Risk Effects of Physical Activity on the Incidence of Stroke with Polygenic Risk Score
Article information
Abstract
PURPOSE
Stroke is influenced by multiple risk factors, including genetic factors summarized by polygenic risk scores (PRS) derived from single nucleotide polymorphisms (SNPs). Physical activity is an important modifiable factor that reduces stroke risk. This study examined the association between physical activity and stroke incidence across different PRS levels.
METHODS
Through genome wide-association study (GWAS) analysis, SNPs with high frequency in stroke patients were classified. The presence and amount of physical activity was assessed based on the weekly physical activity exercise reported in the epidemiological survey data. All participants were classified into three groups based on PRS levels. Logistic regression analysis was performed to evaluate the association between physical activity and stroke incidence according to PRS levels.
RESULTS
Through GWAS (p<1×10−5), 28 SNPs were identified in the stroke patients for calculating PRS. The PRS was calculated using 11 SNPs derived from GWAS through LD-pruning. Each one-standard deviation increase in PRS was associated with an approximately 2.5-fold increased risk of stroke (OR=2.54, p<.0001). Each one-minute increase in physical activity was associated with an approximately 0.3% decrease in stroke risk (OR=0.997, p=.0214). No significant interaction effects were observed between physical activity and PRS levels in relation to stroke incidence (middle PRS×physical activity: OR=1.00, p=.490; high PRS×physical activity: OR=1.00, p=.969).
CONCLUSIONS
This study examined differences in stroke incidence associated with physical activity across PRS levels. Although no significant interaction was observed, the effect of physical activity on stroke risk varied by PRS. Our findings suggest that the stroke prevention effect of physical activity may differ across PRS levels.
INTRODUCTION
Stroke is ranked as the second leading cause of death globally, with approximately 12.2 million cases occurring annually [1]. The increasing number of stroke patients has led to economic issues and societal burdens, including rising medical costs and problems caused by the aftereffects of stroke [2], [3]. Stroke occurs when oxygen and nutrients are not properly delivered to the brain, which leads to brain injury and causes various neurological deficits [4]. In the Republic of Korea, stroke is a major public health challenge as well. Every year, approximately 105,000 people experience a new or recurrent stroke, and more than 26,000 die from stroke, accounting for about 10% of all deaths nationwide. The estimated prevalence among people aged 30 years or older is approximately 795,000, and the nationwide cost of stroke care was estimated at 3,737 billion Korean won, highlighting the substantial socioeconomic burden of stroke [5].
The occurrence of stroke is influenced by diverse modifiable and nonmodifiable factors [6]. Modifiable risk factors include disease-related factors such as hypertension, diabetes, obesity, and hyperlipidemia, as well as lifestyle factors such as smoking, alcohol consumption, physical activity, and diet [7]. Managing these factors plays a crucial role in preventing the occurrence of stroke, reducing recurrence after treatment, and promoting rehabilitation [8]. In contrast, non-modifiable factors are unchangeable and are determined from birth, such as age, sex, family history, and genetic factors [9]. Both types of factors not only independently influence stroke occurrence but also interact with each other, leading to a more complex impact on the risk of stroke occurrence [10]. Therefore, it is essential to examine how the incidence rate varies depending on the status of various risk factors.
Physical activity is a key modifiable factor in stroke prevention, not only to reduce stroke risk but also to improve major stroke-related risk factors such as hypertension, obesity, diabetes, and hyperlipidemia [11-15]. In addition, physical activity influences broader lifestyle behaviors, including diet, smoking, and alcohol consumption [8, 15]. While the protective effects of physical activity on stroke have been well documented, recent research suggests that these effects may vary according to genetic susceptibility. However, few studies have examined whether the benefits of physical activity differ depending on an individual’s polygenic risk for stroke, which is the focus of the present study [16-18]. Understanding this interaction could help tailor more effective, personalized stroke prevention strategies. Therefore, recent studies have been exploring the relationship between physical activity and various other factors [19]. Identifying the relationship between physical activity and other risk factors could help in developing better stroke prevention strategies.
