Background: Lipoprotein(a) (Lp(a)) is a heritable, causal risk factor for coronary heart disease. Both SNPs and structural variants at the LPA locus, namely the 5.5kb kringle IV-2 copy number repeat (KIV2-CN), account for much of the variation in Lp(a) levels. We sought to comprehensively characterize the genetic architecture of Lp(a) and Lp(a)-cholesterol (Lp(a)-C) through whole genome sequence (WGS) association among diverse ethnicities, and understand the relative effects of variant classes on phenotypes.
Methods: We performed deep (~30X) WGS in 8,302 of European and African ancestry: Estonian Biobank, Finland FINRISK, and Jackson Heart Study, and genotyped SNPs, indels, as well as KIV2-CN. Finland variants were subsequently imputed in 27,344 deeply-phenotyped Finnish individuals. We ascertained Lp(a) and cardiometabolic phenotypes and performed (1) common variant, (2) rare variant, and (3) Mendelian randomization analyses.
Results: We analyzed 119M SNPs and 7.2M indels, with a mean total KIV2-CN of 42 (SD 8), and developed a novel KIV2-CN imputation model. Trans-ethnic common variant association yielded 41 variants independent of KIV2-CN for Lp(a), and 5 for Lp(a)-C at the LPA locus. We now observe three additional significant (P<5x10-8) loci; for Lp(a)-C: SORT1 and CETP, for Lp(a): APOE. In modifier analyses of KIV2-CN, three independent variants at the LPA locus demonstrated significant interaction with KIV2-CN’s effect on Lp(a)-C levels. Furthermore, aggregated rare (MAF<1%) coding disruptive mutations within LPA and, separately, rare non-coding variants within regulatory sequence were associated with Lp(a)-C.
Finally, with a panel of 12 clinical and 115 metabolic phenotypes, we perform Mendelian randomization, finding (1) significantly stronger effect of our KIV2-independent genetic risk score (as compared to KIV2-CN) across atherosclerotic cardiovascular diseases, (2) associations with subclinical atherosclerosis in African Americans, and (3) vertical pleiotropy with biomarkers of phospholipid metabolism.
Conclusion: In conclusion, deep WGS-based broad genotyping and population-based imputation permits a comprehensive assessment of the full genetic spectrum contributing to Lp(a) variation and a range of consequential cardiometabolic effects.

We obtained 30X whole genome sequences in 8,368 participants of the Jackson Heart Study (N = 3,246), Framingham Heart Study (N = 4,301), and Amish (N = 1,091). Within each cohort, we associated each variant with plasma lipids (total cholesterol, LDL cholesterol, HDL cholesterol, triglycerides), accounting for lipid-lowering medications, adjusting for age, age2, sex, and an empiric genetic relatedness matrix. Variants with at least 5 carriers within a cohort were meta-analyzed (N ~ 33M per trait) with a fixed effects, inverse variance-weighted approach. We explored the genetic architecture of lipid trait variation across the allelic spectrum.

Plasma lipids are heritable risk factors for cardiometabolic diseases. Prior studies using genotyping arrays and exome sequencing have identified hundreds of associations involving common, non-coding variants or a burden of rare, coding variants. Whole genome sequencing (WGS) allows for testing across the full spectrum of allele frequency and variant type in both non-coding and coding regions. 

We performed deep WGS (>20X) in 16,324 individuals: Amish, Framingham HS, Jackson HS, MESA, FINRISK, and Estonia. 189M variants were identified. We individually tested 32M variants (MAF>0.1%) for association with total cholesterol, LDL-C, HDL-C, and triglycerides. 31 known loci demonstrated association (P<5x10-8). We observed an African American-specific haplotype (MAF 1%) including a variant at the LDLR promoter associated with LDL-C (P=2x10-11) and a 1-bp intronic deletion at 9p24.1 (MAF 2%) with HDL-C (P=1x10-8). We confirmed that rare (MAF<1%) disruptive coding mutations in known Mendelian lipid genes (LDLR, PCSK9, APOE, LCAT, APOC3) associated with lipids (P<2.5x10-6). We found that both rare disruptive coding mutations and an expanded polygenic risk score of 2M variants independently predicted extreme lipid phenotypes, but the effects differed by ancestry.

Finally, we tested the hypothesis that rare, non-coding variants within regulatory sequences contribute to lipid variation. We aggregated rare variants in non-coding sequence using a 3kb sliding window and three other methods leveraging ENCODE and Roadmap annotations for hepatocytes and adipocytes: 1) overlapping enhancers within 20kb of, and promoters within 5kb of transcription start sites, 2) overlapping enhancers co-occurring with gene expression, and 3) overlapping enhancer/Hi-C contacts to transcription start sites. No burden-of-rare-variant association signals were detected in non-coding sequence with any approach.

