Abstract 4147026: Predicting coronary artery disease using an exome-based estimate of KIV-2 repeat count in the Lp(a) gene

Introduction: Lipoprotein a (Lp(a)) levels are an often unmeasured predictor of future risk of cardiovascular disease. Unlike LDL, HDL, and triglycerides, Lp(a) levels are believed to be genetically determined; they are not altered by lipid-lowering medications or dietary changes, and are consistent over a lifetime, making them a potential early-life predictor of adult morbidity. However, genetic predictors of Lp(a) are challenging to design. Repeats of the kringle IV subtype 2 domain (KIV-2) contribute significantly but are extremely challenging to quantify via next-generation sequencing; this variability is particularly impactful in non-European ancestries. Methods: We analyzed ~110K individuals with paired clinical and exome sequencing data from six medical centers. We developed a coverage-based method to determine KIV-2 repeat counts. We validated that these counts associated with clinical Lp(a) measurements derived from the course of care (n = 2039 individuals). We calculated an expected Lp(a) in units of nmol/L using a genetic risk score (GRS) based on 43 variants in the Lp(a) gene. We identified instances of CAD via ICD9CM, ICD10CM and SNOMED terms present in the medical record. Results: Adding the number of KIV-2 repeats to the GRS substantially improved prediction of clinical Lp(a) measurements, for example improving the r-squared for non-Europeans five-fold (from 0.05 to 0.22). Individuals with fewer than 15 estimated KIV-2 repeats and a high GRS-predicted Lp(a) (high risk group) have a mean measured Lp(a) of that is 10-fold that of those in the lowest category of GRS-predicted Lp(a) (low risk group). In the full cohort, most of whom had not been measured for Lp(a), we find that relative to the low risk group, the high risk individuals are 1.6x more likely to have CAD. Conclusion: Lp(a) is a potentially useful biomarker with consistency over a lifetime and implications for future cardiovascular morbidity. We developed a method for estimating the number of KIV-2 repeats, a challenging component of genetic prediction of Lp(a). This estimate can be used in combination with a GRS to predict Lp(a), especially in individuals with non-European ancestry where GRS-based estimates fall short. The final prediction is associated with increased risk of CAD.