Potential Risk Factors and Therapeutic Targets for Dilated Cardiomyopathy Identified Through Mendelian Randomization Analysis

Abstract

Background

Dilated cardiomyopathy (DCM) is a severe form of cardiomyopathy. The study aims to investigate the impact of metabolic and lifestyle factors on DCM and to identify new potential therapeutic targets.

Methods

Both bidirectional Mendelian randomization (MR) and Bayesian colocalization were performed based on summary data from large‐scale, genome‐wide association studies and the FinnGen project. Single nucleotide variant strongly associated with exposure at the genome‐wide significance level were used as instrumental variables. Methods of inverse‐variance weighted, debiased inverse‐variance weighted, weighted median, weighted mode, MR‐Egger, robust‐adjusted profile score, contamination mixture, and constrained maximum likelihood and model averaging were used to validate the robustness of the MR results and the level of multiplicity. Summary‐data‐based MR was used for the analysis of potential drug targets of DCM using data of cis‐expression quantitative trait loci and cis‐protein quantitative trait loci. Steiger filtering and Bayesian colocalization were employed for reverse causality detection to further consolidate the MR findings.

Results

This study found that genetically predicted systolic blood pressure, hypertension, basal metabolic rate, waist circumference, body mass index, and hip circumference were associated with an increased risk of DCM. The inverse‐variance weighted estimate showed that sleep duration and never smoking were protective factors against DCM. The summary‐data‐based MR analysis revealed 11 potential candidate genetic variants and risk genes associated with DCM. Among them, SEMA7A, ETFA, CNTFR, TGM2, GOLGB1, and DNAJB12 were associated with a decreased risk of DCM. Colocalization analysis indicated that the expression quantitative trait loci signals of SLC9B1 and TGM2 share causal variants with the risk of DCM.

Conclusions

This study supports independent causal roles of systolic blood pressure, hypertension, waist circumference, hip circumference, body mass index, basal metabolic rate, sleep duration, and never smoking in DCM. We also identified several potential candidate variants and risk genes involved in the genetic basis of DCM.

Nonstandard Abbreviations and Acronyms

MR

Mendelian randomization

Clinical Perspective.

What Is New?

• This is the first Mendelian randomization study to investigate metabolic factors, lifestyle factors, and potential therapeutic targets for dilated cardiomyopathy.

• This study supports independent causal roles of systolic blood pressure, hypertension, waist circumference, hip circumference, body mass index, basal metabolism rate, sleep duration, and never smoking in dilated cardiomyopathy.

What Are the Clinical Implications?

• Our findings may provide new insights into the risk factors for dilated cardiomyopathy and prioritize modifiable risk reduction in high‐risk populations.

Dilated cardiomyopathy (DCM) is characterized by unilateral or bilateral enlargement of the ventricular chambers and impaired systolic function with normal thickness of the left ventricular wall. The initial manifestations may include arrhythmias, and sudden death may occur at any stage of DCM. ¹ Although DCM was previously considered a rare and orphan disease, its prevalence has risen from 36.5/100 000 (1/2740) in 1985 to 1/250 in recent years, and this figure is still thought to be underestimated due to incomplete case ascertainment. ² , ³ DCM is considered the most frequently diagnosed cardiomyopathy, accounting for ∼40% of patients with heart failure with reduced ejection fraction, hospitalized for heart failure, or with a leading indication for heart transplantation. ⁴ Without treatment, DCM is associated with a 1‐year survival rate of ∼70% to 75% and a 5‐year survival rate of 50%, with the most common cause of death being pump failure (accounting for approximately two thirds of deaths), followed by sudden cardiac deaths. ⁵ Despite some promising advances in the treatment of DCM such as the therapy targeting the renin‐angiotensin‐aldosterone system, novel small‐molecule therapies, and cardiac resynchronization therapy, ⁶ , ⁷ DCM continues to result in a significant number of deaths, with a 10‐year survival rate of ∼60%. ⁸ , ⁹ In addition, the prognosis of DCM appears to be poorer in men and patients of Black African descent than in other populations. ¹⁰ , ¹¹ Therefore, identifying potential risk factors and drug targets for DCM has essential implications for the intervention and management of this disease.

Epidemiological studies have revealed several possible risk factors for DCM, including myocarditis; exposure to alcohol, drugs, or other toxins; and metabolic or endocrine disorders. ¹² It has been found that metabolic risk factors such as increases in lipids, blood pressure (BP), and glucose and modifiable lifestyle behaviors including smoking, drinking, sleeping, and physical activities are also closely associated with cardiovascular diseases (CVDs). However, most of the associations found between risk factors and DCM are ambiguous, inconsistent, or even contradictory. Furthermore, identifying causal associations of lifestyle and metabolic factors with the risk of DCM through observational epidemiological studies is challenging due to biases such as confounding and reverse causality. Although randomized controlled trials perform well on causal inference, the difficulty in conducting large‐scale randomized controlled trials has limited the exploration of such associations. Thus, there has been a great demand for alternative methods to infer the causality between various factors and DCM.

