GWAS-by-Subtraction Resolves Coronary Artery Disease-Related and Model-Defined Residual Genetic Components of Ischaemic Stroke

Figures

• 

  Figure 1. Study workflow and subtraction model. (a) Primary GWAS-by-subtraction analysis, robustness analyses and biological characterisation. The minus and equals signs summarise model-based partitioning rather than arithmetic subtraction of association estimates [6]. (b) Per-SNP Cholesky model. SNP effects were estimated on F1 and F2; arrows from the factors to observed traits show standardised loadings. The primary model fixed Cov(F1,F2) to zero, latent variances to one and observed residual variances to zero. F1 denotes the CAD-related shared component; F2 denotes the model-defined residual AIS component.

• 

  Figure 2. Genome-wide association results for F1 and F2. Manhattan and quantile–quantile plots show two-sided Wald p values for 1,254,765 variants. The dashed line marks p = 5 × 10−8. Red points denote genome-wide significant variants. The LDSC intercepts were 0.9916 for F1 and 1.0047 for F2.

• 

  Figure 3. Robustness of the F2 architecture. (a) Genome-wide F2 Z scores from the primary and alternative-CAD analyses; the plot shows a deterministic sample of 80,000 variants and reports the correlation calculated across all matched valid variants. (b) Effect estimates for six prespecified sentinel variants after replacing the primary CAD GWAS with the Aragam CAD GWAS. (c) The primary analysis and FinnGen F2 effect estimates for 13 exact-matched lead variants. FinnGen is interpreted as a cross-implementation comparison with potential participant overlap because FinnGen contributed to GIGASTROKE.

• 

  Figure 4. Genetic correlations of F1 and F2. Points show LDSC genetic correlations and bars show 95% confidence intervals. Red circles denote F1; teal triangles denote F2. Formal F1–F2 contrasts used the complete three-trait LDSC sampling covariance matrix and were corrected across nine prespecified traits. Atrial fibrillation, marked by an asterisk, was a secondary joint analysis with an unadjusted formal contrast. Candidate VTE/DVT analyses were excluded because the archived external files did not pass phenotype and sample-size provenance quality control.

• 

  Figure 5. Molecular and regulatory characterisation of the F2 component. (a) Bonferroni-significant F2 MAGMA genes. (b) F2 arterial gene–tissue associations passing p(SMR) < 8.149 × 10−7 and p(HEIDI) > 0.01. (c) Maximum variant PIP at each of the 14 FUMA-defined F2 loci; labels give purity-qualified 95% credible-set size. (d) PP4 for selected eQTL and pQTL colocalisation results; the dashed line denotes PP4 = 0.75. These analyses prioritise compatible local signals for subsequent mechanistic testing.