Shared Genetic Liability Between Heart Failure and Myocardial Infarction Revealed by Genome-Wide Cross-Trait Analysis

Abstract

Background

The extent and biological relevance of shared genetic architecture between myocardial infarction (MI) and heart failure (HF) remain incompletely understood.

Methods

We analyzed large-scale European-ancestry genome-wide association studies summary statistics for MI and HF. Genome-wide genetic correlation was estimated using linkage disequilibrium score regression, and polygenic overlap was quantified using MiXeR. Shared loci were identified via conditional and conjunctional false discovery rate (condFDR/conjFDR) approaches. Functional prioritization incorporated Functional Mapping and Annotation-based annotation, Bayesian fine-mapping, transcriptome-wide association studies (TWAS), FOCUS gene fine-mapping, and summary-level Mendelian randomization (SMR) integrating UKB-PPP proteomic data.

Results

Linkage disequilibrium score regression revealed a robust positive genetic correlation between MI and HF (rg = 0.494, P = 1.12 × 10⁻¹⁵). MiXeR demonstrated substantial polygenic overlap, with approximately 90% of MI-associated variants shared with HF and strong concordance in effect direction. The cond/conjFDR analyses identified multiple pleiotropic loci, including novel HF-associated regions. Fine-mapping prioritized rs544366796 within the SLC22A2/SLC22A3 locus as a high-confidence candidate variant for MI based on posterior probability. The TWAS and FOCUS highlighted canonical MI genes (CDKN2B, CELSR2, BRAP, NBEAL1) and identified MYOZ1 as an HF-specific candidate gene. Proteome-wide SMR analysis provided statistical evidence consistent with apolipoprotein E being a shared protein influenced by variants associated with both MI and HF.

Conclusion

The MI and HF share substantial genetic liability characterized by strong polygenic overlap and pleiotropic loci. Our integrative analyses suggest a potential 2-stage genetic framework linking ischemic susceptibility to myocardial remodeling and HF progression, which should be interpreted as a hypothesis-generating conceptual model rather than direct evidence of temporal progression.

Introduction

Myocardial infarction (MI) and heart failure (HF) constitute 2 major global health burdens and represent tightly interconnected stages along the continuum of cardiovascular disease progression. ¹ Myocardial infarction, characterized by irreversible myocardial necrosis resulting from ischemia, frequently precipitates adverse ventricular remodeling and ultimately progresses to HF, a clinical syndrome marked by impaired cardiac function and high mortality.^2,3 Despite substantial advances in reperfusion strategies and pharmacological interventions, the transition from MI to HF remains incompletely understood. Pronounced interindividual variability in disease progression suggests a substantial genetic contribution, underscoring the necessity for systematic identification of genetic determinants regulating susceptibility to MI and HF.^4,5

Epidemiological and clinical evidence indicate that genetic factors substantially influence susceptibility to MI and HF.^6,7 Heritability estimates suggest that HF is moderately heritable, with genetic factors accounting for approximately 15% to 35% of disease risk, whereas MI exhibits higher heritability, with estimates ranging from 50% to 60% across studies and populations. Advances in genome-wide association studies (GWAS) have markedly accelerated the identification of common genetic determinants of cardiovascular disease.^8,9 Large-scale GWAS have identified numerous MI-associated loci, enriched in genes regulating lipid metabolism, vascular inflammation, and endothelial homeostasis, including LPA, SORT1, and PCSK9.^10,11 In contrast, GWAS of HF have revealed a distinct yet partially overlapping set of susceptibility loci, implicating pathways related to myocardial contractile function, calcium handling, and extracellular matrix remodeling.^7,12,13 Collectively, these findings indicate that MI and HF share partially overlapping genetic regulatory architectures, suggesting that their pathogenesis may involve common molecular pathways and biological mechanisms.

