Whole genome sequence analysis of low-density lipoprotein cholesterol across 246 K individuals

Abstract

Background: Rare genetic variation provided by whole genome sequence datasets has been relatively less explored for its contributions to human traits. Meta-analysis of sequencing data offers advantages by integrating larger sample sizes from diverse cohorts, thereby increasing the likelihood of discovering novel insights into complex traits. Furthermore, emerging methods in genome-wide rare variant association testing further improve power and interpretability.

Results: Here, we conduct the largest meta-analysis of whole genome sequencing for low-density lipoprotein cholesterol (LDL-C), a therapeutic target for coronary artery disease, analyzing data from 246 K participants and integrating 1.23B variants from the UK Biobank and the Trans-Omics for Precision Medicine (TOPMed) program. We identify numerous rare coding and non-coding gene associations related to LDL-C, with replication across 86 K participants in All of Us. Our findings are based on single-variant analyses, rare coding and non-coding variant aggregation tests, and sliding window approaches. Through this comprehensive analysis, we identify 704 novel single-variant associations, 25 novel rare coding variant aggregates, 28 novel rare non-coding variant aggregates, and one novel sliding window aggregate.

Conclusions: This study provides a meta-analysis framework for large-scale whole genome sequence association analyses from diverse population groups, yielding novel rare non-coding variant associations.

Fig. 1

Overall results from TOPMed and UKB WGS meta-analysis. A Procedure used to identify novel variants from individual variant GWAS: A total of 21,657 variants were genome significant in the individual variant meta-analysis with a P-value ≤ 5 × 10⁻⁰⁹. These variants were used to define genomic risk locus using FUMA with 1000 Genome phase 3 as reference panel, where independent SNPs (r2 ≥ 0.6) were identified and if LD blocks of independent significant SNPs are closely located to each other (< 250 kb based on the most right and left SNPs from each LD block), they are merged into one genomic locus. Finally, we identified 128 genomic loci from our individual variant GWAS. Comparison of these genomic risk loci with the GLGC summary statistics yielded one locus with a genome-significant variant associated to LDL-C. From the totality of the genome, significant variants identified in the present study, 704 were unique to WGS data and not found in the GLGC summary statistics. B Procedure used to identify novel aggregates from rare variant test: From a total of 20 K genes, rare variant aggregates were assessed for coding (5-masks) and non-coding (7-masks) and 2.6 M regions based on the sliding-window approach. Bonferroni-corrected p-values were used to identify genome-significant rare variant aggregates before and after conditional analysis. The set of aggregates that passed the conditional analysis were replicated in an independent cohort (i.e., AoU). The number of rare variant aggregates passing each step is shown. WGS—Whole Genome Sequencing; GLGC—Global Lipids Genetics Consortium; SNP—Single-Nucleotide Polymorphism; GWAS—Genome-Wide Association Studies; LD—Linkage Disequilibrium; LDL-C—Low-Density Lipoprotein Cholesterol

Fig. 2

MetaSTAAR-O p-value comparison. A MetaSTAAR-O p-value comparison between before and after conditional analysis for gene-centric coding aggregates. B MetaSTAAR-O p-value comparison between before and after conditional analysis for gene-centric non-coding aggregates. In both the plots, the aggregates are ordered based on conditional MetaSTAAR p-value. C Comparison of MetaSTAAR-O p-value for genes that have at least one coding and non-coding signal before conditional analysis. Each dot represents a gene, and the color of the dots represent if the gene is genome significant in either coding or non-coding or both. Most significant gene names each of those categories are mapped

Fig. 3

ABCA6 protein structure with variants identified using the aPC scores. A Protein structure of ABCA6 consists of 2 ABC transporter and transmembrane domains. B Structure of ABCA6 with the 39 SNVs identified from ABCA6-pLoF-ds aggregate set (red) and 30 SNVs identified from ABCA6-missense aggregate set (orange) and the common SNVs mapping to both aggregate sets (pink). C Structure of ABCA6 with the highly scored variants (red) and structural proximity of the stretch of amino acids in ABC transporter 1 domain identified through this analysis. ABC—ATP-binding cassette; SNV—Single- Nucleotide Variations; pLoF-ds—putative loss-of-function and disruptive missense variants