Skip to main content
Journal of the American Heart Association: Cardiovascular and Cerebrovascular Disease logoLink to Journal of the American Heart Association: Cardiovascular and Cerebrovascular Disease
. 2026 Oct 16;15(20):e050689. doi: 10.1161/JAHA.126.050689

Multivariate Genomic Analysis Reveals Shared Genetic Architecture and Potential Therapeutic Targets for Atherosclerotic Cardiovascular Disease

Yifeng Xu 1, Jie Wang 1,✉, Ruoqi Zhang 1, Chengzhi Hou 1, Xuan Sun 1, Qiyang Li 2, Chao Liu 1
PMCID: PMC13769608  PMID: 42885799

Abstract

Background

Atherosclerotic cardiovascular disease (ASCVD) is a leading cause of mortality worldwide and comprises multiple clinical phenotypes, but their shared genetic architecture remains incompletely defined.

Methods

We integrated genome‐wide association study summary statistics for coronary artery disease, peripheral artery disease, large‐artery atherosclerotic stroke, and carotid plaque using genomic structural equation modeling to derive a multivariable ASCVD common factor. We applied fine‐mapping, multi‐marker analysis of genomic annotation, transcriptome‐wide association studies, Mendelian randomization, colocalization, mediation analysis, and virtual screening.

Results

We identified 289 linkage disequilibrium–independent lead single‐nucleotide polymorphisms mapping to 177 genomic risk loci; 13 lacked corresponding genome‐wide significant signals in the input genome‐wide association studies. Eleven had prior genome‐wide association study catalog association records, whereas no relevant records were identified for the loci represented by rs2294777 near SASH1 and rs4843053 near AKAP13. Fine‐mapping prioritized 26 putative causal variants. Integrated functional summary‐based imputation, multi‐tissue transcriptome‐wide association study, and multi‐marker analysis of genomic annotation analyses implicated 123 genes. Enrichment analyses highlighted endothelial cells and H3K27ac‐marked regulatory regions. Lipidomic analyses linked higher triacylglycerols and sterol esters to increased multivariable ASCVD common factor risk. Protein Mendelian randomization analyses in deCODE and Fenland prioritized aldehyde dehydrogenase 2 (ALDH2), phosphodiesterase 5A (PDE5A), HtrAserinepeptidase 1 (HTRA1), and transcription elongation factor A2 (TCEA2) as candidate targets. Colocalization supported a shared causal variant between TCEA2 and multivariable ASCVD common factor. Mediation analyses suggested partial mediation by diastolic blood pressure for ALDH2, PDE5A, and TCEA2 and by triglycerides for TCEA2. Virtual screening identified candidate compounds with favorable predicted binding to PDE5A and ALDH2.

Conclusions

Integrating multivariate genome‐wide association study with multi‐omics and causal inference analyses identified shared genetic signals associated with ASCVD susceptibility and prioritized candidate protein targets, informing future functional studies and target validation.

Keywords: atherosclerotic cardiovascular disease, genome‐wide association studies, genomic structural equation modeling, proteomics, transcriptome‐wide association studies

Subject Categories: Coronary Artery Disease, Vascular Disease, Atherosclerosis


Nonstandard Abbreviations and Acronyms

CAP

carotid plaque

LAS

large‐artery atherosclerotic stroke

MAGMA

multi‐marker analysis of genomic annotation

mvASCVD

multivariable atherosclerotic cardiovascular disease common factor

PheWAS

phenome‐wide association study

pQTL

protein quantitative trait locus

QTL

quantitative trait locus

SEM

structural equation modeling

TWAS

transcriptome‐wide association study

Research Perspective.

What Is New?

  • This study used multivariate genomic structural equation modeling to define an atherosclerotic cardiovascular disease common factor and characterize shared genetic signals across coronary artery disease, peripheral artery disease, large‐artery atherosclerotic stroke, and carotid plaque, identifying 177 genomic risk loci; 13 lacked corresponding genome‐wide significant signals in the 4 input single‐trait genome‐wide association studies, 11 of which had prior genome‐wide association study catalog records, whereas the loci containing rs2294777 near SASH1 and rs4843053 near AKAP13 had no relevant records.

  • Based on the atherosclerotic cardiovascular disease common‐factor genome‐wide association study, this study integrated transcriptomic, proteomic, Mendelian randomization, colocalization, mediation, phenome‐wide association study, and virtual screening analyses to prioritize candidate genes, biological pathways, protein targets (ALDH2, PDE5A, HTRA1, and TCEA2), and candidate small‐molecule compounds.

What Question Should Be Addressed Next?

  • Future studies should use functional genomics, experimental perturbation, and pharmacologic validation to further identify and validate causal variants reflecting shared genetic risk of atherosclerotic cardiovascular disease, test the reproducibility of these genetic signals across diverse ancestry populations, and advance validation of prioritized protein targets (ALDH2, PDE5A, HTRA1, and TCEA2) and translational evaluation of candidate compounds.

Atherosclerotic cardiovascular disease (ASCVD) is a major contributor to the global burden of death and disability. 1 It encompasses key clinical phenotypes, including coronary artery disease (CAD), large‐artery atherosclerotic stroke (LAS), and peripheral artery disease (PAD), as well as subclinical imaging phenotypes of atherosclerosis exemplified by carotid plaque (CAP). Although these phenotypes differ in anatomic context, hemodynamic stress, and acute triggering factors, they share a common pathophysiological foundation: endothelial dysfunction promotes lipid retention and immune–inflammatory cascades that drive plaque formation, progression, and destabilization. 2 , 3 , 4 Despite well‐established preventive and therapeutic strategies for ASCVD, the shared genetic architecture across ASCVD‐related phenotypes remains incompletely characterized. A clearer understanding of this shared genetic basis may improve mechanistic insight into systemic atherosclerotic risk and help prioritize cross‐phenotype molecular pathways and candidate targets for future validation.

In recent years, genome‐wide association studies (GWASs) have identified hundreds of risk loci associated with individual ASCVD phenotypes. For example, large‐scale meta‐analyses of CAD and stroke have each reported dozens of genome‐wide significant regions. 5 , 6 However, single‐phenotype studies treat ASCVD phenotypes as isolated traits, which can obscure the underlying systemic genetic architecture. Moreover, most cross‐phenotype studies have been limited to pairwise genetic correlation estimates or simple overlap of association signals; these approaches often lack power to model latent shared genetic influences spanning multiple phenotypes and do not adequately disentangle shared causative signals from phenotype‐specific effects. Genomic structural equation modeling (SEM) accounts for sample overlap and differences in sample size when modeling the genetic covariance structure across traits. 7 This strategy can integrate GWAS summary statistics from studies with different sample sizes and increase power to detect cross‐trait association signals that may not reach genome‐wide significance in single‐trait GWASs.

Here, we constructed a multivariable ASCVD common factor (mvASCVD) using large‐scale GWAS summary statistics to characterize the shared genetic architecture of ASCVD in individuals of European ancestry. Within this mvASCVD framework, we prioritized putative causal variants and candidate genes, and delineated their biological context using Bayesian fine‐mapping, transcriptome‐wide association studies (TWASs), multi‐marker analysis of genomic annotation (MAGMA), regulatory annotation, and cell‐type enrichment analyses. We further integrated lipidomic and proteomic data and applied Mendelian randomization (MR) and colocalization to prioritize modifiable risk factors and candidate targets, followed by a phenome‐wide association study (PheWAS), virtual screening, and molecular docking of candidate targets to evaluate broad phenotypic associations, potential safety‐related signals, and candidate compounds.

METHODS

All data analyzed in this study were obtained from publicly available summary‐level data sets. GWAS summary statistics for the 4 ASCVD‐related phenotypes, quantitative trait locus (QTL) data sets, lipidomic GWAS summary statistics, proteomic protein QTL (pQTL) summary statistics, and other exposure GWAS data sets are described in detail in the Methods and Tables S1 and S2. The mvASCVD GWAS summary statistics generated in this study are available at Figshare (DOI: http://doi.org/10.6084/m9.figshare.32604783).

Overview of Study Design

We integrated 4 European‐ancestry GWASs of ASCVD‐related phenotypes (CAD, PAD, LAS, and CAP). After unified quality control and harmonization, we used genomic SEM to derive a common factor across the 4 phenotypes and to estimate GWAS summary statistics for this factor. We then conducted systematic functional characterization of common‐factor association signals. Specifically, functional mapping and annotation of genome‐wide association studies (FUMA) was used to define genomic risk loci and lead single‐nucleotide polymorphisms (SNPs), whereas sum of single effects fine‐mapping prioritized putative causal variants. We conducted TWAS using FUSION (functional summary‐based imputation) and MTWAS (multi‐tissue transcriptome‐wide association studies) and performed gene‐based association analysis using MAGMA. We further performed tissue and cell‐type enrichment analyses and partitioned heritability analyses to interrogate the regulatory context. Finally, we integrated 2‐sample MR analyses to identify prioritized targets and further performed colocalization analysis; we also used PheWAS and virtual screening to evaluate potential safety‐related signals and candidate compounds for the prioritized targets. The overall workflow is shown in Figure 1. Ethics approval and informed consent were obtained in the original studies for all contributing GWASs, and rigorous quality control was implemented throughout our analyses.

Figure 1. Research flowchart.

Figure 1

CAD indicates coronary artery disease; CAP, carotid plaque; FUSION, functional summary‐based imputation; H3K27ac, histone H3 lysine 27 acetylation; LAS, large‐artery atherosclerotic stroke; LD, linkage disequilibrium; MAGMA, multi‐marker analysis of genomic annotation; MTWAS, multi‐tissue transcriptome‐wide association study; mvASCVD, multivariable atherosclerotic cardiovascular disease common factor; PAD, peripheral artery disease; PC, phosphatidylcholine; PE, phosphatidylethanolamine; PheWAS, phenome‐wide association study; PPH3, posterior probability of hypothesis 3; PPH4, posterior probability of hypothesis 4; SNP, single‐nucleotide polymorphism; STE, sterol ester; and TAG, triacylglycerol.