Although large-scale GWAS have yielded substantial insights into the genetic architecture of individual cardiovascular diseases, the extent and functional relevance of genetic determinants shared between MI and HF remain incompletely defined. Cross-trait genetic correlation analyses, including linkage disequilibrium (LD) score regression, have demonstrated significant positive genetic correlations across cardiometabolic traits, indicating a partially overlapping polygenic basis. ¹⁴ In line with these findings, accumulating evidence indicates that certain genetic loci exert pleiotropic effects linking atherosclerotic mechanisms with cardiomyocyte structural and metabolic dysfunction. ¹⁵ To more precisely delineate cross-phenotype genetic architecture and identify variants with shared biological effects, recent studies have increasingly adopted integrative analytical strategies, such as multitrait association models and conditional or conjunctional false discovery rate (condFDR/conjFDR) approaches. By exploiting genetic correlations across related phenotypes, these methods improve power to detect shared susceptibility loci and enable deeper insight into convergent molecular pathways underpinning MI and HF pathophysiology. ¹⁶

However, the distribution of shared genetic signals across MI and HF, along with their connection to related biological processes, remains incompletely elucidated. This study integrates a post-GWAS analysis framework to systematically understand the shared genetic architecture of MI and HF, identify polygenic sites and genes, and visualize findings through annotation and pathway enrichment analysis. Collectively, this research provides an integrated genetic perspective on the association between MI and HF.

Method

Data Sources and Quality Control

An overview of the analytical workflow is provided in Figure 1. To minimize bias arising from population stratification and the heterogeneity of LD patterns across ancestries, and because sufficiently powered multiancestry datasets remain limited, we restricted all primary analyses to individuals of European descent. Myocardial infarction GWAS data were derived from the Neale Lab cohort comprising 361,194 participants (7018 cases and 354,176 controls). Case definitions were based on hospital episode statistics and curated diagnostic codes within the UK Biobank framework. The HF GWAS data were obtained from FinnGen (500,348 individuals; 37,653 cases and 462,695 controls), where case ascertainment was based on nationwide registry-linked diagnostic codes within the Finnish healthcare system. ¹⁷ Protein quantitative trait locus (pQTL) information was accessed through the UK Biobank-PPP initiative. For further details, please refer to Supplemental Table S1.

Figure 1.

Flowchart of this study.

Before downstream analyses, all GWAS summary statistics were subjected to stringent quality-control procedures, including alignment to the hg38 reference genome using Phase 3 European data from the 1000 Genomes Project and removal of rare variants with minor allele frequencies below 1%.

All studies contributing to the GWAS datasets received approval from their respective institutional ethics committees, with written informed consent obtained from all participants. The present analyses complied with all applicable regulations governing human subjects research and adhered to the principles of the Declaration of Helsinki.

Genetic Signal Annotation and Pioneer Variant Identification

Summary-level GWAS data, including single-nucleotide polymorphism (SNP) identifiers and LD reference information, were uploaded to the Functional Mapping and Annotation (FUMA) platform. Following platform-implemented quality control procedures to exclude missing or low-confidence variants, lead SNPs associated with MI and HF were identified. Lead variants were defined as genome-wide significant (P < 5 × 10⁻⁸) and independent, exhibiting low LD with any other genome-wide significant SNP (r² < 0.1). Statistical significance was determined after correction for multiple testing using the false discovery rate (FDR), with genes exceeding an FDR-adjusted P < .05 considered significant.

Bayesian Fine-Scale Positioning (Sum of Single Effects)

To refine causal inference at associated loci, we applied the R package echolocatoR to perform Bayesian fine-mapping using the Sum of Single Effects (SuSiE) model. The SuSiE accommodates multiple causal configurations and enables the simultaneous resolution of independent causal variants within regions of high LD. Fine-mapping was conducted for each lead SNP identified by FUMA, ¹⁸ interrogating a 500-kb locus centered on the lead variant (±250 kb). For all variants within each locus, posterior inclusion probabilities (PIPs) were estimated, and 95% credible sets were constructed. Variants with posterior probabilities exceeding 0.95 were designated as high-confidence candidate variants, reflecting strong statistical support within the fine-mapping model rather than definitive biological causality.

