Abstract

Almost 100 genetic loci are known to affect serum cholesterol and triglyceride levels. For many of these loci, the biological function and causal variants remain unknown. We performed an association analysis of the reported 95 lipid loci against 216 metabolite measures, including 95 measurements on lipids and lipoprotein subclasses, obtained via serum nuclear magnetic resonance metabolomics and four enzymatic lipid traits in 8330 individuals from Finland. The genetic variation in the loci was investigated using a dense set of 440 807 directly genotyped and imputed variants around the previously identified lead single nucleotide polymorphisms (SNPs). For 30 of the 95 loci, we identified new metabolic or genetic associations (P < 5 × 10−8). In the majority of the loci, the strongest association was to a more specific metabolite measure than the enzymatic lipids. In four loci, the smallest high-density lipoprotein measures showed effects opposite to the larger ones, and 14 loci had associations beyond the individual lipoprotein measures. In 27 loci, we identified SNPs with a stronger association than the previously reported markers and 12 loci harboured multiple, statistically independent variants. Our data show considerable diversity in association patterns between the loci originally identified through associations with enzymatic lipid measures and reveal association profiles of far greater detail than from routine clinical lipid measures. Additionally, a dense marker set and a homogeneous population allow for detailed characterization of the genetic association signals to a resolution exceeding that achieved so far. Further understanding of the rich variability in genetic effects on metabolites provides insights into the biological processes modifying lipid levels.

INTRODUCTION

A recent genome-wide association meta-analysis of over 100 000 individuals of European ancestry identified 95 genetic loci associated with plasma levels of total cholesterol (TC), low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C) or triglycerides (TG) (1). For most of these loci, the exact causative variant and biological function are unknown. Some of these loci have shown association to two or more lipids or other metabolic phenotypes, such as glucokinase (hexokinase 4) regulator (GCKR) on TG (1), insulin resistance (2), C-reactive protein (3) and ratios between plasmalogens and phosphatidylcholines (4) and the gene cluster containing fatty acid (FA) desaturases 1, 2 and 3 (FADS1-2-3) on TG (1), glucose (2), sphingolipid levels (5) and ratios between acyl-acyl-phosphatidylcholine lipids (4). It is likely that some of the lipid loci are more specific in their effects on metabolic pathways, while others affect a broader spectrum of metabolic traits. However, no thorough metabolic profiling of the lipid genes has so far been made.

During the past decade due to enhances in high-throughput analytical technologies such as nuclear magnetic resonance (NMR) and mass spectrometry, it has become feasible to measure a much broader range of lipoprotein subclasses and other metabolic markers in large cohorts. This has provided a better catalogue of metabolic variation across humans and provided biological insight into various vascular outcomes (6). Applying an NMR method, Chasman et al. (7) recently described genetic profiles underlying 22 plasma lipoprotein and lipid measures.

In this study, we take a complimentary and extended approach to the previous studies and characterize the biological processes behind the 95 known lipid loci by testing their association with the levels of 95 lipoprotein and lipid measures, 22 small metabolites and 99 derived metabolic variables, including selected ratios of metabolites, and the four conventionally measured enzymatic lipids. Additionally, we characterize the loci using a dense set of genotyped and imputed single nucleotide polymorphisms (SNPs) to search for variants more highly associated with the metabolite and lipid traits than the previously reported variants, and thus potentially closer to causal. We utilize the latest NMR-based serum metabolomics platform (6,8) to quantify 216 metabolic variables and assess a dense set of 440 807 genotyped and imputed genetic markers in the 95 previously identified lipid loci, in 5 population-based Finnish cohorts totalling up to 8330 individuals (Supplementary Material, Table S1).

RESULTS

Included in the analyses were 8330 Finnish men and women enrolled in 5 population-based studies; Northern Finland Birth Cohort 1966 (NFBC1966, n = 4703), The Cardiovascular Risk in Young Finns Study (YF, n = 1904), Helsinki Birth Cohort Study (HBCS, n = 708), Health 2000 GenMets Study (H2000, n = 572) and The Dietary, Lifestyle and Genetic determinants of Obesity and Metabolic Syndrome (DILGOM, n = 443). The basic cohort characteristics are presented in Supplementary Material, Table S1. All cohorts were genotyped using Illumina genotyping arrays and imputed to include all the 102 lead SNPs in the 95 lipid loci of Teslovich et al. (1) and a dense set of markers within 1 Mb regions around the lead SNPs. The cohorts underwent the same serum NMR metabolomics platform to quantify a broad metabolite profile of 216 direct and derived metabolic variables and already had data on the 4 conventional enzymatic lipid measures available.

Refining the associations with metabolomics data

We replicated associations between the loci and the enzymatic lipid measures as in Teslovich et al. (1) for 16 of the lead SNPs in 15 out of the 95 previously reported regions at genome-wide significance in our study (P<5 × 10−8). When testing the lead SNPs across all metabolite measures (117 metabolites, 99 derived measures and 4 enzymatic lipid measures), 22 SNPs from 20 loci (Table 1) were associated with between 1 and 70 metabolite measures (see full list of associations in Supplementary Material, Table S2). In only six of the loci was the strongest association to one of the enzymatic lipid measures.

Table 1.

Significant loci after the metabolic refinement: comparison of the associations of the reported lead SNPs to the enzymatic lipid measures and the best metabolite trait