Data Sources and Quality Control

Using GWAS summary statistics for CAD, PAD, LAS, and CAP, we applied genomic SEM to interrogate their shared genetic architecture. To ensure a consistent linkage disequilibrium (LD) reference panel and to mitigate population stratification, we included only GWAS summary statistics from participants of European ancestry. The CAD, 8 PAD, 9 LAS, 10 and CAP 11 GWASs comprised 1 165 690, 511 634, 1 241 207, and 48 434 participants, respectively (Table S1). Genomic SEM accounts for sample overlap and differences in sample size when modeling the genetic covariance structure across traits. 7 We harmonized quality control across the 4 sets of GWAS summary statistics using the European 1000 Genomes Project Phase 3 (1KGP3) reference panel: we restricted variants to the HapMap3 set, excluded SNPs in the major histocompatibility complex region, removed SNPs with minor allele frequency <0.01 or discordant allele alignment, and, when available, excluded SNPs with imputation quality INFO <0.9.

Kaiser–Meyer–Olkin Test and Parallel Analysis

For the 4 quality‐controlled univariate input GWAS summary statistics, we used multivariate LD score regression (LDSC) to estimate the genetic covariance matrix and its sampling variance–covariance matrix. We then used the Kaiser–Meyer–Olkin measure of sampling adequacy to evaluate whether the correlation matrix was suitable for factor analysis. A Kaiser–Meyer–Olkin value >0.50 was considered the minimum acceptable threshold for factor extraction. In addition, we further performed parallel analysis to determine the number of latent factors to retain. Briefly, we computed eigenvalues of the observed genetic correlation matrix and, leveraging the sampling error structure encoded in the sampling variance–covariance matrix, generated a null distribution of eigenvalues via Monte Carlo simulation. We compared observed and simulated eigenvalues in descending order and retained the largest number of factors for which the observed eigenvalue remained greater than the corresponding simulated eigenvalue.

Exploratory and Confirmatory Factor Analysis

Guided by the parallel analysis, we fitted a common factor model within the genomic SEM framework and conducted exploratory factor analysis to estimate each phenotype's loading on the common factor and its trait‐specific residual variance. We also quantified the proportion of variance in the genetic correlation matrix explained by the common factor. We subsequently performed confirmatory factor analysis using diagonally weighted least squares estimation, and evaluated model fit based on the standardized root mean square residual, comparative fit index, and the χ2 test statistic. In general, lower standardized root mean square residual and higher comparative fit index values indicate better fit; we therefore considered standardized root mean square residual <0.08 and comparative fit index >0.95 as indicative of acceptable fit for the common factor model.

Common‐Factor GWAS

After the common‐factor model achieved acceptable fit, we conducted a common‐factor GWAS. Specifically, we incorporated the cross‐phenotype associations of each autosomal SNP into the covariance matrix obtained from multivariate LDSC to construct an expanded covariance matrix that included the SNP. We then fitted the common‐factor model separately for each SNP using diagonally weighted least squares, estimating the SNP effect on the common factor and its standard error. This yielded GWAS results for mvASCVD, reflecting the shared genetic covariance across the 4 phenotypes.

SNP Heterogeneity Assessment

In the common‐factor GWAS, we performed SNP‐level tests of heterogeneity (QSNP) to evaluate whether SNP effects were adequately captured by the common factor model or showed residual heterogeneity across the input traits. Genome‐wide significance was determined using Bonferroni correction across all SNPs tested.

Identification of LD‐Independent Lead SNPs and Genomic Risk Loci

Based on SNP‐level association statistics from the mvASCVD GWAS, we used FUMA 12 , 13 to identify independent significant SNPs, lead SNPs, and genomic risk loci, using the 1KGP3 European population as the LD reference panel. Genome‐wide significant SNPs (P<5×10−8) with pairwise r2<0.6 were defined as independent significant SNPs, among which SNPs with pairwise r2<0.1 were further defined as lead SNPs. LD blocks of independent significant SNPs separated by <250 kb were merged into the same genomic risk locus. In addition, all genome‐wide significant mvASCVD SNPs were subjected to LD‐based clumping in PLINK using the same reference panel, with an r2 cutoff of 0.1 and a 1‐Mb window. Clumping was restricted to SNPs reaching genome‐wide significance, and the SNP with the smallest association P value in each clump was retained as the clump representative. Each PLINK clump representative was subsequently assigned to the corresponding FUMA‐defined genomic risk locus according to its genomic position.

Heterogeneity Assessment of Lead SNPs

To assess whether mvASCVD lead SNPs showed heterogeneity across the input traits, we summarized QSNP heterogeneity statistics for lead SNPs identified by FUMA. Statistical significance was assessed using Bonferroni correction for the number of lead SNPs tested.

Evaluation of Genomic Risk Loci

To compare association signals from the mvASCVD GWAS with those from the 4 input single‐trait GWASs at the locus level, we evaluated each FUMA‐defined mvASCVD genomic risk locus for the presence of a corresponding genome‐wide significant signal in the CAD, PAD, LAS, and CAP GWASs. We extracted all SNPs reaching genome‐wide significance (P<5×10−8) from the 4 input GWASs. An mvASCVD locus was considered to have a corresponding signal in the input GWASs if at least 1 genome‐wide significant SNP from any input GWAS either fell within that locus or was located within ±1 Mb of, and in LD with (r2≥0.1), any PLINK clump representative mapped to that locus. The LD assessment included all PLINK clump representatives mapped to the same FUMA locus and used the 1KGP3 European population as the LD reference panel. A locus was classified as lacking a corresponding genome‐wide significant signal in the input single‐trait GWASs only when no genome‐wide significant SNP satisfying either criterion was identified in any of the 4 input GWASs.

To further assess loci without corresponding genome‐wide significant signals in the 4 input single‐trait GWASs, we obtained previously reported genome‐wide significant associations (P<5×10−8) from the GWAS catalog through FUMA, accessed on December 24, 2025. 14 For each locus, we used PLINK and the 1KGP3 European reference panel to calculate pairwise LD (r2) between all FUMA lead SNPs assigned to that locus and GWAS catalog variants located within the same genomic risk locus, with r2≥0.8 considered to indicate high LD with a previously reported GWAS signal. 15 , 16

Bayesian Fine‐Mapping

We used susie_rss in susieR to perform Bayesian fine‐mapping of mvASCVD loci and prioritize variants most likely to drive association signals within each locus. 17 For each lead SNP, we extracted all variants within a±250‐kb window centered on the lead SNP and calculated the corresponding LD correlation matrix based on the 1KGP3 European reference panel. We then input the single‐variant summary statistics and the LD matrix into susie_rss, set the maximum number of putative causal signals to 10, and output the 95% credible set for each signal and the posterior inclusion probability (PIP) for each variant. Variants with PIP>0.95 and located within the 95% credible set were prioritized as putative causal variants.

Transcriptome‐Wide Association Study

We applied FUSION and MTWAS to conduct TWAS to link mvASCVD association signals with genetically regulated gene expression and prioritize candidate genes. FUSION integrates pretrained cis‐regulated expression prediction weights from an expression QTL reference panel with mvASCVD GWAS summary statistics and uses an LD matrix from an independent reference population to account for the correlation structure, thereby providing gene‐level TWAS association test results from summary statistics. 18

MTWAS performs joint modeling in a multitissue framework by decomposing genetically regulated effects into components shared across tissues and tissue‐specific components, and then aggregating statistical evidence across tissues to improve robustness and sensitivity when the key causal tissue is unknown or when regulatory effects are shared across tissues. 19 In these analyses, FUSION and MTWAS used Genotype‐Tissue Expression v8 precomputed expression prediction weights and Genotype‐Tissue Expression v8 multitissue pretrained weights, respectively. Significance thresholds were set using Bonferroni correction.

For genes that were significant in both FUSION and MTWAS, we then performed gene‐level Bayesian fine‐mapping within each independent risk locus using fine‐mapping of causal gene sets (FOCUS). 20 Within each locus, FOCUS compares models across genes, reports the PIP for each gene, and constructs a credible gene set to prioritize genes most likely to drive the association signal at that locus.

Tissue‐Specificity and Enrichment Analyses

To evaluate the tissue‐specific regulatory context of mvASCVD association signals, we used QTLEnrich to test whether mvASCVD GWAS signals were enriched among expression QTLs and splicing QTLs identified across 49 Genotype‐Tissue Expression v8 tissues. 21 QTLEnrich implements a rank‐based framework with permutation testing and controls for confounding attributable to variant characteristics by constructing matched background variant sets. Background matching accounted for minor allele frequency, distance to the transcription start site, and local LD structure. Enrichment was quantified as adjusted fold enrichment, and statistical significance was assessed using permutation‐derived empirical P values. Results across all tissue×QTL‐type combinations were corrected using the Benjamini–Hochberg procedure. As a complementary analysis, we applied MAGMA 22 to the mvASCVD GWAS summary statistics to perform tissue‐expression analysis and assess concordance with the tissue‐specific signals identified by QTLEnrich. We also conducted MAGMA gene‐level association analysis to identify significant genes, followed by gene‐set enrichment analysis to investigate potentially relevant biological functions; results were corrected using Bonferroni correction.

Cell‐Type Specificity Analysis

To identify cell types potentially relevant to mvASCVD association signals, we integrated mvASCVD GWAS summary statistics with single‐cell transcriptomic data from the Tabula Muris Consortium (adult Mus musculus samples spanning 20 organs/tissues and >100 000 cells). 23 We used the CELLECT pipeline. Briefly, we first applied CELLEX to compute gene‐level expression specificity scores across cell types, 24 and mapped mouse genes to human 1‐to‐1 orthologs to align with human GWAS gene annotation. We then incorporated cell‐type expression specificity as a continuous annotation within the stratified LDSC framework, and assessed cell‐type–specific heritability enrichment for mvASCVD by testing the significance of the corresponding regression coefficients. P values were Bonferroni corrected.

