This study has several limitations. First, our eQTL analyses were limited to single tissues and excluded replicate RNA-seq measurements. A joint analysis with random-effects models68,69 could increase the effective sample size, which would be especially useful for trans-eQTL identification. Second, our GWAS overlap analysis may have failed to identify previously identified genes due to differences in sample size, effect size, variant density, LD structure and imputation quality. For example, our results did not include the MAPT gene for AD because the H1/H2 haplotype separating SNP rs8070723 had an eQTL P value of 1.8 × 10−5 due to our alignment strategy (Supplementary Note). This might have been an issue for other genes as well. Graph-based alignment tools or long-read sequencing methods are required to ultimately determine the true effects on such genes. Third, the GWAS overlap methods we used have known limitations (for example, Supplementary Note). For the MR analysis, we opted to perform single-SNP MR instead of multi-SNP MR (such as inverse-variance weighted70), which requires multiple independent associations per gene. As this was the case for only a limited proportion