Locus Chr Lead SNP Alleles CAF Enzymatic trait P-value Beta SE Varianceexplained Lead trait P-value Beta Se Varianceexplained P-gain 
Significant P-gain with metabolic refinement 
 ANGPTL3a rs2131925 G/T 0.74 TC-lab 1.08 × 10−7 0.092 0.017 0.33% Val/Serum-TG 4.01 × 10−12 −0.123 0.018 0.59% 2.69 × 104 
 GALNT2a rs4846914 G/A 0.54 HDL-C-lab 1.33 × 10−5 0.067 0.015 0.22% M-HDL-L/S-HDL-L 1.67 × 10−12 0.115 0.016 0.66% 7.95 × 106 
 APOB rs1042034 C/T 0.73 LDL-C-lab 3.72 × 10−9 0.101 0.017 0.40% XS-VLDL-TG 9.80 × 10−18 0.150 0.017 0.89% 3.80 × 108 
 MLXIPL rs17145738 C/T 0.12 TG-lab 1.81 × 10−9 −0.140 0.023 0.42% VLDL-D 1.77 × 10−12 −0.169 0.024 0.62% 1.02 × 103 
 LPL rs12678919 A/G 0.09 TG-lab 2.60 × 10−9 −0.156 0.026 0.40% Val/Serum-TG 5.81 × 10−13 0.197 0.027 0.63% 4.47 × 103 
 ABCA1 rs1883025 C/T 0.19 HDL-C-lab 1.85 × 10−9 −0.117 0.019 0.43% Free-C/Est-C 3.04 × 10−11 −0.135 0.020 0.57% 6.07 × 101 
 FADS1-2-3 11 rs174546 C/T 0.42 LDL-C-lab 1.11 × 10−8 −0.089 0.015 0.38% LA/PUFA 4.77 × 10−268 0.563 0.016 15.41% 2.32 × 10259 
 APOA1 11 rs964184 C/G 0.14 TG-lab 2.98 × 10−24 0.220 0.022 1.17% Val/Serum-TG 1.09 × 10−26 −0.239 0.000 1.38% 2.73 × 102 
 LIPC 15 rs1532085 A/G 0.57 HDL-C-lab 2.58 × 10−16 −0.126 0.015 0.78% XL-HDL-TG 5.52 × 10−72 −0.285 0.016 3.97% 4.67 × 1055 
 HPRa 16 rs2000999 G/A 0.19 TC-lab 1.27 × 10−4 0.074 0.019 0.17% Gp/Tot-C 1.47 × 10−13 −0.148 0.020 0.67% 8.67 × 108 
 CILP2a 19 rs10401969 T/C 0.06 TC-lab 5.72 × 10−6 −0.143 0.032 0.24% MobCH 5.21 × 10−9 −0.188 0.032 0.42% 1.10 × 103 
 APOEa 19 rs439401 T/C 0.71 TG-lab 2.28 × 10−6 0.079 0.017 0.26% XS-VLDL-TG 3.12 × 10−9 0.103 0.017 0.43% 7.31 × 102 
 PLTPa 20 rs6065906 T/C 0.15 HDL-C-lab 6.41 × 10−2 −0.039 0.021 0.04% L-HDL-L/M-HDL-L 1.29 × 10−24 −0.221 0.022 1.27% 4.97 × 1022 
Small P-gain with metabolic refinement 
 GCKR rs1260326 T/C 0.65 TG-lab 1.49 × 10−18 −0.138 0.016 0.87% Ala/Gln 1.20 × 10−18 −0.150 0.017 1.03% 1.24 
 PPP1R3B rs9987289 A/G 0.85 TC-lab 1.26 × 10−8 0.120 0.021 0.37% IDL-C 3.20 × 10−9 0.129 0.022 0.42% 3.93 
 APOE 19 rs4420638 A/G 0.25 LDL-C-lab 9.35 × 10−23 0.231 0.023 2.02% S-LDL-L 3.38 × 10−23 0.238 0.001 2.15% 2.76 
No P-gain 
 SORT1 rs629301 G/T 0.79 LDL-C-lab 3.68 × 10−15 0.148 0.019 0.73% – – – – – – 
 APOB rs1367117 G/A 0.30 LDL-C-lab 4.50 × 10−12 0.120 0.017 0.60% – – – – – – 
 HMGCR rs12916 T/C 0.45 LDL-C-lab 1.30 × 10−10 0.100 0.016 0.50% – – – – – – 
 CETP 16 rs3764261 C/A 0.28 HDL-C-lab 6.32 × 10−49 0.251 0.017 2.52% – – – – – – 
 LDLR 19 rs6511720 G/T 0.12 LDL-C-lab 6.28 × 10−22 −0.233 0.024 1.12% – – – – – – 
 HNF4A 20 rs1800961 C/T 0.04 HDL-C-lab 1.03 × 10−8 −0.211 0.037 0.37% – – – – – – 
Locus Chr Lead SNP Alleles CAF Enzymatic trait P-value Beta SE Varianceexplained Lead trait P-value Beta Se Varianceexplained P-gain 
Significant P-gain with metabolic refinement 
 ANGPTL3a rs2131925 G/T 0.74 TC-lab 1.08 × 10−7 0.092 0.017 0.33% Val/Serum-TG 4.01 × 10−12 −0.123 0.018 0.59% 2.69 × 104 
 GALNT2a rs4846914 G/A 0.54 HDL-C-lab 1.33 × 10−5 0.067 0.015 0.22% M-HDL-L/S-HDL-L 1.67 × 10−12 0.115 0.016 0.66% 7.95 × 106 
 APOB rs1042034 C/T 0.73 LDL-C-lab 3.72 × 10−9 0.101 0.017 0.40% XS-VLDL-TG 9.80 × 10−18 0.150 0.017 0.89% 3.80 × 108 
 MLXIPL rs17145738 C/T 0.12 TG-lab 1.81 × 10−9 −0.140 0.023 0.42% VLDL-D 1.77 × 10−12 −0.169 0.024 0.62% 1.02 × 103 
 LPL rs12678919 A/G 0.09 TG-lab 2.60 × 10−9 −0.156 0.026 0.40% Val/Serum-TG 5.81 × 10−13 0.197 0.027 0.63% 4.47 × 103 
 ABCA1 rs1883025 C/T 0.19 HDL-C-lab 1.85 × 10−9 −0.117 0.019 0.43% Free-C/Est-C 3.04 × 10−11 −0.135 0.020 0.57% 6.07 × 101 
 FADS1-2-3 11 rs174546 C/T 0.42 LDL-C-lab 1.11 × 10−8 −0.089 0.015 0.38% LA/PUFA 4.77 × 10−268 0.563 0.016 15.41% 2.32 × 10259 
 APOA1 11 rs964184 C/G 0.14 TG-lab 2.98 × 10−24 0.220 0.022 1.17% Val/Serum-TG 1.09 × 10−26 −0.239 0.000 1.38% 2.73 × 102 
 LIPC 15 rs1532085 A/G 0.57 HDL-C-lab 2.58 × 10−16 −0.126 0.015 0.78% XL-HDL-TG 5.52 × 10−72 −0.285 0.016 3.97% 4.67 × 1055 
 HPRa 16 rs2000999 G/A 0.19 TC-lab 1.27 × 10−4 0.074 0.019 0.17% Gp/Tot-C 1.47 × 10−13 −0.148 0.020 0.67% 8.67 × 108 
 CILP2a 19 rs10401969 T/C 0.06 TC-lab 5.72 × 10−6 −0.143 0.032 0.24% MobCH 5.21 × 10−9 −0.188 0.032 0.42% 1.10 × 103 
 APOEa 19 rs439401 T/C 0.71 TG-lab 2.28 × 10−6 0.079 0.017 0.26% XS-VLDL-TG 3.12 × 10−9 0.103 0.017 0.43% 7.31 × 102 
 PLTPa 20 rs6065906 T/C 0.15 HDL-C-lab 6.41 × 10−2 −0.039 0.021 0.04% L-HDL-L/M-HDL-L 1.29 × 10−24 −0.221 0.022 1.27% 4.97 × 1022 
Small P-gain with metabolic refinement 
 GCKR rs1260326 T/C 0.65 TG-lab 1.49 × 10−18 −0.138 0.016 0.87% Ala/Gln 1.20 × 10−18 −0.150 0.017 1.03% 1.24 
 PPP1R3B rs9987289 A/G 0.85 TC-lab 1.26 × 10−8 0.120 0.021 0.37% IDL-C 3.20 × 10−9 0.129 0.022 0.42% 3.93 
 APOE 19 rs4420638 A/G 0.25 LDL-C-lab 9.35 × 10−23 0.231 0.023 2.02% S-LDL-L 3.38 × 10−23 0.238 0.001 2.15% 2.76 
No P-gain 
 SORT1 rs629301 G/T 0.79 LDL-C-lab 3.68 × 10−15 0.148 0.019 0.73% – – – – – – 
 APOB rs1367117 G/A 0.30 LDL-C-lab 4.50 × 10−12 0.120 0.017 0.60% – – – – – – 
 HMGCR rs12916 T/C 0.45 LDL-C-lab 1.30 × 10−10 0.100 0.016 0.50% – – – – – – 
 CETP 16 rs3764261 C/A 0.28 HDL-C-lab 6.32 × 10−49 0.251 0.017 2.52% – – – – – – 
 LDLR 19 rs6511720 G/T 0.12 LDL-C-lab 6.28 × 10−22 −0.233 0.024 1.12% – – – – – – 
 HNF4A 20 rs1800961 C/T 0.04 HDL-C-lab 1.03 × 10−8 −0.211 0.037 0.37% – – – – – – 