Partitioned Heritability by Functional Genomic Annotation

To evaluate whether mvASCVD SNP heritability was enriched in specific functional genomic elements, we used stratified LDSC to partition mvASCVD SNP heritability and quantify the contribution of functional genomic annotations to the genetic architecture of mvASCVD. Relative to conventional LDSC, stratified LDSC computes annotation‐specific LD scores for each functional category and decomposes SNP heritability by regressing GWAS association statistics on these annotation‐specific LD scores. This yields estimates of heritability enrichment across functional annotations, including coding regions, promoters, enhancers, and conserved elements. Enrichment P values were Bonferroni corrected.

MR Analysis

This study was conducted following the Strengthening the Reporting of Observational Studies in Epidemiology Using MR recommendations. To assess potential causal relationships between lipidomic traits, protein traits, biomarkers, disease phenotypes, lifestyle factors, and mvASCVD, we conducted 2‐sample MR to estimate the putative causal effects of these exposure traits on mvASCVD. GWAS summary statistics for the exposures were obtained from European‐ancestry populations. For each exposure, we selected genome‐wide significant SNPs (P<5×10−8) that were LD independent (r2<0.001 within a 10 000‐kb window) as instrumental variables. We harmonized alleles across data sets and evaluated instrument strength, requiring all instruments to have F statistics >10. P values were Bonferroni corrected. Additional details are provided in Methods S1.

Lipidomics Analysis

Plasma lipidomics GWAS data from the Finnish GeneRISK cohort were used to assess lipid molecular phenotypes related to mvASCVD. The cohort comprised 7174 Finnish individuals. Shotgun lipidomics, conducted using the Lipotype GmbH platform, enabled the detection of 179 lipid molecules, which were categorized into 13 lipid classes encompassing the 4 major lipid types: glycerolipids, glycerophospholipids, sphingolipids, and sterols. Additional information on the lipid molecules and their classifications is provided in the original study. 25

Proteomics Analysis

For genes that were significant in the FUSION, MTWAS, and MAGMA analyses, we further obtained the corresponding protein pQTL GWAS summary statistics from 2 independent large‐scale proteomics studies, deCODE and Fenland, and conducted 2‐sample MR analysis. This analysis was performed to assess whether genetically predicted plasma levels of the corresponding proteins were associated with mvASCVD and to identify potential intervention targets for further evaluation. Proteins showing Bonferroni‐significant MR associations in both the deCODE discovery data set and the Fenland replication data set, with consistent effect directions, were defined as prioritized targets. The deCODE study, based on the Icelandic population (n=35 559), used the SomaScan platform to measure plasma protein levels corresponding to 4907 aptamers, which served as the discovery data set. 26 The Fenland study, conducted with European ancestry participants (n=10 708), measured 4775 proteins using SomaScan and performed genetic association analysis, serving as an independent replication data set. 27

For each prioritized target, we extracted its cis‐pQTL variants as genetic instruments, defined as variants located within ±1 Mb of the transcription start site of the corresponding protein‐coding gene. To further evaluate whether prioritized targets and mvASCVD shared causal variants at the corresponding loci, we performed colocalization analysis using coloc based on the approximate Bayes factor method. We specifically reported the posterior probability of hypothesis 3 (PPH3), posterior probability of hypothesis 4 (PPH4), PPH3+PPH4, and PPH4/PPH3. PPH3+PPH4 >0.8 was used to indicate the coexistence of mvASCVD GWAS and corresponding protein pQTL signals within the tested region; conditional on this criterion, PPH4/PPH3 ≥3 was considered supportive evidence for a shared causal variant. 28 Further details are provided in Methods S2.

Biomarkers, Diseases, and Lifestyle Analysis

To assess potential causal relationships between selected exposure traits and mvASCVD, we performed a large‐scale 2‐sample MR screening informed by the gene‐set enrichment results. This analysis focused on factors related to metabolism, inflammation, coagulation function, clinical comorbidities, and lifestyle, all of which are closely linked to the development and progression of ASCVD. Specifically, we included 34 biomarkers (spanning lipid profiles, blood pressure phenotypes, obesity and fat distribution, glucose and lipid metabolism, insulin function, inflammatory factors, blood cell counts, and coagulation/fibrinolysis markers), 19 disease phenotypes (encompassing metabolic, immune, mental health, respiratory, and renal diseases), and 15 lifestyle factors (including physical activity, smoking, alcohol consumption frequency, sleep duration, and dietary intake). Further details are provided in Table S2.

Mediation Analysis

A 2‐step MR mediation analysis was conducted to investigate whether biomarkers identified in the broad MR screening mediated the associations between prioritized targets and mvASCVD. Mediator variables were limited to biomarkers that showed significant association with mvASCVD after Bonferroni correction in the broad MR screening. First, we used cis‐pQTL variants significantly associated with the prioritized targets as instrumental variables to estimate the causal effect of genetically predicted protein levels on the mediator. Next, we estimated the causal effect of the mediator on mvASCVD. The total effect of protein on mvASCVD (βtotal) was calculated using standard univariable MR. The indirect effect (βindirect) was computed using the product method, and the direct effect (βdirect) was defined as the total effect minus the indirect effect. The mediation proportion was then determined as βindirect/βtotal.

PheWAS Analysis

We conducted a PheWAS analysis using the AstraZeneca PheWAS Portal to assess broad phenotypic associations and potential safety‐related signals related to prioritized targets. This platform uses sequencing data from UK Biobank participants of European ancestry to investigate the relationship between rare protein‐coding variants and both binary and quantitative phenotypes. We performed individual searches for each prioritized target and extracted all available association results, applying the portal's default significance threshold of P<1×10−8 for significant associations.

Virtual Screening and Molecular Docking

Virtual screening and molecular docking were performed using the DrugCLIP platform to identify candidate therapeutic compounds associated with prioritized targets for mvASCVD. DrugCLIP is a deep learning–based virtual screening method that characterizes “protein‐binding pockets—small‐molecule ligands” jointly. It incorporates molecular and pocket encoders, aligning their embedded spaces through contrastive learning. This process increases the similarity of true pocket‐ligand pairs and reduces the similarity of nonmatching pairs. DrugCLIP outperforms traditional docking methods in both speed and accuracy. 29 Protein structures were sourced from the Protein Data Bank (PDB). The compound library used in this study includes ChemDiv, Enamine Screening Collection, FCH Group Screening Collection, Mcule HTS library, Princeton BioMolecular Research, and Vitas‐M. We selected the top 100 ranked candidate molecules before clustering, with DrugCLIP scores >4 indicating high confidence. Subsequently, we performed molecular docking evaluations on the top 100 candidates, using a Vina score of <−7 as the screening threshold.

Software and Package Versions

All analyses were performed using publicly available software, including R v4.5.1 (https://www.r‐project.org/), GenomicSEM v0.0.5 (https://github.com/GenomicSEM/GenomicSEM), LDSC v1.0.1 (https://github.com/bulik/ldsc), FUMA v1.5.2 with embedded MAGMA (https://fuma.ctglab.nl/), PLINK v1.90 (https://www.cog‐genomics.org/plink/), susieR v0.14.2 (https://github.com/stephenslab/susieR), FUSION TWAS v1.4.2 (http://gusevlab.org/projects/fusion/), FOCUS v0.803 (https://github.com/bogdanlab/focus), MTWAS v0.1.0 (https://github.com/szcf‐weiya/MTWAS), coloc v5.1.0 (https://github.com/chr1swallace/coloc), TwoSampleMR v0.6.21 (https://github.com/MRCIEU/TwoSampleMR), ieugwasr v1.1.0 (https://github.com/MRCIEU/ieugwasr), and DrugCLIP (https://drugclip.com/index; version not specified).

RESULTS

Genomic SEM Construction

We used LDSC to assess the SNP heritability (h2) of the 4 input GWAS after quality control. The h2 values for the 4 phenotypes were as follows: CAD 0.118 (SE=0.007), PAD 0.131 (SE=0.016), LAS 0.086 (SE=0.031), and CAP 0.078 (SE=0.016) (Table S3). Across the 4 phenotypes, λGC values ranged from 1.029 to 1.416, mean χ2 values ranged from 1.037 to 1.745, and LDSC intercepts ranged from 0.994 to 1.083 (Table S3). Further bivariate LDSC analysis revealed significant positive genetic correlations between the 4 phenotypes (Figure 2A; Table S4), with the corresponding genetic covariances remaining significant after multiple corrections, supporting the presence of shared genetic components among these ASCVD‐related phenotypes.

Figure 2. Genetic correlations between phenotypes, confirmatory factor analysis, and mvASCVD GWAS.

Figure 2

A, Genetic correlations between ASCVD phenotypes calculated using LDSC. *The genetic covariances between the 2 phenotypes are significant after multiple testing correction. B, Path diagram of the standardized mvASCVD model estimated using genomic SEM. Standard errors are shown in parentheses. The subscript “g” denotes genetic variables, and “u” represents the residual genetic variance in each phenotype not explained by the common factor. C, Manhattan plot of the mvASCVD GWAS. The red dashed line indicates the threshold for genome‐wide significance (P=5×10−8). Labels indicate the nearest gene of the most significant lead SNP within each of the top 15 FUMA‐defined genomic loci. CAD indicates coronary artery disease; CAP, carotid plaque; FUMA, functional mapping and annotation of genome‐wide association studies; GWAS, genome‐wide association study; LAS, large‐artery atherosclerotic stroke; LDSC, linkage disequilibrium score regression; mvASCVD, multivariable atherosclerotic cardiovascular disease common factor; PAD, peripheral artery disease; rg, genetic correlation; ASCVD, atherosclerotic cardiovascular disease; SEM, structural equation modeling; and SNP, single‐nucleotide polymorphism.

