Study design

The study design is illustrated in Fig. 1. First, we conducted a two-sample MR analysis to identify drug target genes utilizing publicly available eQTL datasets in blood and brain as genetic instruments (IVs) and GWAS summary statistics of SBP as the outcome. Second, we employed a two-sample MR analysis to estimate the relationship between each drug target gene expression and the risk of AD. Third, we performed sensitivity analyses to further validate the MR associations, including detecting reverse causality, assessing horizontal pleiotropy, Bayesian colocalization, scanning phenotypes, and protein quantitative trait loci (pQTL) analysis. Finally, we verified our findings with additional eQTL datasets and GWAS summary statistics for AD.

Fig. 1

The MR framework relies on three key assumptions: (i) relevance, where genetic variants should be significantly associated with exposures; (ii) exclusiveness, where genetic variants are not linked to potential confounders; and (iii) independence, where genetic variants affect outcomes only through the exposures. The study adhered to the Strengthening the Reporting of Observational Studies in Epidemiology-Mendelian randomization reporting guidelines (STROBE-MR) [16] (Table S1). The online tool mRnd (https://shiny.cnsgenomics.com/mRnd/) was utilized to assess the statistical power and calculate the results.

Identification of drug target genes

The commonly prescribed antihypertensive medications, including ACEis, ARBs, beta-blockers (BBs), CCBs, diuretics, and other antihypertensive agents, were included in the analysis. Potential therapeutic genes were obtained from the DrugBank (https://go.drugbank.com/) [17] and ChEMBL (https://www.ebi.ac.uk/chembl/) databases. Genomic regions associated with these genes were retrieved from GeneCards (https://www.genecards.org/). To demonstrate that changes in gene expression are associated with reduced blood pressure due to drug exposure, we conducted summary-data-based Mendelian Randomization (SMR) analysis with blood gene expression (from the eQTLGen data) as exposure [12] and systolic blood pressure (SBP) as the positive outcome, with summary data from a GWAS of SBP in 757,601 individuals of European ancestry sourced from the UK Biobank and the International Consortium of Blood Pressure Genome Wide Association Studies (ICBP) [18]. Genes with blood expression associated with SBP at least nominal significance (i.e., P < 0.05) were included in further analysis. The SMR method estimated SBP change per standard deviation (SD) increment in gene expression.

Blood and brain expression quantitative trait loci

Our analysis utilized publicly available data from the eQTLGen consortium (comprising 31,684 individuals) to identify significant (minor allele frequency > 1%; p < 5 × 10− 8) single-nucleotide polymorphisms (SNPs) associated with the expression of antihypertensive drug target genes in blood. Only cis associations are available in the eQTLGen data (distance between SNP and gene is < 1 MB). The eQTL data are scaled to a 1-SD change in gene expression per additional effect allele. The strength of SNP instruments was assessed using the F statistic. The discovery brain eQTL data were obtained from the PsychENCODE consortium [14], which included 1,387 prefrontal cortex, predominantly of European samples. In addition, we used whole blood and brain-related eQTLs from the Genotype-Tissue Expression (GTEx) project V8 [13] and BrainMeta V2 [15], including 2,865 cortex European samples, to validate our findings. Further details on the eQTLs are provided in Table S2.

GWAS summary statistics of Alzheimer’s disease

We obtained publicly available case-control GWAS summary statistics for AD, including 111,326 clinically diagnosed cases and 677,663 controls [19]. This dataset includes contributions from various European GWAS consortia and a new dataset from 15 European countries. Moreover, an AD GWAS from the FinnGen cohort, which included 9,301 cases and 367,976 controls, was used to validate our finding [20]. Details for accessing the summary statistics used in the current analyses are provided in Table S3.

Summary-level mendelian randomization

We conducted a two-sample MR analysis to estimate the association between target gene expression and the risk of AD. The association between AD and each target gene was analyzed after harmonizing the genetic data from the aforementioned AD GWAS. The SMR approach involved selecting the most significant or multiple associated eQTL SNP (located near the target gene, i.e., cis-eQTL SNP) as an instrument. When a gene probe has more than five instruments, the Heterogeneity in Dependent Instruments (HEIDI) test determines whether the observed association is due to two distinct SNPs in linkage disequilibrium, each independently associated with gene expression and AD risk. A HEIDI P value threshold of ≥ 0.05 indicated the reliability of the gene probe.