The three panels of the table list the significant associations ranked by their P-gain. In the top panel are the loci where the phenotype refinement resulted in a significant P-gain (P-gain >47, see Materials and Methods for details). In the middle panel are the loci where the lead SNP had a stronger association to a metabolite trait than to the enzymatic lipid but the P-gain was not significant. In the bottom panel are the loci where the enzymatic lipid trait was the most associated trait.

The gene name in ‘Locus’ column is the name used in Teslovich et al. (1). Chr; chromosome. Lead SNP is the reported strongest SNP in the locus. Column ‘Alleles’ has the non-coded and coded alleles, respectively, and CAF is the frequency of the coded allele. Enzymatic trait is the one of the 4 enzymatic lipid traits that had the strongest association with the lead SNP. Beta is the effect size in units of standard deviation, modelled as an additive effect of the coded allele, and SE the standard error. Lead trait is the trait with the smallest P-value across all 216 metabolite measures and the 4 enzymatic lipids for the lead SNP.

aThe associations of the lead SNPs with the enzymatic lipids in these loci did not reach genome-wide significance. See the Supplementary Material, Table S10 for metabolite trait abbreviations.

We then compared the P-values between the SNPs and enzymatic and metabolite phenotypes using the P-gain statistic (the ratio of the P-values of association of the SNP to the enzymatic and metabolite measures). In total, 13 lead SNPs (CILP2, APOE, APOB, LIPC, FADS1-2-3, MLXIPL, ABCA1, PLTP, GALNT2, HPR, ANGPTL3, APOA1 and LPL) with significant associations to the metabolites had P-gain >47 (multiple testing corrected significance threshold) (see Table 1 for details and Supplementary Material, Table S3 for P-gain comparisons of all the 102 lead SNPs). The proportion of variance explained by each associated SNP was considerably higher for the detailed metabolite measures (range 0.42–15.41%, median 0.66%) than for the enzymatic lipid measures (range 0.04–2.02%, median 0.39%) (Table 1).

The extensive lipoprotein subclass profile describes the particles involved in lipoprotein metabolism in more detail than the enzymatic lipid measures, which are composite measures of the cholesterol or TG carried in various circulating lipoprotein particles. Therefore, by investigating the associations of the loci across the subclass measures, we were able to map in detail the parts of the apoB cascade and HDL metabolism affected by the loci. Figure 1 presents the associations between the lead SNPs and lipoprotein subclasses (associations between all the metabolite and enzymatic lipid measures and the lead SNPs are provided in Supplementary Material, Table S2). Based on their associations to the subclass measures, the genes clustered into three main categories that exhibit distinct patterns of effects; genes mainly associated with (i) TG-rich VLDL particles (upper part of Fig. 1), with (ii) the cholesterol-carrying intermediate-density lipoprotein (IDL) and LDL particles and smallest VLDLs (top and middle sections of Fig. 1) or with (iii) the HDL particles (the bottom section of Fig. 1). These overall patterns of associations are in line with the previously reported associations to the enzymatic lipids (1). However, the association patterns within each group of genes were not uniform, but the associations to other subclass measures varied within each group. The associations with HDL particles were particularly heterogeneous; some loci demonstrated opposite associations between the measures of smaller and larger HDL particles and in several cases the association to the TG content of very large and small HDL particles followed the association of VLDL particles being the opposite to other HDL measures.

Figure 1.

The associations of the lead SNPs of known lipid loci with the lipoprotein subclasses. The effect estimates of the associations between the lead SNPs and the lipoprotein subclass measures plotted as heat maps; each row represents a SNP and each column a subclass measure. The lipoprotein subclasses are in size order, the largest VLDLs on the left, IDLs and LDLs in the middle and the smallest HDLs on the right. The order of the SNPs is based on the similarities in their effects on the subclass measures. The strength and direction of the effect is illustrated with the colour scale, blue indicating a negative and red a positive association with respect to the SNP effect on the trait the SNP had the most significant association to in the genome-wide scan (1). The units are standard deviations. A star (★) denotes genome-wide significant association (P-value < 5 × 10−8) and a dot (•) suggestive, i.e. nominally significant association (P<0.05). Seven loci, including APOB and APOE, had two reported lead SNPs, thus the double references to these loci in the locus names on the left.

