐β facilitates the removal of Aβ deposits by promoting microglial phagocytosis. ⁴¹ , ⁴² In our PPI analysis, CD4 was identified as the key protein linking stroke and AD. A previous Mendelian randomization study revealed that CD4 count was strongly associated with AD risk. ⁴³ Among APOE ε4 noncarriers, NEFL played the largest proportion of mediating effect in the association between stroke and ACD. As a neuronal cytoskeletal protein, NEFL is released into the cerebrospinal fluid and blood in response to neuronal damage or degenerative diseases. ⁴⁴ In recent years, several large cohort studies have confirmed a strong association between NEFL and dementia. ⁴⁵ , ⁴⁶ In conclusion, these findings underscore that CD4 and NEFL may serve as critical factors or potential markers for dementia in individuals with a history of stroke, particularly in APOE ε4 noncarriers.

The present study has several strengths. First, the large sample size of UKB participants, coupled with its prospective design and long‐term follow‐up, provided robust statistical power to detect the association between interaction of stroke by APOE ε4 and incident dementia. Second, we performed bioinformatics analyses based on proteomics data, allowing for a deeper exploration of the molecular clues underlying our observations. Third, to minimize statistical errors in identifying differentially expressed proteins, we applied the Benjamini–Hochberg method for P value adjustment, resulting in corrected FDR. However, some limitations must be acknowledged. First, the use of self‐reportable definitions might have resulted in underreporting or misclassification of stroke cases. Second, although proteomics data were collected at baseline in our study, repeated collection of such data over time is warranted to further investigate the role of relevant proteins in the association between stroke and dementia. Third, the biological clues revealed by proteomic analyses cannot reflect causal relationships and should be further validated in longitudinal studies and animal experiments. This limits the generalizability of our findings to more diverse populations.

Conclusions

In conclusion, this large prospective cohort study highlights the association of stroke with an elevated risk of dementia, for which neuroinflammation‐related signaling pathways may be a potential mechanisms. There was a significant interaction between stroke and APOE ε4 on dementia risk, and neuroinflammation due to imbalance in the TGF‐β signaling pathway may be associated with a higher risk of dementia in APOE ε4 noncarriers with stroke history. Our study provides new targets for the prevention and treatment of dementia in patients with a history of stroke.

Sources of Funding

This study was supported by grants from the Taishan Scholar Project (No. tsqn202211375).

Disclosures

The authors declare that they have no conflict of interest.

Supporting information

Tables S1–S16

Figures S1–S9