Background: Genome-wide association studies (GWAS) have identified susceptibility loci for a wide range of traits and diseases, yet translating this rapidly expanding body of genetic associations into disease aetiology remains challenging. Because genetic variants often have pleiotropic effects, repeated overlap of independent susceptibility loci across traits may indicate shared biological mechanisms. However, there is currently no systematic, hypothesis-free framework to identify and prioritise non-random overlap across thousands of traits.

Objective: To introduce VIPER (Variant-Informed Epidemiology for Predicting and Exploring Risk Factors), a framework for hypothesis-free identification of non-random genetic overlap across traits, and to demonstrate its application to glioma for the discovery of novel associated traits.

Methods: VIPER integrates GWAS results across studies and represents each trait by independent association signals defined by sentinel variants and their linkage disequilibrium (LD) proxies. Genetic overlap between traits is quantified as the LD-weighted sum of shared susceptibility loci and evaluated against an empirical null distribution to generate a Z-score that accounts for differences in locus number. Prioritised traits are then evaluated using complementary approaches, including two-sample Mendelian randomisation (MR) and prospective Cox models.

Results: VIPER identified intracellular volume fraction (ICVF), a diffusion MRI-derived proxy for neurite density (with lower ICVF indicating lower density), as sharing more susceptibility loci with glioma than expected by chance, with the strongest overlap in the genu of the corpus callosum. Mendelian randomisation analyses indicated that genetically predicted higher ICVF was associated with lower glioma risk (OR = 0.63, 95% CI: 0.51–0.79, p = 4 × 10??). Consistently, prospective analyses in the UK Biobank imaging cohort (57,503 participants with 30 incident cases) showed that lower pre-diagnostic ICVF was associated with increased glioma risk up to five years before diagnosis (HR per SD increase = 0.60, 95% CI: 0.42–0.85, p = 0.004).

Conclusion: We developed VIPER, a variant-informed framework for identifying and prioritising traits showing non-random genetic overlap with disease, and for their downstream evaluation. Applied to glioma, VIPER identified neurite density as a candidate trait, with convergent genetic and prospective evidence linking lower pre-diagnostic neurite density in the genu to increased glioma risk. These findings illustrate how our framework can generate and evaluate biologically meaningful hypotheses and could be applied to other diseases with poorly understood aetiology.