We performed genome-wide tests for association between haplotype clusters and each of 9 metabolic traits in a cohort of 5402 Northern Finnish individuals genotyped for 330 000 single-nucleotide polymorphisms. The metabolic traits were body mass index, C-reactive protein, diastolic blood pressure, glucose, high-density lipoprotein (HDL), insulin, low-density lipoprotein (LDL), systolic blood pressure, and triglycerides. Haplotype clusters were determined using Beagle. There were LDL-associated clusters in the chromosome 4q13.3-q21.1 region containing the albumin (ALB) and platelet factor 4 (PF4) genes. This region has not been associated with LDL in previous genome-wide association studies. The most significant haplotype cluster in this region was associated with 0.488 mmol/l higher LDL (95% CI: 0.361-0.615 mmol/l, P-value: 6.4 Â 10 À14 ). We also observed three previously reported associations: Chromosome 16q13 with HDL, chromosome 1p32.
INTRODUCTIONThe identification of genetic factors that influence quantitative traits such as low-density lipoprotein (LDL) has clinical importance. For example, higher LDL levels are associated with increased risk of cardiovascular health disease, and the discovery of the association between PCSK9 mutations and LDL led to the development of PCSK9 inhibitors as a novel class of LDL-reducing drugs. 1 Notably, in the cohort study that found the LDL-associated PCSK9 mutations, each mutation was present in fewer than 2% of study individuals. 2 The standard single-SNP analysis commonly employed in genome-wide association studies (GWAS) has low power for detecting such low frequency causal variants. By employing haplotypic analysis in combination with single-SNP analysis, we can improve power above that of single-SNP analysis alone for detecting causal variants with low minor allele frequency. 3 The standard approach for association analysis of genome-wide single-nucleotide polymorphism (SNP) array data in population samples is to test each SNP individually for association with the trait. This approach can have high power to detect an ungenotyped causal variant when the causal variant is common and correlated with one or more genotyped variants on the array. However, the single-SNP approach has lower power to detect low frequency ungenotyped causal variants. 3 To improve power for low frequency variants, one could impute them and test for association with the trait, but imputation of low frequency variants suffers from poor accuracy. 4 Moreover, variants that are unique to the population of interest