Multi‐Omic Insight Into the Molecular Networks in the Pathogenesis of Coronary Artery Disease

Abstract

Background

Genome‐wide association studies have revealed numerous loci associated with coronary artery disease (CAD). However, some potential causal/risk genes remain unidentified, and causal therapies are lacking.

Methods and Results

We integrated multi‐omics data from gene methylation, expression, and protein levels using summary data‐based Mendelian randomization and colocalization analysis. Candidate genes were prioritized based on protein‐level associations, colocalization probability, and links to methylation and expression. Single‐cell RNA sequencing data were used to assess differential expression in the coronary arteries of patients with CAD. TAGLN2 (Transgelin 2), APOB (Apolipoprotein B), and GIP (Glucose‐dependent insulinotropic polypeptide) were identified as the genes most strongly associated with CAD, with TAGLN2 exhibiting the most significant association. Higher methylation levels of TAGLN2 at specific Cytosine‐phosphate‐Guanine sites were negatively correlated with its gene expression and associated with a lower risk of CAD, whereas higher circulating TAGLN2 protein levels were positively associated with CAD risk (odds ratio,1.66 [95% CI, 1.32–2.08). These results suggest distinct regulatory mechanisms for TAGLN2. In contrast, APOB and GIP showed positive associations with CAD risk, whereas DHX58 (DExH‐box helicase 58) and SWAP70 (Switch‐associated protein 70) were associated with decreased risk.

Conclusions

Our findings provide multi‐omics evidence suggesting that TAGLN2, APOB, GIP, DHX58, and SWAP70 genes are associated with CAD risk. This work provides novel insights into the molecular mechanisms of CAD and highlights the potential of integrating multi‐omics data to uncover potential causal relationships that cannot be fully captured by traditional genome‐wide association studies.

Nonstandard Abbreviations and Acronyms

CARDIoGRAMplusC4D

Coronary Artery Disease Genetics Consortium

eQTLs

expression quantitative trait loci

HEIDI

heterogeneity in dependent instruments

mQTLs

methylation quantitative trait loci

MR

Mendelian randomization

PPH4

posterior probability of H4

pQTLs

protein quantitative trait loci

QTL

quantitative trait loci

SMR

summary‐data‐based Mendelian randomization

Clinical Perspective.

What Is New?

• This study integrates multi‐omics data from gene methylation, expression, and protein abundance, identifying TAGLN2, APOB, GIP, DHX58, and SWAP70 as genes associated with coronary artery disease.

What Are the Clinical Implications?

• Identifying TAGLN2, APOB, GIP, DHX58, and SWAP70 as potential risk or protective factors for coronary artery disease provides new molecular targets for personalized diagnosis and therapy.

• Understanding the protective role of TAGLN2 methylation in coronary artery disease and the risk‐enhancing roles of APOB and GIP aids in developing gene regulation‐based prevention and treatment strategies.

The incidence of coronary artery disease (CAD) is increasing annually worldwide and is the leading cause of mortality. ¹ CAD results from complex interactions between genetic predispositions, environmental factors, and personal lifestyle choices. ² Despite the availability of various treatment methods and secondary prevention therapies for CAD, a significant residual risk remains for some patients. This underscores the need for a deeper understanding of the pathophysiological mechanisms of CAD and the identification of effective therapeutic targets. ³ The availability of comprehensive human genetic data sets presents a valuable opportunity for advancing our understanding of the genetic foundations of CAD. ⁴ , ⁵ In recent years, several extensive genome‐wide association studies (GWAS) have identified numerous single‐nucleotide polymorphisms (SNPs) associated with increased risk of CAD. ⁶ , ⁷ However, GWAS often fail to clearly identify causal genes and therapeutic targets related to CAD, because the biological implications of the identified SNPs are not immediately clear. ⁸ , ⁹

During the development of CAD, the body undergoes complex functional and structural changes involving the genetic, epigenetic, and protein changes that interact extensively. ¹⁰ Methylation quantitative trait loci (mQTLs) elucidate how genetic variations shape DNA methylation, a key regulatory mechanism for gene expression affecting critical cardiovascular processes like endothelial function and lipid metabolism. ¹¹ Expression quantitative trait loci (eQTLs) link specific genetic variants with gene expression levels across different tissues. This connection helps in understanding how genetic variations can affect gene function and contribute to disease phenotypes. ¹² , ¹³ Integrating protein quantitative trait loci (pQTLs) and GWAS data can illuminate the causal role of the proteome in diseases and address the limitations of GWAS in directly identifying therapeutic targets. ¹⁴ Integrating multi‐omics data, such as eQTLs, pQTLs, and mQTLs, provides a comprehensive framework for understanding the complex molecular mechanisms of diseases like CAD.