Mendelian randomization (MR) is an approach of genetic instrumental variable analysis that typically uses single nucleotide variants (SNVs) from genome‐wide association studies (GWAS) to estimate the causal impact of exposures on outcomes. ¹³ With multiomics research, large‐scale data on protein quantitative trait loci (pQTLs) and expression quantitative trait loci (eQTLs) have been produced, and the correlation of genotypes with protein abundance and gene expression has been demonstrated. Thus, MR has been increasingly applied in recent years not only for causal inference but also for the prediction of the efficacy and safety of new drugs and drug repurposing for additional indications. ¹⁴ A study showed that a drug target with genetic support and guided selection was twice as likely to be approved as those without such support. ¹⁵ In addition, the MR analysis can simulate the characteristics of a randomized trial and reduce unobserved confounding. As genetic variants are randomly categorized at the time of conception, they are unrelated to adaptive lifestyle factors and behaviors, which are often considered confounders in the association between exposures and outcomes. Another significant advantage of MR analysis is that it can be used to reveal lifetime exposure to a risk factor.

Therefore, we focused on cardiometabolic factors and lifestyle exposures with established mechanistic links to myocardial remodeling in existing literature, such as increased lipids, BP, blood glucose, smoking, drinking, sleep, and physical activities. This bidirectional MR analysis based on GWAS summary data aimed to explore the potential causal relationship between genetic predisposition to common lifestyle and metabolic risk factors and DCM. We also conducted a summary‐data‐based Mendelian randomization analysis using pQTLs and eQTLs data to identify new therapeutic targets for DCM (Figure 1).

Figure 1. Flow chart of the MR study to assess the risk factors and identify therapeutic targets for DCM.

cis‐eQTL indicates cis‐expression quantitative trait loci; cis‐pQTL, cis‐protein quantitative trait loci; DCM, dilated cardiomyopathy; dIVW, debiased IVW; GWAS, genome‐wide association studies; HDL, high‐density lipoprotein; IVW, inverse‐variance weighted; LDL, low‐density lipoprotein; and MR, Mendelian randomization.

METHODS

Data Availability Statement

GWAS data on DCM are publicly available at https://www.nature.com/articles/s41588‐021‐00931‐x#data‐availability and https://pmc.ncbi.nlm.nih.gov/articles/PMC9849126/#notes4.

GWAS data on hypertension, basal metabolic rate, hip circumference, waist circumference, tea intake, coffee intake, alcohol intake frequency, average weekly spirit consumption, smoking status, sleep, and physical activity are publicly available at http://www.nealelab.is/uk‐biobank.

GWAS data on systolic BP (SBP), diastolic BP, and pulse pressure are publicly available at https://pmc.ncbi.nlm.nih.gov/articles/PMC7610439/#S33.

GWAS data on type 2 diabetes are publicly available at https://pmc.ncbi.nlm.nih.gov/articles/PMC6063971/#_ad93.

GWAS data on fasting glucose are publicly available at https://pmc.ncbi.nlm.nih.gov/articles/PMC7610958/#S45.

GWAS data on total cholesterol levels are publicly available at https://pmc.ncbi.nlm.nih.gov/articles/PMC8349845/#_ad93.

GWAS data on triglycerides, low‐density lipoprotein cholesterol and high‐density lipoprotein cholesterol are publicly available at https://pmc.ncbi.nlm.nih.gov/articles/PMC7089422/#notes1.

GWAS data on body mass index are publicly available at https://pmc.ncbi.nlm.nih.gov/articles/PMC6488973/#_ad93.

GWAS data on alcoholic drinks are publicly available at https://pmc.ncbi.nlm.nih.gov/articles/PMC6358542/#_ad93.

GWAS data on cis‐eQTL are publicly available at https://www.eqtlgen.org/.

GWAS data on cis‐pQTL are publicly available at https://www.nature.com/articles/s41588‐021‐00978‐w#data‐availability.

Data Sources and Selection of Genetic Instruments

For the primary analysis, summary statistics of DCM were retrieved from the largest GWAS data set of DCM, ¹⁶ which included 355 381 individuals (n_case=1444, n_control=353 937) of European ancestry. For external replication, summary statistics were obtained from the FinnGen study ¹⁷ (n_case=1754, n_control=340 815, R10 release). SNVs associated with metabolic risk factors including hypertension, ¹⁸ SBP, ¹⁹ diastolic BP, ¹⁹ pulse pressure, ¹⁹ type 2 diabetes, ²⁰ fasting glucose, ²¹ total cholesterol, ²² triglycerides, ²³ low‐density lipoprotein cholesterol, ²³ high‐density lipoprotein cholesterol, ²³ body mass index (BMI), ²⁴ basal metabolic rate (BMR), ¹⁸ waist circumference, ¹⁸ and hip circumference, ¹⁸ as well as lifestyle factors including alcohol consumption, ²⁵ coffee drinking, ¹⁸ tea drinking, ¹⁸ insomnia, ¹⁸ sleep duration, ¹⁸ smoking, ¹⁸ , ²⁵ moderate physical activity, ¹⁸ and vigorous physical activity, ¹⁸ were used as instrumental variables. Summary statistics of the individuals were extracted from the UKB (UK Biobank), MVP (Million Veteran Program), deCODE (Diabetes Epidemiology: Collaborative Analysis of Diagnostic Criteria in Europe), and GENOA (Genetic Epidemiology Network of Arteriopathy) studies. All of the GWAS data mentioned have minimal or no participant overlap with the DCM data, thereby maintaining the assumption of independence for 2‐sample MR.