Figure 1.

The associations of the lead SNPs of known lipid loci with the lipoprotein subclasses. The effect estimates of the associations between the lead SNPs and the lipoprotein subclass measures plotted as heat maps; each row represents a SNP and each column a subclass measure. The lipoprotein subclasses are in size order, the largest VLDLs on the left, IDLs and LDLs in the middle and the smallest HDLs on the right. The order of the SNPs is based on the similarities in their effects on the subclass measures. The strength and direction of the effect is illustrated with the colour scale, blue indicating a negative and red a positive association with respect to the SNP effect on the trait the SNP had the most significant association to in the genome-wide scan (1). The units are standard deviations. A star (★) denotes genome-wide significant association (P-value < 5 × 10−8) and a dot (•) suggestive, i.e. nominally significant association (P<0.05). Seven loci, including APOB and APOE, had two reported lead SNPs, thus the double references to these loci in the locus names on the left.

There were four genes where the aggregate nature of the conventional lipid measures appeared to be a disadvantage not capturing the heterogeneous associations to HDL measures or associations limited to specific subclass particles. The variants in PLTP had no association with HDL-C but a number of strong associations to the subclass measures existed, with the effects to large and small HDL particles being in the opposite direction and the strongest association being to the ratio of large and medium HDL particles. These associations point to the role of PLTP in modulating HDL particle size (9,10). Similarly, LIPC showed opposite associations between the smallest and larger HDL particles, the strongest association being to the TG content of very large HDLs, with the associations to average HDL particle size and HDL size ratios among the associations with the smallest P-values and largest P-gains. The LIPC SNP also associated to all very small VLDL and IDL measures. These detailed associations reveal important information not captured by the conventional lipid measures as there were no significant associations to either LDL-C or TG. Together these associations of LIPC to both IDL and HDL likely reflect hepatic lipase's (HL) dual role as a TG hydrolase in the conversion of IDL to LDL (11) and from HDL2 to smaller HDL3 (12). Similarly, neither TG nor LDL-C is a perfect descriptor of the associations of one APOB and one APOE SNP that affected the particles in the borderline of VLDL and LDL metabolism, the SNPs having the strongest associations to the TG content of very small VLDL particles.

The six loci for which an enzymatic measure was the most-associated trait showed consistent associations across subclass particles in certain lipoprotein class; the lead SNPs in HDL genes CETP and HNF4A associated coherently to all except one HDL measure, and LDL-C associated SNPs in LDLR, APOB, SORT1 and HMGCR, associated similarly yet with varying strength to the IDL and LDL measures. As HDL-C and LDL-C sum up the cholesterol carried in HDL and LDL particles, respectively, the association to the aggregate measure is the strongest, although the associations to individual subclass measures were not in all cases significant.

Interestingly, the associations of the lipid loci were not limited to lipoproteins but rather several loci showed associations to other lipid measures, small molecules or ratios derived from these measures. Among the genome-wide significant loci was GCKR that associated significantly to 70 metabolite measures. The associations span the whole NMR measurable metabolome; the SNP rs1260326 displayed associations to lipoproteins, FA measures and small molecule traits such as amino acids and their ratios, including the strongest association of the SNP to the ratio of alanine with glutamine (Supplementary Material, Table S2). Notably, in our data, the GCKR SNP did not associate with glucose concentration (P = 0.404) and the strongest association in the GCKR region to this trait only had suggestive evidence (2-27502746, P = 2.36 × 10−3). As expected, the FADS1-2-3 SNP rs174546 associated with several measures of FAs and their degree of saturation. These traits closely reflect the pathways the genes act on, and as such the SNP explains a high proportion of the variance in these traits (15.4% for the lead trait, the ratio of linoleic acid with other polyunsaturated FAs).

Combining the refined phenotypes and the dense map of SNPs

We hypothesized that additional SNPs in the 95 lipid loci may better tag the causative variants in homogeneous Finnish population samples and that the regions may harbour multiple signals. Thus, we sought further stronger associations by testing for all metabolite–SNP combinations within 1 Mb of the reported lead SNPs of the lipid loci taking advantage of the dense map of variants from the 1000 Genomes project. Altogether, 2928 SNPs within the lipid loci showed significant associations together to 181 metabolite traits giving in total 39901 unique pairs of association (Supplementary Material, Table S4). We identified stronger associations than the reported lead SNP for 33 SNPs mapping to 27 loci [Table 2, see Supplementary Material, Tables S5 and S6 for the comparisons between the reported lead SNPs and the most-associated SNPs for the lead traits (S5) and between the reported lead SNPs and the best metabolite-SNP pairings (S6)]. These include 11 regions (LDLRAP1, PCSK9, ABCG5/8, C6orf106, ABO, LRP4, LRP1, HNF1A, SCARB1, LCAT and LIPG) where the original lead SNP was not associated with any of the metabolite traits in our data. Notably, there were only four loci (CETP, HNF4A, APOA1 and GCKR) where the original SNP with association at the genome-wide significance level remained as the most strongly associated regional marker.

Table 2.

Significant loci after combining the refined metabolite traits and the dense map of variants: comparison of the associations of the reported lead SNP and lead metabolite trait to region's strongest association across all metabolite traits and variants