Subsequently, using multivariate LDSC, we estimated the empirical genetic covariance matrix for the 4 phenotypes and applied this correlation structure to the Kaiser–Meyer–Olkin test and parallel analysis. The overall Kaiser–Meyer–Olkin value was 0.735 (CAD 0.897, PAD 0.753, LAS 0.664, and CAP 0.703), indicating good sampling adequacy for factor modeling (Table S5). Parallel analysis suggested retaining 1 potential common factor (Figure S1). In exploratory factor analysis, the single‐factor solution showed high loadings for all 4 phenotypes (CAD 0.677, PAD 0.765, LAS 0.997, and CAP 0.794), with the factor explaining 66.7% of the variance (Table S6). Further confirmatory factor analysis within the genomic SEM framework demonstrated that the single common factor model fit well: comparative fit index=1.000, standardized root mean square residual=0.048 (Table S7). The standardized loadings of this common factor for the 4 phenotypes were CAD 0.748, PAD 0.827, LAS 0.930, and CAP 0.703 (Table S7, Figure 2B), supporting the presence of a shared genetic factor across the 4 ASCVD‐related phenotypes.

Common Factor GWAS

We conducted a multivariate GWAS based on genomic SEM and obtained 4 592 424 SNP association results for the mvASCVD factor (Figure 2C), with an effective sample size of 320 561. LDSC was then applied to assess the SNP heritability and examine the source of inflation in the test statistics for this GWAS. The h2 of the mvASCVD factor SNPs was 0.122 (SE=0.007), mean χ2=1.873, and λGC=1.551. The LDSC intercept was 1.082 (SE=0.017), with a ratio of 0.094 (SE=0.019), suggesting that only a small proportion of the observed inflation in test statistics was attributable to potential confounding bias, with most attributed to polygenic genetic signals. Further assessment of QSNP heterogeneity statistics showed that, after genome‐wide Bonferroni correction (P<0.05/4 592 424), 12 SNPs showed significant QSNP heterogeneity. Among the 11 446 genome‐wide significant SNPs associated with mvASCVD, only 1 showed significant QSNP heterogeneity.

Identification of LD‐Independent Lead SNPs and Genomic Risk Loci

Using FUMA, we identified 289 LD‐independent lead SNPs that mapped to 177 genomic risk loci (Table S8). These lead SNPs were primarily located in intronic and intergenic regions (Table S8). Similarly, SNPs in LD with the independent significant SNPs were predominantly located in intronic and intergenic regions (Figure S2). An independent PLINK clumping analysis yielded 291 clump representatives, all of which mapped to the same 177 FUMA‐defined genomic risk loci. Further locus‐level comparison showed that 164 loci had a corresponding genome‐wide significant signal in at least 1 input single‐trait GWAS, whereas the remaining 13 loci had no corresponding signal in any of the 4 input GWASs (Table, Table S9). The FUMA lead SNPs corresponding to these 13 loci showed no Bonferroni‐significant QSNP heterogeneity (P<0.05/289) (Table S8).

Table 1.

mvASCVD Genomic Risk Loci and Corresponding LD‐Independent Lead SNPs

FUMA genomic risk locus (chromosome: start–end) LD‐independent lead SNP Effect allele Other allele MAF β SE P value Nearest gene
3:171 759 410–171 833 266 rs4894803 A G 0.406 −0.0161 0.0027 1.41×10−9 FNDC3B
5:122 650 224–122 680 189 rs2081914 C G 0.475 0.0137 0.0025 4.25×10−8 CEP120
6:36 614 272–36 649 401 rs73730105 C T 0.072 −0.0258 0.0047 3.43×10−8 RNU1‐88P
6:148 790 741–148 828 240 rs2294777 T C 0.100 −0.0248 0.0045 4.36×10−8 SASH1
8:10 576 753–10 732 050 rs28613656 G A 0.447 0.0158 0.0025 3.45×10−10 SOX7; PINX1
10:33 476 141–33 519 019 rs1331316 C T 0.314 −0.0146 0.0027 4.27×10−8 NRP1
11:16 481 576–16 826 436 rs4396254 G C 0.179 0.0200 0.0033 1.47×10−9 SOX6
11:54 867 322–56 279 562 rs10897179 C T 0.391 −0.0142 0.0026 4.64×10−8 OR5D3P
11:107 053 437–107 321 156 rs1388168 G A 0.496 −0.0146 0.0025 5.53×10−9 RP11‐819C21.1
12:57 511 734–57 534 912 rs11172113 T C 0.379 −0.0139 0.0025 2.78×10−8 LRP1
15:85 824 618–86 000 503 rs4843053 T G 0.486 0.0143 0.0025 1.07×10−8 AKAP13
17:65 822 573–66 096 529 rs2365468 C T 0.217 −0.0172 0.0031 2.64×10−8 BPTF
19:16 415 993–16 443 718 rs3745318 T C 0.232 0.0176 0.0031 1.10×10−8 KLF2

These mvASCVD genomic risk loci had no corresponding genome‐wide significant signal in any of the 4 input single‐trait GWASs. The LD‐independent lead SNPs were both FUMA‐defined lead SNPs and PLINK clump representatives. Effect estimates correspond to the effect allele. FUMA indicates functional mapping and annotation of genome‐wide association studies; LD, linkage disequilibrium; MAF, minor allele frequency; mvASCVD, multivariable atherosclerotic cardiovascular disease common factor; and SNP, single‐nucleotide polymorphism.

GWAS catalog annotation showed that 11 of the 13 loci had previously reported association records, whereas no relevant records were identified for the 2 loci represented by rs2294777 near SASH1 and rs4843053 near AKAP13 (Table S10). Among the 11 loci with available records, the FUMA lead SNPs in 9 loci were in high LD (r2≥0.8) with previously reported variants, whereas the 2 loci represented by rs1331316 near NRP1 and rs10897179 near OR5D3P did not meet the high‐LD threshold (Table S11).

Fine‐Mapping

We performed Bayesian fine‐mapping analysis on a ±250‐kb region surrounding the 289 identified lead SNPs. Using sum of single effects‐RSS, we identified 26 putative causal variants (PIP>0.95, located within the 95% credible set) (Table S12). Notably, rs10846744 (SCARB1, intronic) on chr12, rs11066320 (PTPN11, intronic), rs10455872 (LPA, intronic) on chr6, and rs7090277 (CDC123, intronic) on chr10 exhibited stable, high‐confidence signals. In addition, we identified several intergenic variants, including rs11813268 (OBFC1, intergenic) and rs148812085 (WDR12, intergenic). The regional association plots for these loci showed distinct peaks (Figures S3 and S4).

Transcriptome‐Wide Association Study

We performed a TWAS using FUSION based on the summary statistics of the mvASCVD GWAS to evaluate the gene‐level association between genetically predicted expression and disease risk. After correcting for multiple testing using the Bonferroni method, we identified 306 significantly associated genes (Table S13). As a supplementary analysis, we also conducted an MTWAS, identifying 677 significant genes under the same correction method (Table S14). To prioritize high‐confidence genes, we intersected the significant gene sets from both analyses, resulting in 167 significant genes (Table S14). We then performed a FOCUS fine‐mapping analysis and selected 46 genes with PIP>0.95 from the 167 genes, including SLC22A3, ADAMTS7, NBEAL1, FGD6, MAP1S, and IL6R. These genes are more likely to drive the TWAS association signals at the corresponding loci (Table S15).

Tissue‐Specific and Enrichment Analysis

We used the QTLEnrich system to evaluate the enrichment of mvASCVD GWAS association signals in expression QTL and splicing QTL loci. Overall, mvASCVD signals were broadly enriched in both types of QTL annotations. In expression QTLs, the highest enrichment was observed in whole blood, followed by tibial artery, cultured fibroblasts, transverse colon, coronary artery, and aorta (Table S16, Figure 3A). In splicing QTLs, greater enrichment was found in esophageal and colon tissues, including esophagus gastroesophageal junction, colon sigmoid, and esophagus mucosa. Significant enrichment was also observed in several brain and arterial tissues, such as brain frontal cortex BA9, brain cerebellum, tibial artery, coronary artery, and aorta (Table S17). These results suggest that mvASCVD GWAS signals are disproportionately concentrated in regulatory variants active in vascular and blood‐related tissues, with possible contributions from the gastrointestinal tract and several brain regions. Additionally, tissue‐specific analysis using MAGMA confirmed that the association signals were significantly enriched in vascular tissues, with the coronary artery, aorta, and tibial artery showing the most prominent enrichment (Figure 3B), which was consistent with the QTLEnrich results.

Figure 3. Enrichment analysis of mvASCVD GWAS signals in different tissues, cells, and pathways.

Figure 3

A, Enrichment of eQTL and sQTL signals in different tissues. B, Tissue‐specific analysis using MAGMA. C, Gene set enrichment analysis. D, Cell type enrichment analysis. *Statistically significant enrichment after Bonferroni correction. eQTL indicates expression quantitative trait locus; GO, Gene Ontology; GWAS, genome‐wide association study; MAGMA, multi‐marker analysis of genomic annotation; mvASCVD, multivariable atherosclerotic cardiovascular disease common factor; and sQTL, splicing quantitative trait locus.

Furthermore, after Bonferroni multiple comparison correction, gene‐level association analysis identified 375 significant genes (Table S18). Gene set enrichment analysis (Figure 3C) showed that the significant signals were primarily concentrated in processes related to lipid/lipoprotein metabolism and clearance, with the plasma lipoprotein assembly, remodeling, and clearance pathway being the most significant. Additionally, pathways such as statin inhibition of cholesterol production, cholesterol metabolism, and the metabolic pathway of low‐density lipoprotein, high‐density lipoprotein, and triglycerides, including diseases, also showed significant enrichment. Core genes overlapping in these metabolic pathways included PCSK9, APOB, LDLR, and APOE, among others. Beyond lipid metabolism pathways, significant enrichment was also observed in pathways related to angiopoietin receptor, transforming growth factor‐β, platelet‐derived growth factor signaling, Gene Ontology biological process terms, such as circulatory system development and response to endogenous stimulus, as well as cell types like fibroblasts and endothelial cells (Table S19). Additionally, factors related to low‐density lipoprotein cholesterol, high‐density lipoprotein cholesterol, blood pressure, and pulse pressure were also found to be significantly enriched (Table S19, Figure S5).

