A protein-prioritised, colocalisation-informed genetic framework for safety hypothesis generation across inflammatory bowel disease therapeutic pathways

Human genetics can help prioritise target–outcome safety hypotheses before broad therapeutic exposure, but the strength and translational meaning of such signals depend on molecular proxy type, locus specificity and instrument architecture. We developed a protein-prioritised, colocalisation-informed framework for inflammatory bowel disease (IBD) therapeutic pathways. Twenty-five cis-pQTL or blood cis-eQTL sets represented 18 IBD therapeutic genes across 76 prespecified FinnGen R13 outcomes. Because stringent cis-QTL selection yielded sparse instruments, the workflow combined single- or few-variant Mendelian randomisation (MR) with false-discovery-rate (FDR) control, objective follow-up, sensitivity analyses, instrument-centred colocalisation and variant annotation. Global Benjamini–Hochberg correction covered 1,824 estimates; corrections within 40 narrow outcome domains were secondary. Estimates with global q < 0.05, a risk-increasing direction under target lowering and location outside the major histocompatibility complex entered follow-up. Colocalisation used coloc::coloc.abf() in ± 250-kb windows, with posterior probability of a shared causal variant (PP.H4) > 0.5 as the criterion. Protein and expression results were classified as P1/P2 and E1/E2. Among 1,900 planned tasks, 1,824 were analysable; 1,368 (75.0%) were single-variant Wald ratios and 456 were inverse-variance-weighted estimates. Global FDR identified 100 estimates; 20 more met narrow-domain FDR only. Fifteen tasks representing 13 target–outcome pairs entered follow-up. Protein-level P1 hypotheses were ITGB7–multiple sclerosis/demyelination (OR 2.35, global q = 0.003, PP.H4 = 0.935) and TNFRSF1A–myocardial infarction in UKB-PPP (OR 1.56, q = 0.021, PP.H4 = 0.643) and deCODE (OR 1.37, q = 0.021, PP.H4 = 0.677); both TNFRSF1A estimates used rs4149584/R92Q. Expression-level E1 hypotheses were IL23A–psoriasis (OR 4.13, q = 7.47 × 10⁻⁹, PP.H4 = 0.919) and TYK2–psoriasis (OR 1.53, q = 1.10 × 10⁻¹¹, PP.H4 = 0.984). IL12B–psoriasis and TNFSF15–lower-extremity deep-vein thrombosis were P2 because PP.H4 was below 0.5. Integrating sparse-instrument MR with colocalisation and variant annotation produced an auditable, molecular-layer-stratified set of target–outcome hypotheses. The principal protein-level and expression-level loci define focused priorities for mechanistic validation and evaluation in clinical safety datasets. Not applicable.