Among all the identified variants of each gene, only SNVs significantly associated with exposure factors (P<5×10⁻⁸), with F‐statistic >10, and clumped to a linkage disequilibrium threshold of r² <0.001 were considered candidate proxies. All data used in this MR study are publicly available (Table).

Table 1.

Overview of the Data Sources of the Instrumental Variables Used in the MR Study

Risk factor	SNVs	Sample size	Ancestry	Authors
Blood pressure
Hypertension	204	462 933	European	Mitchell et al. 18
Systolic blood pressure	196	810 865	European	Surendran et al. 19
Diastolic blood pressure	189	810 865	European	Surendran et al. 19
Pulse pressure	194	810 865	European	Surendran et al. 19
Glucose
Type 2 diabetes	115	655 666	European	Xue et al. 20
Fasting glucose	60	200 622	European	Chen et al. 21
Lipids
Total cholesterol levels	183	437 878	European	Barton et al. 22
Triglycerides	291	441 016	European	Richardson et al. 23
Low‐density lipoprotein cholesterol	167	440 546	European	Richardson et al. 23
High‐density lipoprotein cholesterol	336	403 943	European	Richardson et al. 23
Anthropometric measurements
Basal metabolic rate	500	454 874	European	Mitchell et al. 18
Body mass index	488	681 275	European	Yengo et al. 24
Hip circumference	386	462 117	European	Mitchell et al. 18
Waist circumference	339	462 166	European	Mitchell et al. 18
Diets
Tea intake	39	447 485	European	Mitchell et al. 18
Coffee intake	38	428 860	European	Mitchell et al. 18
Alcoholic drinks	34	335 394	European	Liu et al. 25
Alcohol intake frequency	92	462 346	European	Mitchell et al. 18
Average weekly spirit consumption	4	326 565	European	Mitchell et al. 18
Smoking
Smoking status: current	116	359 706	European	Mitchell et al. 18
Smoking status: never	71	359 706	European	Mitchell et al. 18
Cigarettes consumed per day	22	337 334	European	Liu et al. 25
Sleep
Sleeplessness/insomnia	39	462 341	European	Mitchell et al. 18
Sleep duration	66	460 099	European	Mitchell et al. 18
Physical activity
Vigorous physical activity	11	440 512	European	Mitchell et al. 18
Moderate physical activity	16	440 512	European	Mitchell et al. 18

MR indicates Mendelian randomization; and SNVs, single nucleotide variants.

Selection of Cis‐pQTLs and Cis‐eQTLs

The pQTL data set was obtained from a large‐scale integration study by Ferkingstad et al., ²⁶ who provided the pQTLs of 4907 plasma proteins from 35 559 Icelanders. Only pQTLs that met the following criteria were included: (1) showing significant genome‐wide association (P<5×10⁻⁸), (2) located outside the major histocompatibility complex region (chr6, 26–34 Mb), (3) meeting the assumption of independence ([linkage disequilibrium] clustering r² <0.001), (4) being a cis‐acting pQTL, and (5) not being a weak instrument (F‐statistic >10). The blood eQTL data set was obtained from eQTLGen (https://www.eqtlgen.org/, accessed on 1 January 2024). This data set contains the cis‐eQTLs of 16 987 genes from 31 684 blood samples from healthy European pedigrees. ²⁷ The selected proteins were significantly associated (P<5×10⁻⁸) with corresponding gene frequencies (minor allele frequency; >0.01) in the blood. Similar to the primary MR analysis, only strong (F‐statistic >10) eQTLs (within a 1000 kb window from the encoding gene) were included.

Statistical Analysis

Statistical Analysis for MR

In this study, the MR analysis was performed on “TwoSample MR” (https://github.com/MRCIEU/TwoSampleMR) using the R package, with SNVs associated with metabolic factors and lifestyle factors being the exposure and DCM being the outcome. The Wald ratio was used if only 1 instrument was available for exposure. When ≥2 genetic instruments were available, inverse‐variance weighted (IVW) MR and heterogeneity analysis would be applied. For the primary analysis, Bonferroni correction was used to adjust for multiple testing, and a P value of 0.05/26 (0.0019) was used to prioritize the results for further analysis. To avoid the “winner’s curse” effect where selecting only SNVs meeting statistical significance criteria for assessing the odds ratios (ORs) probability density of these selected SNVs, we employed conditional maximum likelihood estimation and quantile‐based CI methods to correct for biased ORs and CIs using the method by Zhong et al. ²⁸ In addition, the power calculations of the instruments selected for risk factors was performed ²⁹ (Table S1). The MR analysis was performed only on the preliminarily identified exposure for external replication, with a P value of 0.05.

Subsequently, we used the IVW approach with random effects and a multiplicative random effects model as the main analytical methods to estimate the MR association of metabolic factors and lifestyle factors with the risk of DCM. The IVW approach assumes that all SNVs are valid instrumental variables and that these estimates can reflect the total effect of exposure. ³⁰ For each genetic variant, a Wald ratio estimate was obtained by dividing the SNV‐outcome association by the SNV–exposure association. For single‐SNV instruments, causal estimates were calculated using the Wald ratio (ie, the SNV‐outcome association divided by the SNV–exposure association). For multi‐SNV instruments, we applied IVW meta‐analysis under a multiplicative random‐effects model to account for between‐instrument heterogeneity. However, the IVW estimator could be biased in the presence of weak IVs, which is a common challenge in MR studies. To overcome this challenge, we used the debiased IVW estimator, which is obtained by multiplying the IVW estimator by a bias correction factor.

