Plasma proteome-wide Mendelian randomization reveals the association of extracellular matrix proteins with abdominal aortic aneurysm

Abstract

Objective: Abdominal aortic aneurysm (AAA) is a common and life-threatening vascular disease. Genetic studies have identified numerous risk loci, many potentially encoding plasma proteins. However, the causal effects of plasma proteins on AAAs have not been thoroughly studied. We used genetic causal inference approaches to identify plasma proteins that have a potential causal impact on AAAs.

Methods: Causal inference was performed using two-sample Mendelian randomization (MR). For AAAs, we utilized recently published summary statistics from a multi-population genome-wide association meta-analysis including 39,221 individuals with and 1,086,107 individuals without AAAs from 14 cohorts. We used protein quantitative trait loci (protein quantitative trait loci) identified in two large-scale plasma-proteomics studies (deCODE and UKB-PPP) to generate genetic instruments. We tested 2783 plasma proteins for possible causal effects on AAAs using two-sample MR with inverse variance weighting with common sensitivity analyses.

Results: MR identified 90 plasma proteins associated with AAAs at a false discovery rate <0.05, with 25 supported by colocalization analysis. Among those supported by both MR and colocalization were proteins such as PCSK9 (odds ratio [OR], 1.3; 95% confidence interval [CI], 1.2-1.4; P < 1e-10), LTBP4 (OR, 3.4; 95% CI, 2.6-4.6; P < 1e-10), and COL6A3 (OR, 0.6; 95% CI, 0.5-0.7; P < 1e-6). Gene Ontology analysis revealed enrichment of proteins (extracellular matrix; OR, 7.8; P < 1e-4), some with maximal mRNA levels in aortic tissue. Bi-directional MR suggested plasma level changes were not caused by liability to AAA itself. Colocalization analysis showed that an aortic expression quantitative trait locus for COL6A3, and a splicing quantitative trait locus for LTBP4 colocalized with their respective plasma pQTLs and AAA signals.

Conclusions: Our results highlight proteins and pathways with potential causal effects on AAAs, providing a foundation for future functional experiments. These findings suggest a possible causal pathway whereby genetic variation affecting extracellular matrix proteins expressed in the aortic wall cause their levels to change in blood plasma, influencing development of AAAs.

Keywords: AAA; COL6A3; ECM; LTBP4; Mendelian randomization.

Fig 1

Mendelian randomization (MR) identifies 90 proteins associated with abdominal aortic aneurysms (AAAs). Instruments were generated and associations were tested for 2783 plasma proteins across the deCODE and UKB proteomics studies. The effect estimate is the natural logarithm of the odds ratio (OR). Significance was determined at false discovery rate (FDR) <0.05 in each dataset. Proteins with P-adj<1e-6 are labeled.

Fig 2

Colocalization analysis provides additional evidence for a causal association between protein levels and abdominal aortic aneurysms (AAAs). Bayesian colocalization results for all plasma proteins with significant associations with AAA (false discovery rate [FDR] <0.05). The hypotheses are as follows: H0, no causal variant for either trait; H1, causal variant for plasma levels only; H2, causal variant for AAA only; H3, two distinct causal variants; and H4, one shared causal variant for plasma levels and AAAs. Twenty-five distinct plasma proteins had posterior probability for H4 >0.7, showing that the plasma protein levels and AAAs likely share causal variants.

Fig 3

Biological properties of proteins associated with abdominal aortic aneurysms (AAAs). A, The 25 proteins supported by Mendelian randomization (MR) and colocalization were tested for significant overrepresentation of pathways and gene ontology (GO) terms. The most significant reactome pathway was “Molecules Associated with Elastic Fibers” (odds ratio [OR], 29; P < 1e-3), whereas the most significant GO cellular component term was “Collagen-Containing Extracellular Matrix” (OR, 7.8; P < 1e-4). All terms and pathways shown were significant at false discovery rate [FDR] <0.05. B, Protein-protein interactions (PPIs) among proteins supported by both MR and colocalization. The PPIs were identified in the STRING database. Increasing edge thickness indicates increasing strength of supporting evidence for PPI. C, The expression levels of extracellular matrix (ECM)-associated proteins identified by MR and colocalization across tissues. Expression levels are shown relative to expression levels in the tissue with highest expression (among tissues shown). Median mRNA levels of the genes (transcripts per million [TPM]) in each tissue were Log₂(TPM + 1) transformed and divided by the Log₂(TPM + 1) transformed median TPM of the tissue with maximal expression. A value of 1 indicates that the tissue has the highest expression level for that gene of the 18 tissues analyzed. Several of the ECM proteins show some of their highest expression levels in aortic tissue.

Fig 4

Regional association plots for COL6A3. Plots demonstrating the colocalization between variants associated with mRNA levels of COL6A3 in aorta, plasma levels of COL6A3, and abdominal aortic aneurysms (AAAs). The posterior probability for the colocalization of all three traits is 0.84.

Fig 5

Regional association plots for LTBP4. Regional association plots demonstrating colocalization of splicing quantitative trait locus (sQTL) and protein quantitative trait loci (pQTL) variants for LTBP4, with variants associated with abdominal aortic aneurysms (AAAs). The posterior probability for the colocalization of all four traits is 0.89.

Fig 6

A causal model for abdominal aortic aneurysms (AAAs) through extracellular matrix (ECM) proteins. Genetic variation alters the expression or splicing of ECM proteins, which leads to changes in the aortic wall. These aortic ECM changes lead to AAAs, while simultaneously causing ECM proteins to leak out into the aortic lumen. Alternatively, the ECM proteins in circulation themselves cause AAA, rather than the aortic ECM changes. Both possibilities, or a combination of the two, are consistent with our findings and the assumptions of Mendelian randomization (MR). pQTL, Protein quantitative trait loci. This figure was generated using BioRender.