Integrated WGS-based noncoding, splicing, and coding rare-variant burden in familial hypercholesterolemia genes for prediction of lipid levels and ASCVD risk

Background Rare variants in familial hypercholesterolemia (FH) genes play an important role in lipid dysregulation and cardiovascular risk; however, their combined effects across noncoding regulatory, splicing, and coding regions remain insufficiently characterized at the population level. We aimed to develop an integrated whole-genome sequencing (WGS)–based rare-variant burden framework in major FH genes and evaluate its contribution to lipid levels and atherosclerotic cardiovascular disease (ASCVD) risk. Methods Using WGS data from a large population-based cohort (UK Biobank), rare variants (allele frequency <0.0006) within LDLR, APOB, and PCSK9 (±5 kb) were aggregated into multi-layer genetic burden scores capturing noncoding regulatory (Phase A), splicing-related (Phase B; SpliceAI and CADD-informed), and coding pathogenic variation (Phase C; VEP-based annotation). Elastic-net models were applied to predict LDL cholesterol and apolipoprotein B using an 80/20 train–test framework with five-fold cross-validation. Incremental predictive value beyond clinical covariates was assessed, and a supervised integrated genetic burden score was derived. Effect modification of genetic burden on associations between lipid traits and incident ASCVD and mortality outcomes was evaluated using multivariable Cox proportional hazards models, with subgroup analyses by glycemic status, lipid-lowering therapy, and sex. Results Elastic-net models incorporating multi-layer genetic burden features improved out-of-sample prediction of LDL cholesterol and apolipoprotein B compared with clinical covariates alone (Fig 1.a). Sequential inclusion of gene- and phase-specific components demonstrated incremental gains in model performance (Fig 1.b,c), supporting complementary effects of noncoding regulatory, splice-altering, and coding variation. Permutation importance analysis further indicated that genetic burden components ranked among the most informative predictors after established clinical variables (Fig 1.d,e). In prospective analyses, integrated genetic burden significantly modified the association between lipid traits and ASCVD mortality after multivariable adjustment (Fig 2.b). The adverse effect of higher genetic burden increased progressively with higher lipid exposure, suggesting amplification of lipid-related cardiovascular risk among genetically susceptible individuals (Fig 2.c). Associations between genetic burden and lipid traits were largely consistent across clinical subgroups, with modest heterogeneity observed according to glycemic status and lipid-lowering therapy use (Fig 2.a). Conclusion Integrated WGS-based rare-variant burden modeling improves lipid prediction and identifies genetic susceptibility that modifies lipid-related cardiovascular risk, supporting the integration of multi-layer rare-variant information for improved cardiovascular risk stratification.