Mendelian randomization (MR) analysis uses genetic variants as instrumental variables to investigate the causal relationships between exposures and outcomes. ¹⁵ Summary‐data‐based MR (SMR) analysis is a statistical method based on the principles of MR, similarly using genetic variants (such as SNPs) as instrumental variables to assess the relationship between exposures and outcomes. SMR analysis is particularly suitable for relevance inference between genes and complex diseases or traits, especially when direct randomized controlled trials are not feasible. ¹⁶ Compared with traditional MR analysis, SMR analysis relies on summary results from GWAS rather than individual‐level data, making it more advantageous in terms of privacy protection and data sharing. Additionally, SMR analysis can integrate multi‐omics data, helping researchers explore the potential associations between specific drug targets and diseases. The increasing availability of large‐scale GWAS and molecular quantitative trait loci (QTL) data enables us to explore the causal connections between gene regulation, from perspectives such as methylation, expression, and protein abundance, and CAD. SMR‐based strategies have already identified plasma proteins associated with CAD, colorectal cancer, and hemorrhagic stroke, facilitating the identification of potential therapeutic targets. ¹⁷ , ¹⁸ , ¹⁹ Moreover, by integrating multi‐omics evidence, the SMR method has also identified immune and inflammation‐related genes in intracranial aneurysms. ²⁰ Here, we use SMR to study the potential associations between gene methylation, expression, and protein abundance with the risk of CAD.

METHODS

This study adhered to Strengthening the Reporting of Observational studies in Epidemiology – Mendelian Randomization (STROBE‐MR) guidelines for reporting findings from observational studies using MR. ²¹ All QTL, GWAS, and single‐cell gene sequencing data were sourced from published studies, as detailed in Table S1. Because the data were derived from publicly available sources, informed consent from individual participants was not required.

In this study, we extracted extensive QTL data as instrumental variables from 3 biological levels: methylation, gene expression, and protein abundance. A 2‐stage SMR analysis (discovery and replication sets) was conducted to explore the associations between variations at different biological levels and CAD. To further validate the causal inference, colocalization analysis was conducted. By integrating the MR analysis results from the levels of methylation, gene expression, and protein abundance, we identified candidate genes with potential association. There was no overlap in populations between the exposures and outcomes.

Selection of Instrumental Variables

The mQTL, eQTL, and pQTL data used in this study were sourced from established databases, applying consistent criteria across all 3 biological levels for selecting instrumental variables. mQTL data were obtained from McRae et al's meta‐analysis, which analyzed 1980 individuals and identified >50 000 mQTLs using the Illumina Infinium HumanMethylation450 BeadChip. ²² eQTL data came from the eQTLGen Consortium, which included 31 684 individuals and 16 987 genes. ¹² pQTL data were drawn from the deCODE Genetics study, which measured 4907 proteins in 35 559 Icelanders using SomaScan ²³ ; data from Sun et al were used as a replication set to further validate findings. ²⁴ For each QTL level, cis‐SNPs within ±1000 kbp of Cytosine‐phosphate‐Guanine sites, genes, or proteins were selected as instrumental variables, with a significance threshold of P<5×10⁻⁸. SNPs without clear allele information were excluded, and redundant SNPs were removed through linkage disequilibrium clumping (r ²>0.2, 250 kbp window). The top SNP for each exposure, representing the most significant association, was chosen, and each SNP's strength was evaluated using the F statistic $\begin{array}{c}\left(R^2=\left(2\times {\mathrm{\beta }}^2\times \mathrm{MAF}\times \left(1-\mathrm{MAF}\right)\right)/\right.\hfill \end{array}$ $\begin{array}{c}\left(2\times {\mathrm{\beta }}^2\times \mathrm{MAF}\times \left(1-\mathrm{MAF}\right)+{\left(\mathrm{SE}\times \beta \right)}^2\times 2\mathrm{N}\times \mathrm{MAF}\times \left(1-\mathrm{MAF}\right)\right)\hfill \end{array}$;$\left.\mathrm{F}=\left(R^2\times \left(N-2\right)\right)/\left(1-R^2\right)\right)$ to ensure robust instrumental variables (F>10), where MAF refers to Minor Allele Frequency. ²⁵

CAD Outcome Data Sets

Summary‐level data on CAD are derived from the global genome‐wide replication and meta‐analysis by the CARDIoGRAMplusC4D (Coronary Artery Disease Genetics Consortium) as part of the discovery analysis. ⁶ This consortium has conducted a meta‐analysis of GWAS data from 48 studies, involving 60 801 cases and 123 504 controls, predominantly of European descent (77%). The diagnostic criteria for CAD included inclusive definitions such as myocardial infarction, acute coronary syndrome, chronic stable angina, or coronary artery narrowing >50%. The FinnGen study provided summary‐level genetic data for replication analysis, sourced from the latest R10 data release, including 46 959 cases and 365 222 controls. ²⁶ Diagnoses of CAD in the Finnish cohort were aligned with International Classification of Diseases, Tenth Revision (ICD‐10) definitions including unstable angina, myocardial infarction, subsequent myocardial infarction, chronic ischemic heart disease, and cardiac arrest. Another study by Harst et al also conducted a meta‐analysis of CAD GWAS data primarily from the UK Biobank, incorporating 122 733 cases and 424 528 controls, which served as additional outcome data sources for replication analysis. ²⁷ Diagnoses in this context included ICD‐10 codes I20 to I25, covering a spectrum of conditions from angina to various forms of chronic ischemic heart disease. All data used in our study are publicly available, and data sources can be found in the References.

