Common variants in Cholesteryl ester transfer protein (CETP) are associated with differences in HDL-cholesterol (HDL) levels in most ancestral groups. However, no studies have examined whether genetic variation in CETP associates with HDL among people of Polynesian ancestry.

An HDL GWAS using the Illumina CoreExome platform was conducted in 2,594 people of Polynesian ancestry (Aotearoa/New Zealand [A/NZ] M?ori, Cook Island M?ori, S?moan, Tongan, Niu?an and Pukapukan) living in A/NZ. Using 55 CETP gene sequences, genetic variants common in people of A/NZ Polynesian ancestry (>5%) but extremely rare (<0.001%) in gnomAD were selected for follow-up. A set of 2,851 S?moans (cohort I) using imputed genotypes based on S?moans sequenced via TOPMed and a set of 1,510 directly genotyped S?moans and American S?moans (cohort II) were used to replicate initial findings. The A/NZ analysis was adjusted for age, sex and principal components of ancestry, whereas the S?moan cohorts were adjusted for age, sex, principal components of ancestry and relatedness. Selection was calculated via the haplotype-based statistic (nSL) using the selscan program.

Three variants within CETP (rs183130, rs247617, rs9939224) were genome-wide significant in the A/NZ Polynesian. Closer inspection found a novel Polynesian-specific variant (NC_000016.10:g56971035C>T, CETP:p.P177L) associated with higher HDL levels in the A/NZ (β= 11.3 [9.1, 13.5] mg/dL, p=9.1×10−24), S?moan cohort I (β=8.7 [7.4, 10.1] mg/dL, p=3.0×10−37) and cohort II (β=8.3 [6.7, 9.9] mg/dL, p=4.5×10−25). The MAF for the A/NZ Polynesian, S?moan cohorts I and II were 3.4%, 4.4% and 4.9% respectively but this variant is unobserved in gnomAD. The variances in HDL explained by the variant in the A/NZ cohort and S?moan cohort I were 7.7% and 5.1% respectively. Analysis of this variant showed moderate evidence of positive selection (nSL score of 1.52 with an empirical percentile of 92%) in the S?moan cohort I. A CADD score of 25.6 places the variant in the top 1% of predicted deleteriousness. Secondary analysis conditional on this variant shows that all other variants associated with HDL within or close to CETP did not reach genome-wide significance.

Our results show that a unique variant in CETP contributes strongly to the phenotypic variation of HDL in people of Polynesian ancestry. This improved understanding of CETP variation in people of Polynesian ancestry allows insights into differences in control of HDL in this population.

Current understanding of lipid genetics has come mainly from studies in European-ancestry populations; limited effort has focused on Polynesian populations, whose unique population history may provide insight into the biological foundations of variation in lipid levels. Here, we performed an association study of 4 lipids traits: total cholesterol, high-density lipoprotein (HDL), low-density lipoprotein, and triglycerides (TG) and genetic variants in 5q35, a region that has shown suggestive association with HDL in Micronesian and Polynesian peoples and no previous evidence of association in Europeans.
We performed association testing between variants on 5q35 and lipid levels in the discovery cohort comprising 2,851 Samoans using linear mixed modeling with imputed genotypes based on Samoans sequenced via TOPMed and adjusting for age, sex, principal components of ancestry, and relatedness. Fine-mapping analyses highlighted an association between a stop-gained variant (rs200884524; c.652C>T, p.R218*) in BTNL9 and HDL (discovery cohort: β=-1.99mg/dL, p=4.49×10-8; replication cohort of 1,522 Samoans and American Samoans: β=-1.35mg/dL, p=0.0017; replication cohort of 2,378 Māori and Pacific individuals living in Aotearoa/New Zealand: β=-0.90mg/dL, p=0.003; meta-analysis of all cohorts: β=-1.35mg/dL, p=4.75×10-11). rs200884524 is also associated with TG (discovery cohort: β=14.38mg/dL, p=3.63×10-6, Samoan replication cohort: β=8.29mg/dL, p=0.061, Aotearoa/New Zealand replication cohort: β=6.67mg/dL, p=3.51×10-5; meta-analysis: β=8.3mg/dL, p=1.02×10-9). This variant accounts for 1.02% and 0.7% of the variation in HDL and TG respectively in the discovery cohort.
rs200884524 has a MAF of <0.0002 in gnomAD; of the 50 alleles seen in gnomAD, 45 are from East Asians. However Samoans have a MAF of 0.218 in the discovery cohort (0.215 and 0.101 in the Samoan and Aotearoa/New Zealand replication cohorts respectively). Consistent with these disparate allele frequencies, this variant showed evidence of positive selection (PBS score of 0.477, 99.6 percentile in the Samoan genome; nSL score of 1.78, 98.3 percentile in the Samoan genome).
BTNL9 is expressed in adipose, breast, and lung tissue (GTEx). Additional work is necessary to characterize the relationship between variation in BTNL9 and lipid levels; however, these initial findings fine map the suggestive association in 5q35, providing evidence of a new contributor to the genetic architecture of lipids in Polynesian peoples.