Locus Chr SNP Alleles CAF Lead trait P-value Beta SE Variance explained (%) New SNP Alleles CAF New trait P-value Beta SE Variance explained (%) r2 Distance Cor 
LDLRAP1a rs12027135 A/T 0.54 LDL-C-lab 1.46 × 10−6 0.076 0.016 0.29 rs35346083 C/A 0.42 LDL-C-lab 4.95 × 10−8 −0.091 0.017 0.40 0.93 12692 1.00 
PCSK9a rs2479409 G/A 0.72 LDL-C-lab 3.02 × 10−5 −0.077 0.019 0.24 1-55892749 C/A 0.02 LDL-C-lab 1.34 × 10−21 −0.598 0.063 1.46 0.02 615511 1.00 
ANGPTL3 rs2131925 G/T 0.74 Val/Serum-TG 4.01 × 10−12 −0.123 0.018 0.59 rs1168029 G/A 0.70 MobCH 1.18 × 10−13 0.129 0.017 0.71 0.88 56540 −0.83 
SORT1 rs629301 G/T 0.79 LDL-C-lab 3.68 × 10−15 0.148 0.019 0.73 rs660240 T/C 0.79 LDL-C-lab 1.94 × 10−15 0.149 0.019 0.75 0.99 468 1.00 
GALNT2 rs4846914 G/A 0.54 M-HDL-L/S-HDL-L 1.67 × 10−12 0.115 0.016 0.66 rs11122454 C/T 0.56 M-HDL-L/S-HDL-Lb 8.36 × 10−13 0.120 0.017 0.71 0.91 8360 1.00 
APOB rs1042034 C/T 0.73 XS-VLDL-TG 9.80 × 10−18 0.150 0.017 0.89 rs4665710 A/C 0.73 XS-VLDL-TG 9.17 × 10−18 0.150 0.017 0.89 1.00 4246 1.00 
APOB rs1367117 G/A 0.30 LDL-C-lab 4.50 × 10−12 0.120 0.017 0.60 rs4665710 A/C 0.73 XS-VLDL-TG 9.17 × 10−18 0.150 0.017 0.89 0.13 42865 0.83 
GCKR rs1260326 T/C 0.65 Ala/Gln 1.20 × 10−18 −0.150 0.017 1.03 rs1260326 T/C 0.65 Ala/Gln 1.20 × 10−18 −0.150 0.017 1.03 1.00 1.00 
ABCG5/8a rs4299376 G/T 0.78 Lac/Pyr 2.06 × 10−4 0.071 0.019 0.17 rs6756629 G/A 0.09 LDL-C-lab 1.88 × 10−10 −0.172 0.027 0.47 0.04 7486 −0.46 
HMGCR rs12916 T/C 0.45 LDL-C-lab 1.30 × 10−10 0.100 0.016 0.50 rs7703051 C/A 0.43 LDL-C-lab 3.86 × 10−11 0.103 0.016 0.52 0.93 31052 1.00 
C6orf106a rs2814982 C/T 0.16 M-HDL-PL 1.47 × 10−3 −0.068 0.021 0.12 6-34803686 G/A 0.19 M-HDL-CE 2.26 × 10−8 0.161 0.029 0.81 0.06 149148 0.92 
C6orf106a rs2814944 A/G 0.79 M-HDL-PL 1.22 × 10−3 0.062 0.019 0.13 6-34803686 G/A 0.19 M-HDL-CE 2.26 × 10−8 0.161 0.029 0.81 0.11 142911 0.92 
TYW1Ba,c rs13238203 T/C 0.94 S-HDL-P 2.33 × 10−2 0.137 0.004 0.20 rs13247874 C/T 0.20 VLDL-D 4.78 × 10−14 −0.161 0.021 0.83 0.07 880775 0.14 
MLXIPL rs17145738 C/T 0.12 VLDL-D 1.77 × 10−12 −0.169 0.024 0.62 rs13247874 C/T 0.20 VLDL-D 4.78 × 10−14 −0.161 0.021 0.83 0.67 27568 1.00 
PPP1R3B rs9987289 A/G 0.85 IDL-C 3.20 × 10−9 0.129 0.022 0.42 rs983309 T/G 0.83 IDL-C 2.43 × 10−9 0.122 0.021 0.42 0.90 5626 1.00 
LPL rs12678919 A/G 0.09 Val/Serum-TG 5.81 × 10−13 0.197 0.027 0.63 8-19956650 G/A 0.10 M-VLDL-PL 2.28 × 10−15 −0.222 0.028 0.90 0.77 68148 −0.96 
ABCA1 rs1883025 C/T 0.19 Free-C/Est-C 3.04 × 10−11 −0.135 0.020 0.57 rs2575876 G/A 0.19 Free-C/Est-C 1.63 × 10−11 −0.137 0.020 0.58 0.99 1438 1.00 
ABOa rs635634 C/T 0.22 XS-VLDL-L 8.43 × 10−7 0.095 0.019 0.31 rs11244035 C/T 0.10 LDL-C-lab 3.87 × 10−9 0.163 0.028 0.49 0.27 73681 0.92 
LRP4a 11 rs3136441 T/C 0.23 HDL-C-lab 3.14 × 10−4 0.065 0.018 0.15 rs3758673 C/T 0.41 HDL-C-lab 7.99 × 10−11 0.100 0.015 0.49 0.31 535670 1.00 
FADS1-2-3 11 rs174546 C/T 0.42 LA/PUFA 4.77 × 10−268 0.563 0.016 15.41 rs174547 T/C 0.42 LA/PUFA 1.31 × 10−269 0.569 0.016 15.72 0.99 953 1.00 
APOA1 11 rs964184 C/G 0.14 Val/Serum-TG 1.09 × 10−26 −0.239 0.000 1.38 rs964184 C/G 0.14 Val/Serum-TG 1.09 × 10−26 −0.239 0.000 1.38 1.00 1.00 
LRP1a 12 rs11613352 C/T 0.24 Ile/Tyr 1.07 × 10−3 −0.061 0.019 0.13 rs2638315 G/C 0.17 Gln/Glc 2.38 × 10−36 −0.290 0.023 2.42 0.00 927524 −0.53 
HNF1Aa 12 rs1169288 A/C 0.36 Phe/Tyr 6.78 × 10−5 −0.069 0.017 0.22 rs58706475 T/A 0.29 Tyr 2.07 × 10−8 0.112 0.020 0.51 0.13 349379 −0.52 
ZNF664ac 12 rs4765127 G/T 0.29 LDL-C-lab 1.35 × 10−5 −0.073 0.017 0.22 12-123911977 A/T 0.13 HDL-C-lab 1.68 × 10−8 0.147 0.026 0.49 0.00 885857 −0.55 