SMR Analysis and Statistical Analysis

SMR is a sophisticated analytical method used in genetics and epidemiology to investigate associations between traits, typically between QTLs and diseases. ¹⁶ We used SMR to assess the associations between gene methylation, expression, and protein abundance on the risk with CAD, considering SNPs with SMR P values <0.05 as nominally significant. To account for multiple testing, we applied the Benjamini‐Hochberg method with a false discovery rate (FDR) threshold of 0.05, thereby reducing the risk of false positives in multiple comparisons. To distinguish between associations driven by pleiotropy and those caused by linkage, we used the heterogeneity in dependent instruments (HEIDI) test. The HEIDI test evaluates the heterogeneity in effect estimates across multiple SNPs associated with the same exposure. We applied a linkage disequilibrium threshold of 0.05 to 0.9, with HEIDI P values >0.01 indicating the absence of pleiotropy. SMR and HEIDI tests were conducted using version 1.3.1 of the SMR software via command line (https://yanglab.westlake.edu.cn/software/smr/#Overview).

Bayesian Colocalization Analysis

To refine our findings, we conducted a colocalization analysis using the coloc R package on 2 traits to estimate the posterior probabilities of shared variants. This analysis at various molecular‐level loci provided 5 mutually exclusive hypotheses on the evidence: (1) Neither trait exhibits a causal variant (H0). (2) Only gene expression is influenced by a causal variant (H1). (3) Only disease risk is influenced by a causal variant (H2). (4) Each trait is influenced by different causal variants (H3). (5) Both traits share the same causal variant (H4). For each major SNP in the GWAS database for hypertension, all SNPs within 500 kbp, 1000 kbp, and 1000 kbp upstream and downstream were retrieved in mQTL, eQTL, and pQTL, respectively, for colocalization analysis. We set the prior probabilities for SNPs associated only with the first trait (p1) at 1×10⁻⁴, only with the second trait (p2) at 1×10⁻⁴, and with both traits (p12) at 1×10⁻⁵. The posterior probability of H4 (PPH4) >0.7 was used as the cutoff threshold for strong evidence of colocalization between GWAS and QTL associations.

Integrating Results From Multi‐Omics Level of Evidence

We integrated results from methylation, gene expression, and protein abundance to explore the regulation of genes across different molecular levels and their association with CAD. Given that proteins are the executors of biological functions, we prioritized candidate genes with significant protein‐level associations (FDR <0.05 after Benjamini‐Hochberg correction) and assessed additional evidence from colocalization probabilities (PPH4) and methylation or gene expression data, specifically, (1) genes with high colocalization probabilities (PPH4 >0.7), and significant associations at both methylation and gene expression levels after Benjamini‐Hochberg correction; (2) genes with high colocalization probabilities (PPH4 >0.7), but significant associations at only 1 level (methylation or gene expression), or lacking eQTL data for SMR analysis; (3) genes with lower colocalization probabilities (PPH4 between 0.5 and 0.7), and nominal associations at methylation and expression levels (P<0.05, uncorrected). We also conducted MR analyses between gene methylation and expression, and between gene expression and protein abundance. Additional colocalization analyses were performed to validate these causal effects.

Two‐Sample MR Analysis

To further validate the robustness of the candidate gene identified through multi‐omics SMR screening, we used 2‐sample MR to assess the association between the target protein, as the functional effector, and CAD. A fixed‐effects inverse‐variance weighted approach was used as the primary method. To ensure the robustness of the causal estimates to potential violations of MR assumptions, we performed several diagnostic tests. The MR‐Egger intercept was used to assess directional horizontal pleiotropy, whereas the Cochran Q test was applied to estimate between‐SNP heterogeneity in the causal effect. Additionally, we used MR‐PRESSO (Mendelian Randomization Pleiotropy RESidual Sum and Outlier) to detect and assess outliers driven by pleiotropy and to adjust for their influence on the causal estimates when appropriate. Finally, the MR‐Steiger directionality test was conducted to evaluate whether the genetic variants are valid instruments for the exposure (protein levels), confirming that the genetic variants are strongly associated with the exposure rather than the outcome. This provides evidence supporting the assumed causal direction in this MR analysis.

Single‐Cell RNA Sequencing Analysis of Coronary Arteries

We used single‐cell RNA sequencing data from the Gene Expression Omnibus and Zenodo databases to evaluate the differential expression of target genes across various cell types and disease states in human coronary arteries (Table S1). Diseased samples were obtained from transplanted hearts (n=8), ²⁸ whereas control samples came from patients with end‐stage heart failure but no significant atherosclerotic lesions (n=5). ²⁹ Data processing and analysis were conducted using the Seurat package in R (version 4.3.2). ³⁰ Cells with <300 unique features or a mitochondrial gene percentage >10%, and hemoglobin gene percentage >5%, were excluded. Data were normalized and scaled using the NormalizeData and ScaleData functions. Dimensionality reduction was first performed using principal component analysis, and batch effects were corrected with the Harmony method. The top 25 significant principal components were selected for further analysis based on the ElbowPlot (Figure S1). Cell clustering was conducted using the FindNeighbors and FindClusters functions, with a resolution of 0.3 selected based on clustering tree analysis using the Clustree tool (Figure S2). Cells were visualized in 2 dimensions using uniform manifold approximation and projection and t‐distributed stochastic neighbor embedding, both of which were performed after principal component analysis (Figure S3). These visualization methods enabled clustering of cells with similar features and further validation of the relationships between cell clusters. Differential expression analysis between disease and control groups was performed using the Wilcoxon rank sum test. Genes with an average log2 fold change >0.25 and an adjusted P value <0.05 were considered differentially expressed.