Transcriptome-Wide Association Study

Following the identification of putative causal variants, we conducted a Transcriptome-Wide Association Study (TWAS) to prioritize genes whose genetically regulated expression was associated with MI and HF. The TWAS was performed using the FUSION framework, incorporating 37,920 precomputed expression quantitative trait locus (eQTL) models derived from the GTEx v8 reference panel. Tissue-specific eQTL weights were used to infer genetically predicted gene expression levels, which were then tested for association with the phenotypes. Genes surpassing the Bonferroni-corrected genome-wide significance threshold (P < 1.31 × 10⁻⁶) were considered TWAS-significant and carried forward for downstream analyses. ¹⁹

Fine-Mapping of Genes Based on TWAS (FOCUS)

To further refine candidate genes implicated by TWAS, we applied FOCUS, a statistical fine-mapping framework tailored for transcriptome-wide association signals. The FOCUS integrates GWAS summary statistics with expression reference panels to estimate PIPs for each gene, thereby quantifying the likelihood that a given gene accounts for the observed association. Genes that were both TWAS-significant and exhibited high posterior support in FOCUS (PIP > 0.9) were prioritized as putative candidate genes with high statistical support for explaining the observed association signal.

Genome-Wide Genetic Correlation Assessment

We applied LD score regression to quantify the genome-wide genetic correlation between MI and HF. This approach leverages GWAS summary statistics to estimate genetic correlation while remaining robust to bias introduced by sample overlap.^20,21 For each variant, the product of z-scores from the primary and secondary traits was regressed on the corresponding LD score to derive the genetic covariance. Genetic correlation was subsequently obtained by normalizing this covariance by the SNP-based heritability of each trait. A nominal threshold of P < .05 was considered suggestive of association.

MiXeR Analysis

We first applied univariate MiXeR to delineate the fundamental properties of the genetic architectures underlying MI and HF. ²² We then extended the analysis using the bivariate MiXeR framework to quantify the magnitude of their polygenic overlap. MiXeR models GWAS summary statistics using a Gaussian mixture framework, allowing estimation of the proportion of non-null variants shared between traits and delineation of their joint genetic architecture. Because MiXeR is sensitive to LD patterns, we used the 1000 Genomes Project European reference panel to ensure accurate LD matching. ²³ Model stability was further assessed using the Akaike Information Criterion (AIC), comparing the full MiXeR specification with a reduced model to evaluate the robustness of the inferred genetic overlap. ²⁴

Cond/Conjunctive FDR Analysis

We applied condFDR and conjFDR frameworks to improve the detection of polygenic architecture in GWAS. The condFDR procedure reranks association statistics by leveraging cross-trait enrichment. The conjFDR metric imposes joint posterior significance across both traits, enabling the identification of pleiotropic loci that may not be fully captured in traditional pairwise analyses of MI and HF.^19,25

Proteome-wide MR Analysis

To delineate putative shared causal mechanisms linking MI and HF at the proteomic level, we applied summary-level Mendelian randomization (SMR) to evaluate protein-disease associations. ²⁵ This framework leveraged cis-acting pQTL (cis-pQTL) identified in the UKB-PPP resource, which comprises plasma proteomic profiles for 2940 proteins measured in 34,557 individuals of European ancestry using the Olink Explore platform. To further assess whether the pQTL signals reflect a shared causal variant underlying both protein abundance and disease risk, we performed HEIDI analyses on SNPs located within ±500 kb of each lead pQTL. HEIDI P-value greater than .05 was interpreted as evidence consistent with the possibility of a shared underlying variant influencing both protein abundance and disease risk.

Result

Functional Mapping and Annotation and Bayesian Fine Mapping (SuSiE)

Genome-wide association studies statistics for MI and HF were submitted to the FUMA platform, from which the respective lead SNPs were identified (Supplemental Table S2). The SuSiE fine-mapping analysis statistically prioritized rs544366796 as a high-confidence candidate variant (PP = 0.97) associated with MI within the SLC22A2 locus on chromosome 6. Notably, this variant resides within the contiguous SLC22A3 locus region and exhibits a strong association signal (PP = 0.93) in this model. Additionally, within the SLC22A3 locus, we detected an independent minor signal driven by variant rs140570886 (PP = 0.89), which did not meet our preset 95% credible set threshold, suggesting that the association signal in this region may be contributed by multiple independent genetic effects (Supplemental Table S3). In the analysis of HF traits, SuSiE did not identify any causal variants meeting the high posterior probability criterion (Figure 2).