Recent research has increasingly focused on the impact of genetic factors, which are non-modifiable factors that mainly affect the risk of stroke occurrence. Genome-wide association studies (GWAS) have identified single nucleotide polymorphisms (SNPs) that contribute to stroke risk, revealing significant associations between these genetic variations and stroke occurrence [20]. SNPs represent variations at a single nucleotide position, and SNPs located near genes associated with a specific disease can influence the function of those genes, potentially increasing the risk of disease [21]. One study has revealed 35 SNP loci that are strongly associated with stroke [22]. Since these SNPs can vary across regions and ethnicities [23], additional research is needed to classify stroke-associated SNPs for each population.
SNPs identified through GWAS are used to calculate polygenic risk scores (PRS), which serve as an indicator of an individual’s genetic risk for disease [24]. The PRS is calculated by multiplying the number of SNPs associated with the disease by the effect size of each SNP [25]. This score quantifies genetic susceptibility, with a higher PRS indicating a higher risk of disease. However, because allele frequencies and effect sizes of SNPs vary substantially across ethnic groups, PRS derived from other populations may not accurately reflect genetic risk in Koreans [26, 27]. Calculating an individual’s PRS allows for the prediction of disease risk, enabling personalized preventive measures and encouraging lifestyle modifications [28]. Therefore, there is a critical need to examine whether the effects of physical activity on stroke risk differ across polygenic risk levels in the Korean population, which could provide insights for population-specific preventive strategies.
Previous studies have demonstrated that both genetic risk and lifestyle factors independently influence the risk of stroke. Although physical activity does not alter underlying genetic variants, recent studies on genetically influenced diseases, such as obesity, have shown that individuals with higher PRS exhibit greater differences in disease risk according to lifestyle behaviors [17]. Additionally, analyses using additive scales have demonstrated that the absolute risk reduction was larger among individuals with higher PRS [18]. These findings emphasize the importance of examining whether the effects of physical activity differ across PRS levels. However, research investigating how physical activity influences stroke risk across different levels of genetic susceptibility remains limited. Therefore, we investigated whether the effect of physical activity on stroke incidence differs according to genetic risk measured by PRS.
METHODS
1. Participants
This study was conducted with the cardiovascular disease association study (CAVAS) in the Korean Genome and Epidemiology Study (KoGES) cohort, including 12,502 adults living in rural areas of Korea, including Yangpyeong in Gyeonggi Province, Goryeong in Gyeongsangbukdo, Namwon in Jeollanam-do, Wonju and Pyeongchang in Gangwon Province, and Ganghwa in Incheon City. The baseline survey was conducted from 2005 to 2011 and the follow-up data were collected from 2007 to 2016. The epidemiological and genetic data in this study were obtained on November 3, 2023. Epidemiological and clinical data were collected through questionnaires and examinations after consent from the participants. All participants were anonymized and assigned identification codes.
All participants were enrolled between the ages of 40 and 69. Participants without genetic data were excluded (n= 4,397). Those with missing values for stroke diagnosis, amount of physical activity, or stroke-related factors were excluded (n=119). Additionally, participants classified as stroke patients in the baseline survey with no previous physical activity data were excluded (n=196). Finally, the analysis was conducted on a total of 7,790 individuals (Fig. 1) (184 stroke patients, 7,606 controls).
Flow chart of sample selection.
2. Genotyping quality control and imputation
The genomic DNA of each participant was isolated from whole blood and the genotypes were identified with a Korean Biobank Array Chip (Korean Biobank, Cheong Ju, Korea). The Korean Disease Control and Prevention Agency provided the genetic variation data. Genetic quality control (QC) was performed as follows. Samples with a call rate <98% were excluded. SNPs failing QC thresholds were removed, including variants with missingness ≥ 2%, Hardy–Weinberg equilibrium (HWE) p<1×10–6, minor allele frequency (MAF) <0.05, or non-biallelic status. Genotype data were aligned to the GRCh37/hg19 genome build, phased using SHAPEIT4, and imputed for autosomal variants with IMPUTE4 using the 1000 Genomes Project Phase 3 reference panel. Post-imputation QC excluded variants with an INFO score <0.8, while applying the same call rate, HWE, MAF, and biallelic filters. These procedures were implemented to ensure the accuracy and reproducibility of downstream analyses. Overall, 5,428,572 SNPs across 22 autosomal chromosomes were retained for analysis.