We conducted several sensitivity analyses including weighted median, weighted mode, MR‐Egger, robust‐adjusted profile score, contamination mixture, and constrained maximum likelihood and model averaging to test the robustness of the association and for horizontal pleiotropy of exposures detected using SNVs (the applicability of different MR analysis methods is detailed in Table S2). The weighted median method was used to check for bias in invalid instruments, thereby providing consistent estimates even if >50% of the information came from invalid or weak instruments. The MR‐Egger analysis was used to detect unbalanced horizontal pleiotropy through its intercept to provide estimates after correction for pleiotropic effects, but it could be limited by low statistical power. Then, we evaluated the presence of outlier SNVs using the MR polytropic residual sum and outliers (MR‐PRESSO) method, which was also used for outlier analysis and multieffects analysis of the SNVs, with outliers excluded from the instrumental variables. It is important to emphasize that these methods depend on specific assumptions—such as the InSIDE condition in MR‐Egger regression—that may not always hold true in real‐world studies. Additionally, their performance can be influenced by factors including the number of instrumental variables and the presence of measurement error. To evaluate the statistical power of the pleiotropy detection methods used in our main analysis, we conducted a comprehensive simulation analysis (Table S3).

Sensitivity Analyses to Evaluate Reverse Causality

For potential associations between exposure and outcome identified in the previous analyses, we conducted 2 sensitivity analyses, bidirectional MR ³¹ and Steiger filtering, ³² to identify results that might result from reverse causation.

Drug Target Analysis

Summary‐data‐based MR was used for potential drug target analysis. Using cis‐pQTLs and cis‐eQTLs as the exposure, we employed methods of summary‐data‐based MR and heterogeneity in dependent instrument to test for the pleiotropic association between the expression of a gene and a complex trait of DCM using summary‐level data from GWAS. Specifically, the significance thresholds for cis‐eQTL and cis‐pQTL were set at P<5.00E‐04 and P<5.00E‐03, respectively. The eQTLs and pQTLs that met these adjusted thresholds were defined as “candidate” signals and subsequently included in colocalization analyses and other functional association studies.

Phenotype Scanning

We also performed phenotype scanning to search previous GWAS for associations between identified instruments and other traits. Phenotype scanning was conducted using the “PhenoScanner” ³³ to determine whether a phenotype was directly related to an outcome; if a direct association were identified, the corresponding phenotype would be regarded as a confounding SNV and removed before reanalysis. The version of PhenoScanner used (v2), the SNVs queried and the thresholds for trait association (P<1×10⁻⁵), the filtering for cis‐ versus trans‐associations, and the traits used to exclude SNVs with known pleiotropic effects (Table S4).

Bayesian Colocalization Analysis

Bayesian colocalization analyses were used to assess the probability of 2 traits sharing the same causal variant using the “colo” package (https://github.com/chr1swallace/coloc) with default arguments. We further used different types of colocalization analyses (GWAS‐GWAS ³⁴ and eQTL‐DCM ³⁵ ) to test whether risk factors, eQTLs, pQTLs, and DCM share the same causal variant in a given region. Bayesian colocalization provides the posterior probability for 5 hypotheses on whether a single variant is shared between 2 traits. In the present study, we tested the posterior probability of Hypothesis 3—both the protein and DCM were associated with the region through different variants, and the posterior probability of Hypothesis 4—both the protein and DCM were associated with the region through shared variants. ³⁶ For this analysis, a P value <0.05 indicated potential interaction, and posterior probability of Hypothesis 4 >0.8 indicated potential colocalization.

RESULTS

Genetically Determined Exposures and the Risk of DCM

At the significance threshold after Bonferroni correction (P<0.0019), the MR analysis revealed significant causal effects of SBP, hypertension, BMR, BMI, waist circumference, and hip circumference on DCM. Specifically, SBP (OR_‐IVW per SD increase, 1.876 [95% CI, 1.341–2.624], P=2.41E‐04), hypertension (OR_‐IVW per SD increase, 7.183 [95% CI, 3.591–14.371], P=2.50E‐08), BMR (OR_‐IVW per SD increase, 1.706 [95% CI, 1.322–2.203], P=4.07E‐05), and waist circumference (OR_‐IVW per SD increase, 1.996 [95% CI, 1.526–2.612], P=4.67E‐07) were found associated with an increased risk of DCM. Notably, the causal effects were confirmed by sensitivity analyses using debiased IVW, weighted median, weighted mode, contamination mixture, robust‐adjusted profile score, and constrained maximum likelihood and model averaging methods. BMI (OR_‐IVW per SD increase, 1.659 [95% CI, 1.339–2.057], P=3.85E‐06) and hip circumference (OR_‐IVW per SD increase, 1.942 [95% CI, 1.569–2.403], P=1.04E‐09) were also found to be associated with an increased risk of DCM, which was validated using weighted median, contamination mixture, robust‐adjusted profile score, debiased IVW, and constrained maximum likelihood and model averaging methods (Table S5). Sensitivity analyses found no significance for heterogeneity of effect sizes (leave‐one‐out; IVW Cochran Q statistic, P>0.05) or for pleiotropy (MR‐Egger intercept, P>0.05) among these factors (Tables S6 and S7).