Reverse Transcriptase–Quantitative Polymerase Chain Reaction and Western Blot Analysis

To investigate the vascular expression of the identified genes/proteins in arteries with varying atherosclerotic states, we used an apolipoprotein E‐deficient mouse model of atherosclerosis. Eight‐week‐old male apolipoprotein E‐deficient mice (n=30) were randomly divided into 2 groups: 15 mice were fed a high‐fat diet for 12 weeks to induce atherosclerosis, whereas the other 15 mice were fed a normal diet as the control group. The success of the atherosclerosis model was confirmed through Oil Red O staining of both the aorta and aortic valve tissues. The mRNA expression levels of the identified genes in aortic tissues were analyzed using reverse transcriptase–quantitative polymerase chain reaction. Total RNA was extracted from aortic tissues using TRIzol Reagent (Life Technologies), followed by reverse transcription into complementary DNA. Reverse transcriptase–quantitative polymerase chain reaction was performed with gene‐specific primers, and the relative mRNA expression levels were calculated using the 2^−ΔΔCT method, with β‐actin serving as the internal control. The sequences of the primers are listed in Table S2. The protein expression levels of DHX58 (DExH‐box helicase 58), SWAP70 (Switch‐associated protein 70), and TAGLN2 (Transgelin 2) in aortic tissues were evaluated using Western blot. Total protein was extracted from aortic tissues using standard protocols, and equal amounts of protein were separated by SDS‐PAGE, followed by transfer to polyvinylidene fluoride membranes. Membranes were incubated with the following primary antibodies: DHX58 (1:500, Proteintech, 10 494‐1‐AP), SWAP70 (1:1000, Abclonal, A14857), TAGLN2 (1:2000, Proteintech, 10 234‐2‐AP), and GAPDH (1:20000, Proteintech, 10 494‐1‐AP), which served as the loading control. An enhanced chemiluminescence kit (Boster Biological Technology, Wuhan, China) was used to visualize the protein, and Image J software (National Institutes of Health, Bethesda, MD) was used to quantify band intensity. In each group, 3 mice were used for Oil Red O staining of the aorta and aortic valves, 6 mice were used for RNA extraction and reverse transcriptase–quantitative polymerase chain reaction, and 6 mice were used for protein extraction and Western blot. All animal protocols were approved by the Animal Research Ethics Committee of Chongqing Medical University.

RESULTS

SMR Analysis of Genome‐Wide Cis‐mQTLs and CAD

A total of 100 440 methylation probe sites mapped with 15 494 genes were included. After conducting SMR and HEIDI tests, we identified 6975 methylation sites associated with the expression of 3061 independent genes that are significantly related to CAD. To control for type I errors across the genome, multiple corrections were made, resulting in 149 methylation sites near 74 genes. Further colocalization analysis was performed to exclude confounding factors caused by linkage disequilibrium. Among the identified signals, 59 methylation sites showed strong colocalization evidence (PPH4 >0.7), supporting an association with CAD, related to the expression changes of 26 genes. Figure 1 displays the methylation sites with top significance in P values and their corresponding genes, including MORF4L1 (cg15571903, cg00540400), LIPA (cg12555086), ICA1L (cg13521797), and ADAMTS7P3 (cg04896959). Detailed results are presented in Table S3. Causal estimates were expressed using odds ratios (ORs); for example, an increase of 1 SD in methylation at the site cg16107628 associated with the TAGLN2 gene correlates with a decreased risk of CAD (OR, 0.91 [95% CI, 0.88–0.95]; P=5.93e‐6). Among the 149 significant methylation sites identified in the discovery set, 109 were replicated as significantly associated with CAD in the Finnish study and 113 in another meta‐analysis. Notably, although not all sites reached statistical significance in the replication sets, the direction of the β values was consistent with those found in the discovery set (Table S4).

Figure 1. Associations of genetically predicted gene methylation with coronary artery disease in summary‐based Mendelian randomization analysis.

OR indicates odds ratio; and PPH4, posterior probability of H4.

