Systematic disease-agnostic identification of therapeutically actionable targets using the genetics of human plasma proteins

Abstract

Proteome-wide Mendelian randomization (MR) has emerged as a promising approach in uncovering novel therapeutic targets. However, genetic colocalization analysis has revealed that a third of MR associations lacked a shared causal signal between the protein and disease outcome, raising questions about the effectiveness of this approach. The impact of proteome-wide MR, stratified by cis-trans status, in the presence or absence of genetic colocalization, on therapeutic target identification remains largely unknown. In this study, we conducted genome-wide MR and cis/trans-genetic colocalization analyses using proteomic and complex trait genome-wide association studies. Using two different gold-standard datasets, we found that the enrichment of target-disease pairs supported by MR increased with more p-value stringent thresholds MR p-value, with the evidence of enrichment limited to colocalizing cis-MR associations. Using a phenome-wide proteogenetic colocalization approach, we identified 235 unique targets associated with 168 binary traits at high confidence (at colocalization posterior probability of shared signal > 0.8 and 5% FDR-corrected MR p-value). The majority of the target-trait pairs did not overlap with existing drug targets, highlighting opportunities to investigate novel therapeutic hypotheses. 42% of these non-overlapping target-trait pairs were supported by GWAS, interacting protein partners, animal models, and Mendelian disease evidence. Additionally, as previously reported, these high confidence target-trait pairs assisted with causal gene identification and helped uncover translationally informative novel biology, especially from trans-colocalizing signals, such as the association of lower intestinal alkaline phosphatase with a higher risk of inflammatory bowel disease in FUT2 non-secretors. Beyond target identification, we used MR of colocalizing signals to infer therapeutic directions and flag potential safety concerns. For example, we found that most genetically predicted therapeutic targets for inflammatory bowel disease could potentially worsen allergic disease phenotypes, except for TNFRSF6B where we observed directionally consistent associations for both phenotypes. Our results are publicly available (https://mk31.shinyapps.io/mr_app2/), enabling others to use proteogenomic evidence to appraise therapeutic targets.

Extensive co-regulation of neighbouring genes complicates the use of eQTLs in target gene prioritisation

“…We also found that precision can be improved (up to 81%) when combining multiple colocalising QTL signals in a Mendelian randomization framework to explicitly take into account the concordance of the causal effect estimates provided by independent genetic variants. This agrees with other recent studies, affirming that combining colocalisation with Mendelian randomisation reduces confounding by LD and improves the sensitivity and specificity of identifying biologically relevant targets [20][21][22].…”

Trans-eQTL mapping prioritisesUSP18as a negative regulator of interferon response at a lupus risk locus

“…Thus, plasma levels of CRP do not seem to have a direct causal effect on CAD risk, but can still act as a molecular readout (biomarker) of the IL6R-mediated inflammatory response that does seem to have a causal effect 63 . These observations suggest that widespread horizontal pleiotropy in gene regulatory networks could be a general property of trans-QTLs and could help explain why using trans-pQTL signals in Mendelian randomisation analysis has had low specificity for identifying known drug targets 64,65 . Instead, we propose that target genes identified from large-scale trans-QTL studies could be better thought of as drug response biomarkers for drugs targeting the cis gene responsible for the trans association 8 .…”