The IVW estimate showed that sleep duration (OR_‐IVW per SD decrease, 0.339 [95% CI, 0.157–0.734], P=6.02E‐03) and never smoking (OR_‐IVW per SD decrease, 0.216 [95% CI, 0.076–0.615], P=4.08E‐03) were protective factors against DCM, and similar results were also found with the use of debiased IVW and constrained maximum likelihood and model averaging methods (Figure 2).

Figure 2. The associations of metabolic factors and lifestyle behaviors with the risk of DCM.

Genome‐wide significantly associated (P<5×10⁻⁸) independent (linkage disequilibrium r²=0.001, clumping distance=10 000 kb) SNVs were used as instruments. OR indicates odds ratio; and SNV, single nucleotide variants. *P<0.0019 (0.05/26).

Sensitivity Analysis for the Risk of DCM

All 6 risk factors and the potential protective factors (sleep duration and never smoking) revealed by the MR analysis were identified by the sensitivity analysis. The bidirectional MR analysis showed no causal effect of DCM on the 6 identified factors, and the directionality was further ensured by Steiger filtering (Table S8).

In the Finn cohort, results showed significant associations between several risk factors and DCM, including the BMR (OR_‐IVW per SD increase, 1.8235 [95% CI, 1.399–2.377], P=8.97E‐06), BMI (OR_‐IVW per SD increase, 1.504 [95% CI, 1.215–1.861], P=1.73E‐04), waist circumference (OR‐_IVW per SD increase, 1.4041 [95% CI, 1.075–1.834], P=1.27E‐02), and hip circumference (OR_‐IVW per SD increase, 1.507 [95% CI, 1.203–1.889], P=3.64E‐04) (Table S9). Sensitivity analyses found no significance for heterogeneity of effect sizes (leave‐one‐out; IVW Cochran Q statistic, P>0.05) or for pleiotropy (MR‐Egger intercept, P>0.05) (Tables S10 and S11).

Potential Drug Targets and Bayesian Colocalization Analysis

MR of cis‐pQTLs revealed 4 proteins associated with DCM, namely SEMA7A (semaphorin 7A), TMEM132A (transmembrane protein 132A), ETFA (electron transfer flavoprotein subunit alpha), and CNTFR (ciliary neurotrophic factor receptor subunit alpha). Specifically, increased SEMA7A (OR_{‐Wald ratio}, 0.215 [95% CI, 0.084–0.552], P=1.40E‐03), ETFA (OR_{‐Wald ratio}, 0.131 [95% CI, 0.0338–0.506], P=3.23E‐03), and CNTFR (OR_{‐Weighted median}, 0.705 [95% CI, 0.554–0.897], P=4.44E‐03) were associated with a decreased risk of DCM, whereas elevated TMEM132A (OR_{‐Weighted median}, 1.551 [95% CI, 1.168–2.060], P=2.42E‐03) was associated with an increased risk of DCM (Data S1). No heterogeneity was found for these proteins, but the evidence was challenged by the absence of support from the Bayesian colocalization analysis.

Based on the summary‐data‐based MR analysis, 7 eQTLs—rs6533030 (solute carrier family 9 member B1, SLC9B1), rs7244254 (tubulin beta 6 class V, TUBB6), rs12624511 (transglutaminase 2, TGM2), rs11646956 (CTD‐3088G3.4), rs59987610 (GOLGB1), rs6822658 (CDGSH iron sulfur domain 2, CISD2), and rs1245555 (DNAJ heat shock protein family member B12, DNAJB12) —showed at least a statistically significant effect (P<0.0005). Among them, TGM2, GOLGB1, and DNAJB12 were associated with a decreased risk of DCM, and the remaining 4 were associated with an increased risk of DCM (Data S1). The colocalization analysis revealed that the eQTL signals of SLC9B1 and TGM2 share causal variants with the risk of DCM (posterior probability of Hypothesis 4 >0.8), prioritizing these 2 genes as potential therapeutic targets (Table S12 and Figure 3).

Figure 3. Illustration of the colocalization results.

TGM2 gene (A, ENSG00000198959 and DCM, PPH₄ >0.8) and SLC9B1 (B, ENSG00000246560 and DCM, PPH₄ >0.8) colocalization results. DCM indicates dilated cardiomyopathy; eQTL, expression quantitative trait loci; GWAS, genome‐wide association studies; and PPH₄, posterior probability of Hypothesis 4.

Ethics Approval

The use of data from all GWAS was approved by the respective institutional review boards, and all participants provided written informed consent.

DISCUSSION

To the best of our knowledge, this is the first MR study to investigate risk factors and therapeutic targets for DCM. We found that genetically predicted hypertension and increased SBP, BMI, BMR, waist circumference, and hip circumference were associated with an increased risk of DCM, whereas long sleep duration and never smoking might reduce the risk of DCM. The standard MR analysis showed no significant associations of blood glucose, blood lipids, physical activity, coffee consumption, alcohol consumption, and smoking with DCM. We also found 4 and 7 proteins from cris‐pQTLs and eQTLs, respectively, that could potentially serve as drug targets for DCM, but the findings were yet to be verified by Bayesian colocalization analysis.