Figure 2.

Identify the most likely causal variant single-nucleotide polymorphisms (SNPs) associated with myocardial infarction (MI) using Sum of Single Effects (SuSiE).

Transcriptome-Wide Association Study and FOCUS Fine-Mapping

Next, we performed TWAS analysis using FUSION to identify gene-level associations with MI and HF. After multiple comparison correction (Bonferroni correction), we identified 22 genes significantly associated with MI (Figure 3). Furthermore, we performed FOCUS fine-mapping analysis on the TWAS-identified susceptibility genes to infer the most likely causal genes. FOCUS identified 6 genes with high posterior probability (pip > 0.9) as statistically prioritized candidate genes likely contributing to the observed MI association signal. To further validate these high-confidence gene-level associations, we performed an intersection analysis of TWAS and FOCUS results, obtaining a gene set supported by both types of evidence. This included multiple genes such as CDKN2B, CELSR2, BRAP, NBEAL1, and the long noncoding RNA RP11-378J18.8 (Supplemental Table S4). Among these, RP11-378J18.8 demonstrated exceptionally strong colocalization evidence (COLOC.PP4 = 0.989) and high causation probability (pip = 0.99), suggesting this noncoding RNA may represent an underappreciated regulator of MI risk. Notably, although CDKN2B showed a highly significant TWAS association (Z = −11.4) and the highest causal gene probability (pip = 1), its colocalization analysis with eQTLs did not support sharing the same causal variant (PP3 = 0.993). This suggests the presence of a pathogenic variant within this locus that is tightly linked to the primary eQTL but functionally independent, potentially acting through alternative mechanisms (eg, affecting mRNA splicing or protein function). Directionality analysis of these genes revealed TWAS Z-scores greater than 0 for both BRAP and NBEAL1, indicating that predicted increases in their expression levels are positively correlated with MI risk. Conversely, TWAS Z-scores for CDKN2B, CELSR2, and RP11-378J18.8 were all less than 0, suggesting their predicted gene expression levels are negatively correlated with MI risk.

Figure 3.

Identification of putative causal genes for myocardial infarction (MI) and heart failure (HF) via Transcriptome-Wide Association Study (TWAS) and FOCUS fine-mapping. Significantly associated genes were subsequently fine-mapped using FOCUS to calculate the posterior inclusion probabilities (PIP) for each gene, representing the probability of it being the causal gene driving the genome-wide association studies (GWAS) signal. The larger the dot, the higher the PIP value (ranging from 0.0 to 1.0) .

In the HF analysis, we identified 11 genes significantly associated with HF. FOCUS further identified MYOZ1 (COLOC.PP4 = 0.991, pip=0.982) as a causal gene for HF. This gene's TWAS Z-score of 5.006 suggests that its predicted high expression is associated with increased HF risk (Figure 4).

Figure 4.

Transcriptome-wide association analysis using FUSION. The x-axis represents chromosomal location (chromosome 1-22), while the y-axis represents the Z-score for association strength (positive and negative values indicate positive and negative directions of genetic effect, respectively). The inset in the top right-hand corner is a QQ plot, which illustrates the comparison between the observed -log₁₀ P values and the expected distribution, and is used to assess systematic bias in the test and the presence of genomic inflation. (A) Regional Z-score evaluation of significant gene loci for myocardial infarction (MI). (B) Regional Z-score evaluation of significant gene loci for heart failure (HF).

