Genome-wide analysis of heart failure yields insights into disease heterogeneity and enables prognostic prediction in the Japanese population

Figures

• 

  Fig. 1. Overview of the study design. Flowchart of the study, which encompasses the Japanese GWAS with the BioBank Japan, a trans-ancestry meta-analysis with large-scale European and other population GWAS followed by the downstream analyses. AFR, African; AMR, American; EAS, East Asian; GBMI, Global Biobank Meta-analysis Initiative; GWAS, genome-wide association study; HF, heart failure; HFpEF, heart failure with preserved ejection fraction; HFrEF, heart failure with reduced ejection fraction; H-MAGMA, Hi-C-coupled MAGMA; MAGMA, Multi-marker Analysis of GenoMic Annotation; MIGWAS, miRNA–target gene networks; MRI, magnetic resonance imaging; MVP, Million Veteran Program; NIHF; non-ischemic heart failure; scDRS, single-cell disease relevance score; TWAS, transcriptome-wide association study; UKBB, UK Biobank; VEGAS, versatile gene-based association study.

• 

  Fig. 2. Results of the Japanese GWAS. a Sample overlap between each HF phenotype in the Japanese GWAS. b Miami plot of Japanese GWAS. -log10(P value) on the y-axis is shown against the genomic positions (hg19) on the x-axis. Association signals that reached a genome-wide significance level (P < 5 × 10−8) are shown in red if they are new loci and in blue if previously reported only in MTAG but not in GWAS. Two-sided uncorrected P values were calculated using a logistic regression model, and the genome-wide significance level was applied. Dashed lines represents the genome-wide significance threshold. c Genetic correlation between HF GWAS. * indicates P < 0.05, and ** P < 0.0083 (0.05/6). Two-sided uncorrected P values were derived from the LDSC software. d Genetic correlation between HF and risk factors. * indicates P < 0.05 and ** indicates P < 0.00125 (0.05/40 tested pairs). Two-sided uncorrected P values were derived from the LDSC software. AF, atrial fibrillation; BBJ, Biobank Japan; BMI, body mass index; CAD, coronary artery disease; DBP, diastolic blood pressure; DM, diabetes mellitus; EUR, European; GWAS, genome-wide association study; HDL, high-density lipoprotein cholesterol; HF,…

• 

  Fig. 3. Prioritized genes. a The odds ratios for HF development of the independent signals in the Japanese GWAS, cross-ancestry meta-analysis, and MTAG. The color of each point indicates the corresponding HF phenotype. The size of each point indicates -log10(P value). The shape of each point indicates the type of analysis, BBJ, Meta, or MTAG. The marker inside each point indicates novelty status. b Prioritized genes for each HF phenotype according to the types of analyses. CACNA1D was prioritized in both BBJ and cross-ancestry meta-analysis and is shown in the “Meta-analysis” row. BBJ, BioBank Japan; CM, cardiomyopathy; eQTL, expression quantitative trait loci; HF, heart failure; HFpEF, heart failure with preserved ejection fraction; HFrEF, heart failure with reduced ejection fraction; H-MAGMA, Hi-C-coupled MAGMA; MAGMA, Multi-marker Analysis of GenoMic Annotation; Meta, meta-analysis; MTAG, multi-trait analysis of genome-wide association study; NIHF, non-ischemic heart failure; PoPs, polygenic priority score; sQTL, splicing quantitative loci.

• 

  Fig. 4. Pathway analysis for each HF phenotype. Enriched terms (FDR < 0.05) for each HF phenotype from GO, KEGG, PI, Reactome, and WP. Up to the top 20 enriched terms in each HF phenotype gene set are labeled. The color of the dots indicates the pathway. There was no enriched pathway for HFpEF. CDK, Cyclin-dependent kinases; DDR, DNA damage response; GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes; PI, Pathway Interaction database; PRL, prolactin; SASP, Senescence-Associated Secretory Phenotype; WP, Wiki Pathways.

• 

  Fig. 5. Effects of common variants in cardiomyopathy genes on left ventricular function and long-term HF mortality. a Minor allele frequency of lead SNPs in cardiomyopathy genes for East Asian and European populations according to gnomAD. b Effect size comparison of rs148116 (TTN) between BBJ (4254 cases and 197,577 controls) and European (19,495 cases and 258,943 controls) for HFrEF. Data are presented as estimated coefficients and their 95% confidence interval (CI). Two-sided uncorrected P values were calculated using a logistic regression model, and the Bonferroni-corrected significance level was applied. c Effects of lead variants in cardiomyopathy genes on LVEF in BBJ cohorts (n = 13,416 [TTN, All-cause HF], 8675 [TTN, NIHF], 13,458 [BAG3, All-cause HF], 8711 [BAG3, NIHF], 12,970 [NEBL, All-cause HF], 8384 [NEBL, NIHF]; left panel) and UKBB cohorts (n = 36,041; right panel). Data are presented as estimated coefficients and their 95% CI. The shape of the points in c and d indicates the HF subtypes. Two-sided uncorrected P values were calculated using a logistic regression model, and the Bonferroni-corrected significance level was applied. Red points in c and d indicate statistically…

• 

  Fig. 6. Performance and clinical impact of the HF-PRS. a Each point indicates Nagelkerke’s pseudo R2 for all-cause HF case-control status in the Japanese independent cohort (1837 cases and 11,319 controls). PRS-derivation cohorts are indicated on the x-axis. Two-sided uncorrected P values were calculated using a Cochran-Armitage trend test. b Association between HF-PRS and the onset age of HF. The onset age of HF in individuals with data available (n = 10,810) is shown based on the HF-PRS tertiles. The number of individuals in each tertile is 3603 to 3604. The center line of the box plot indicates the median, the bounds represent the first and third quartiles, and the whiskers reach to 1.5 times the interquartile range. Two-sided uncorrected P values were calculated using a linear regression model. c Kaplan–Meier estimates of cumulative events from cardiovascular mortality (left) and HF death (right) are shown with a band of 95%CI. Individuals are classified into high (top tertile), intermediate (middle tertile), and low PRS (bottom tertile) among individuals without HF at baseline (n = 67,991). Two-sided uncorrected P values were calculated for PRS by Cox proportional hazard analysis…