In line with some prior observational studies, the present study demonstrated an association between hypertension and DCM. Studies have shown that the prevalence of hypertension in patients with DCM is approximately one third higher than the prevalence of primary hypertension in the general population; patients with hypertension and DCM also have increased heart rates at rest than those with normotensive DCM as well as a higher prevalence of positive family history of primary hypertension. ³⁷ The mechanisms by which DCM develops in patients with hypertension have not been elucidated. It has been hypothesized that increased sympathetic nerve activity may be one of the early pathophysiological abnormalities in DCM patients with compensated or early cardiac insufficiency. ³⁸ In addition, researchers have found that an SBP >120 mm Hg may be an independent risk factor for 5‐ and 15‐year mortality in patients with DCM. ³⁹ , ⁴⁰ According to a single‐center prospective study, SBP was significantly associated with increased mortality of DCM, with an SBP of 90 to 110 mm Hg associated with the lowest mortality rate, followed by SBP of 111 to 120, 121 to 130, and >130 mm Hg; the mortality rate was found to be quadrupled when the SBP was >130 mm Hg. ⁴¹ This may be related to increased afterload caused by higher SBP in patients with DCM, which further compromises myocardial contractility, leading to poor outcomes. Together, these findings suggested that strict management of BP, especially SBP, might help to reduce the morbidity and mortality of DCM.

The present study found genetic evidence of a causal association of increases in BMI, BMR, waist circumference, and hip circumference with an increased risk of DCM. ⁴² BMI, calculated by body weight in kilograms divided by the square of height in meters, is one of the simplest indices to indicate overweight and obesity in adults ⁴³ , ⁴⁴ , ⁴⁵ ; however, compared with other approaches measuring adiposity, BMI cannot accurately reflect the cardiometabolic risk related to obesity as it cannot distinguish fat mass from fat‐free mass. Thus, we also included waist circumference and hip circumference as additional indicators in this study. Waist circumference is an outstanding measure of excess visceral fat and independently predicts health risks when BMI is not significantly elevated. The waist‐to‐hip ratio is also a sensitive predictor of health risks, as it assesses the risk of CVDs in a population when the majority is considered healthy as measured by BMI. ⁴⁶ , ⁴⁷ Several studies demonstrated that waist circumference and waist‐to‐hip ratio were good predictors of the presence of central adiposity and dyslipidemia and had high specificity and accuracy in predicting CVDs. ⁴⁸ , ⁴⁹

Obesity is also an independent risk factor for myocardial dysfunction. It has been found that obesity is associated with left ventricular wall thickening, dilatation, and remodeling, as well as decreased left ventricular systolic function. ⁵⁰ , ⁵¹ Patients with obesity may present with a range of disorders from asymptomatic left ventricular dysfunction to DCM. Excessive adipose tissue may lead to increased cardiac output and blood volume, resulting in decreased systemic vascular resistance and ventricular dilatation. ⁵² , ⁵³ Sustained ventricular dilatation reduces myocardial contractility, eventually resulting in systolic dysfunction. ⁵⁴ It is also believed that the high metabolic activity of adipose tissues and the increase in lean mass may lead to increased circulatory preload and afterload, which in turn predisposes an individual to ventricular remodeling and dilatation and to increased myocardial wall stress and oxygen consumption, ultimately resulting in left ventricular dysfunction. Several recent studies have shown that patients with DCM and overweight or obesity may have a surprisingly better prognosis than those with a normal BMI if the DCM has progressed to heart failure. ⁵⁵ , ⁵⁶ , ⁵⁷ , ⁵⁸ This unexpected finding is often referred to as the “obesity paradox,” but the underlying cause of this phenomenon is unclear. The paradox may be related to lead‐time bias, which occurs when patients with overweight or obesity develop DCM earlier in their lifetime or are assessed for DCM earlier than those with normal weight, resulting in their earlier diagnosis and treatment. In the present study, although obesity and DCM were found to be strongly associated, the correlation of lipids including total cholesterol, triglycerides, low‐density lipoprotein cholesterol, and high‐density lipoprotein cholesterol with DCM was not found; this may be related to different metabolic risk profiles of obesity and hyperlipidemia. In addition, ectopic fat deposition in the pericardial and epicardial cavities may directly affect cardiac structure and function through the myocardium and vascular system rather than exerting the effect indirectly through obesity‐associated metabolic comorbidities, such as hyperlipidemia.

It is worth noting that, in contrast to conventional views, the present study showed no correlation between alcohol consumption and DCM. Although the J‐ or U‐shaped relationship between alcohol consumption and the total mortality of DCM has been well established, conflicting evidence remains regarding the role of alcohol consumption in CVDs. ⁵⁹ , ⁶⁰ Previous studies have generally concluded that alcohol consumption decreases cardiac contractility while increasing the diastolic BP and that chronic and excessive alcohol use may lead to poor patient outcomes, including DCM; for instance, a study showed that up to 36% of DCM cases in Western countries were associated with alcohol abuse. ⁶¹ In contrast, a recent prospective cohort study showed that alcohol use was not associated with adverse cardiovascular outcomes despite the risk of adverse cardiac remodeling. ⁶² In line with the present study, no association was found between alcohol use and DCM, which might be related to the potential protective role of alcohol consumption, the mediation of the impact of alcohol consumption on outcomes by an unknown or unmeasured confounder, or insufficient statistical power of the study. The lack of precision in the self‐report of alcohol consumption might also lead to measurement errors, which might have affected risk estimates. Furthermore, most participants in the present study were of European ancestry. Some variants of specific genes, such as ALDH2‐rs671, are common only in East Asian populations ⁶³ ; this gene markedly slows down the breakdown of acetaldehyde, and the resultant build‐up of acetaldehyde can lead to severe discomfort. Thus, future research involving East Asian populations or genetically specific populations may help us better understand the relationship between alcohol consumption and DCM.