Cell Type and Partitioned Heritability Analysis

Cell type enrichment analysis was performed using CELLECT, and the results showed that mvASCVD heritability was significantly enriched in endothelial cell‐related annotations. This signal was consistently observed across multiple organs, including tracheal endothelial cells, adipose endothelial cells, and cardiac endothelial cells (Table S20, Figure 3D). Stratified LDSC analysis indicated that the heritability signals were primarily concentrated in regulatory‐related annotations, with the H3K27ac‐marked regions explaining a large proportion of the overall h2 (0.83–0.85). Additionally, conserved elements (Lindblad‐Toh) showed the highest relative enrichment. Other annotations, such as superenhancers, open chromatin, and coding regions, also displayed some degree of signal concentration (Table S21).

Lipidomics and mvASCVD Association Analysis

We evaluated 179 lipid species, of which 154 had valid genetic instruments and were included in the MR analysis (Tables S22 and S23). After Bonferroni correction, multiple lipid species were associated with mvASCVD risk (Figure 4A). For example, triacylglycerol (54:4) levels (odds ratio [OR]=1.10 [95% CI, 1.08–1.12]), triacylglycerol (53:4) levels (OR=1.10 [95% CI, 1.07–1.14]), triacylglycerol (52:4) levels (OR=1.10 [95% CI, 1.06–1.13]), and triacylglycerol (54:6) levels (OR=1.11 [95% CI, 1.07–1.16]) were all positively associated with increased mvASCVD risk. Additionally, sterol esters and diacylglycerols were also associated with an elevated mvASCVD risk, such as sterol ester (27:1/20:2) levels (OR=1.10 [95% CI, 1.08–1.12]) and diacylglycerol (16:0_18:1) levels (OR=1.17 [95% CI, 1.13–1.22]). In contrast, phosphatidylcholine (O‐17:0_17:1) levels and phosphatidylethanolamine (O‐16:1_18:2) levels were negatively associated with mvASCVD risk (OR=0.88 [95% CI, 0.84–0.92]; OR=0.94 [95% CI, 0.91–0.97]) (Table S22). Instrument information is provided in Table S23.

Figure 4. Lipidomics, proteomics, and mediation analysis.

Figure 4

A, Association analysis of lipidomics and mvASCVD. *Statistically significant associations after Bonferroni correction. B, Intersection of significant genes from MAGMA, FUSION, and MTWAS analyses. C, Association analysis of proteomics and mvASCVD (deCODE). D, Association analysis of proteomics and mvASCVD (Fenland). In panels C and D, horizontal dashed lines indicate the Bonferroni‐corrected significance threshold. E–H, Mediation analysis of ALDH2 (E), PDE5A (F), and TCEA2 (G, H) with mvASCVD. FUSION indicates functional summary‐based imputation; MAGMA, multi‐marker analysis of genomic annotation; MTWAS, multi‐tissue transcriptome‐wide association study; mvASCVD, multivariable atherosclerotic cardiovascular disease common factor; OR, odds ratio; and SNP, single‐nucleotide polymorphism.

Proteomics and mvASCVD Association Analysis

Among the 123 genes jointly supported by FUSION, MTWAS, and MAGMA, proteins with available cis‐pQTL instruments were evaluated by 2‐sample MR in the deCODE and Fenland data sets (Figure 4B; Tables S24–S27). Instrument information is provided in Tables S24 and S25. After Bonferroni correction, 6 significantly associated proteins were identified in the deCODE data set, including ALDH2, HTRA1, DHX58, PDE5A, STAT3, and TCEA2 (Table S26, Figure 4C). In the Fenland replication data set, 7 significantly associated proteins were identified, including SERPINH1, ALDH2, NT5C2, PDE5A, HTRA1, IL6R, and TCEA2 (Table S27, Figure 4D). Notably, ALDH2, PDE5A, HTRA1, and TCEA2 remained significantly associated with mvASCVD in both deCODE and Fenland, with consistent effect directions; ALDH2 showed a positive association with mvASCVD, whereas PDE5A, HTRA1, and TCEA2 showed negative associations. Therefore, these 4 proteins were defined as prioritized targets for further evaluation. Subsequently, we performed colocalization analysis, and all 4 targets met the locus‐level signal coexistence criterion (PPH3+PPH4 >0.8). However, the degree of shared‐variant support differed across targets. Only TCEA2 showed supportive evidence for a shared causal variant in both proteomics data sets (PPH4/PPH3=7.07 in deCODE and 3.97 in Fenland) (Table S28, Figures S6 and S7).

Biomarkers, Diseases, and Lifestyle Analysis

In the MR analysis, Bonferroni correction was applied separately to biomarkers, disease phenotypes, and lifestyle factors, with instrument information provided in Tables S29 to S31; Cochran's Q and MR‐Egger intercept tests were used as sensitivity analyses (Tables S32–S34). Among the 34 biomarkers, 9 showed significant positive associations with mvASCVD after Bonferroni correction and no evidence of directional horizontal pleiotropy based on the MR‐Egger intercept test (all P≥0.05): apolipoprotein B, triglycerides, remnant cholesterol, diastolic blood pressure, pulse pressure, body mass index, waist/hip ratio, body fat percentage, and 2‐hour glucose. High‐density lipoprotein cholesterol, systolic blood pressure, and urate levels also reached Bonferroni significance in inverse‐variance weighting MR, but showed evidence of directional horizontal pleiotropy based on the MR‐Egger intercept test (P<0.05) (Table S32). In the disease and lifestyle factor analyses, hypertension, major depression, and smoking initiation remained positively associated with mvASCVD after Bonferroni correction, with no evidence of directional horizontal pleiotropy based on the MR‐Egger intercept test (all P≥0.05) (Tables S33 and S34). The Cochran Q indicated heterogeneity for the main significant associations (all P<0.05), suggesting variability across SNP‐specific causal estimates.

Mediation Analysis

We used a 2‐step MR approach to evaluate the potential mediating roles of the 9 biomarkers associated with mvASCVD in the associations between prioritized targets and mvASCVD. First, we assessed the associations of ALDH2, PDE5A, HTRA1, and TCEA2 with these 9 biomarkers, followed by Bonferroni correction. The results showed that ALDH2 was significantly associated with multiple biomarkers: it was negatively associated with apolipoprotein B (β=−0.208) and remnant cholesterol (β=−0.225), and positively associated with diastolic blood pressure (β=4.048). Additionally, PDE5A was negatively associated with diastolic blood pressure (β=−0.636), and TCEA2 was negatively associated with triglycerides (β=−0.067) and diastolic blood pressure (β=−1.144). No significant association was observed between HTRA1 and the biomarkers included in the analysis (Table S35).

Next, we estimated the indirect effects and mediation proportions for each biomarker. The results suggested that blood pressure and lipid‐related biomarkers could explain part of the protein effects on mvASCVD: for ALDH2 and mvASCVD, the indirect effect mediated by diastolic blood pressure was significant (β=0.079, P=4.91×10−13), with a mediation proportion of 37.4% (Figure 4E); for PDE5A, the mediation proportion of diastolic blood pressure was 17.2% (Figure 4F); for TCEA2, the effects were partially mediated by diastolic blood pressure (27.1%) (Figure 4G) and triglycerides (12.9%) (Figure 4H) (Table S36).

Phenome‐Wide Association Study

To systematically assess broad phenotypic associations and potential safety‐related signals for the 4 prioritized target genes, we performed a PheWAS analysis. The results showed that HTRA1 was significantly associated with multiple phenotypes, primarily involving traits related to physical development and respiratory function (eg, height, forced vital capacity, and forced expiratory volume in the first second), with consistently negative effect directions. Additionally, it was significantly associated with certain neuroimaging phenotypes, such as white matter hyperintensity volume, suggesting that HTRA1 may exhibit pleiotropy. In contrast, the significant association for PDE5A involved plasma levels of its encoded protein (Table S37, Figure S8). No significant phenotypic associations were detected for ALDH2 or TCEA2 (Figure S8).

Virtual Screening and Molecular Docking Evaluation of Potential Therapeutic Compounds

To identify potential therapeutic compounds targeting the 4 prioritized targets (ALDH2, HTRA1, PDE5A, and TCEA2) for mvASCVD, we conducted a systematic virtual screening and molecular docking study. Among these, PDE5A (PDB: 8W4S) yielded the most high‐confidence candidate molecules (DrugCLIP score>4), with overall strong docking affinities. Top‐ranked compounds included F668‐0303 (DrugCLIP=4.15; Vina=−10.22), Z2644973547 (4.10; −9.91), and Z284132542 (4.05; −9.82) (Figure S9), with detailed compound information provided in Table S38. For ALDH2 (PDB: 8DR9), the candidate molecules had relatively lower DrugCLIP scores (1.83–2.33), but the docking results indicated strong binding tendencies. Representative molecules included Z8382539376 (2.00; −11.09), STK603391 (1.84; −10.70), and Z9241054594 (1.83; −10.34) (Table S39, Figure S9). In contrast, no candidate molecules for HTRA1 (PDB: 3NZI) or TCEA2 (PDB: 2LW4) met the docking threshold (Tables S40 and S41). These results suggest that PDE5A and ALDH2 are more likely to yield small‐molecule candidates with strong binding potential within the current compound space.