SCARB1a 12 rs838880 C/T 0.58 HDL-C-lab 1.15 × 10−5 −0.069 0.016 0.23 12-123911977 A/T 0.13 HDL-C-lab 1.68 × 10−8 0.147 0.026 0.49 0.05 84431 1.00 
LIPC 15 rs1532085 A/G 0.57 XL-HDL-TG 5.52 × 10−72 −0.285 0.016 3.97 rs35853021 G/T 0.39 XL-HDL-TG 7.11 × 10−76 0.306 0.017 4.46 0.81 2723 1.00 
CETP 16 rs3764261 C/A 0.28 HDL-C-lab 6.32 × 10−49 0.251 0.017 2.52 rs3764261 C/A 0.28 HDL-C-lab 6.32 × 10−49 0.251 0.017 2.51 1.00 1.00 
LCATa 16 rs16942887 G/A 0.15 HDL-C-lab 4.98 × 10−7 0.105 0.021 0.29 16-66448807 C/G 0.10 HDL-C-lab 2.34 × 10−8 0.149 0.027 0.40 0.65 36736 1.00 
HPR 16 rs2000999 G/A 0.19 Gp/Tot-C 1.47 × 10−13 −0.148 0.020 0.67 rs4788815 A/T 0.64 Phe/Tyr 7.37 × 10−18 0.147 0.017 0.98 0.04 473282 −0.13 
LIPGa 18 rs7241918 G/T 0.83 M-HDL-L/S-HDL-L 3.99 × 10−6 0.097 0.021 0.27 rs7228085 A/G 0.46 XL-HDL-TG 4.34 × 10−11 0.109 0.016 0.59 0.16 139 −0.01 
LDLR 19 rs6511720 G/T 0.12 LDL-C-lab 6.28 × 10−22 −0.233 0.024 1.12 19-11058749 A/G 0.10 LDL-C-lab 4.01 × 10−24 −0.286 0.028 1.45 0.88 4557 1.00 
LOC55908a,c 19 rs737337 T/C 0.09 PC 3.85 × 10−4 −0.099 0.028 0.16 19-11058749 A/G 0.10 LDL-C-lab 4.01 × 10−24 −0.286 0.028 1.45 0.00 149744 0.56 
CILP2 19 rs10401969 T/C 0.06 MobCH 5.21 × 10−9 −0.188 0.032 0.42 rs17216588 C/T 0.06 MobCH 1.04 × 10−9 −0.195 0.032 0.45 0.91 256359 1.00 
APOE 19 rs439401 T/C 0.71 XS-VLDL-TG 3.12 × 10−9 0.103 0.017 0.43 rs7412 C/T 0.04 LDL-C-lab 2.75 × 10−64 −0.752 0.044 4.62 0.01 2372 0.83 
APOE 19 rs4420638 A/G 0.25 S-LDL-L 3.38 × 10−23 0.238 0.001 2.15 rs7412 C/T 0.04 LDL-C-lab 2.75 × 10−64 −0.752 0.044 4.62 0.05 10867 0.93 
HNF4A 20 rs1800961 C/T 0.04 HDL-C-lab 1.03 × 10−8 −0.211 0.037 0.37 rs1800961 C/T 0.04 HDL-C-lab 1.03 × 10−8 −0.211 0.037 0.37 1.00 1.00 
PLTP 20 rs6065906 T/C 0.15 L-HDL-L/M-HDL-L 1.29 × 10−24 −0.221 0.022 1.27 rs6065904 G/A 0.24 L-HDL-L/M-HDL-L 1.49 × 10−32 −0.221 0.019 1.77 0.60 19364 1.00 
Locus Chr SNP Alleles CAF Lead trait P-value Beta SE Variance explained (%) New SNP Alleles CAF New trait P-value Beta SE Variance explained (%) r2 Distance Cor 
LDLRAP1a rs12027135 A/T 0.54 LDL-C-lab 1.46 × 10−6 0.076 0.016 0.29 rs35346083 C/A 0.42 LDL-C-lab 4.95 × 10−8 −0.091 0.017 0.40 0.93 12692 1.00 
PCSK9a rs2479409 G/A 0.72 LDL-C-lab 3.02 × 10−5 −0.077 0.019 0.24 1-55892749 C/A 0.02 LDL-C-lab 1.34 × 10−21 −0.598 0.063 1.46 0.02 615511 1.00 
ANGPTL3 rs2131925 G/T 0.74 Val/Serum-TG 4.01 × 10−12 −0.123 0.018 0.59 rs1168029 G/A 0.70 MobCH 1.18 × 10−13 0.129 0.017 0.71 0.88 56540 −0.83 
SORT1 rs629301 G/T 0.79 LDL-C-lab 3.68 × 10−15 0.148 0.019 0.73 rs660240 T/C 0.79 LDL-C-lab 1.94 × 10−15 0.149 0.019 0.75 0.99 468 1.00 
GALNT2 rs4846914 G/A 0.54 M-HDL-L/S-HDL-L 1.67 × 10−12 0.115 0.016 0.66 rs11122454 C/T 0.56 M-HDL-L/S-HDL-Lb 8.36 × 10−13 0.120 0.017 0.71 0.91 8360 1.00 
APOB rs1042034 C/T 0.73 XS-VLDL-TG 9.80 × 10−18 0.150 0.017 0.89 rs4665710 A/C 0.73 XS-VLDL-TG 9.17 × 10−18 0.150 0.017 0.89 1.00 4246 1.00 
APOB rs1367117 G/A 0.30 LDL-C-lab 4.50 × 10−12 0.120 0.017 0.60 rs4665710 A/C 0.73 XS-VLDL-TG 9.17 × 10−18 0.150 0.017 0.89 0.13 42865 0.83 
GCKR rs1260326 T/C 0.65 Ala/Gln 1.20 × 10−18 −0.150 0.017 1.03 rs1260326 T/C 0.65 Ala/Gln 1.20 × 10−18 −0.150 0.017 1.03 1.00 1.00 
ABCG5/8a rs4299376 G/T 0.78 Lac/Pyr 2.06 × 10−4 0.071 0.019 0.17 rs6756629 G/A 0.09 LDL-C-lab 1.88 × 10−10 −0.172 0.027 0.47 0.04 7486 −0.46 
HMGCR rs12916 T/C 0.45 LDL-C-lab 1.30 × 10−10 0.100 0.016 0.50 rs7703051 C/A 0.43 LDL-C-lab 3.86 × 10−11 0.103 0.016 0.52 0.93 31052 1.00 
C6orf106a rs2814982 C/T 0.16 M-HDL-PL 1.47 × 10−3 −0.068 0.021 0.12 6-34803686 G/A 0.19 M-HDL-CE 2.26 × 10−8 0.161 0.029 0.81 0.06 149148 0.92 
C6orf106a rs2814944 A/G 0.79 M-HDL-PL 1.22 × 10−3 0.062 0.019 0.13 6-34803686 G/A 0.19 M-HDL-CE 2.26 × 10−8 0.161 0.029 0.81 0.11 142911 0.92 