SMR Analysis of Genome‐Wide Cis‐eQTLs and CAD

In our analysis of the relationship between gene expression and CAD, we applied FDR correction and HEIDI tests, ultimately identifying 56 genes significantly associated with CAD out of the 15 619 genes analyzed. After colocalization filtering, 38 unique genetic loci were found to have associations with CAD (Table S5). Figure 2 displays the top 15 genes ranked by P value significance. Additionally, we observed that some loci identified at the gene expression level also showed significant associations with CAD risk at the methylation level. Eleven such genes were identified: TAGLN2 (OR, 1.21 [95% CI, 1.12–1.31]; P=3.19e‐6), LIPA (OR, 1.08 [95% CI, 1.05–1.1]; P=5.48e‐12), HOXC4 (OR, 0.71 [95% CI, 0.61–0.83]; P=1.48e‐5), FES (OR, 0.86 [95% CI, 0.82–0.91]; P=1.61e‐7), VAMP8 (OR, 0.88 [95% CI, 0.85–0.92]; P=1.13e‐9), IPO9 (OR, 0.76 [95% CI, 0.66–0.87]; P=7.66e‐5), CNNM2 (OR, 2.44 [95% CI, 1.58–3.78]; P=6.06e‐5), BCAR1 (OR, 0.77 [95% CI, 0.69–0.86]; P=7.72e‐6), TSPAN14 (OR, 0.82 [95% CI, 0.74–0.9]; P=2.04e‐5), SF3A3 (OR, 0.89 [95% CI, 0.84–0.94]; P=1.58e‐5), and SMARCA4 (OR, 1.38 [95% CI, 1.22–1.56]; P=3.79e‐7). Following multiple corrections, 23 genes were replicated in a Finnish study and 29 in another meta‐analysis (Table S6), with 18 genes replicated in both data sets (B9D2, BCAR1, CCDC97, FES, HOXC4, LIPA, LPL, MAPKAPK5‐AS1, PSRC1, RNU6‐510P, SF3A3, SH2B3, SMARCA4, TAF1A, TMEM150A, UTP11, and VAMP8).

Figure 2. Associations of genetically predicted gene expression with coronary artery disease in summary‐based Mendelian randomization analysis.

OR indicates odds ratio; and PPH4, posterior probability of H4.

SMR Analysis of Genome‐Wide Cis‐pQTLs and CAD

A total of 1743 proteins were included in the SMR analysis. For the association between circulation proteins and CAD, 11 convincing loci were identified after multiple adjustments (Figure 3, Table S7). Colocalization analysis identified GIP (Glucose‐dependent Insulinotropic Polypeptide; OR, 2.54 [95% CI, 1.52–4.24]; P=3.85e‐4), APOF (Apolipoprotein F; OR, 1.28 [95% CI, 1.13–1.45]; P=1.33e‐4), APOC3 (Apolipoprotein C3; OR, 1.27, 95% CI, 1.13–1.43]; P=8.8e‐5), APOB (Apolipoprotein B; OR, 2.19 [95% CI, 1.54–3.11]; P=1.4e‐5), and TAGLN2 (Transgelin 2; OR, 1.66 [95% CI, 1.32–2.08]; P=1.39e‐5) sharing causal variants with increased risk for CAD. In the replication set, using pQTL data from Sun et al as exposures, 5 proteins (IL6R [Interleukin 6 Receptor], MST1 [Macrophage Stimulating 1], APOF, FABP2[Fatty Acid Binding Protein 2], SWAP70 [Switch‐associated Protein 70]) were validated. Although GIP did not reach significance in the replication set, the direction of its positive association with CAD was consistent with that observed in the discovery set (Table S8). In the Finnish study, the associations between IL6R and APOC3 with CAD were replicated (corrected P value <0.05). Nine proteins were also replicated in the research conducted by Harst et al (Table S8).

Figure 3. Associations of genetically predicted gene encoded protein with coronary artery disease in summary‐based Mendelian randomization analysis.

OR indicates odds ratio; and PPH4, posterior probability of H4.

Integrating Evidence From Multi‐Omics Levels

Through the integration of multi‐omics data, TAGLN2 was identified as a gene supported by robust multi‐omics evidence (Figure S4), including significant protein‐level associations (FDR <0.05), high colocalization probabilities (PPH4 >0.7), and consistent associations at both methylation and gene expression levels (FDR <0.05). As such, TAGLN2 is prioritized as a high‐confidence gene associated with CAD. For APOB and GIP, significant protein‐level associations (FDR <0.05) and high colocalization probabilities (PPH4>0.7) were observed, but gene expression data were unavailable for these genes. Consequently, their prioritization was based on evidence from protein abundance and methylation levels (Figures S5 and S6). DHX58 and SWAP70 showed protein‐level associations with CAD (FDR <0.05) and colocalization probabilities between 0.5 and 0.7, with nominally significant associations (P<0.05, unadjusted) at the methylation and expression levels. These genes were considered to have moderate evidence supporting their association with CAD (Table 1, Figures S7 and S8). Although not all associations achieved statistical significance in replication analyses, our primary concern was whether they increased or decreased the risk of CAD, specifically if the direction of β values align with the discovery set. The direction of most associations' β values in the replication set was consistent with that of the discovery set (Table S9).

Table 1.