Linkage Disequilibrium Score Regression and MiXeR Analysis

Linkage disequilibrium score regression demonstrated a robust positive genetic correlation between MI and HF (rg = 0.494, SE = 0.062, P = 1.124 × 10⁻¹⁵). Univariate MiXeR analyses indicate that both MI and HF are highly polygenic traits. Approximately 55.9% of genetic variants influencing MI overlap with those affecting HF, with HF appearing to be under stronger genetic regulation (Supplemental Table S5). Consistent with this, the estimated number of causal variants for HF was markedly greater than for MI (1258 vs 93; Supplemental Table S5). Bivariate MiXeR modeling identified extensive shared genetic architecture, comprising 862 ± 172 causal variants, which accounted for 90.2% of MI-associated and 40.7% of HF-associated signals (Supplemental Table S6). In line with these findings, bivariate MiXeR analysis revealed a positive genetic correlation between MI and HF (r_g = 0.43; Figure 5). Shared variants also showed strong concordance in effect direction (rho_beta = 0.73; Supplemental Table S5). Stratified quantile-quantile plots demonstrated close agreement between MiXeR-predicted and observed cross-trait enrichment patterns (Figure 5), and log-likelihood profiles confirmed an excellent model fit (Figure 5). Moreover, MI-associated variants were significantly enriched in HF association statistics in a manner proportional to effect size, as reflected by enrichment curves consistently exceeding the null expectation (Figure 5). Model comparison further supported robustness, with differences in Akaike and Bayesian information criteria between the optimal and minimal models remaining < 3 (best vs min_AIC = 1.37; best vs min_BIC = −8.16), indicating that the model captures the shared genetic architecture of MI and HF without evidence of overfitting (Supplemental Table S6).

Figure 5.

Bivariate MiXeR analysis of genetic architecture between myocardial infarction (MI) and heart failure (HF).

(A) The Venn diagram illustrates the estimated quantities of nonzero variation shared between MI and HF, as well as those unique to each: the figures within the circles denote the amount of genetic variation in thousands. The left circle represents MI, the right circle represents HF, and the overlapping center section denotes shared variation. Circle size indicates polygenicity, with larger circles reflecting stronger polygenicity. Estimated genetic correlations are also displayed below the Venn diagram, where the sign of rg indicates the direction of genetic correlation. (B) Conditional Q-Q plot (MI | HF): displays deviations in −log10(P-value) for MI when SNPs show stronger association with HF across different significance thresholds (P < .1, .01, .001). Deviations to the upper left relative to the global null hypothesis (indicated by the black dashed line) suggest that variants significant for HF also tend to show stronger association with MI, implying a shared genetic basis. (C) Conditional Q-Q plot (HF | MI): Similarly, the observed −log10 (P values) for HF were assessed across different MI significance thresholds (P < .1, .01, .001). Results again showed significant deviation from the diagonal, further supporting genetic overlap between MI and HF. (D) Evaluating bivariate MiXeR model fit using log-likelihood plots. (E) Density plot of additive causal effects underpinning the model predictions. (F) Density plot of observed GWAS sign test statistics. (G) Corresponding density plot predicted from the fitted MiXeR model.

Cond/ConjFDR Analysis

To enhance the identification of genetic loci shared between HF and MI, we performed cond/conjFDR analysis. Results showed that when MI was used as the a priori trait, 24 loci significantly associated with HF were identified at a CondFDR threshold <0.01, including 3 novel loci not previously reported. Conversely, when HF was used as the a priori trait, 22 MI-associated loci were identified, but no novel loci were discovered. Conditional Q-Q plot analysis indicated enrichment of MI-associated SNPs in HF and vice versa (Supplemental Tables S7-S9).

Shared Causal Proteins Between HF and MI

Using SMR, we evaluated the causal relationships between plasma proteins profiled in the UKB-PPP cohort and the risks of MI and HF. After applying FDR correction, 9 proteins showed significant associations with HF and 9 with MI (adjusted P < .05) (Supplemental Table S10). Notably, Apolipoprotein E (APOE) exhibited robust associations with both conditions and satisfied the HEIDI test (P > .05), indicating that the observed signals were unlikely to be driven by linkage (Supplemental Table S10).

Discussion

In this study, we performed an integrative cross-trait post-GWAS analysis to systematically delineate the shared genetic architecture between MI and HF. By combining genome-wide genetic correlation analysis, polygenic overlap modeling, pleiotropic locus discovery, transcriptome-wide gene prioritization, fine-mapping, and proteome-wide MR, we provide convergent evidence that MI and HF share extensive yet biologically structured genetic liability. These findings advance current understanding of how inherited susceptibility contributes to the progression from ischemic myocardial injury to chronic cardiac dysfunction.