TYW1Ba,c rs13238203 T/C 0.94 S-HDL-P 2.33 × 10−2 0.137 0.004 0.20 rs13247874 C/T 0.20 VLDL-D 4.78 × 10−14 −0.161 0.021 0.83 0.07 880775 0.14 
MLXIPL rs17145738 C/T 0.12 VLDL-D 1.77 × 10−12 −0.169 0.024 0.62 rs13247874 C/T 0.20 VLDL-D 4.78 × 10−14 −0.161 0.021 0.83 0.67 27568 1.00 
PPP1R3B rs9987289 A/G 0.85 IDL-C 3.20 × 10−9 0.129 0.022 0.42 rs983309 T/G 0.83 IDL-C 2.43 × 10−9 0.122 0.021 0.42 0.90 5626 1.00 
LPL rs12678919 A/G 0.09 Val/Serum-TG 5.81 × 10−13 0.197 0.027 0.63 8-19956650 G/A 0.10 M-VLDL-PL 2.28 × 10−15 −0.222 0.028 0.90 0.77 68148 −0.96 
ABCA1 rs1883025 C/T 0.19 Free-C/Est-C 3.04 × 10−11 −0.135 0.020 0.57 rs2575876 G/A 0.19 Free-C/Est-C 1.63 × 10−11 −0.137 0.020 0.58 0.99 1438 1.00 
ABOa rs635634 C/T 0.22 XS-VLDL-L 8.43 × 10−7 0.095 0.019 0.31 rs11244035 C/T 0.10 LDL-C-lab 3.87 × 10−9 0.163 0.028 0.49 0.27 73681 0.92 
LRP4a 11 rs3136441 T/C 0.23 HDL-C-lab 3.14 × 10−4 0.065 0.018 0.15 rs3758673 C/T 0.41 HDL-C-lab 7.99 × 10−11 0.100 0.015 0.49 0.31 535670 1.00 
FADS1-2-3 11 rs174546 C/T 0.42 LA/PUFA 4.77 × 10−268 0.563 0.016 15.41 rs174547 T/C 0.42 LA/PUFA 1.31 × 10−269 0.569 0.016 15.72 0.99 953 1.00 
APOA1 11 rs964184 C/G 0.14 Val/Serum-TG 1.09 × 10−26 −0.239 0.000 1.38 rs964184 C/G 0.14 Val/Serum-TG 1.09 × 10−26 −0.239 0.000 1.38 1.00 1.00 
LRP1a 12 rs11613352 C/T 0.24 Ile/Tyr 1.07 × 10−3 −0.061 0.019 0.13 rs2638315 G/C 0.17 Gln/Glc 2.38 × 10−36 −0.290 0.023 2.42 0.00 927524 −0.53 
HNF1Aa 12 rs1169288 A/C 0.36 Phe/Tyr 6.78 × 10−5 −0.069 0.017 0.22 rs58706475 T/A 0.29 Tyr 2.07 × 10−8 0.112 0.020 0.51 0.13 349379 −0.52 
ZNF664ac 12 rs4765127 G/T 0.29 LDL-C-lab 1.35 × 10−5 −0.073 0.017 0.22 12-123911977 A/T 0.13 HDL-C-lab 1.68 × 10−8 0.147 0.026 0.49 0.00 885857 −0.55 
SCARB1a 12 rs838880 C/T 0.58 HDL-C-lab 1.15 × 10−5 −0.069 0.016 0.23 12-123911977 A/T 0.13 HDL-C-lab 1.68 × 10−8 0.147 0.026 0.49 0.05 84431 1.00 
LIPC 15 rs1532085 A/G 0.57 XL-HDL-TG 5.52 × 10−72 −0.285 0.016 3.97 rs35853021 G/T 0.39 XL-HDL-TG 7.11 × 10−76 0.306 0.017 4.46 0.81 2723 1.00 
CETP 16 rs3764261 C/A 0.28 HDL-C-lab 6.32 × 10−49 0.251 0.017 2.52 rs3764261 C/A 0.28 HDL-C-lab 6.32 × 10−49 0.251 0.017 2.51 1.00 1.00 
LCATa 16 rs16942887 G/A 0.15 HDL-C-lab 4.98 × 10−7 0.105 0.021 0.29 16-66448807 C/G 0.10 HDL-C-lab 2.34 × 10−8 0.149 0.027 0.40 0.65 36736 1.00 
HPR 16 rs2000999 G/A 0.19 Gp/Tot-C 1.47 × 10−13 −0.148 0.020 0.67 rs4788815 A/T 0.64 Phe/Tyr 7.37 × 10−18 0.147 0.017 0.98 0.04 473282 −0.13 
LIPGa 18 rs7241918 G/T 0.83 M-HDL-L/S-HDL-L 3.99 × 10−6 0.097 0.021 0.27 rs7228085 A/G 0.46 XL-HDL-TG 4.34 × 10−11 0.109 0.016 0.59 0.16 139 −0.01 
LDLR 19 rs6511720 G/T 0.12 LDL-C-lab 6.28 × 10−22 −0.233 0.024 1.12 19-11058749 A/G 0.10 LDL-C-lab 4.01 × 10−24 −0.286 0.028 1.45 0.88 4557 1.00 
LOC55908a,c 19 rs737337 T/C 0.09 PC 3.85 × 10−4 −0.099 0.028 0.16 19-11058749 A/G 0.10 LDL-C-lab 4.01 × 10−24 −0.286 0.028 1.45 0.00 149744 0.56 
CILP2 19 rs10401969 T/C 0.06 MobCH 5.21 × 10−9 −0.188 0.032 0.42 rs17216588 C/T 0.06 MobCH 1.04 × 10−9 −0.195 0.032 0.45 0.91 256359 1.00 
APOE 19 rs439401 T/C 0.71 XS-VLDL-TG 3.12 × 10−9 0.103 0.017 0.43 rs7412 C/T 0.04 LDL-C-lab 2.75 × 10−64 −0.752 0.044 4.62 0.01 2372 0.83 
APOE 19 rs4420638 A/G 0.25 S-LDL-L 3.38 × 10−23 0.238 0.001 2.15 rs7412 C/T 0.04 LDL-C-lab 2.75 × 10−64 −0.752 0.044 4.62 0.05 10867 0.93 
HNF4A 20 rs1800961 C/T 0.04 HDL-C-lab 1.03 × 10−8 −0.211 0.037 0.37 rs1800961 C/T 0.04 HDL-C-lab 1.03 × 10−8 −0.211 0.037 0.37 1.00 1.00 
PLTP 20 rs6065906 T/C 0.15 L-HDL-L/M-HDL-L 1.29 × 10−24 −0.221 0.022 1.27 rs6065904 G/A 0.24 L-HDL-L/M-HDL-L 1.49 × 10−32 −0.221 0.019 1.77 0.60 19364 1.00 