Genetically Predicted Methylation, Expression, and Protein of Candidate Gene With Coronary Artery Disease in Summary‐Data‐Based Mendelian Randomization Analysis

Gene	Probe	Per SD increase* in methylation	P † for mQTL	Per SD increase in gene levels	P value for eQTL	Per SD increase in protein levels	P value for pQTL	PPH4
TAGLN2	cg13892570	0.92	5.0E‐6	1.21	3.2E‐6	1.66	1.4E‐5	0.99
	cg15641364	0.96	2.6E‐6
	cg16107628	0.91	5.9E‐6
	cg22339338	0.94	3.6E‐6
APOB	cg00673290	1.11	1.2E‐7	…	…	2.19	1.4E‐5	0.94
	cg16723488	1.11	1.7E‐7
	cg25035485	1.1	1.4E‐7
GIP	cg27201301	1.07	3.8E‐5	…	…	2.54	3.9E‐4	0.77
DHX58	cg20291162	0.97	2.0E‐4	0.92	4.6E‐4	0.81	2.5E‐4	0.54
SWAP70	cg03108697	0.9	2.5E‐3	0.84	3.4E‐7	0.89	1.3E‐6	0.63
	cg10649130	0.96	6.4E‐3

eQTL indicates expression quantitative trait loci; mQTL, methylation quantitative trait loci; PPH4, posterior probability of H4; pQTL, protein quantitative trait loci.

^*Per SD increase represents the odds ratio for coronary artery disease per SD increase in the respective trait (methylation, gene expression, or protein levels).

^† P values refer to the P values from summary‐data‐based Mendelian randomization analyses, testing the genetically predicted effects of mQTL, eQTL, and pQTL on coronary artery disease.

Epigenetic regulation and the central dogma indicate that methylation, as an epigenetic mechanism, can alter gene expression levels, subsequently affecting protein levels and functionality. Therefore, we used SMR analysis to investigate the causal relationships between gene methylation and expression, and between gene expression and circulating protein levels. Consistent with the analyses above, methylation at the TAGLN2 loci cg22339338, cg15641364, cg13892570, and cg16107628 showed a negative correlation with gene expression, which correlates with a decreased risk of CAD at these sites (Figure 4). Similarly, methylation is associated with higher expression levels of DHX58 and SWAP70, aligning with the protective effects of upregulated gene expression and methylation against CAD (Table S10). Furthermore, colocalization analysis revealed strong evidence of colocalization between mQTL, eQTL, and pQTL within TAGLN2.

Figure 4. A chromosome map of TAGLN2 illustrates the effect values of SNPs within 500 kbp upstream and downstream regions, identified as eQTLs and mQTLs, in the context of coronary artery disease GWAS data.

The x axis denotes the chromosomal position coordinates, whereas the y axis represents the negative logarithm of P values (including eQTLs, mQTLs, GWAS data, and SMR). The red dashed line indicates the P selection threshold of 1e‐5 for eQTLs analyzed with SMR, with genes exhibiting P<1e‐5 represented by red diamonds; the blue dashed line represents the P selection threshold of 0.001 for mQTLs analyzed with SMR, with methylated sites displaying P<0.001 denoted by blue solid circles. eQTL indicates expression quantitative trait loci; GWAS, genome‐wide association study; mQTL, methylation quantitative trait loci; SMR, summary‐data‐based Mendelian randomization; and SNPs, single‐nucleotide polymorphisms.

MR Analysis of Identified Proteins and CAD

Given that proteins are the ultimate executors of biological function, we conducted a 2‐sample MR analysis to assess the causal relationship between 5 proteins (TAGLN2, APOB, DHX58, SWAP70, GIP) and CAD, further validating the robustness of our results. All instrumental variables included in the MR analysis had F statistics >10 (Table S11), ensuring sufficient instrument strength. The MR results demonstrated significant associations between these 5 proteins and CAD, with effect directions consistent with those observed in the SMR analysis (Figure 5, Figure S9). The MR‐Egger intercept test indicated no evidence of horizontal pleiotropy for any of the proteins. However, MR‐Egger regression and Cochran Q test suggested potential heterogeneity in the instrumental variables for APOB (Table S12). Further analysis using MR‐PRESSO, which corrected for outlier SNPs, revealed that the association between APOB and CAD remained significant (P=1.17e‐7), underscoring the robustness of this result (Table S13). No evidence of horizontal pleiotropy or heterogeneity was found for TAGLN2, DHX58, SWAP70, or GIP. Additionally, MR Steiger directionality tests did not identify any genetic variants as potential outliers (Table S13).

Figure 5. Associations of genetically predicted gene encoded protein with coronary artery disease in 2‐sample Mendelian randomization analysis.

OR indicates odds ratio.

Cell‐Specific Expression of Identified Genes in Atherosclerotic Plaque Tissue

To explore the cell‐specific enrichment of the 5 identified genes in atherosclerotic plaques, we performed a single‐cell expression analysis using publicly available single‐cell RNA sequencing data. Uniform manifold approximation and projection (Figure 6A) and t‐distributed stochastic neighbor embedding (Figure S10) were used to visualize the distribution of cell clusters, identifying 17 clusters that were annotated into 8 cell types: fibroblasts, macrophages, smooth muscle cells, T/NK (Natural Killer) cells, endothelial cells, B cells, plasma cells, and neurons. GIP showed no detectable expression across any cell type, whereas TAGLN2, APOB, DHX58, and SWAP70 exhibited no significant cell‐specific enrichment (Figure 6B and 6C).