DISCUSSION

In this study, we used genomic SEM to construct a common genetic factor, mvASCVD, to evaluate the shared genetic architecture across CAD, PAD, LAS, and CAP. This common factor explained >66% of the genetic variance across the 4 phenotypes, with factor loadings ranging from 0.703 to 0.930. These findings suggest that different ASCVD‐related phenotypes share substantial genetic susceptibility and may partly reflect common biological mechanisms. This provides genomic‐level support for considering ASCVD as a systemic vascular disease with shared genetic components.

In the mvASCVD GWAS, 13 of the 177 FUMA‐defined genomic risk loci lacked a corresponding genome‐wide significant signal in any of the 4 input single‐trait GWASs. GWAS catalog annotation showed that 11 of these 13 loci had prior association records and that the FUMA lead SNPs at 9 loci were in high LD (r2≥0.8) with previously reported variants. The loci represented by rs1331316 near NRP1 and rs10897179 near OR5D3P did not meet the high‐LD threshold, whereas no relevant GWAS catalog records were identified for the loci represented by rs2294777 near SASH1 and rs4843053 near AKAP13. Previous studies support biological links between atherosclerosis and both NRP1 and SASH1. Endothelial‐specific deletion of Nrp1 increased atherosclerotic plaque area in a mouse model. 30 A previous study reported higher SASH1 expression in atherosclerotic carotid tissue from smokers than in tissue from nonsmokers; SASH1 silencing also increased endothelial cell proliferation, migration, and angiogenesis. 31 In contrast, functional evidence linking OR5D3P and AKAP13 to ASCVD remains limited.

Our cell type enrichment analysis highlights the central role of endothelial cells in mvASCVD and suggests that the genetic risk of mvASCVD is more likely to act through cross‐organ shared endothelial regulatory programs. This finding is consistent with previous genetic studies, which showed that GWAS signals for CAD converge on specific gene programs or pathways in endothelial cells and are associated with changes in endothelial barrier function and cytoskeletal remodeling. 32 Additionally, single‐cell atlases have identified endothelial subgroups with inflammatory features during atherosclerosis, providing further evidence for the widespread endothelial enrichment we observed. 33 Gene set enrichment analysis emphasized the central role of dysregulated plasma lipoprotein assembly, remodeling, and clearance in the pathogenesis of mvASCVD, involving genes such as PCSK9, LDLR, APOB, and APOE, which work synergistically in lipid homeostasis regulation and subsequently affect atherosclerosis risk. 34 Beyond lipid metabolism, this study also identified pathways involving angiopoietin receptors, platelet‐derived growth factor, and transforming growth factor‐β that are worthy of attention. Among these, Tie2, as the main endothelial angiopoietin receptor, has been genetically supported for its atheroprotective effects, 35 whereas platelet‐derived growth factor signaling is believed to be closely related to processes such as vascular wall thickening, restenosis, and fibrotic remodeling. 36 The transforming growth factor‐β axis has been associated with plaque matrix stability and matrix degradation markers. 37 Overall, these pathway‐level results implicate lipid homeostasis, endothelial receptor signaling, and growth factor–mediated vascular remodeling in the shared genetic susceptibility captured by mvASCVD.

Through MR analysis of the lipidome and mvASCVD, we revealed that multiple triacylglycerols (eg, triacylglycerol 54:4 and triacylglycerol 53:4), diacylglycerols, and sterol esters exhibited consistent positive associations with mvASCVD. Given that triacylglycerols are primarily present in triglyceride‐rich lipoproteins and their metabolically derived remnants, these subtype‐specific signals may reflect an increased burden of triglyceride‐rich lipoproteins and related remnant particles. Notably, apolipoprotein B–containing lipoprotein particles, such as very‐low‐density lipoprotein remnants and chylomicron remnants, can enter and be retained within the arterial wall; the cholesterol they carry accumulates locally, thereby promoting macrophage foam‐cell formation and accelerating plaque progression. 38 , 39 Diacylglycerols serve not only as metabolic intermediates but also as second messengers that activate signaling such as protein kinase C, which can promote atherosclerosis through endothelial dysfunction, oxidative stress, and inflammation. 40 In addition, we observed potential protective effects of certain phosphatidylcholines and phosphatidylethanolamines, which may be related to their biological roles in maintaining cell‐membrane fluidity and suppressing oxidative stress. 41

In the disease‐related MR analysis, depression showed a significant positive association with mvASCVD. Mechanistically, the relationship between depression and ASCVD likely involves several interconnected pathways. First, depression‐associated hyperactivation of the sympathetic nervous system disrupts endothelial function, promotes platelet activation, and increases thrombotic propensity. 42 Moreover, hypothalamic–pituitary–adrenal axis dysregulation in depression can elevate cortisol levels, contributing to dysregulated lipid metabolism. 43 Inflammation may also play a key role: patients with depression often exhibit increased levels of inflammatory mediators, including CRP (C‐reactive protein), interleukin‐6, tumor necrosis factor‐α, and interleukin‐1β, and interleukin‐6 and CRP have been reported as shared risk markers for depression and CAD. 44 Tumor necrosis factor‐α, interleukin‐1β, and CRP may further promote foam‐cell formation and exacerbate plaque pathology, thereby facilitating atherosclerotic progression. 45 Therefore, the role of depression in ASCVD risk deserves further attention.

By integrating TWAS, MAGMA, and protein MR analyses, this study prioritized 4 potential intervention targets: ALDH2, PDE5A, HTRA1, and TCEA2. Among them, ALDH2 is a key mitochondrial enzyme involved in the detoxification of reactive aldehydes. Although previous studies have often linked aldehyde accumulation induced by alcohol intake, ischemia, or oxidative stress to increased cardiovascular risk, aldehyde accumulation alone is insufficient to fully explain the complex interplay between aldehydes and cardiovascular disease. 46 Evidence indicates that ALDH2 not only participates in regulating platelet adhesion, 47 but also modulates the synthesis of cholesterol and high‐density lipoprotein through nonenzymatic mechanisms. 48 Meanwhile, it is closely associated with an increased risk of ischemic stroke. 49 , 50 In addition, genetic evidence at the ALDH2 locus has been linked to both CAD and hypertension. 51 HTRA1 is a structurally complex and functionally diverse serine protease. 52 Our results showed that HTRA1 levels were negatively associated with mvASCVD risk, consistent with previous findings that lower circulating HTRA1 levels were associated with increased risk of ischemic stroke and CAD. 53 For PDE5A, in the context of certain aortic diseases, downregulation or inhibition of PDE5A may excessively enhance cGMP/PKG‐mediated smooth muscle relaxation, thereby exacerbating aortic aneurysm and dissection. 54 This is directionally consistent with our findings and suggests that PDE5A inhibition may warrant caution in ASCVD‐related contexts. As a transcription elongation factor, TCEA2 may lower mvASCVD risk by regulating triglycerides and diastolic blood pressure. Notably, colocalization analysis provided supportive evidence for a shared causal variant between TCEA2 and mvASCVD.

A major strength of this study lies in the use of genomic SEM to derive a common factor from the genetic covariance among multiple ASCVD phenotypes and evaluate common‐factor association signals at the locus level. We further integrated multi‐omics data and combined MR with colocalization analyses to prioritize candidate targets; 2‐step MR was used to examine potential mediating pathways, whereas PheWAS and virtual screening were used to assess the broad phenotypic associations of these targets and screen candidate compounds, respectively.

Nevertheless, this study still has several limitations. First, participants were predominantly of European ancestry. Although this reduces population stratification and enables the use of consistent LD reference panels, it may limit the generalizability of our findings to other ancestries, particularly East Asian and admixed populations. Second, heterogeneity may exist across the contributing GWASs in phenotype definitions, diagnostic criteria, case ascertainment, and control selection, and misclassification in the original GWAS phenotypes cannot be fully excluded. Although genomic SEM can leverage the genetic covariance structure to extract cross‐phenotype shared components and thereby attenuate these differences to some extent, mvASCVD, as a latent variable, remains dependent on model specification and fitting assumptions. Therefore, loci with stronger effects in 1 input phenotype may still contribute to the common‐factor GWAS signal, and phenotype‐specific genetic effects or mechanistic signals may remain. To assess this possibility, we reported QSNP heterogeneity statistics and compared mvASCVD genomic risk loci with genome‐wide significant signals from the 4 input single‐trait GWASs at the locus level. Third, in the MR analyses of biomarkers, disease phenotypes, and lifestyle factors, although MR‐Egger intercept tests did not suggest directional horizontal pleiotropy for the main significant findings, the Cochran Q indicated heterogeneity for several significant associations. This heterogeneity may be related to differences in sample composition, measurement, and phenotype definition across the source GWASs. In addition, many lipid species and proteins had only a limited number of independent genome‐wide significant instruments, limiting power to detect pleiotropy and heterogeneity. Fourth, virtual screening and molecular docking primarily generate hypotheses about potential binding rather than direct evidence of efficacy or safety. Protein conformational dynamics, pocket plasticity, and in vivo pharmacokinetic and toxicological factors may all contribute to discrepancies between in silico predictions and true biological effects, underscoring the need for further validation.

CONCLUSIONS

In conclusion, we leveraged a multivariate GWAS framework to elucidate the shared genetic basis of ASCVD‐related phenotypes and identified lead SNPs, putative causal variants, genes, regulatory contexts, lipid traits, and protein targets associated with the mvASCVD common factor, including ALDH2, PDE5A, HTRA1, and TCEA2. These findings improve our understanding of shared genetic susceptibility to ASCVD and may inform future mechanistic and target‐focused studies.

Sources of Funding

This work was supported by the National Natural Science Foundation of China (Key Program) (No. 82230124), the National Natural Science Foundation of China for Young Scientists Fund (No. 82305221), and the High‐Level Chinese Medical Hospital Promotion Project (No. HLCMHPP2023083).

Disclosures

None.

