Abstract 40: Score Based on Immunoglobulin G N-glycans for Cardiovascular Risk Prediction: Secondary Analysis of Two Randomized Trials of Primary and Secondary Prevention

Introduction: Immunoglobulin G (IgG) are post-translationally modified proteins with the addition of complex carbohydrate molecules (glycans). These glycans can modulate IgG inflammatory capacity and determine the transition from healthy to diseased tissue. Hypothesis: A glycan score based on certain IgG glycosylation patterns is associated with CVD risk and can improve model prediction. Methods: IgG glycosylation profiles were measured on baseline plasma samples from nested CVD case-control participants from 2 studies: JUPITER (NCT00239681; Npairs=162; discovery) and the TNT (NCT00327691; Npairs=367, validation). Lasso regression was used to select IgG glycan peaks (GPs). The linear combination of selected IgG-GPs that significantly associated with CVD in a mutually adjusted conditional regression comprised the glycan score (IgG GS ). CVD prediction using IgG GS was investigated controlling for clinical risk factors. Using a parametric approach, we calculated the area under the curve (AUC) with and without IgG GS . Results: From 24 IgG-GPs, Lasso selected 4 IgG-GPs that were also associated with incident CVD (PGS ; Fig1A), which was significantly associated with incident CVD in JUPITER after adjusting for clinical risk factors (model 2 hazard ratio [HR]: 2.1, 95% CI 1.55 - 2.84), with significant validation in TNT (HR = 1.2, 95% CI 1.03 - 1.4; Fig. 1B). The AUC was higher for the model with IgG GS (0.74, 95%CI = 0.69 - 0.81) than without (0.69, 95%CI = 0.67 - 0.71) in JUPITER; replicating in TNT: 0.72 [0.71 - 0.83] vs 0.64, [0.62 - 0.65]. The P-value for the likelihood ratio test comparing models with and without IgG GS was 5.9x10 -8 in JUPITER and .02 in TNT. Conclusions: An IgG glycan score with 4 IgG N-glycans was positively associated with incident CVD in the JUPITER primary prevention population, which replicated in TNT, a secondary prevention cohort, and improved model prediction performance.