3. Linkage disequilibrium (LD) analysis and LD pruning
Previous studies have shown that SNPs in high LD are often highly correlated, and including multiple SNPs from the same LD block can introduce redundancy and inflate genetic effect estimates [29]. To address this, LD pruning was performed using PLINK (version 2.0), and only one representative SNP from each LD block was retained. Within each block, the SNP with the lowest p-value in the GWAS was selected as the representative variant [30]. Using these representative SNPs reduced redundancy and enabled a more robust calculation of the PRS.
4. Exposures
1) Polygenic risk score for stroke
To calculate the PRS, SNPs with p-values less than 1×10–5, which satisfied the suggestive threshold of statistical significance in the GWAS results, were selected. We performed LD pruning, which is a process of choosing SNPs that are close together and have similar effects. The SNP with the lowest p-value was selected as the representative SNP among those located within the same gene. For each participant, the PRS was calculated by summing the weighted values of these SNPs, where the weight was the number of alleles for each SNP multiplied by its log odds ratio from GWAS analysis. The PRS calculation was performed using PLINK software (version 2.0). Subsequently, participants were divided into three groups based on their PRS values: high PRS (top 25%), middle PRS (middle 50%), and low PRS (bottom 25%) [31].
2) Physical activity
Participants were first divided into two groups based on the presence of physical activity using the survey question in KoGES epidemiological data, “Do you engage in regular exercise that makes you sweat?”: PA X (weekly average physical activity= 0 min) and PA O (weekly average physical activity >0 min). For participants in the PA O group, the weekly average duration of moderate physical activity was calculated by multiplying responses to the items ‘How many times a week do you exercise?’ and ‘Average amount of physical activity per session’. For the control groups, the average weekly moderate physical activity amount was calculated as the mean of the weekly average physical activity levels measured during the baseline and follow-up surveys, while for the patient group it was based on the reported levels prior to stroke onset. Afterward, participants were grouped based on the ACSM guideline of 150 minutes/week of moderate physical activity [32]: low PA (0 min <weekly average physical activity <150 min), and high PA (weekly average physical activity ≥150 min) [33]. Weekly average physical activity was quantified and classified according to ACSM guidelines to improve the reliability of the physical activity measurement. All physical activity data in KoGES were collected through standardized questionnaires validated in previous cohort studies.
3) Incidence of stroke
Stroke incidence was measured using survey data on stroke diagnosis from the epidemiological data. The survey responses were collected as binary variables with “Yes” and “No” options. In this study, participants who reported a physician-diagnosed ischemic or hemorrhagic stroke in the baseline KoGES survey were classified as stroke cases.
5. Statistical analysis
Statistical analyses were performed using GraphPad Prism ver. 10.2.2 (GraphPad Software Inc., California, USA) and PLINK 2.0. All participants were divided into stroke patients and control groups. For comparison, age, sex, body mass index (BMI), hypertension, diabetes, smoking, alcohol consumption, systolic blood pressure, diastolic blood pressure, white blood cell count, total cholesterol, and blood glucose levels were assessed. Continuous data such as age, BMI, systolic blood pressure, diastolic blood pressure, white blood cell count, total cholesterol, and glucose were presented as means and standard deviations, while categorical data, including sex, hypertension, diabetes, smoking, and alcohol consumption, were expressed as counts and proportions. The significance of continuous data was verified using independent t-tests. Categorical data were analyzed using chi-square tests. Principal component analysis (PCA) was conducted to control for population structure and background genetic variation, minimizing bias and improving the accuracy of gene-disease association results [34]. A logistic regression model was used for the GWAS analysis, with age, sex, smoking, alcohol consumption, hypertension, and 10 principal components (PCs) derived from the PCA analysis included as covariates [35]. The threshold for statistical significance in GWAS was 5×10–8, with a suggestive threshold of 1×10–5. The PRS was calculated by summing the number of risk alleles for all SNPs and then categorized into three groups. The risk of stroke incidence associated with a single genotype was estimated using odds ratios (OR) and 95% confidence intervals (CIs). The relationship between PRS, standardized as z-scores, and physical activity levels with stroke risk was estimated using logistic regression. An OR greater than 1 indicates an increased risk of disease, while an OR less than 1 indicates a reduced risk. The statistical significance level (α) was set at 5%, and all data were presented as means and standard deviations.