Apolipoprotein E emerged from our SMR framework as a protein showing statistical evidence consistent with shared genetic regulation between MI and HF. This finding is highly biologically plausible given the central role of APOE in lipoprotein clearance and cholesterol homeostasis, processes that are fundamental to atherogenesis and ischemic myocardial injury. The 3 common APOE isoforms (E2, E3, and E4) differ in receptor-binding affinity and lipid-transport efficiency, with the E4 allele consistently associated with elevated low-density lipoprotein cholesterol and accelerated atherosclerotic plaque formation, thereby increasing susceptibility to MI.^26,27 Beyond its effects on coronary atherosclerosis, accumulating evidence suggests that APOE modulates postinfarction myocardial remodeling through inflammation, oxidative stress, and extracellular matrix turnover, which are key biological processes underlying the transition from MI to HF. ²⁸ Population-based prospective studies have further demonstrated that circulating APOE levels and APOE genotypes are associated with the risk of ischemic heart disease and long-term cardiovascular outcomes. ²⁹ In patients with established HF, APOE variants have also been implicated in cognitive impairment and adverse prognosis, highlighting a broader systemic role of this protein in chronic cardiovascular disease states.^30,31 Collectively, these lines of evidence, together with our proteogenomic results, support a model in which APOE integrates lipid metabolic and inflammatory pathways, thereby linking coronary plaque instability with maladaptive ventricular remodeling.

Beyond proteomic prioritization implicating APOE, our integrative post-GWAS framework resolves genetic architecture at both the locus and gene levels, refining the shared susceptibility landscape between MI and HF. Bayesian fine-mapping at single-variant resolution identified rs544366796 within the SLC22A2 locus as a high-confidence candidate variant for MI, while prior evidence linking rs140570886 to coronary artery disease further supports the pathogenic relevance of this locus in ischemic phenotypes. ³² Members of the organic cation transporter family encoded by SLC22A genes are expressed in vascular endothelial cells and cardiomyocytes, where they orchestrate the cellular uptake of endogenous metabolites and xenobiotics; functional studies additionally implicate SLC22A3 in catecholamine handling and local inflammatory responses within ischemic myocardium. ³³ Collectively, these findings position the SLC22A locus at the intersection of metabolic flux and inflammatory signaling during ischemic stress, suggesting that its perturbation may modulate infarct size and subsequent ventricular remodeling, thereby providing a mechanistic continuum linking acute ischemic injury to chronic HF progression.

Notably, although SuSiE did not resolve a single high-posterior-probability causal variant for HF, the substantial polygenic overlap delineated by MiXeR and conjunctional FDR analyses indicates that HF susceptibility is predominantly driven by the cumulative effects of numerous variants of modest effect, rather than a limited set of large-effect loci.^34–36 Consistent with this architecture, HF-associated variants are preferentially enriched within regulatory elements controlling myocardial contractility, calcium handling, and extracellular matrix organization, underscoring the centrality of gene regulatory perturbation in preserving cardiac structural and functional integrity.^37,38 Strikingly, more than 90% of MI-associated statistically prioritized variants colocalize with HF-associated signals, reinforcing a shared genetic substrate in which inherited susceptibility to ischemic injury predisposes to maladaptive ventricular remodeling and progressive HF.

