Identification of genetic heterogeneity in coronary heart disease stratification by integrating phenotypic and genetic features

To the Editor: Coronary heart disease (CHD) remains a critical public health issue, significantly impacting populations worldwide. The etiology of CHD is often complex and involves a diverse set of risk factors. Although individuals with CHD share a common endophenotype, they generally exhibit distinct molecular or signaling pathways.[1] Clustering, a commonly employed exploratory and hypothesis-generating technique, is vital for examining heterogeneous subtypes that may aid in accurate cardiovascular phenotyping and disease prevention. However, prior studies have predominantly focused on phenotype-only or genetic-only clustering, with limited integration of genetic and non-genetic factors in exploring heterogeneity. In this study, we performed a clustering analysis by integrating genetic and phenotypic data, further examining stratified CHD risk and the heterogeneity within the clusters. This prospective study utilized data from the Fangshan Family-based Ischemic Stroke Study in China (FISSIC),[2] with approval from the Ethics Committee of the Peking University Health Science Centre (No. IRB00001052-13027). All participants provided written informed consent. We analyzed baseline data from the second phase of the FISSIC study, collected through a face-to-face questionnaire that covered sociodemographic information, education, occupation, dietary habits, lifestyle behaviors, and medical history. Two-year incident CHD assessments were obtained from the 2014 follow-up. Detailed cohort characteristics and outcome definitions are available in the supplementary materials. Supplementary Figure 1, https://links.lww.com/CM9/C610 outlines the flowchart of eligible participants included in the study. In brief, we included 2514 participants who had genotyping data and were free from CHD at baseline. We initially conducted reverse genome-wide association studies (RGWAS) analysis to identify subtypes based on baseline phenotypic and genetic features. Feature inclusion was guided by missing rate, variable correlation, and established disease knowledge. After clustering using RGWAS, we characterized the distinct clusters. We then employed Cox regression analysis to assess CHD risk stratification. To further investigate the genetic heterogeneity underlying the clusters and their potential mechanisms related to CHD, we selected an initial subset of single-nucleotide polymorphisms (SNPs) that were distributed significantly differently among clusters (false discovery rate <0.05). We tested the association of these SNPs with CHD both in the overall cohort and within each cluster. Gene expression analysis was conducted using dataset GSE64554 from the Gene Expression Omnibus (https://www.ncbi.nlm.nih.gov/geo/) database. Among the 2514 participants, 146 were diagnosed with incident CHD in 2014, yielding an overall incidence of 5.8%. Basic characteristics are provided in Supplementary Table 1, https://links.lww.com/CM9/C610. For the RGWAS, we included six binary traits, hypertension, type 2 diabetes, dyslipidemia, smoking, drinking, preference for salty taste, and eight quantitative traits, namely, fasting blood glucose (FBG), body mass index (BMI), low-density lipoprotein cholesterol, high-density lipoprotein cholesterol, triglycerides, systolic blood pressure (SBP), diastolic blood pressure (DBP), and serum creatinine (SCr). Age and the use of medications for hypertension, glucose, and lipid control were included as covariates. Genetic features were represented by the top 10 genetic principal components. We identified three clusters. Cluster 1 displayed relatively better metabolic health and healthier behavioral habits. Compared with the other two clusters, Cluster 1 showed optimal management of blood glucose, blood lipids, and SBP, with notably lower rates of smoking and drinking. Cluster 2, termed the risk behavior group, had the highest proportions of smoking and drinking. A self-reported preference for a salty diet was observed in 46.3% (295/637) of participants in this cluster, significantly higher than in the other two clusters, along with notably higher SBP and the highest prevalence of hypertension. Cluster 3 was marked by a dominance of hyperglycemia, accompanied by elevated BMI. Although participants in Cluster 3 were younger and generally reported better behavioral habits, this cluster exhibited the poorest metabolic status, with the highest prevalence of type 2 diabetes and elevated FBG [Figure 1A]. Furthermore, obesity, as measured by BMI, was most prevalent in this group [Supplementary Figure 1, https://links.lww.com/CM9/C610]. The 2-year incidence rate of CHD was lowest in Cluster 1 (4.1%) and highest in Cluster 3 (12.7%). Compared with Cluster 1, the hazard ratio for Cluster 2 was 1.90 (95% confidence interval (CI): 1.29–2.78, P = 0.001), and for Cluster 3, it was 3.38 (95% CI: 2.14–5.26, P <0.001) [Figure 1B, Supplementary Table 2, https://links.lww.com/CM9/C610].Figure 1: Comparison of incident CHD risk, characteristics of health conditions and lifestyle factors, and SNP associations across the three clusters. (A) Radar plot of health conditions and lifestyle factors across three clusters. Lines in grey, blue, and red correspond to cluster 1, cluster 2, and cluster 3, respectively. (B) Kaplan–Meier survival curve of the incidence of CHD. (C) Heatmap showing distinct SNP effect patterns on CHD risk across three clusters. CHD: Coronary heart disease; OR: Odds ratio; SNP: Single-nucleotide polymorphism.Beyond common cardiometabolic characteristics, Scr, a widely used marker of renal function, emerges as a key feature that improves CHD classification clustering. The increased CHD risk observed in Clusters 2 and 3 may be partially attributed to decreased renal function, as previous studies have demonstrated a link between even modest elevations in SCr and increased cardiovascular risk.