Supporting information

Methods S1–S2

Tables S1–S41

Figures S1–S9

References 55–61

STROBE‐MR Checklist

JAH3-15-e050689-s001.pdf (181.4KB, pdf)

This manuscript was sent to Michelle H. Leppert, MD, MBA, Associate Editor, for review by expert referees, editorial decision, and final disposition.

For Sources of Funding and Disclosures, see page 15.

References

  • 1. Roth GA, Mensah GA, Johnson CO, Addolorato G, Ammirati E, Baddour LM, Barengo NC, Beaton AZ, Benjamin EJ, Benziger CP, et al. Global burden of cardiovascular diseases and risk factors, 1990‐2019: update from the GBD 2019 study. J Am Coll Cardiol. 2020;76:2982–3021. doi: 10.1016/j.jacc.2020.11.010 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2. Libby P. The changing landscape of atherosclerosis. Nature. 2021;592:524–533. doi: 10.1038/s41586-021-03392-8 [DOI] [PubMed] [Google Scholar]
  • 3. Roy P, Orecchioni M, Ley K. How the immune system shapes atherosclerosis: roles of innate and adaptive immunity. Nat Rev Immunol. 2022;22:251–265. doi: 10.1038/s41577-021-00584-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4. Soehnlein O, Libby P. Targeting inflammation in atherosclerosis ‐ from experimental insights to the clinic. Nat Rev Drug Discov. 2021;20:589–610. doi: 10.1038/s41573-021-00198-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5. Tcheandjieu C, Zhu X, Hilliard AT, Clarke SL, Napolioni V, Ma S, Lee KM, Fang H, Chen F, Lu Y, et al. Large‐scale genome‐wide association study of coronary artery disease in genetically diverse populations. Nat Med. 2022;28:1679–1692. doi: 10.1038/s41591-022-01891-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6. Malik R, Chauhan G, Traylor M, Sargurupremraj M, Okada Y, Mishra A, Rutten‐Jacobs L, Giese AK, van der Laan SW, Gretarsdottir S, et al. Multiancestry genome‐wide association study of 520,000 subjects identifies 32 loci associated with stroke and stroke subtypes. Nat Genet. 2018;50:524–537. doi: 10.1038/s41588-018-0058-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7. Grotzinger AD, Rhemtulla M, de Vlaming R, Ritchie SJ, Mallard TT, Hill WD, Ip HF, Marioni RE, McIntosh AM, Deary IJ, et al. Genomic structural equation modelling provides insights into the multivariate genetic architecture of complex traits. Nat Hum Behav. 2019;3:513–525. doi: 10.1038/s41562-019-0566-x [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8. Aragam KG, Jiang T, Goel A, Kanoni S, Wolford BN, Atri DS, Weeks EM, Wang M, Hindy G, Zhou W, et al. Discovery and systematic characterization of risk variants and genes for coronary artery disease in over a million participants. Nat Genet. 2022;54:1803–1815. doi: 10.1038/s41588-022-01233-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9. van Zuydam NR, Stiby A, Abdalla M, Austin E, Dahlström EH, McLachlan S, Vlachopoulou E, Ahlqvist E, di Liao C, Sandholm N, et al. Genome‐wide association study of peripheral artery disease. Circ Genom Precis Med. 2021;14:e002862. doi: 10.1161/CIRCGEN.119.002862 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10. Mishra A, Malik R, Hachiya T, Jürgenson T, Namba S, Posner DC, Kamanu FK, Koido M, le Grand Q, Shi M, et al. Stroke genetics informs drug discovery and risk prediction across ancestries. Nature. 2022;611:115–123. doi: 10.1038/s41586-022-05165-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11. Franceschini N, Giambartolomei C, de Vries PS, Finan C, Bis JC, Huntley RP, Lovering RC, Tajuddin SM, Winkler TW, Graff M, et al. GWAS and colocalization analyses implicate carotid intima‐media thickness and carotid plaque loci in cardiovascular outcomes. Nat Commun. 2018;9:5141. doi: 10.1038/s41467-018-07340-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12. Watanabe K, Taskesen E, van Bochoven A, Posthuma D. Functional mapping and annotation of genetic associations with FUMA. Nat Commun. 2017;8:1826. doi: 10.1038/s41467-017-01261-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13. Watanabe K, Umićević Mirkov M, de Leeuw CA, van den Heuvel MP, Posthuma D. Genetic mapping of cell type specificity for complex traits. Nat Commun. 2019;10:3222. doi: 10.1038/s41467-019-11181-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14. Buniello A, MacArthur JAL, Cerezo M, Harris LW, Hayhurst J, Malangone C, A MM, Morales J, Mountjoy E, Sollis E, et al. The NHGRI‐EBI GWAS catalog of published genome‐wide association studies, targeted arrays and summary statistics 2019. Nucleic Acids Res. 2019;47(D1):D1005–d1012. doi: 10.1093/nar/gky1120 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15. Caron B, Patin E, Rotival M, Charbit B, Albert ML, Quintana‐Murci L, Duffy D, Rausell A; the Milieu Intérieur Consortium , Abel L, et al. Integrative genetic and immune cell analysis of plasma proteins in healthy donors identifies novel associations involving primary immune deficiency genes. Genome Med. 2022;14:28. doi: 10.1186/s13073-022-01032-y [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16. Halama A, Zaghlool S, Thareja G, Kader S, Al Muftah W, Mook‐Kanamori M, Sarwath H, Mohamoud YA, Stephan N, Ameling S, et al. A roadmap to the molecular human linking multiomics with population traits and diabetes subtypes. Nat Commun. 2024;15:7111. doi: 10.1038/s41467-024-51134-x [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17. Zou Y, Carbonetto P, Wang G, Stephens M. Fine‐mapping from summary data with the “sum of single effects” model. PLoS Genet. 2022;18:e1010299. doi: 10.1371/journal.pgen.1010299 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18. Gusev A, Ko A, Shi H, Bhatia G, Chung W, Penninx BW, Penninx BWJH, Jansen R, Geus EJC, Boomsma DI, et al. Integrative approaches for large‐scale transcriptome‐wide association studies. Nat Genet. 2016;48:245–252. doi: 10.1038/ng.3506 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19. Song S, Wang L, Hou L, Liu JS. Partitioning and aggregating cross‐tissue and tissue‐specific genetic effects to identify gene‐trait associations. Nat Commun. 2024;15:5769. doi: 10.1038/s41467-024-49924-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20. Mancuso N, Freund MK, Johnson R, Shi H, Kichaev G, Gusev A, Pasaniuc B. Probabilistic fine‐mapping of transcriptome‐wide association studies. Nat Genet. 2019;51:675–682. doi: 10.1038/s41588-019-0367-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21. The GTEx consortium atlas of genetic regulatory effects across human tissues. Science. 2020;369:1318–1330. doi: 10.1126/science.aaz1776 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22. de Leeuw CA, Mooij JM, Heskes T, Posthuma D. MAGMA: generalized gene‐set analysis of GWAS data. PLoS Comput Biol. 2015;11:e1004219. doi: 10.1371/journal.pcbi.1004219 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23. Single‐cell transcriptomics of 20 mouse organs creates a tabula Muris. Nature. 2018;562:367–372. doi: 10.1038/s41586-018-0590-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24. Timshel PN, Thompson JJ, Pers TH. Genetic mapping of etiologic brain cell types for obesity. elife. 2020;9:e55851. doi: 10.7554/eLife.55851 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25. Ottensmann L, Tabassum R, Ruotsalainen SE, Gerl MJ, Klose C, Widén E; FinnGen , Simons K, Ripatti S, Pirinen M. Genome‐wide association analysis of plasma lipidome identifies 495 genetic associations. Nat Commun. 2023;14:6934. doi: 10.1038/s41467-023-42532-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26. Ferkingstad E, Sulem P, Atlason BA, Sveinbjornsson G, Magnusson MI, Styrmisdottir EL, Gunnarsdottir K, Helgason A, Oddsson A, Halldorsson BV, et al. Large‐scale integration of the plasma proteome with genetics and disease. Nat Genet. 2021;53:1712–1721. doi: 10.1038/s41588-021-00978-w [DOI] [PubMed] [Google Scholar]
  • 27. Pietzner M, Wheeler E, Carrasco‐Zanini J, Cortes A, Koprulu M, Wörheide MA, Oerton E, Cook J, Stewart ID, Kerrison ND, et al. Mapping the proteo‐genomic convergence of human diseases. Science. 2021;374:eabj1541. doi: 10.1126/science.abj1541 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28. Pan S, Kang H, Liu X, Li S, Yang P, Wu M, Yuan N, Lin S, Zheng Q, Jia P. COLOCdb: a comprehensive resource for multi‐model colocalization of complex traits. Nucleic Acids Res. 2024;52(D1):D871–d881. doi: 10.1093/nar/gkad939 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29. Jia Y, Gao B, Tan J, Zheng J, Hong X, Zhu W, Tan H, Xiao Y, Tan L, Cai H, et al. Deep contrastive learning enables genome‐wide virtual screening. Science. 2026;391:eads9530. doi: 10.1126/science.ads9530 [DOI] [PubMed] [Google Scholar]
  • 30. Bosseboeuf E, Chikh A, Chaker AB, Mitchell TP, Vignaraja D, Rajendrakumar R, Khambata RS, Nightingale TD, Mason JC, Randi AM, et al. Neuropilin‐1 interacts with VE‐cadherin and TGFBR2 to stabilize adherens junctions and prevent activation of endothelium under flow. Sci Signal. 2023;16:eabo4863. doi: 10.1126/scisignal.abo4863 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31. Weidmann H, Touat‐Hamici Z, Durand H, Mueller C, Chardonnet S, Pionneau C, Charlotte F, Janssen KP, Verdugo R, Cambien F, et al. SASH1, a new potential link between smoking and atherosclerosis. Atherosclerosis. 2015;242:571–579. doi: 10.1016/j.atherosclerosis.2015.08.013 [DOI] [PubMed] [Google Scholar]