Figure 6. Single‐cell type expression in atherosclerotic plaque for the 5 genes identified by Mendelian randomization.

A, A total of 17 cell clusters and 8 cell types based on UMAP visualization. B through C, Expression patterns of 4 genes (TAGLN2, SWAP70, DHX58, and APOB) across cell‐type clusters visualized using UMAP and dot plots. D, Differential expression of 3 genes (TAGLN2, SWAP70, and DHX58) between atherosclerotic plaques and controls, shown in 8 cell types at average Log2FC >0.25 and FDR <0.05 level. FDR indicates false discovery rate; Log2FC, log2 fold change; SMC, smooth muscle cells; T/NK, T or Natural Killer cells and UMAP, uniform manifold approximation and projection.

Next, we conducted differential expression analysis between atherosclerotic and control groups. TAGLN2 was significantly upregulated in macrophages and fibroblasts in atherosclerotic plaques, whereas its expression was significantly downregulated in B cells and T/NK cells (Figure 6D). SWAP70 was upregulated in endothelial and dendritic cells, and DHX58 was upregulated in smooth muscle cells. No significant differential expression was observed for APOB (Table S14).

Expression of Identified Genes in the Aortas of Atherosclerotic Mice

Oil Red O staining of the aorta and aortic valve confirmed the formation of atherosclerotic plaques, whereas Masson's trichrome staining revealed more pronounced fibrosis in the aortic valves of mice in the atherosclerosis group compared with controls (Figure 7A). Western blot showed that DHX58 expression was significantly downregulated in the aortas of atherosclerotic mice, whereas TAGLN2 expression was also markedly reduced (Figure 7B and 7C). Although SWAP70 exhibited a trend toward downregulation, the change was not statistically significant. Consistent with protein expression, reverse transcriptase–quantitative polymerase chain reaction analysis showed that the mRNA levels of SWAP70 were downregulated, whereas those of TAGLN2 were upregulated in atherosclerotic aortas, both consistent with their respective protein expression trends (Figure 7D). In contrast, DHX58 mRNA levels showed an upward trend, which was inconsistent with its protein expression but did not reach statistical significance.

Figure 7. Validation of gene and protein expression in the aortas of AS and control mice.

A, Sequential panels show Oil Red O staining of the aorta, Oil Red O staining of the aortic valve, hematoxylin and eosin staining of the aortic valve, and Masson's trichrome staining of the aortic valve. B, Western blot DHX58, SWAP70, and TAGLN2 protein expression in the aortic tissues of AS and control mice, with GAPDH used as a loading control. C, Quantification of DHX58, SWAP70, and TAGLN2 protein levels in the aortic tissues of AS and control mice. Data are presented as mean±SE (n=6 per group, analyzed using the Mann‐Whitney U test). D, Quantification of DHX58, SWAP70, and TAGLN2 mRNA levels in the aortic tissues of AS and control mice. Data are presented as mean±SE (n=6 per group, analyzed using the Mann‐Whitney U test). AS indicates atherosclerotic. KD, kiloDalton.

DISCUSSION

In this study, we explored the genetic susceptibility to CAD from the perspectives of epigenetics, whole‐genome, and whole‐proteome analyses, and identified 5 genes related to CAD: TAGLN2, APOB, GIP, DHX58, and SWAP70. Validation using an atherosclerotic mouse model further supported the involvement of these genes, with consistent expression changes observed at both the mRNA and protein levels. Our study provides multi‐omics evidence for the genetic susceptibility to CAD, offering new insights into the potential mechanisms and therapeutic targets involving genetic loci, gene expression, and methylation.

Our study demonstrates that TAGLN2 is associated with an increased risk of CAD, and its expression is significantly upregulated in macrophages and fibroblasts within atherosclerotic plaques. Macrophages, the primary component of foam cells, and fibroblasts, which are observed within atherosclerotic plaques, contribute to plaque stability by synthesizing and depositing collagen fibers, and influence local inflammation through the secretion of cytokines and chemokines. ³¹ , ³² A previous publication has suggested the important role of TAGLN2 in macrophage phagocytosis ³³ in atherosclerosis. More studies are needed to investigate the role of TAGLN2 in macrophages and fibroblasts. More studies are needed to investigate the role of TAGLN2 in macrophages and fibroblasts. Additionally, the downregulation of TAGLN2 in B cells and T/NK cells, key players in atherosclerosis, might contribute to impaired cellular function, further promoting disease progression. ³⁴ , ³⁵ Overall, speculation exists that TAGLN2 plays a broad role in shaping the immune microenvironment within plaques, thereby contributing to the pathogenesis of atherosclerosis and increasing CAD risk. However, further research is required to confirm the precise role of TAGLN2 in the pathological process of CAD.

