Abstract Background The proteome is a key source of therapeutic targets. We conducted a comprehensive Mendelian randomization analysis across the proteome to identify potential protein markers and therapeutic targets for abdominal aortic aneurysm (AAA). Methods and Results Our study used plasma proteomics data from the UK Biobank, comprising 2923 proteins from 54 219 individuals, and from deCODE Genetics, which measured 4907 proteins across 35 559 individuals. Significant proteomic quantitative trait loci were used as instruments for Mendelian randomization. Genetic associations with AAA were sourced from the AAAgen consortium, a large‐scale genome‐wide association study meta‐analysis involving 37 214 cases and 1 086 107 controls, and the FinnGen study, which included 3869 cases and 381 977 controls. Sequential analyses of colocalization and summary‐data‐based Mendelian randomization were performed to verify the causal roles of candidate proteins. Additionally, single‐cell expression analysis, protein–protein interaction network analysis, pathway enrichment analysis, and druggability assessments were conducted to identify cell types with enriched expression and prioritize potential therapeutic targets. The proteome‐wide Mendelian randomization analysis identified 34 proteins associated with AAA risk. Among them, 2 proteins, COL6A3 and PRKD2, were highlighted by colocalization analysis, summary‐data‐based Mendelian randomization, and the heterogeneity in an independent instrument test, providing the most convincing evidence. These protein‐coding genes are primarily expressed in macrophages, smooth muscle cells, and mast cells within abdominal aortic aneurysm tissue. Several causal proteins are involved in pathways regulating lipid metabolism, immune responses, and extracellular matrix organization. Nine proteins have already been targeted for drug development in diabetes and other cardiovascular diseases, presenting opportunities for repurposing as AAA therapeutic targets. Conclusions This study identifies causal proteins for AAA, enhancing our understanding of its molecular cause and advancing the development of therapeutics. Keywords: abdominal aortic aneurysm, biomarker, drug target, plasma protein, proteome‐wide Mendelian randomization Subject Categories: Aneurysm __________________________________________________________________ Nonstandard Abbreviations and Acronyms HEIDI heterogeneity in independent instrument MR Mendelian randomization pQTL protein quantitative trait loci SMR summary‐data‐based Mendelian randomization STRING Search Tool for the Retrieval of Interacting Genes UKB‐PPP UK Biobank Pharma Proteomics Project Clinical Perspective. What Is New? * The current study identified 34 circulating plasma proteins that were causally linked to abdominal aortic aneurysms. * COL6A3 and PRKD2 were supported by Mendelian randomization, colocalization analysis, and summary‐data‐based Mendelian randomization analysis as plasma circulating proteins strongly associated with the risk of abdominal aortic aneurysm. What Are the Clinical Implications? * These findings further support the evaluation of COL6A3 and PRKD2 as potential pharmacological targets for abdominal aortic aneurysm treatment. Abdominal aortic aneurysm (AAA) is a severe vascular disease characterized by a permanent dilation of the arterial wall, measuring 50% or more compared with the normal diameter.[34] ^1 , [35]^2 As the disease progresses, it can lead to aortic rupture, which has a mortality rate >80%.[36] ^1 Despite significant advancements in surgical and endovascular treatments for AAAs over the past decade, no proven pharmacological interventions exist to slow AAA growth or prevent aortic rupture.[37] ^3 Population‐based and clinical studies have identified obesity, smoking, unhealthy diets, and sedentary lifestyles as major modifiable risk factors for AAA.[38] ^4 , [39]^5 , [40]^6 To enhance genetic understanding, several large‐scale genome‐wide association studies (GWASs) have been conducted, revealing >140 loci associated with AAA, thereby increasing the usefulness of genetic prediction.[41] ^4 , [42]^6 , [43]^7 With the advent of high‐throughput techniques for serum protein detection and quantification, numerous studies have explored protein associations with AAA risk to elucidate the molecular pathological basis.[44] ^4 , [45]^8 A population‐based study identified 118 plasma proteins linked to AAA in the discovery phase, but only 2 proteins demonstrated a causal relationship with AAA.