RESULTS
1. Baseline characteristics and confounders for GWAS
The baseline characteristics of participants are summarized in Table 1. Data were obtained from 184 stroke cases and 7,606 controls. The mean age was higher in the stroke group (62.2±7.6 years) than in the control group (58.3±8.8 years) (p <.0001). The proportion of men to women showed a significant difference between the stroke group (45.6% men, 54.4% women) and the control group (37.1% men, 62.9% women) (p=.0173). The prevalence of hypertension was higher in the stroke group (33.2%) compared to the control group (26.3%) (p=.0368). Similarly, smoking prevalence was higher in the stroke group (37.0%) than in the control group (27.7%) (p =.0057). Additionally, alcohol consumption was more common in the stroke group (54.4%) than in the control group (46.2%) (p =.0283). The stroke group had higher systolic and diastolic blood pressure than the control group (p=.0004 and p=.0445, respectively). In blood markers, the white blood cell count was significantly higher in the stroke group (6.9±2.2) compared to the control group (6.3±1.8) (p<.0001).
Characteristics of participants at baseline
2. SNPs frequently observed in stroke patients were identified through GWAS
Twenty-eight statistically significant SNPs associated with stroke between stroke patients and controls were identified through GWAS [36]. All participants were divided into the stroke group (n=184) and the control group (n=7,606). Although no significant SNPs were identified at a threshold of 5×10–8, 28 SNPs were identified at a threshold of 1×10–5 (Fig. 2A, B) [25]. To confirm the characteristics of the identified SNPs, we examined the chromosome, base pair, and associated genes. Significant SNPs were found on chromosome 1 (n= 6), 2 (n=1), 5 (n=1), 7 (n=3), 9 (n= 4), 10 (n=3), 12 (n=8), 13 (n=1), and 18 (n=1). Among the 28 SNPs, 1 SNP was located in both FAAP20 (Gene ID: 199990) and LOC 124903823 (Gene ID: 124903823), 3 SNPs in SYT14 (Gene ID: 255928), 1 in RYR2 (Gene ID: 6262), 1 in LINC01470 (Gene ID: 101927134), 3 in DGKB (Gene ID: 1607), 3 in OR13C4 (Gene ID: 138804), 3 in LOC107987105 (Gene ID: 107987105), 8 in PPM1H (Gene ID: 57460), 1 in HS6ST3 (Gene ID: 266722), and 1 in ASXL3 (Gene ID: 80816).
Distribution of significant SNPs by GWAS analysis and LD pruning. (A) Manhattan plot of GWAS results with stroke (184 cases and 7,606 controls). All SNPs analyzed through the GWAS are represented as dots, with the X-axis showing the chromosomal position of each SNP and the Y-axis representing the log-transformed p-values of each SNP. The significance threshold for GWAS (5×10–8) is marked with a red horizontal line, while the suggestive significance threshold (1×10–5) is shown in green. (B) QQ plot of the GWAS analysis for stroke. (C) Heatmap of LD analysis. The color of the cells reflects the degree of LD between SNPs. Red indicates high LD, while blue indicates low LD, with darker colors representing stronger levels. All SNPs are labeled by chromosome, base pair position, and allele. QQ, Quantile-Quantile; GWAS, Genome Wide Association Study; LD, Linkage Disequilibrium; SNP, Single Nucleotide Polymorphism.
Furthermore, among the genes harboring these SNPs, PPM1H and RYR2 have been previously identified as stroke-related genes [37,38]. Additionally, SNPs were found in genes associated with stroke-related diseases, such as atrial fibrillation (DGKB) [39], Parkinson’s disease (ASXL3) [40], and obesity (HS6ST3) (Supplementary Table 1) [41].
3. SNPs for PRS calculation were selected through LD pruning
Among stroke-associated SNPs, LD pruning retained the variants with the lowest p-value within each LD block (Fig. 2C). As a result, 11 representative SNPs that had a significant impact on stroke were selected and used for PRS calculation (Supplementary Table 2).
