Locus 9p21.3 CAD Risk Across Diverse Populations: LD and Allelic Heterogeneity Explain Variation Across Ancestries and Reduced Replication in Africans

Abstract

BackgroundThe 9p21.3 locus was the first genome-wide significant signal for coronary artery disease (CAD) and replicates across multiple non-African populations, yet is absent in African ancestry cohorts. We hypothesized that ancestry-specific linkage disequilibrium (LD) and haplotype structure, rather than allele frequency or power alone, explain this discrepancy. MethodsWe analyzed multi-ancestry data from European, East Asian, South Asian, Middle Eastern, African, and Admixed American groups. Within each ancestry, we performed CAD common-variant associations at 9p21.3, rare-variant tests in whole-genome-sequenced Europeans, local-ancestry inference (LAI) stratified associations, conditional and haplotype analyses, and pleiotropy assessments. ResultsCAD associations at 9p21.3 were robust in Europeans, East Asians, Middle Eastern, and Admixed Americans ancestry groups, nominal in South Asians, and absent in Africans despite similar allele frequencies for lead-associated SNPs in other ancestry groups. LAI-stratified analyses showed strong signals in African heterozygotes carrying 9p21.3 European chromosomes but not in 9p21.3-homozygous Africans. The association is due to highly common variants; rare variants did not explain the locus signal. LD blocks were extended in non-Africans but fragmented into smaller blocks in Africans, with marked divergence by fixation index. Fine-mapping identified multiple independent signals, including an East Asian-specific haplotype absent in Europeans and Africans. Hamming-distance analyses revealed that risk alleles are dispersed across multiple haplotypes in homozygous African groups, consistent with weaker LD and restricting risk variation at this locus in that group. PheWAS mirrored these ancestry-specific patterns across metabolic traits. ConclusionsThe non-replication of 9p21.3 in African ancestry is not due to absent risk alleles or poorer imputation of missing risk alleles due to weaker LD; rather, it reflects greater haplotype diversity that disperses risk alleles across configurations, attenuating case-control contrast. These findings illustrate how ancestry-specific LD and multi-causal haplotype architecture modulate association detectability, with broader implications for initial association detection and fine-mapping.