  • 32. Schnitzler GR, Kang H, Fang S, Angom RS, Lee‐Kim VS, Ma XR, Zhou R, Zeng T, Guo K, Taylor MS, et al. Convergence of coronary artery disease genes onto endothelial cell programs. Nature. 2024;626:799–807. doi: 10.1038/s41586-024-07022-x [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33. Zhao Z, Dong Y, Zhao Z, Xiahou Z, Sun C. Single‐cell atlas of endothelial cells in atherosclerosis: identifying C1 CXCL12+ ECs as key proliferative drivers for immunological precision therapeutics in atherosclerosis. Front Immunol. 2025;16:1569988. doi: 10.3389/fimmu.2025.1569988 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34. Mehta A, Shapiro MD. Apolipoproteins in vascular biology and atherosclerotic disease. Nat Rev Cardiol. 2022;19:168–179. doi: 10.1038/s41569-021-00613-5 [DOI] [PubMed] [Google Scholar]
  • 35. Anisimov A, Fang S, Hemanthakumar KA, Örd T, van Avondt K, Chevre R, Toropainen A, Singha P, Gilani H, Nguyen SD, et al. The angiopoietin receptor Tie2 is atheroprotective in arterial endothelium. Nat Cardiovasc Res. 2023;2:307–321. doi: 10.1038/s44161-023-00224-y [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36. Folestad E, Kunath A, Wågsäter D. PDGF‐C and PDGF‐D signaling in vascular diseases and animal models. Mol Asp Med. 2018;62:1–11. doi: 10.1016/j.mam.2018.01.005 [DOI] [PubMed] [Google Scholar]
  • 37. Edsfeldt A, Singh P, Matthes F, Tengryd C, Cavalera M, Bengtsson E, Dunér P, Volkov P, Karadimou G, Gisterå A, et al. Transforming growth factor‐β2 is associated with atherosclerotic plaque stability and lower risk for cardiovascular events. Cardiovasc Res. 2023;119:2061–2073. doi: 10.1093/cvr/cvad079 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38. Ginsberg HN, Packard CJ, Chapman MJ, Borén J, Aguilar‐Salinas CA, Averna M, Ference BA, Gaudet D, Hegele RA, Kersten S, et al. Triglyceride‐rich lipoproteins and their remnants: metabolic insights, role in atherosclerotic cardiovascular disease, and emerging therapeutic strategies‐a consensus statement from the European Atherosclerosis Society. Eur Heart J. 2021;42:4791–4806. doi: 10.1093/eurheartj/ehab551 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39. Borén J, Packard CJ, Binder CJ. Apolipoprotein B‐containing lipoproteins in atherogenesis. Nat Rev Cardiol. 2025;22:399–413. doi: 10.1038/s41569-024-01111-0 [DOI] [PubMed] [Google Scholar]
  • 40. Lien CF, Chen SJ, Tsai MC, Lin CS. Potential role of protein kinase C in the pathophysiology of diabetes‐associated atherosclerosis. Front Pharmacol. 2021;12:716332. doi: 10.3389/fphar.2021.716332 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41. Dean JM, Lodhi IJ. Structural and functional roles of ether lipids. Protein Cell. 2018;9:196–206. doi: 10.1007/s13238-017-0423-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42. Lima BB, Hammadah M, Kim JH, Uphoff I, Shah A, Levantsevych O, Almuwaqqat Z, Moazzami K, Sullivan S, Ward L, et al. Association of Transient Endothelial Dysfunction Induced by mental stress with major adverse cardiovascular events in men and women with coronary artery disease. JAMA Cardiol. 2019;4:988–996. doi: 10.1001/jamacardio.2019.3252 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43. Huang L, Zhang L, Liu C, Xu Q, Qiu K. Interaction and mechanisms of depression and cardiovascular disease: a mini‐review. PeerJ. 2025;13:e20148. doi: 10.7717/peerj.20148 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44. Khandaker GM, Zuber V, Rees JMB, Carvalho L, Mason AM, Foley CN, Gkatzionis A, Jones PB, Burgess S. Shared mechanisms between coronary heart disease and depression: findings from a large UK general population‐based cohort. Mol Psychiatry. 2020;25:1477–1486. doi: 10.1038/s41380-019-0395-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 45. Kalkman HO. The association between vascular inflammation and depressive disorder. Causality, biomarkers and targeted treatment. Pharmaceuticals (Basel). 2020;13:92. doi: 10.3390/ph13050092 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 46. Qiao J, Wang Q, Zhao Y, Chang M, Sun S, Zhang P, Yao K, Chen M, Zheng L, Xing X, et al. Contribution of leukocyte telomere length to cardiovascular disease onset from genome‐wide cross‐trait analysis. Nat Commun. 2025;16:8677. doi: 10.1038/s41467-025-63707-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 47. Zhang C, Qin F, Li X, Du X, Li T. Identification of novel proteins for lacunar stroke by integrating genome‐wide association data and human brain proteomes. BMC Med. 2022;20:211. doi: 10.1186/s12916-022-02408-y [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 48. Zhang J, Guo Y, Zhao X, Pang J, Pan C, Wang J, Wei S, Yu X, Zhang C, Chen Y, et al. The role of aldehyde dehydrogenase 2 in cardiovascular disease. Nat Rev Cardiol. 2023;20:495–509. doi: 10.1038/s41569-023-00839-5 [DOI] [PubMed] [Google Scholar]
  • 49. Fan M, Li N, Huang L, Chen C, Dong X, Gao W. Exploring potential drug targets in multiple cardiovascular diseases: a study based on proteome‐wide mendelian randomization and colocalization analysis. Cardiovasc Ther. 2025;2025:5711316. doi: 10.1155/cdr/5711316 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 50. Loci associated with ischaemic stroke and its subtypes (SiGN): a genome‐wide association study. Lancet Neurol. 2016;15:174–184. doi: 10.1016/S1474-4422(15)00338-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 51. Liu Z, Xu J, Tan J, Li X, Zhang F, Ouyang W, Wang S, Huang Y, Li S, Pan X. Genetic overlap for ten cardiovascular diseases: a comprehensive gene‐centric pleiotropic association analysis and Mendelian randomization study. iScience. 2023;26:108150. doi: 10.1016/j.isci.2023.108150 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 52. Shi H, Yuan M, Cai J, Lan L, Wang Y, Wang W, Zhou J, Wang B, Yu W, Dong Z, et al. HTRA1‐driven detachment of type I collagen from endoplasmic reticulum contributes to myocardial fibrosis in dilated cardiomyopathy. J Transl Med. 2024;22:297. doi: 10.1186/s12967-024-05098-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 53. Malik R, Beaufort N, Li J, Tanaka K, Georgakis MK, He Y, Koido M, Terao C, Japan BB, Anderson CD, et al. Genetically proxied HTRA1 protease activity and circulating levels independently predict risk of ischemic stroke and coronary artery disease. Nat Cardiovasc Res. 2024;3:701–713. doi: 10.1038/s44161-024-00475-3 [DOI] [PubMed] [Google Scholar]
  • 54. Feng Y, Xue Y, Feng X, Li Z, Ren L, Guo W, Hou Y, Shu T, Zhang W, Yang Y, et al. Phosphodiesterase 5 inactivation in vascular smooth muscle cells aggravates aortic aneurysm and dissection. J Pathol. 2025;266:144–159. doi: 10.1002/path.6411 [DOI] [PubMed] [Google Scholar]
  • 55. Burgess S, Bowden J, Fall T, Ingelsson E, Thompson SG. Sensitivity analyses for robust causal inference from mendelian randomization analyses with multiple genetic variants. Epidemiology. 2017;28:30–42. doi: 10.1097/EDE.0000000000000559 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 56. Bowden J, Del Greco MF, Minelli C, Davey Smith G, Sheehan NA, Thompson JR. Assessing the suitability of summary data for two‐sample Mendelian randomization analyses using MR‐egger regression: the role of the I2 statistic. Int J Epidemiol. 2016;45:1961–1974. doi: 10.1093/ije/dyw220 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 57. Bowden J, Davey Smith G, Haycock PC, Burgess S. Consistent estimation in Mendelian randomization with some invalid instruments using a weighted median estimator. Genet Epidemiol. 2016;40:304–314. doi: 10.1002/gepi.21965 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 58. Yavorska OO, Burgess S. MendelianRandomization: an R package for performing Mendelian randomization analyses using summarized data. Int J Epidemiol. 2017;46:1734–1739. doi: 10.1093/ije/dyx034 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 59. Hartwig FP, Davey Smith G, Bowden J. Robust inference in summary data Mendelian randomization via the zero modal pleiotropy assumption. Int J Epidemiol. 2017;46:1985–1998. doi: 10.1093/ije/dyx102 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 60. Skrivankova VW, Richmond RC, Woolf BAR, Yarmolinsky J, Davies NM, Swanson SA, VanderWeele TJ, Higgins JPT, Timpson NJ, Dimou N, et al. Strengthening the reporting of observational studies in epidemiology using Mendelian randomization: the STROBE‐MR statement. JAMA. 2021;326:1614–1621. doi: 10.1001/jama.2021.18236 [DOI] [PubMed] [Google Scholar]
  • 61. Skrivankova VW, Richmond RC, Woolf BAR, Davies NM, Swanson SA, VanderWeele TJ, Timpson NJ, Higgins JPT, Dimou N, Langenberg C, et al. Strengthening the reporting of observational studies in epidemiology using mendelian randomisation (STROBE‐MR): explanation and elaboration. BMJ. 2021;375:n2233. doi: 10.1136/bmj.n2233 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Methods S1–S2

Tables S1–S41

Figures S1–S9

References 55–61

STROBE‐MR Checklist

JAH3-15-e050689-s001.pdf (181.4KB, pdf)

Articles from Journal of the American Heart Association: Cardiovascular and Cerebrovascular Disease are provided here courtesy of Wiley

RESOURCES