In summary, we present a large-scale WGS analysis of plasma lipids in 16,324 ethnically diverse participants. Common, non-coding variants and rare, coding variants contribute to plasma lipid variation; however, association signals for rare, non-coding mutations were not detectable.

Plasma lipids are heritable risk factors for cardiometabolic diseases. Prior studies using genotyping arrays and exome sequencing have identified hundreds of associations involving common, non-coding variants or a burden of rare, coding variants. Whole genome sequencing (WGS) allows for testing across the full spectrum of allele frequency and variant type in both non-coding and coding regions. 

We performed deep WGS (>20X) in 16,324 individuals: Old Order Amish, Framingham Heart Study, Jackson Heart Study, Multi-Ethnic Study of Atherosclerosis, FINRISK, and Estonian Biobank. 189M unique variants were identified. We individually tested 32M variants (MAF>0.1%) for association with total cholesterol, low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), and triglycerides. Common variants in 31 known loci demonstrated association (P<5x10-8). We observed an African American-specific haplotype (1% frequency) including a variant at the LDLR promoter associated with 28 mg/dl lower LDL-C (P=2x10-11) and a 1-bp intronic deletion at 9p24.1 (MAF 2%) with HDL-C (P=1x10-8). We confirmed that rare (MAF<1%) disruptive coding mutations in known Mendelian lipid genes (LDLR, PCSK9, APOE, LCAT, APOC3) associated with blood lipids (P<2.5x10-6). We found that both rare disruptive coding mutations and an expanded polygenic risk score of 2M variants independently predicted extreme lipid phenotypes, but the effects differed by ancestry.

Finally, we tested the hypothesis that rare, variants in regulatory (non-coding) sequence contribute to plasma lipid variation. We aggregated rare variants in non-coding sequence using a 3kb sliding window approach as well as three other methods which leverage ENCODE and Roadmap annotations for hepatocytes and adipose tissue: 1) regions overlapping enhancers within 20kb and promoters within 5kb of transcription start sites, 2) overlapping enhancers co-occurring with gene expression, and 3) overlapping enhancer/Hi-C contacts to transcription start sites. No burden-of-rare-variant association signals were detected in non-coding sequence with any of these approaches.

In summary, we present a large-scale whole genome sequence analysis of plasma lipids in 16,324 ethnically diverse participants. Common variants in non-coding sequence as well as rare variants in coding sequence contribute to plasma lipid variation; however, association signals for rare mutations in non-coding sequence were not detectable.

Mendelian pathogenic variants resulting in severe hypercholesterolemia influences cardiovascular disease risk but a causal mutation is only present in ~1 in 50 with severe hypercholesterolemia. Whole genome sequencing (WGS) enables assessment of variation in both non-coding and coding regions across the full allele frequency spectrum. We estimated the simultaneous contribution of monogenic and polygenic determinants of the extremes of LDL-C, HDL-C, and triglycerides, separately, for the 8,394 participants of the Old Order Amish (OOA), Framingham Heart Study (FHS), and Jackson Heart Study (JHS) that underwent deep-coverage (>30x) WGS as part of Phase 1 of the NHLBI TOPMed program. We catalogued presence of monogenic mutations in known lipid Mendelian genes. We applied a genome-wide significant restricted (35-66 variants) polygenic risk score (PRS) and the best performing expanded PRS in 25,534 unrelated Nord-Trøndelag Health Study (HUNT) individuals utilizing 2M variants from prior GWAS summary statistics within OOA, FHS, and JHS and identified individuals in the top 20th risk percentile for each cohort. Across the three cohorts, 3.8% of individuals carried a LDL monogenic mutation. We found that rare disruptive coding mutations and a PRS of 2M variants independently predicted extreme lipid phenotypes, but the effects differed by cohort and phenotype. For example, in FHS, individuals with a monogenetic mutation had a 3.4 increased odds of high LDL and those in the top 20th percentile had an 18.4 increased odds of high LDL. Risk conferred by the expanded PRS (OR: 18.4) was greater than from a restricted PRS (OR: 3.9). As expected, since summary statistics for PRS generation were derived from European ancestry cohorts, stronger associations were observed among individuals of European ancestry compared with those of African ancestry (18.4 vs. 2.4). In conclusion, both monogenic and polygenic variants additively contribute to the extremes of plasma lipids.