4. The PRS levels were different between stroke patients and controls
The PRS levels were significantly higher in the stroke group compared to the control group (stroke=2.970, control= 1.869) (Supplementary Fig. 1A). Additionally, to validate the reliability of the PRS, the cohort was expanded by including previously excluded participants (383 cases and 7,715 controls). Consequently, the PRS levels were significantly higher in the stroke group compared to the control group as well (stroke = 2.368, control=1.870) (Supplementary Fig. 1B).
5. The stroke incidence varies depending on the amount of physical activity across different PRS levels
We confirmed the differences in stroke incidence based on the presence or absence of physical activity (Table 2). The stroke incidence was lower in the presence of physical activity in the high PRS and middle PRS groups. There was no significant difference in the low PRS group. Regarding the amount of physical activity, stroke incidence was significantly different based on the amount of physical activity in the high and middle PRS groups. However, there was no significant difference in stroke incidence in the low PRS group (Supplementary Table 3).
Stroke prevention effect of physical activity by PRS level
6. The effect of physical activity on stroke incidence was consistent across PRS levels
We examined the effect of physical activity on stroke incidence across PRS levels using logistic regression (Fig. 3). In the analysis using standardized continuous PRS (Fig. 3A), higher PRS was associated with increased stroke risk (OR=2.54, 95% CI=1.59–4.06, p<.0001), and physical activity showed a modest protective effect (OR=0.997, 95% CI= 0.994–0.9995, p =.0214). The interaction between PRS and physical activity was not significant (OR=1.000, 95% CI= 0.9986–1.0021, p =.704). When PRS was categorized into middle- and high-risk groups (Fig. 3B), stroke incidence was higher in individuals with middle PRS (OR= 3.27, 95% CI=1.52–7.05, p =.0025) and high PRS (OR=11.48, 95% CI= 5.25–25.13, p<1×10–9). Physical activity ×PRS interactions were not significant in the middle (OR= 0.998, 95% CI= 0.989–1.006, p =.49) or high PRS groups (OR=1.000, 95% CI= 0.992–1.008, p =.969).
The effect of physical activity on stroke incidence across PRS levels. (A) Forest plot of logistic regression results using standardized continuous PRS. (B) Forest plot of logistic regression results using categorical PRS groups (Middle and High, with Low PRS as the reference). Odds ratios (ORs) and 95% confidence intervals (CIs) are presented. All models were adjusted for age, sex, and clinical / lifestyle factors, including hypertension, diabetes, smoking status, and alcohol consumption. The analysis includes the main effects of PRS and physical activity, as well as their interaction term (PRS x physical activity). *p<.05, **p<.01, ****p<.0001). PRS, Polygenic Risk Score; OR, Odds Ratio; CI, Confidence Interval.
DISCUSSION
SNPs identified through GWAS for a disease are associated with the genetic risk of the condition. By calculating PRS based on these SNPs, an individual’s genetic predisposition to disease can be quantified. Identifying a disease-specific PRS in advance provides preventive benefits by enabling individuals to adjust their lifestyle habits and respond to potential health risks more effectively. Physical activity is a key preventive factor for stroke, with its protective effect potentially varying based on an individual’s stroke risk factors. However, there is a lack of research examining stroke incidence in relation to physical activity according to strokerelated genetic factors. Therefore, this study aimed to examine the differences in stroke incidence in relation to physical activity across PRS levels.
First, 28 SNPs were identified through GWAS using stroke incidence as the outcome variable. To determine whether the identified SNPs are associated with stroke, the genes harboring these SNPs were identified. From the 28 SNPs, 11 genes were identified. Among them, 1 gene was associated with stroke, and 3 genes were associated with other diseases related to stroke. RYR2 is a gene known to regulate cardiac rhythm and calcium signaling. It has been identified as being associated with atrial fibrillation, and has also been found in a previous GWAS study related to stroke [38]. Additionally, DGKB, HS6ST3, and PPM1H have been identified in previous GWAS studies as being linked to diseases associated with stroke, including type 2 diabetes, obesity, and ADHD, respectively. DGKB is a gene that encodes a metabolic enzyme involved in signaling pathways and vascular function. It has been identified in a previous GWAS study of type 2 diabetes, a major risk factor for stroke [39]. HS6ST3 is a gene involved in heparan sulfate metabolism. It has been found in a previous GWAS study of obesity and high triglyceride levels [41]. The PPM1H gene encodes a protein phosphatase associated with cell signaling [37]. It has been shown to be associated with attention-deficit/hyperactivity disorder, a neurodevelopmental disorder in a previous GWAS study.