In this study, current smoking status and the amount of cigarettes smoked per day were not found to be associated with the onset of DCM. Smoking is a well‐known risk factor for CVDs that leads to vasoconstriction and exacerbates ischemia and hypoxia. ⁶⁴ However, a cohort study showed that smoking and alcohol consumption had no effect on the all‐cause mortality in patients with DCM. ⁶⁵ In addition, the 2023 European Society of Cardiology guidelines indicated that there are no data showing an interaction between tobacco and cardiomyopathy. ⁶⁶ For the MR analysis, genes associated with smoking are more associated with a predisposition to an amount of smoking once smoking has started but may not fully capture lifetime exposure, duration, or cessation effects. Interestingly, genetically proxied “never smoking” showed a suggestive association with reduced DCM risk (P=0.0037) before multiple testing correction. However, this association did not survive the rigorous Bonferroni correction applied to control the family‐wise error rate in our primary analysis. This nonsignificance after correction has 2 equally valid interpretations. (1) The initial association may represent a false positive result, appropriately mitigated by the conservative correction. Our stringent Bonferroni threshold is designed precisely to reduce such type I errors. (2) Alternatively, it could indicate a true protective effect of never smoking against DCM that the correction rendered nonsignificant due to reduced statistical power. This potential protective effect might relate to the complex nature of smoking‐related genetic instruments. They may influence DCM risk through multiple pathways beyond simply the absence of tobacco exposure. For instance, individuals genetically predisposed to never smoking may also have correlated healthier lifestyles (eg, diet, exercise) or other inherent protective factors resulting from lifelong avoidance of tobacco’s harmful constituents. ⁶⁷ Disentangling the direct effect of nonsmoking from these pleiotropic effects is challenging.

The present study suggests that long sleep duration has a protective effect against DCM. Although the correlation between long sleep duration and DCM has not been confirmed in observational studies, studies have shown that individuals with a short sleep duration are at a higher risk of atherosclerotic CVDs. ⁶⁸ , ⁶⁹ Sleep deprivation can lead to increased sympathetic nervous system activity, which stimulates the renin‐angiotensin‐aldosterone system to produce more catecholamines and vasoconstriction, ⁶⁸ , ⁷⁰ resulting in a range of diseases such as coronary artery disease, hypertension, and ischemic stroke. The mechanism by which long sleep duration prevents DCM may be its association with lower risks of immune impairment, sympathetic nervous activity, oxidative stress response, and endocrine or metabolic disruption. ⁷¹ , ⁷²

Despite the recent development of new therapies, the treatment options for DCM remain limited. On this basis, we combined summary‐data‐based MR and colocalization analysis to search for potential drug targets for DCM. Because causality assessed by the MR analysis might be interfered by reverse causality, we performed Steiger filtering analysis to ensure the directionality of causal effects. To minimize the bias caused by horizontal pleiotropy, we selected cis‐pQTLs and eQTLs as IVs and performed Bayesian colocalization analysis. ⁷³ We found 11 potential candidate genetic variants and risk genes associated with DCM (ie, TGM2, SEMA7A, ETFA, CNTFR, TMEM132A, SLC9B1, TUBB6, CTD‐3088G3.4, GOLGB1, CISD2, and DNAJB12). Among them, TGM2, SEMA7A, ETFA, CNTFR, GOLGB1, and DNAJB12 were associated with a decreased risk of DCM. The colocalization analysis revealed that the eQTL signals of SLC9B1 and TGM2 share causal variants with the risk of DCM.

TGM2 (transglutaminase 2), also known as tissue transglutaminase, is expressed in all types of cells and tissues; however, its expression increases only under stress. ⁷⁴ TGM2 is highly expressed in human atherosclerotic plaques, where it may affect plaque stability. ⁷⁵ Myocardial fibrosis promotes cardiomyocyte hypertrophy and ventricular dilatation, exacerbating DCM. Studies have demonstrated that TGM2 deficiency exacerbates myocardial fibrosis, which is related to the role of TGM2 in maintaining vascular stability and repair. ⁷⁶ SEMA7A is a glycosylphosphatidylinositol‐anchored membrane protein that participates in a variety of biological processes including neural development, immune regulation, angiogenesis, ⁷⁷ and atherosclerosis. ⁷⁸ Studies have shown that SNVs in SEMA7A are correlated with metabolic parameters in the human population. SEMA7A deficiency is associated with diet‐induced obesity in mice by increasing adipogenic gene expression. In vitro studies showed that SEMA7A suppressed adipogenesis in preadipocytes and lipogenesis in mature adipocytes. ⁷⁹ However, paradoxically, a study showed that SEMA7A aggravated coxsackievirus B3‐induced viral myocarditis by increasing α1β1‐integrin macrophages and through subsequently enhanced inflammatory response. ⁸⁰ Given the different roles of SEMA7A in various diseases, further research is needed to clarify its specific mechanisms in various pathological conditions to provide a stronger scientific basis for the prevention and treatment of DCM. Mitochondria is the core of energy metabolism in cardiac cells, and the excessive production of reactive oxygen species resulting from mitochondrial dysfunction and altered mitochondrial dynamics collectively contribute to the progression of DCM. ⁸¹ ETFA, a protein that mediates the transfer of electrons from a series of mitochondrial flavoenzymes to the respiratory chain, plays a critical role in mitochondrial energy metabolism in cardiomyocytes. ⁸² Impaired electron transfer in the respiratory chain leads to excessive reactive oxygen species production, leading to mitochondrial DNA mutations and ultimately inducing structural changes in cardiomyocytes. ⁸³ It has been found that ETFA helps to maintain vascular development and angiogenesis by regulating the expression of mitochondrial fatty acid oxidation and hypoxia‐inducible factor‐1α in endothelial cells. ⁸⁴ Thus, treatment targeting ETFA may inhibit oxidative stress and inflammatory responses, thereby attenuating the development of hypertension and atherosclerosis. ⁸⁵ CNTFR is a specific receptor for CNTF, and its binding leads to multiple activation of signaling cascades. ⁸⁶ It has been demonstrated that CNTFR activates the Nrf2 (NFE2‐related factor 2) signaling, inhibiting oxidative stress and programmed necrosis in myocardial cells. ⁸⁷ CNTFR also helps to reduce body weight partly by regulating nuclear respiratory factor 1 and mitochondrial transcription factor A. ⁸⁸