[46] ^8 Another larger study found causal relationships between 23 circulating protein levels and genetic susceptibility to AAA.[47] ^4 However, these studies often had limitations, such as focusing on a limited number of candidate proteins, using observational designs, or having small sample sizes, thereby constraining the understanding of the causal role of protein markers in AAA risk. The increasing use of high‐throughput proteomics platforms in large‐scale genotyped biobanks presents new opportunities for deriving biological insights from GWAS data.[48] ^9 , [49]^10 Notably, the Mendelian randomization (MR) approach facilitates the identification of causal relationships between exposures and diseases, effectively mitigating the potentially confounding effects of environmental factors.[50] ^11 , [51]^12 In this study, we conducted the largest investigation to date using MR to identify potential causal effects of circulating proteins on AAA phenotypes. We used blood‐based proteomics data from 2 large, independent cohorts (UK Biobank and deCODE Genetics) and well‐powered GWASs for AAA. Recognizing that MR alone may be insufficient for pinpointing credible proteins in causal pathways, we also used colocalization, summary‐data‐based MR (SMR), and the heterogeneity in independent instrument (HEIDI) test. Furthermore, single‐cell‐type expression analysis was used to detect the enrichment of these proteins in specific cell types within AAA tissue. Finally, we performed druggability evaluations to explore their potential as therapeutic targets for AAA. METHODS Data Availability The GWAS summary statistics for AAA based on the AAAgen Consortium can be accessed at [52]https://csg.sph.umich.edu/willer/public/AAAgen2023/. The data of FinnGen can be accessed at [53]https://www.finngen.fi/en. The protein quantitative trait loci (pQTL) data can be accessed at [54]https://registry.opendata.aws/ukbppp/ and [55]https://www.decode.com/summarydata/. The single‐cell RNA sequencing data of AAA tissue can be accessed at [56]https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE166676. Proteomic Data Source and the Processing of Genetic Instruments Two large‐scaled proteomic studies, UKB‐PPP (UK Biobank Pharma Proteomics Project)[57] ^13 and deCODE Genetics,[58] ^14 were used to extract summary statistics of genetic associations with plasma proteins. The UKB‐PPP involves profiling plasma pQTL data for 2923 proteins using the Olink Explore 3072 platform from a cohort of 54 219 UK Biobank participants. For deCODE Genetics, 4907 proteins in 35 559 individuals were measured to obtain their plasma pQTL data using the SOMAscan version 4 assay (Table [59]S1). We used pQTL from circulating plasma from the 2 studies above as genetic instruments for analysis. The platform identification for each protein from each study was mapped to the gene symbol and unified based on annotations provided by the original studies and manual review.[60] ^15 To obtain pQTLs for MR analysis, all summary statistics were subjected to a quality control (QC) filtering process that excluded insertion–deletion variants, variants with a minor allele frequency <0.001, palindromic variants with minor allele frequency >0.42, and variants in trans‐pQTL, including the extended major histocompatibility complex region (Chr6: 25–34 Mb). The inclusion of trans‐pQTL helps to enhance statistical capacity and evaluate and control potential pleiotropic effects through sensitivity analysis. We selected variants that achieved genome‐wide significance (P>5×10^−8) for linkage disequilibrium pruning and then divided genome‐wide significant variants into cis‐pQTL and trans‐pQTL, where cis‐pQTLs are variants within 1 Mb upstream and downstream of the associated protein‐coding genes (ensemble 108 annotations). QC‐filtered genome‐wide significant variants outside of the cis region are considered trans‐pQTLs.[61] ^16 , [62]^17 The single‐nucleotide polymorphisms (SNPs) corresponding to each pQTL variant were unified according to the human genome Build 37 (National Center for Biotechnology Information GRCh37) genome coordinates, based on annotations provided by the original study and manual reviews ([63]https://github.com/HaobinZhou/Get_MR/blob/main/2.0/Get_MR2.0.r). Linkage disequilibrium clumping was then conducted to identify independent pQTLs for each protein (r ^2<0.001). The R ^2 and F statistic (R ^2=2×Expected average frequency (EAF)×(1−EAF)×β; F=R ^2×(N−2)/(1−R ^2)) were used to estimate the strength of genetic instruments, where R ^2 was the proportion of the variability of the protein levels explained by each genetic instrument.