This finding is consistent with previous research, which suggests that SNPs identified through GWAS for specific diseases tend to be located in genes associated with those diseases [42]. SNPs located in disease-related genes may affect gene function and disease incidence. Therefore, additional research is needed to confirm the impact of SNPs on the mechanisms of stroke involving these genes. Nevertheless, the significance of this study lies in the fact that stroke-related SNPs were identified through GWAS using the KoGES dataset from a Korean population, suggesting that these SNPs may have the potential to increase the risk of stroke occurrence [43].
Second, the PRS levels of all participants were analyzed using 11 representative SNPs. Consistent with a previous study [44], the results showed that the patient group had higher PRS levels compared to the control group, indicating that the PRS based on the SNPs identified in this study may predict stroke risk. Moreover, a validation analysis was conducted to ensure the reliability of the PRS in predicting stroke. The results showed that patients still had higher PRS levels than the control group. This is consistent with prior research suggesting that higher PRS levels are associated with greater vulnerability to stroke [45]. These results suggest that the PRS in this study may reflect the genetic risk for stroke, and confirm that an individual’s genetic risk of stroke can be estimated using PRS.
Third, the preventive effect of physical activity on stroke is well established [46]. In this study, we aimed to investigate whether this effect varies across different PRS levels. However, no significant effect was observed in the low, middle, and high PRS groups. A previous study reported that, in the case of type 2 diabetes, the absolute risk reduction due to lifestyle improvements was significantly greater in individuals with high PRS [18]. Since physical activity is also a key lifestyle factor, it is possible that the risk reduction due to physical activity in stroke may be relatively smaller in those with low PRS. To understand the cause of this phenomenon, it is necessary to include a sufficient number of participants with low PRS and further investigate their characteristics. Nevertheless, since stroke is a condition where post-onset management is challenging and moreover, the prognosis and treatment outcomes often remain unfavorable, highlighting the importance of preventive approaches is essential [47]. Therefore, confirming that the preventive effect of physical activity on stroke may vary according to PRS levels is significant, as it can help in developing effective stroke prevention strategies. Next, we examined the differences in stroke incidence based on the amount of physical activity within each PRS level. While no significant difference was observed in the low PRS group, differences were found in the middle and high PRS groups. When examining the stroke incidence with increasing amounts of physical activity, the middle PRS group showed a decrease in stroke incidence as the amount of physical activity increased. On the other hand, in the high PRS group, stroke incidence was higher in the group with more physical activity compared to the group with less physical activity. However, these results should be interpreted with caution due to the relatively small sample size in this subgroup, which may make the findings unstable. Some previous studies suggest that excessive endurance exercise may have a negative impact on cardiovascular health [48]. Furthermore, one study revealed that excessive physical activity duration could be a risk factor for hypertension in middle-aged populations [49]. Physical activity plays a critical role in stroke prevention by regulating key risk factors such as obesity and hypertension [50]. Although physical activity does not change the genetic sequence itself, it can influence protein levels and gene expression [51]. This study suggests that the preventive effect of physical activity may vary depending on genetic risk, but there is insufficient evidence to support this result. Therefore, the underlying mechanisms for the observed association remain speculative, and further studies are needed to clarify the relationship between high PRS, physical activity, and stroke risk, including investigations into potential biological mechanisms.
Furthermore, logistic regression analysis was conducted to examine whether the effect of physical activity on stroke prevention differs according to PRS levels. Both PRS and physical activity were significantly associated with stroke incidence. However, the interaction between PRS and physical activity was not statistically significant, suggesting that the protective effect of physical activity was consistent across different genetic risk groups. One possible explanation is that the beneficial impact of physical activity could operate similarly regardless of underlying genetic predisposition. Alternatively, the absence of interaction may reflect limited statistical power to detect effect modification across PRS levels. Importantly, these results support the public health relevance of physical activity as an effective preventive strategy, even among individuals with elevated genetic risk.