At the gene expression level, TWAS and FOCUS analyses prioritized numerous known and novel candidate genes differentially associated with MI and HF. The CDKN2B and CELSR2 genes, located on chromosome 9 short-arm segment 21 (9p21) and chromosome 1 long-arm segment 13 (1p13), have been repeatedly demonstrated to be associated with coronary artery disease and MI through their regulation of vascular smooth muscle cell proliferation and lipid metabolism.^39–42 Notably, despite robust TWAS and FOCUS support for CDKN2B, colocalization analysis suggests its disease association may not be mediated by steady-state mRNA expression. Such findings underscore the importance of integrating fine-mapping with colocalization analysis to avoid oversimplified causal interpretations. Notably, while TWAS and FOCUS fine-mapping provided statistical support for the CDKN2B gene at the CAD susceptibility locus on chromosome 9p21, colocalization studies revealed that this GWAS signal did not persistently overlap with steady-state cis-eQTL effects of CDKN2B expression in relevant vascular tissues. ⁴³ Instead, stronger evidence supports its colocalization with the regulatory effects of the adjacent long noncoding RNA, CDKN2B-AS1, particularly mediated through splicing QTLs in proliferative vascular smooth muscle cells. ⁴⁴ These findings suggest the pathogenic mechanism at the 9p21 locus may extend beyond simple alterations in CDKN2B mRNA abundance. Potential mechanisms may include: enhancer-mediated cell-type-specific transcriptional regulation within the accessible chromatin region of 9p21; ANRIL isoform-specific regulation affecting polycomb repressive complex recruitment at the CDKN2A/B locus and downstream cell cycle control; allele-specific chromatin looping reshaping 3-dimensional genomic architecture; and 9p21 risk haplotype-associated epigenetic modifications. These modifications promote vascular smooth muscle cell phenotypic switching and proliferative responses.^39,44–46 Collectively, these data support a model in which the 9p21 risk locus exerts its effects through complex regulatory and epigenomic mechanisms, rather than solely through the steady-state expression of CDKN2B.

The MI-associated genes BRAP and NBEAL1, identified through TWAS Z-score positive screening, further highlight the role of intracellular signaling and vesicle transport in ischemic heart disease.^47,48 Studies confirm that BRAP interacts with the MAPK signaling cascade and inflammatory pathways, both of which are activated during plaque rupture and myocardial ischemia.^49,50 The NBEAL1 gene, involved in vesicle formation and lipid metabolism, has been implicated in endothelial dysfunction and coronary artery disease risk. ⁵¹ Consequently, genetic upregulation of these genes may exacerbate inflammatory and metabolic stress responses during acute MI, indirectly increasing the risk of adverse remodeling and progression to HF.

Notably, our analysis also revealed the role of noncoding RNA biology in MI susceptibility. The long noncoding RNA RP11-378J18.8 demonstrated strong colocalization evidence and a high PIP, suggesting a potential biological role warranting functional investigation. Although functional data for this transcript remain limited, emerging literature indicates that long noncoding RNAs are key regulators of cardiomyocyte apoptosis, fibrosis, and angiogenesis following ischemic injury.^52,53 Consequently, our findings identify RP11-378J18.8 as a potential target for future experimental validation.

Unlike MI, HF-specific gene prioritization identified MYOZ1 as a potential disease-causing gene. MYOZ1 encodes myozin-1, a Z-band protein that interacts with calmodulin and plays a crucial role in sarcomere integrity and hypertrophic signaling pathways.^54,55 Genetic predictions indicate that elevated MYOZ1 expression correlates with increased HF risk. ³⁵ Notably, MYOZ1 has not been definitively linked to strong associations with atherosclerosis or MI, supporting the notion that HF risk is influenced by intrinsic myocardial pathways potentially operating independently of ischemic burden.

Collectively, these findings suggest a conceptual 2-stage genetic model linking ischemic susceptibility and myocardial remodeling. However, because our analyses are based on cross-trait genetic associations rather than longitudinal individual-level data, this framework should be interpreted as a hypothesis-generating model rather than direct evidence of temporal progression from MI to HF. In the first stage, variants affecting lipid metabolism, vascular inflammation, and endothelial function (as demonstrated by loci such as APOE, CDKN2B, CELSR2, and SLC22A) predispose individuals to MI. In the second stage, another set of partially overlapping yet distinct variants (eg, MYOZ1 and related myogenic genes) regulates the transition from myocardial injury to chronic HF by influencing cardiac cell structure, calcium handling, and stress response signaling pathways. The significant genetic correlation between MI and HF reflects the shared contribution of upstream ischemic mechanisms, while the partial overlap highlights the importance of downstream myocardial-specific processes.