[64] ^18 Data Sources for AAA Data on the associations of protein‐associated SNPs with AAA were obtained from the AAAgen Consortium[65] ^4 and the FinnGen study ([66]https://r10.finngen.fi/). The AAAgen Consortium included 14 studies with a total of 37 214 cases and 1 086 107 controls of European descent. We used the latest release data on AAA from the FinnGen study R10 in this analysis, which comprised 3869 cases and 381 977 controls. There were no sample overlaps between the 2 outcome data sets. In MR analysis, we treated the AAAgen Consortium as the discovery study and the FinnGen R10 study as the replication. The basic information for these data sets is shown in Table [67]S2. Statistical Analysis Proteome‐Wide MR Analysis We performed MR analyses using the TwoSampleMR package[68] ^19 , [69]^20 within the R environment. When only 1 instrument was available for a particular protein, we applied the Wald ratio method. The inverse‐variance weighted method was used to obtain the MR effects estimates for proteins with >1 instrument. The heterogeneity test was performed to assess the heterogeneity of the genetic instruments based on the Q statistic.[70] ^21 To estimate horizontal pleiotropy, the MR‐Egger intercept test was used. Bonferroni correction was used for multiple testing correction, with P<2.07×10^−5 (0.05/2410) as the significance level for cis‐pQTLs and with P<1.70×10^−5 (0.05/2938) as the significance level for trans‐pQTLs. Replication MR analysis was further performed for the identified proteins based on AAA GWAS summary data from the FinnGen study. A false discovery rate (FDR) <0.05 was defined as the significance level for replication. The Benjamini‐Hochberg procedure was applied for multiple comparisons to deal with type I error. Colocalization Analysis Colocalization analysis was used to test whether the identified associations of proteins with AAA were driven by linkage disequilibrium, providing further validation of the MR results. We used summary statistics of proteins and AAAgen meta‐GWASs to perform Bayesian colocalization analysis using the coloc package.[71] ^22 For each protein, SNPs within ±500 kb upstream and downstream of their corresponding gene were examined for colocalization with AAA. The analysis considered 5 hypotheses: (1) no causal variant for either protein or AAA in the locus (H0); (2) association with protein only (H1); (3) association with AAA only (H2); (4) both protein and AAA associated, but with distinct causal variants (H3); and (5) both protein and AAA associated, sharing the same causal variant (H4).[72] ^23 A posterior probability for H4 >80% under different priors and windows was considered strong evidence of colocalization. The LocusCompareR package was used to visualize colocalization results.[73] ^24 SMR Analysis SMR analysis was conducted as complementary evidence to verify the causal associations between proteins and AAA.[74] ^25 , [75]^26 The HEIDI test, using multiple SNPs (up to 20 SNPs) in a region, was used to distinguish proteins associated with AAA risk due to a shared genetic variant rather than genetic linkage.[76] ^25 , [77]^26 The SMR and HEIDI tests were performed using SMR software (SMR version 1.3.1). A P value <1.47×10^−3 (0.05/34) was defined as the significance level for SMR. A P value of the HEIDI test >0.05 indicated that the association of protein and AAA was not driven by linkage disequilibrium. Single‐Cell‐Type Expression Analysis The cell‐type‐specific expression of target genes with evidence for a potential causal effect on AAA at the plasma protein levels was further evaluated using single‐cell RNA sequencing data of human AAA profiled from the Gene Expression Omnibus database.[78] ^27 The raw single‐cell RNA‐seq data were analyzed using the Seurat package. Cells meeting these criteria were removed: (1) <500 genes (low quality) or >5000 genes detected (potential doublets) and (2) >10% of unique molecular identifiers originating from the mitochondrial genome. Data normalization and batch effect removal were performed using the Seurat package. Cell clusters were annotated based on the SingleR package and references to relevant articles.[79] ^28 , [80]^29 , [81]^30 To examine