Three of the druggable genes identified in our analyses (GOLGB1, DNAJB12, and TMEM132A) were found to be associated with endoplasmic reticulum (ER) stress. In the classical pathogenic process of DCM, the innate immune response is activated first, followed by an adaptive immune response and a chronic phase that might persist for several months or years, which results in myocardial fibrosis and remodeling, culminating in DCM. ⁸⁹ Accumulating studies suggests that ER stress is a key pathogenetic component in the progression of myocardial fibrosis and cardiac remodeling. It promotes the development of autoimmune phenotypes that lead to DCM and are associated with poor prognosis. ⁹⁰ GOLGB1 affects the protein transport and intracellular calcium homeostasis in the ER‐Golgi and influences the endoplasmic reticulum stress response. ⁹¹ A study showed that the SNVs rs3732410 in GOLGB1 appeared to be strongly associated with a decreased risk of hemorrhagic stroke. ⁹² DNAJB12 is a type II transmembrane Hsp40 containing the J structural domain that protrudes from the ER into the cytoplasm and regulates ER cytoplasmic face enzyme activity, which has recently been reported to be a key member of selective autophagy degradation on the ER membrane. ⁹³ , ⁹⁴ TMEM132A was originally identified as a binding protein for the ER resident chaperone 78‐kDa glucose‐regulated protein, and knockdown of TMEM132A increases the apoptosis related to serum starvation‐induced ER stress. ⁹⁵ Despite that the evidence for potential drug targets is challenged by the lack of support from pQTL co‐localization analyses, the identified proteins may still represent promising therapeutic candidates for DCM, given that some of these targets have been demonstrated to be involved in cardiovascular mechanisms in prior studies. However, further experimental validation is needed to confirm their biological functions in DCM.

Limitations

Our findings demonstrated the causal role of some risk factors in the pathogenesis of DCM, which might help to bridge the evidence gap between observational associations and genetic epidemiology. The MR analysis strengthened the causal inference using the genetic variation as an instrumental variable in populations exhibiting substantial genetic and phenotypic heterogeneity while minimizing the confounding bias inherent in observational studies. Despite the strengths of the present study, some limitations still need to be noted. First, as interactions between genes and the environment cannot be ignored but are difficult to estimate using summary‐level genetic statistics, our findings might have been affected by epigenetic bias. Second, since self‐reported data were used in our analyses, misclassification of exposures was inevitable, which might lead to underestimation of the observed associations. Third, given the complexity of DCM biomechanisms and the limited capability of current methods to detect pleiotropy, further studies are warranted, even though the instrumental variables for most risk factors in this study did not exhibit evidence of horizontal pleiotropy. Fourth, it is important to note that the limited number of DCM cases in the FinnGen data set reduces statistical power and makes replication of findings challenging in cohorts with small sample sizes. Fifth, although our study identified several potential drug targets, it should be noted that the analytical results are suggestive rather than conclusive. Finally, most participants in this study were of European ancestry, which may limit the generalizability of the results to other populations.

CONCLUSIONS

Through the strong genetic associations and genetic overlap between lifestyle factors, metabolic factors, and DCM, we found that genes related to hypertension, SBP, BMR, BMI, waist circumference, and hip circumference were associated with an increased risk of DCM, whereas long sleep duration and never smoking were associated with a decreased risk of DCM. We also identified 11 proteins that might serve as drug targets for DCM. Our findings may contribute to a better understanding of the risk factors for DCM, which may facilitate the development of targeted prevention and intervention strategies for DCM. However, further research is still needed to validate these findings.

Sources of Funding

This study was supported by the National Natural Science Foundation of China (Grant No. 82074392 to XL), the Regional Joint project of Natural Science Foundation of Hunan Province (Grant No. 2023JJ50036 to XL), the Key project of Hunan Provincial Department of Education (Grant No. 23A0283 to XL), and the Postgraduate Research Innovation Project of Hunan Province (Grant No. CX20240746 to HW).

Disclosures

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Supporting information

Data S1–S2

Tables S1–S12