[3] However, despite similar SCr levels, the pathogenic pathways diverge between Clusters 2 and 3. In Cluster 2, a preference for salty foods appears to be a contributing factor, along with elevated SBP, a major driver of target organ damage. The harmful effects of dietary salt intake are also associated with salt sensitivity, local hormone interactions, and growth factors that stimulate collagen synthesis, as reviewed by Hosohata et al.[4] By contrast, the CHD risk in Cluster 3 is predominantly driven by hyperglycemia. Here, elevated SCr levels may reflect phenotype–genotype interactions involving renal function and diabetes. Genome-wide association study (GWAS) on SCr-based estimated glomerular filtration rates in individuals with diabetes identified 2 established and 27 potentially relevant loci. Nordeidet et al[5] also identified genetic variants associated with cardiorespiratory fitness, blood creatinine levels, diabetes risk, and endocarditis. Thus, SCr holds promise as a biomarker that merits further investigation for its regulatory connections with blood pressure and diabetes in the context of CHD. We identified 280 SNPs that were distributed significantly differently among the three clusters [Figure 1C]. These SNPs were primarily annotated to two genes, with arginine-glutamic acid dipeptide repeats (RERE) showing notably higher expression levels in CHD cases, particularly in subcutaneous adipose tissue [Supplementary Figure 3, https://links.lww.com/CM9/C610]. Association analysis revealed substantial heterogeneity between the clusters, with 76.1% (213 out of 280, binomial test P <0.0001) of SNPs displaying inconsistent effect directions across the clusters [Supplementary Table 3, https://links.lww.com/CM9/C610]. Two SNPs were identified as risk factors for CHD among the overall participants (rs147647658, odds ratio [OR] = 2.31, 95% CI: 1.17–4.53, P = 0.015; rs73228757, OR = 1.59, 95% CI: 1.09–2.32, P = 0.016), although only rs73228757 was significant in Cluster 3, with a stronger effect (OR = 4.41, 95% CI: 1.45–13.45, P = 0.0091). Cluster 1 exhibited a specific protective variant, rs142821158 (OR = 0.33, 95% CI: 0.12–0.95, P = 0.039), while Cluster 3 displayed a distinct risk variant, rs3019076 (OR = 3.48, 95% CI: 1.33–9.11, P = 0.011). Cluster 2 showed no cluster-specific variants, aligning with the profile of this group, as CHD risk here was primarily influenced by unhealthy lifestyle factors, with minimal genetic contribution. Interestingly, rs142821158 provided protective effects in Cluster 1, whereas it tended to pose a risk in Clusters 2 and 3. rs3019076 posed a significant risk in Cluster 3 but was not evident in the overall participant analysis, as its effect was offset by protective effects in other clusters. rs73228757 was associated with CHD risk across all participants and in Cluster 3. While prior evidence of its association with CHD is lacking, rs73228757 has been linked with alcohol consumption, smoking initiation, triglyceride levels, and apolipoprotein B in phenome-wide association study (PheWAS) analysis. At the gene level, RERE has also been associated with smoking cessation. Although adrenoceptor alpha 2A (ADRA2A) did not display differential expression, its most significant phenotypic associations included diastolic blood pressure, SCr, and mean arterial pressure. Our study has several limitations. First, only baseline data were used to construct clusters, and the cross-sectional data on lifestyle exposures are susceptible to recall bias, which may be increased by the inherently subjective measure of salty taste preference. Adopting more objective methods to assess sodium intake, such as 24-h urine samples, would be beneficial. Second, the two-year follow-up period is relatively short for capturing incident CHD cases and may introduce misclassification bias. The small sample size may also limit the statistical power of our association analysis between SNPs and CHD. Longer-term prospective studies are needed to confirm these findings. Third, we used exome-wide SNPs rather than genome-wide data to define the genetic features of the clusters. While efficient for data-driven analysis, this approach may overlook significant variants in intergenic regions. Finally, our study is limited to the northern Han Chinese population. Additional validation in other ancestries is needed to ensure the generalizability of these CHD risk subgroups. In conclusion, in the northern Han Chinese population, we identified three distinct clusters with stratified CHD risk by simultaneously integrating genetic and phenotypic features, revealing heterogeneous genetic effects within the clusters. Further research is required to validate these clusters and to clarify the biological mechanisms of the associated key genes. Funding This work was supported by grants from the National Natural Science Foundation of China (Nos. 81872692 and 82073642), the High-Performance Computing Platform of Peking University and the grants from China Scholarship Council (No. 202306010285). Conflicts of interest None.