This study has several limitations. First, the number of incident stroke cases was limited, which reduced statistical power and restricted the use of more advanced analyses, such as PRS × physical activity interaction models or time-to-event analyses. The small sample size in certain subgroups, including the high-PRS and high-physical-activity group as well as the low-PRS group, also made the findings more vulnerable to random variation or residual confounding. In addition, because the study cohort consisted of individuals residing in rural areas of Korea, lifestyle patterns specific to rural living may have influenced the results, thereby limiting the generalizability of the findings to the broader Korean population. Further studies incorporating non-rural populations and cross-country comparisons are needed to contextualize these observations.
There are also methodological considerations related to the assessment of physical activity and the construction of the PRS. Physical activity was measured using a self-reported questionnaire, which may have introduced recall bias or misclassification; more detailed or objective measures, such as the IPAQ or accelerometer-based monitoring, would allow for a more accurate evaluation of activity levels and dose–response relationships. With respect to PRS construction, LD pruning was performed by excluding SNPs with r² ≥ 0.5, and SNP selection relied solely on p-values, which may have allowed correlated or non-functional variants to remain in the model. Moreover, the PRS was generated using SNP effect sizes derived from a GWAS conducted in the same cohort, raising the possibility of overfitting that could reduce predictive performance in independent populations. External validation was not possible due to the use of a single cohort and the limited availability of comparable datasets.
Taken together, these limitations highlight the need for replication in larger, independently recruited cohorts, the inclusion of objective physical activity measurements, and the use of external datasets for PRS validation. Such efforts will be essential for improving the robustness, reliability, and generalizability of our findings.
In conclusion, this study suggests that stroke incidence in relation to physical activity may differ according to PRS levels. Additionally, our results suggest that further research is needed to explore the relationships among risk factors for stroke. To the best of our knowledge, this is the first study to examine the differential impact of physical activity on genetic risk in stroke incidence. However, additional research that overcomes the limitations of this study and includes larger cohorts is required to provide new perspectives on stroke prevention.
CONCLUSION
In conclusion, 28 stroke-associated SNPs were identified using KoGES data, and the PRS calculated from these SNPs was higher in the patient group compared to the control group. Physical activity and PRS were each associated with stroke incidence, but no interaction between physical activity and PRS was observed. These findings should be considered hypothesis-generating, highlighting the need for further research using larger and independent cohorts to validate the observed associations and to better understand the interplay between genetic risk and physical activity in stroke prevention.
Supplementary Material
Supplementary Table 1.
List of significant stroke-associated SNPs in GWAS
Supplementary Table 2.
List of 11 representative stroke-associated SNPs identified after LD pruning
Supplementary Table 3.
Stroke incidence depending on the amount of physical activity across PRS levels
Supplementary Fig. 1.
Comparison of PRS levels between the control and stroke groups. (A) Existing participants (184 cases and 7,606 controls). (B) Extended cohort (383 cases and 7,715 controls) includes the previously excluded participants (199 cases and 109 controls). The statistical analysis was performed using an independent t-test. Data were presented as mean±SEM (****p<.0001).
Notes
ACKNOWLEDGMENTS
This study was conducted with bioresources from National Biobank of Korea, the Center for Disease Control and Prevention, Republic of Korea (NBK-D03-B, NBK-D03-F01, NBK-D03-F02, NBK-D03-F03, NBK-D03-F04). The authors have no relevant financial or non-financial interests to disclose.
CONFLICT OF INTEREST
The authors have no relevant financial or non-financial interests to disclose.
AUTHOR CONTRIBUTIONS
Conceptualization: YS Cho, HY Moon; Data curation: YS Cho; Formal analysis: YS Cho, JE Lee, JH Park, K Tanisawa, HY Moon; Funding acquisition: HY Moon; Methodology: YS Cho, TY Kim, K Yuk, YS Kim, K Tanisawa, HY Moon; Project administration: YS Cho, HY Moon; Visualization: YS Cho; Writing - original draft: YS Cho; Writing - review & editing: YS Cho, TY Kim, K Yuk, YS Kim, MC Lee, JP Jeon, I Jeong, HY Moon.