Cross-omic dissection reveals locus-specific heterogeneity and antagonistic pleiotropy between Alzheimer’s disease and type 2 diabetes

Abstract Observational studies associate type 2 diabetes (T2D) with increased dementia risk; however, the specificity of this relationship to Alzheimer’s disease (AD) and its biological basis remains unresolved. We apply an integrative cross-omic framework to dissect genetic links between AD and T2D. Genome-wide analyses reveal a modest positive genetic correlation and robust polygenic sign concordance of AD with T2D. High-resolution analyses demonstrate locus-specific heterogeneity, with coexisting positive and predominantly negative correlations, and strong inverse associations at APOE and HLA . Cross-trait GWAS meta-analyses indicate that most genome-wide significant signals reflect trait-specific effects, with only a limited set of variants supported in both AD and T2D. Colocalisation reveals distinct causal variants at most shared loci. Gene-based analyses highlight convergence at functional genes, including PLEKHA1, VKORC1, ACE, and APOE , without implying concordant variant-level effects. Bidirectional Mendelian randomisation (MR) shows no evidence of a causal relationship between AD and T2D in either direction. Summary-data MR prioritises genes whose expression or methylation affects both AD and T2D, mostly with opposing effects. Only PLEKHA1 (eQTL) and CAMTA2 (mQTL) show concordant positive associations. Five genes, GALNT10, HSD3B7, BCKDK, KAT8 , and ACE , are supported across both regulatory layers, while numerous signals cluster within a regulatory hotspot at 16p11.2, supporting convergent transcriptional and epigenetic involvement, despite directional divergence. These results refine the AD–T2D relationship; rather than a simple shared-risk model, overlap reflects locus-specific heterogeneity and cross-omic convergence often showing opposing effects on AD versus T2D risk, consistent with antagonistic pleiotropy.