A large electronic-health-record-based genome-wide study of serum lipids

Abstract

A genome-wide association study (GWAS) of 94,674 ancestrally diverse Kaiser Permanente members using 478,866 longitudinal electronic health record (EHR)-derived measurements for untreated serum lipid levels empowered multiple new findings: 121 new SNP associations (46 primary, 15 conditional, and 60 in meta-analysis with Global Lipids Genetic Consortium data); an increase of 33-42% in variance explained with multiple measurements; sex differences in genetic impact (greater impact in females for LDL, HDL, and total cholesterol and the opposite for triglycerides); differences in variance explained among non-Hispanic whites, Latinos, African Americans, and East Asians; genetic dominance and epistatic interaction, with strong evidence for both at the ABO and FUT2 genes for LDL; and tissue-specific enrichment of GWAS-associated SNPs among liver, adipose, and pancreas eQTLs. Using EHR pharmacy data, both LDL and triglyceride genetic risk scores (477 SNPs) were strongly predictive of age at initiation of lipid-lowering treatment. These findings highlight the value of longitudinal EHRs for identifying new genetic features of cholesterol and lipoprotein metabolism with implications for lipid treatment and risk of coronary heart disease.

Figure 1

Epistasis between SNPs at FUT2 and ABO (n=94,674). SNP rs601338, determines secretor status; Se/Se and Se/se are secretors (dominant), and se/se are non-secretors. SNP rs2519093 at ABO, C allele is recessive. (A) Residual LDL by genotype combination, where the vertical line represents 95% confidence intervals. (B) ABO locus plot for secretors (Se/Se or Se/se), around rs2519093 (linear regression on LDL). (C) ABO locus plot for non-secretors around rs2519093 (linear regression on LDL). All tests are two-sided.

Figure 2

Tissue specific expression quantitative trait locus (eQTL) analysis in the 44 GTEx tissues. An upper-tailed p-value for enrichment of the count was calculated with a Z-score using the overall median tissue proportion and the standard deviation of the null distribution of that tissue. Tissues with P<0.05 are labeled, and the red line indicates the P-value threshold for Bonferroni significance. (A) All SNPs (N=284). (B) All SNPs removing liver eQTLs and rerunning the analysis in 43 GTEx tissues. (C) Exclusive HDL SNPs only (N=63). (D) Exclusive LDL SNPs only (N=61). (E) Exclusive TG SNPs only (N=39). All tests are two-sided.

Figure 3

Time-to-initiation of lipid-lowering treatment by LDL and TG genetic risk score quintiles. Kaplan Meier curves for (A) non-Hispanic while females (n=44,856), (B) non-Hispanic white males (n=31,771). Other groups are in the Supplementary material. The shaded areas represent the 95% CIs around the estimated curves. All tests are two-sided.