Data for 293 461 autosomal SNPs identified with Affymetrix were available for analysis in up to 11 685 individuals from five studies. 14 SNPs were statistically associated with circulating LDL-cholesterol concentrations at the genome-wide level (p<1·0×10−7
; ). The 14 SNPs showed directionally consistent signals in all five studies (webtable 3
). These SNPs were broadly confined to three distinct genomic regions (), two of which were close to known loci involved in LDL-cholesterol metabolism (including those encompassing the APOB
Notably, two SNPs—rs599839 (p=1·7×10−15
) and rs4970834 (p=3·0×10−11
)—showed evidence for statistical association and were both located at chromosomal region 1p13.3. For these SNPs, there was no material evidence for heterogeneity between studies after adjustment for multiple testing ().
Statistical associations (p<1·0×10−7) between Affymetrix SNPs and circulating concentrations of LDL cholesterol in a genome-wide meta-analysis of five study populations consisting of up to 11 685 participants
Data for 290 140 SNPs identified with Illumina were available for analysis across the three studies that were assessed with these chips. For these analyses, we had up to 4337 participants with a measure of LDL-cholesterol concentration. Seven SNPs were statistically associated with circulating LDL-cholesterol concentrations at the genome-wide level (p<1·0×10−7
; and webtable 4
). Six of these SNPs were located in genomic regions previously linked to LDL-cholesterol metabolism. However, we also found another SNP located at chromosomal region 1p13.3 (rs646776; p=4·3×10−9
Statistical associations (p<1·0×10−7) between Illumina SNPs and circulating concentrations of LDL cholesterol in a genome-wide meta-analysis of three UK study populations consisting of up to 4337 participants
Linkage disequilibrium plots of the three SNPs located at 1p13.3 implicated a region spanning several genes (webfigure
). The strongest statistically associated SNP (rs599839) lay 3′ to the CELSR2
genes (the two genes are in a tail-to-tail orientation) in a 98 kb region of fragmented linkage disequilibrium. This region contained several recombination hotspots and was situated between two blocks of strong linkage disequilibrium (webfigure
). SNP rs4970834 also lay in this region and in our studies was correlated with SNP rs599839 (r2
=0·79; webtable 5
). SNP rs646776 was also colocalised with these SNPs and was highly correlated (r2
=0·94) with SNP rs599839 (webtable 5
). For all three SNPs, the minor allele, with a frequency of around 19–21%, was associated with lower LDL-cholesterol concentrations ( and ).
Likelihood ratio tests of up to 10 310 participants showed that, for the Affymetrix SNPs, assuming that SNP rs599839 was the causal variant or in near complete linkage disequilibrium with the causal variant(s), inclusion of SNP rs599839 as a covariable explained the other observed SNP associations in this region (webtable 6
). In an exploratory and equivalent analysis on a small subset of samples with both Affymetrix and Illumina SNP data (up to 3007 participants), we found that both SNPs rs599839 and rs646776, which are highly correlated, equally explained the association signals in this region (webtable 6
). Thus, when conditioning on SNP rs599389, our results indicated that the three statistically associated SNPs might be characterising identical genetic variant(s) in this region.
Imputation of all SNPs with a MAF of 1% or more from our Affymetrix array and HapMap II data for this 98 kb region for 9988 participants allowed us to assess whether additional association signals might be present in this region. On the basis of these imputed data, the strongest evidence for association was found for SNP rs646776 (p=3·0×10−14; and ). Indeed, within this 98 kb region, the strongest association signals from both imputed and genotyped data were localised to a 14 kb region containing a group of seven highly correlated SNPs that included the 3′ untranslated region (UTR) of the CELSR2 gene ( and ). By use of data from HapMap, we found that SNPs rs599839 and rs646776 tag six of these seven SNPs with an r2 between 0·96 and 1·00. However, in view of the strong linkage disequilibrium, the specific source of the association signal is unlikely to be reliably differentiated between these SNPs in our data.
Imputed SNPs showing genome-wide statistical association (p<1·0×10−7) with circulating concentrations of LDL cholesterol; meta-analysis of four study populations consisting of up to 9988 participants
To validate associations found in our genome-wide association screens and imputational analysis, we genotyped rs599839 and rs646776 in the two replication cohorts (webtable 7
). There was again evidence of statistical association in each of these studies (webtable 7
), thus corroborating our imputed results.
Lastly, we conducted a meta-analysis and pooled analysis of all available studies by use of a comparable analytical approach. Meta-analysis of data for up to 16 571 participants showed evidence for statistical association for SNP rs599839 with LDL-cholesterol concentrations (p=1·2×10−33
; webtable 7
). On the basis of data for up to 9282 participants, we found evidence for statistical association for SNP rs646776 with LDL-cholesterol concentrations (p=4·8×10−20
). These associations were directionally consistent across all studies (), with no heterogeneity between studies (p=0·43 for SNP rs599839 and p=0·88 for rs646776).
Association between SNPs at the 1p13.3 locus and circulating concentrations of LDL cholesterol
A pooled analysis of all studies in which we had individual participant data, which consisted of 15 196 individuals for SNP rs599839 and 7952 individuals for SNP rs646776, suggested that SNPs rs599839 and rs646776 both explained around 1% of the variation in circulating LDL-cholesterol concentrations and were associated with about 15% of an SD change in LDL-cholesterol concentrations per allele (). Further adjustment for age and sex did not alter these findings (data not shown).