This study offers the following insights: Shared genetic loci and proteins such as APOE may represent promising candidates for future functional investigation and therapeutic exploration. Although our findings do not establish immediate clinical applicability, they highlight molecular pathways that could potentially inform early intervention strategies pending experimental validation. Similarly, the identification of HF-specific genes may provide a conceptual foundation for future precision medicine approaches, contingent upon longitudinal validation and mechanistic studies. Finally, integrating GWAS with transcriptomic and proteomic data provides an expandable analytical framework for deciphering the genetic architecture underlying complex cardiovascular disease progression.

Although our findings delineate shared genetic architecture between MI and HF at the population level, they do not directly translate into clinically actionable predictions for individual patients. In particular, the present cross-trait framework cannot determine which individuals with MI are at elevated risk of subsequently developing HF. Addressing such questions would require longitudinal, individual-level genetic and phenotypic data, enabling time-to-event analyses, mediation modeling, or the evaluation of polygenic risk scores within post-MI cohorts. Accordingly, the translational implications of our results should be viewed as hypothesis-generating with respect to risk stratification and precision medicine, pending validation in prospective datasets.

Several limitations merit consideration. First, although the MI and HF GWAS datasets were derived from independent national cohorts (UK Biobank and FinnGen), HF frequently develops on the background of prior or silent MI. As a result, phenotype-level nonindependence may exist, reflecting the biological and epidemiological continuum between ischemic injury and ventricular dysfunction. Our cross-trait analyses, therefore, quantify shared inherited liability across related cardiovascular phenotypes rather than fully independent disease entities. Second, partial participant overlap cannot be excluded in the SMR analysis integrating UKB-PPP proteomic data, as both the MI GWAS and the pQTL dataset originate from UK Biobank. Although summary-level SMR is widely applied, sample overlap may introduce bias or inflate association statistics. Accordingly, proteogenomic findings, including those related to APOE, should be interpreted as statistically supported candidates pending independent validation. Third, all primary analyses were restricted to individuals of European ancestry, which may limit generalizability to other populations. Fourth, although fine-mapping and colocalization strengthen causal inference, functional validation in experimental systems remains necessary to confirm the biological roles of prioritized variants and genes. Finally, because our analyses rely on independent summary-level GWAS data rather than longitudinal individual-level cohorts, we are unable to assess predictive performance or clinical risk stratification for HF following MI. Future studies integrating prospective follow-up data will be necessary to determine whether the shared genetic components identified here can improve individualized risk prediction or guide targeted intervention strategies.

Conclusion

In conclusion, this integrative cross-trait post-GWAS analysis demonstrates a robust positive genetic correlation and substantial polygenic overlap between MI and HF, supporting a shared inherited susceptibility underlying ischemic injury and adverse ventricular remodeling. Proteogenomic and transcriptomic integration statistically prioritizes APOE as a convergent candidate mediator, pending functional validation. While canonical loci such as CDKN2B and CELSR2 reflect upstream atherosclerotic mechanisms, and MYOZ1 highlights myocardial-specific pathways influencing HF progression. Together, these findings suggest a potential 2-stage genetic framework characterizing shared and partially distinct components of genetic susceptibility to MI and HF. This model offers a comprehensive explanation of the trans-trait genetic architecture that generates new hypotheses and warrants validation in longitudinal and mechanistic studies.

Supplemental Material

sj-xlsx-1-cpt-10.1177_10742484261452728 - Supplemental material for Shared Genetic Liability Between Heart Failure and Myocardial Infarction Revealed by Genome-Wide Cross-Trait Analysis

Supplemental material, sj-xlsx-1-cpt-10.1177_10742484261452728 for Shared Genetic Liability Between Heart Failure and Myocardial Infarction Revealed by Genome-Wide Cross-Trait Analysis by Ruikang Liu, MS, Chiyun Sun, MS, Nan Jiang, PhD, Yang Liu, PhD, Jun Li, PhD, Fuyuan Zhang, PhD, Cong Chen, PhD, Yiying Liu, PhD, Xiaodi Qi, MS, Bingting Guo, MS, and Kai Yang, PhD in Journal of Cardiovascular Pharmacology and Therapeutics