Cardiovascular diseases (CVD), and in particular cerebrovascular and ischemic heart diseases, are leading causes of death globally. Lowering circulating lipids is an important treatment strategy to reduce risk. However, some pharmaceutical mechanisms of reducing CVD may increase risk of fatty liver disease or other metabolic disorders. To identify potential novel therapeutic targets, which may reduce risk of CVD without increasing risk of metabolic disease, we focused on the simultaneous evaluation of quantitative traits related to liver function and CVD. Using a combination of low-coverage (5x) whole-genome sequencing and targeted genotyping, deep genotype imputation based on the TOPMed reference panel, and genome-wide association study (GWAS) meta-analysis, we analyzed 12 liver-related blood traits (including liver enzymes, blood lipids, and markers of iron metabolism) in up to 69,479 participants in the population-based Nord-Trøndelag Health Study (HUNT) in Norway. Following fine-mapping of genomic regions using step-wise conditional analyses and trans-ancestry meta-analyses in up to 203,476 people, we identified 88 likely causal protein-altering variants that were associated with one or more liver-related blood traits. We identified several loss-of-function (LoF) variants reducing low-density lipoprotein cholesterol (LDL-C) or risk of CVD without increased risk of liver disease or diabetes, including variants in known lipid genes (e.g. APOB, LPL). A novel LoF variant, ZNF529:p.K405X, was associated with decreased levels of LDL-C (P=1.3x10-8) but demonstrated no association with liver enzymes or non-fasting blood glucose levels. Silencing of ZNF529 in human hepatocytes resulted in upregulation of LDL receptor (LDLR) and increased LDL uptake in the cells, suggesting that inhibition of ZNF529 or its gene product could be used for treating hypercholesterolemia and hence reduce the risk of CVD. Taken together, we demonstrate that simultaneous consideration of multiple phenotypes and a focus on rare protein-altering variants may identify promising therapeutic targets.

For genetic association studies, whole-exome sequences provide two distinct advantages: a focus on rare variants in coding sequence and the ability to pinpoint a causal gene. Here, we assembled blood lipid phenotypes and exome sequences in >170,000 participants from multiple ancestries emerging from four data sources: The Myocardial Infarction Genetics Consortium (n=44,208), the AMP-T2D-GENES exomes (n=32,486), the Trans-Omics for Precision Medicine (TOPMed) project restricted to the exome (n=44,101) and UK Biobank exomes (n=51,275). Our study included individuals of African-American (n=16,507), East Asian (n=10,420), European (n=97,493), Hispanic (n=16,440), Samoan (n=1,182) and South Asian (n=30,025) ancestries. We performed extensive quality control within each data source and additionally removed duplicates and first- and second-degree relatives across the data sources. We annotated ~24.5 million variants and selected ~580,000 loss-of-function (LOF) variants and over 1.1 million missense variants predicted to be damaging. We performed association analysis within each data source by ancestry and case-status using linear mixed models incorporating relatedness for 6 traits (total, LDL, HDL, and non-HDL cholesterol and triglycerides and triglyceride-HDL ratio). Gene-based meta-analysis across data sources was performed in all subjects, by race, and by case-status using RAREMETALS. We replicated known associations with LDL cholesterol (PCSK9, LDLR, and APOB), triglycerides (APOC3, ANGPTL3 and ANGPTL4), and HDL cholesterol (CETP and SCARB1). We additionally identified several novel genes associated with blood lipids including SRSF2 with total cholesterol (-29 mg/dl, p=7×10-8), ALB with total (33 mg/dl, p=1×10-8) and LDL cholesterol (30 mg/dl, p=8×10-9), NR1H3 with HDL cholesterol (2.1 mg/dl, p=1×10-9), PLA2G12A with HDL cholesterol (-2.3 mg/dl, p=8×10-14) and triglycerides (6%, p=1×10-8) and CREB3L3 with triglycerides (12%, p=5×10-14). We observed significant heterogeneity (P<0.002) among different ancestries for LDLR and APOB with LDL cholesterol and LIPG and ABCA1 with HDL cholesterol. Our findings demonstrate the importance of large exome sequencing studies in multiple ancestries for the discovery of novel genes associated with blood lipids that may be valuable targets for cardiovascular disease prevention.

Introduction
Large-scale whole genome sequencing studies have enabled the analysis of rare variants (RVs) associated with complex human traits. Commonly used RV association tests (RVATs) have limited scope to leverage the functions of variants.
Methods
We propose STAAR (variant-Set Test for Association using Annotation infoRmation), a scalable and powerful method to increase the power of RVATs by effectively incorporating both variant functional categories and multiple complementary functional annotations using a dynamic weighting scheme. For the latter, we introduce “annotation Principal Components” (aPCs), multi-dimensional summaries of in-silico coding and noncoding variant functional annotations. STAAR accounts for population structure and relatedness, and can be used for analyses of continuous and dichotomous traits. We highlighted STAAR-O, the omnibus test which aggregates multiple annotation-weighted tests including the burden test, SKAT, and ACAT-V in the STAAR framework. We focused on two types of WGS RV association analysis using STAAR-O: gene-centric functional element-based analysis by grouping variants into functional categories for each protein-coding gene and agnostic genetic region analysis using sliding windows. We applied STAAR-O to identify RV-sets associated with four quantitative lipid traits (LDL-C, HDL-C, TG and TC) in 12,316 discovery samples and 17,822 replication samples from the NHLBI Trans-Omics for Precision Medicine (TOPMed) program.
Results
In RV gene-centric analysis, STAAR-O identified 25 significant associations with lipids traits. After conditioning on known lipids-associated variants, 13 out of the 25 associations remained significant and could be validated in replication phase, including the association between NPC1L1 missense RVs and LDL-C, which was not detected by the conventional variant-set tests and has not been observed in previous studies. In RV sliding window analysis, STAAR-O detected 59 significant 2kb sliding windows associated with lipid traits. 12 sliding windows remained significant after conditioning on known variants and could be validated in replication phase, including associations between an intergenic region near APOC1P1 and LDL-C that were not detected using conventional tests.
Conclusion
By incorporating multiple variant functional annotations via aPCs, STAAR-O empowers rare variant association analysis and detected novel rare variants association with lipid traits using the TOPMed WGS Freeze 5 data.