SMR estimates were calculated using SMR software version 1.3.1 (https://cnsgenomics.com/software/smr/#Overview), with a p-value threshold of < 0.0024 considered strong evidence (Bonferroni correction for association testing of 21 genes).

Sensitivity analysis

First, Steiger filtering was used to determine the directionality in the eQTL-GWAS association, with significance set at P < 0.05. Second, MR estimates may be biased if genetic variants influence the outcome through a pleiotropic pathway unrelated to the exposure of interest (i.e., target gene expression). This can distort MR tests, leading to inaccurate causal estimates, reduced statistical power, and potential false-positive causal relationships. Therefore, we assessed horizontal pleiotropy by examining the available associations with other nearby genes (within a 2 MB window) for each genetic instrument. We then performed SMR analysis to investigate whether the significant association of these nearby genes with the genetic instrument was related to the risk of AD. Only cis-genes were considered for the analysis because of their proximity to the gene expression. Third, we identified traits and diseases associated with the causal variants using ‘phenoscanner V2’ for the significant genes and examined the relevance of these phenotypes to AD risk in the literature. Traits were selected based on a p-value less than 5 × 10− 8 and an r2 > 0.8 with European ancestry. Fourth, we employed Bayesian co-localization analyses with the ‘coloc’ package to assess the probability of shared causal variants between traits. The focus was on hypothesis 4 (PPH4), which suggests a shared variant association for both eQTL and Alzheimer’s disease. A gene-based PPH4 > 80% indicated significant co-localization. Finally, we validated our findings using pQTL data for the gene of interest (a significant gene in the association between eQTL and AD). pQTL data was obtained from the UK Biobank Pharma Proteomics Project (UKB-PPP), which includes 54,219 UK Biobank individuals [21].

MR analysis to estimate the association between blood pressure and Alzheimer’s disease

We also performed a two-sample MR analysis to estimate the relationship between blood pressure and the risk of AD using the TwosampleMR package version 0.5.6 in R. Exposure IVs were selected based on common SNPs (minor allele frequency [MAF] > 1%) that exhibited significant associations with BP at a genome-wide level of significance (p < 5 × 10− 8 ) but were not associated with AD risk (p > 0.05). These SNPs were then subjected to linkage disequilibrium (LD) clumping using a stringent threshold (r2 < 0.001; distance threshold, 10,000 kb) with the ieugwasr package version 0.1.5 in R. For palindromic SNPs, forward strand alleles were determined using allele frequency information. IVs of BP were listed in Table S4. After applying the above filtering criteria, the SNPs were considered as IVs of SBP and analyzed further. The primary analysis utilized inverse variance weighted (IVW), while secondary analyses included Mendelian randomization Egger Regression (MR Egger), weighted median, weighted mode, and Mendelian Randomization Pleiotropy RESidual Sum and Outlier (MR-PRESSO). IVW provided the most accurate estimates of causality in the absence of horizontal pleiotropy for all SNPs. The weighted median and weighted mode provide robust estimates in the presence of invalid instruments, while MR-Egger and MR-PRESSO (NbDistribution = 5000, SignifThreshold = 0.05) account for horizontal pleiotropy in genetic variation and tests for pleiotropy and heterogeneity in MR estimates.

Data availability

Data are available in public, open-access repositories. Blood pressure GWAS statistics can be downloaded at https://grasp.nhlbi.nih.gov/FullResults.aspx. Alzheimer’s disease meta-analysis GWAS statistics are stored at https://www.ebi.ac.uk/gwas/ under accession no. GCST90027158. GWAS summary statistics of the FinnGen (R9 release) were obtained from https://www.finngen.fi/en/access_results. eQTLGen data can be accessed at https://www.eqtlgen.org/. GTEx V8 data can be accessed at https://gtexportal.org/. PsychENCODE data and BrainMeta V2 data can be accessed at https://yanglab.westlake.edu.cn/software/smr/#Overview. ACE Protein level data can be accessed at http://ukb-ppp.gwas.eu.