Fig. 5. Cross-population comparison of the LDL-C, HDL-C, and TG marginal effect sizes of the variants reported in Graham et al.³.

A EUR vs AFR, no $\mathit{G}\times \mathit{E}$ interaction or mediation. B EUR vs AFR, $\mathit{G}\times \mathit{E}$ interaction or mediation. C EUR vs HIS, no $\mathit{G}\times \mathit{E}$ interaction or mediation. D EUR vs HIS, $\mathit{G}\times \mathit{E}$ interaction or mediation. E EUR vs EAS, no $\mathit{G}\times \mathit{E}$ interaction or mediation. F EUR vs EAS, $\mathit{G}\times \mathit{E}$ interaction or mediation. G AFR vs HIS, no $\mathit{G}\times \mathit{E}$ interaction or mediation. H AFR vs HIS, $\mathit{G}\times \mathit{E}$ interaction or mediation. I AFR vs EAS, no $\mathit{G}\times \mathit{E}$ interaction or mediation. J AFR vs EAS, $\mathit{G}\times \mathit{E}$ interaction or mediation. K HIS vs EAS, no $\mathit{G}\times \mathit{E}$ interaction or mediation. L HIS vs EAS, $\mathit{G}\times \mathit{E}$ interaction or mediation. The variants with no $\mathit{G}\times \mathit{E}$ interaction or mediation are those not included in Supplementary Data S3a. The variants with $\mathit{G}\times \mathit{E}$ interaction or mediation are those in Supplementary Data S3a. We only included independent variants. The shadow error bands represent the 95% confidence intervals. Clearly the variants without $\mathit{G}\times \mathit{E}$ interactions or mediations have substantially larger cross-population correlations than the variants with $\mathit{G}\times \mathit{E}$ interactions or mediations, suggesting that $\mathit{G}\times \mathit{E}$ interactions or mediations contribute the marginal effect size heterogeneity across populations. (European (EUR), African (AFR), Hispanics (HIS), Eastern Asian (EAS)).