APOB plays a critical role in CAD through lipid metabolism, and various therapies targeting APOB have been shown to significantly reduce the incidence of cardiovascular events in patients with coronary heart disease. ³⁶ , ³⁷ Our study further demonstrates that APOB exerts a significant influence on CAD risk at different molecular levels, providing additional validation for the reliability of our findings. However, single‐cell sequencing analysis revealed no differential expression of APOB in atherosclerotic plaques, which may be explained by the fact that APOB's primary site of metabolism is the liver. ³⁶ Current research on the impact of GIP on cardiovascular disease risk is inconsistent. One study links higher circulating GIP levels to increased all‐cause mortality and cardiovascular diseases. ³⁸ However, another study associates lower circulating GIP levels with worse outcomes in acute myocardial infarction. ³⁹ Furthermore, tirzepatide, a dural GIP and glucagon‐like peptide‐1 receptor agonist, has shown mixed cardiovascular effects in type 2 diabetes. ⁴⁰ , ⁴¹ Despite these findings, our research provides new evidence that GIP may increase the risk of CAD, but its specific effects and mechanisms require further investigation.

In our study, DHX58 and SWAP70 were found to be negatively associated with CAD risk. However, single‐cell analysis revealed that DHX58 is upregulated in smooth muscle cells within atherosclerotic plaques, whereas SWAP70 is upregulated in endothelial cells and macrophages, suggesting distinct roles in different cell types. Previous research indicates that the activation of DHX58 can regulate ferroptosis in hepatocytes. ⁴² We speculate that DHX58 may promote the acquisition of a macrophage‐like phenotype in smooth muscle cells through the regulation of ferroptosis, contributing to foam cell formation and plaque vulnerability. ⁴³ Additionally, SWAP70 may be involved in the maturation and differentiation of immune cells, inhibiting lipid deposition and inflammation, thereby exerting a protective effect within atherosclerotic plaques. ⁴⁴ , ⁴⁵

The main strength of this study lies in the integration of multi‐omics data and the use of SMR to identify associations between different molecular levels (protein, gene, methylation) and CAD, along with multiple sensitivity analyses to confirm the robustness of the results. However, this study has several limitations that may affect the interpretation and generalizability of the findings. First, because the QTL effects are derived from population‐based samples of unaffected individuals (ie, healthy individuals) rather than directly from case–control studies, SMR cannot conclusively determine causality. Therefore, in interpreting our results, we consider the identified relationships as associations with CAD rather than direct causal relationships. Second, the SMR analysis included only the most significant top SNPs, which may influence pleiotropy analysis and heterogeneity testing. Nevertheless, we confirmed the strength of the instrumental variables (F>10) through F‐statistic calculations, indicating minimal bias from weak instruments. Additionally, the HEIDI test was used to identify and exclude false‐positive associations due to horizontal pleiotropy and linkage disequilibrium. To further strengthen causal inference, we validated the identified probes using external data and conducted 2‐sample MR analysis of the identified proteins and CAD. MR‐PRESSO analysis and Steiger tests were performed to confirm pleiotropy, heterogeneity, and directionality. These methods are valuable tools for addressing potential violations of MR assumptions; however, their statistical power may be limited in detecting subtle pleiotropy or in the presence of weak instruments. For instance, phenotype scanning (phenosanner website) results revealed that SWAP70 (rs415895) is also associated with reticulocyte count, mean corpuscular volume, and hypertension; APOB (rs563290) with hematocrit and sphingomyelin; and TAGLN2 (rs2789422) with mean platelet volume. Although colocalization analysis confirmed that the identified proteins and CAD share causal variants, we cannot entirely exclude the possibility that confounding factors such as hypertension and mean corpuscular volume may influence the associations between the identified proteins and CAD. Therefore, further studies are required to confirm the causal relationships between the identified proteins and CAD. In addition, although the animal experiments provide additional biological evidence supporting the identified genes, it is important to note that findings in the mouse model may not fully capture the complexity of human atherosclerosis. Further studies are warranted to validate these findings in clinical settings. Moreover, functional studies and mechanistic investigations are needed to further explore the roles and underlying mechanisms of these genes in atherosclerosis and coronary artery disease. Last, our study population primarily consisted of individuals of European ancestry, which limits the generalizability of our findings to other populations. Moreover, despite including all available QTL data, some genes lacked comprehensive multi‐omics data, and the available QTL data sets did not include genetic variants associated with gene expression or methylation levels on the X chromosome, Y chromosome, or mitochondrial genome, thus limiting the scope of our comprehensive assessment.

In individuals of European ancestry, this MR study supports the hypothesis that genetically predicted TAGLN2, APOB, and GIP are associated with an increased risk of CAD, whereas DHX58 and SWAP70 are associated with a reduced risk of CAD. However, further research is needed to confirm the direct causal relationships and to elucidate the underlying biological mechanisms behind these associations.

Sources of Funding

The work was supported by Major Scientific Instrument Development Project of the National Natural Science Foundation of China [grant number 32127802], National Natural Science Foundation of China [grant number 82170445] and the National Major Scientific Instruments and Equipments Development Project of National Natural Science Foundation of China [grant number 2015BAI01B00].

Disclosures

None.

Supporting information

Tables S1–S14

Figures S1–S10