The gene name in ‘Locus’ column is the name used in Teslovich et al. (1). Chr; chromosome. Lead SNP is the reported strongest SNP in the locus. Column ‘Alleles’ has the non-coded and coded alleles, respectively, and CAF is the frequency of the coded allele. Lead trait is the trait with the smallest P-value across all 216 metabolite measures and the 4 enzymatic lipids for the lead SNP. Beta; the effect size in units of standard deviation, modelled as an additive effect of the coded allele. SE; the standard error. New SNP and new trait are the most-associated variant and trait pair of the region. r2; the square of the correlation coefficient between the lead and new SNP Distance, distance (in base pairs) between the lead and best SNP. Cor, the correlation between the lead and new traits.

aThe associations of the reported lead SNPs with the enzymatic lipids in these loci did not reach genome-wide significance.

bResult missing from one cohort.

cThe new SNP and trait pair for these loci is the same as for the neighbouring lipid locus and most likely tags variation from that locus. See the Supplementary Material, Table S10 for metabolite trait abbreviations.

For 17 loci, the metabolite with strongest association was the same as for the previously reported lead SNP, but for 10 loci a completely new SNP–metabolite pairing was identified. For 11 loci (LDLRAP1, PCSK9, SORT1, ABCG5/8, HMGCR, ABO, LRP4, SCARB1, LCAT, LDLR and APOE), the best trait was an enzymatic lipid measure (HDL-C or LDL-C) and for 16 a more refined phenotype. Six of the new variants arise from the 1000 Genomes reference panel and two have minor allele frequency (MAF) <5% (for 1-55892749 MAF = 0.021 and for rs7412 MAF = 0.043). The proportion of variance explained by each of the 33 new variants was considerably higher (range 0.40–15.72%, median 0.81%) than for the corresponding original lead SNPs (range 0.12–15.41%, median 0.42%). Supplementary Material, Figure S1 shows the regional association plots for two loci (LIPC and PLTP) as examples of regions where significant improvement in the explained variance was obtained.

Majority of the new variants were in moderate or strong correlation with the reported lead SNP (r2 > 0.50) and thus these associations likely arise from the same association signal but the new SNPs may tag the causative variant better in Finnish population. However, for nine loci, the identified lead variants correlate only weakly (r2<0.2) with the reported lead SNP of that locus and therefore these may represent another signal. Interestingly, for two loci, the regions' strongest associations appear to point to a different gene. A variant 930 kb from the LRP1 SNP in recently reported glutamine locus (13) associated with the ratio of glutamine with glucose, explaining 2.42% of the variance of this measure, and a SNP 470 kb from HPR lead SNP associated with the ratio of phenylalanine to tyrosine, which is close to the tyrosine aminotransferase gene.

In order to evaluate whether the newly identified SNPs are more likely to be functional than the reported variants, we compared the SNP associations to leukocyte expression in a subset of the sample (n = 585) (Supplementary Material, Table S7). In two significant loci, the cis-expression quantitative trait locus (eQTL) for the most-associated SNP was considerably stronger than the corresponding eQTL for the reported lead SNP; the strongest signal in the C6orf106 locus showed a significant eQTL to DEF6 (P = 2.31 × 10−8), a guanine nucleotide exchange factor, while the two reported lead SNPs of the locus showed no significant association to the expression levels of this gene (P = 0.221 and P = 0.238 for rs2814944 and rs2814982, respectively). For the LRP1 locus, the new SNP had a strong eQTL for SPRYD4 (P = 3.79 × 10−13).

Since many of the regions showed a large number of associated SNPs, we tested for statistical independence of the signals by conditioning on the originally reported SNP in each locus and recursively for the most significantly associated variant until no further genome-wide significant associations were found. In total, 12 of the loci had more than one independent association (PCSK9, APOB, TYW1B, FADS1-2-3, LRP1, LIPC, HPR, CETP, LOC55908, CILP2, APOE and PLTP) and four regions (APOB, LIPC, HPR and APOE) had 3 or more independently associated SNPs (Supplementary Material, Table S8 and Fig. S2).

As an example of both genetic and metabolic refinement, a locus with multiple signals was APOB, in which there are two previously identified independent (r2 = 0.118) lead SNPs; one (rs1367117) associated with TC and LDL-C and the other (rs1042034) with TG and HDL-C. We found one further independent SNP (rs562338) in the region and an additional variant (rs4665710), which is in LD with rs1042034, responsible for the strongest association signals in the region. These four variants showed distinct profiles in their associations with the lipoprotein subclass measures (Fig. 2). Two of the SNPs (rs1042034 and rs4665710) associated primarily to the TG-rich medium, small and very small VLDL particles with weak opposite associations to largest HDLs, while the two other SNPs showed an association profile shifted towards smaller apoB lipoproteins. These SNPs associated specifically with increased very small VLDL and IDL measures, the previously reported lead SNP also associating with increases in LDL measures, but having merely suggestive associations to HDL and other VLDL particles. The SNPs with similar association profiles cluster together in terms of genomic positions; the VLDL-associated SNPs are positioned at the 3′ end of the APOB gene close to the LDL receptor-binding region of the gene while the three other variants are in the 5′ end. In terms of associations and positions the SNPs seem to relate to functionally distinct domains of the APOB gene. The 5′ end of the gene encodes the lipovitellin-like domain of the apoB protein that is important during apoB-containing lipoprotein particle assembly. The other end of the protein harbours the domain that binds to LDL receptor and an α-helix that is suggested to modulate the surface pressure as the lipoprotein particle size decreases during lipoprotein metabolism (14,15).

Figure 2.

The associations of the independent and region's strongest variant in the APOB locus to the lipoprotein subclass measures. The effect estimates of the associations between the SNPs in the APOB locus and lipoprotein subclasses plotted as heat maps; each row represents a SNP and each column a metabolite. The grid on the right next to the heat map illustrates conditional model that led to the discovery of the SNP or indicates whether the SNP is the strongest variant in the region with symbols described in the figure legend. The strength and direction of the effect is illustrated with the colour scale, blue indicating a negative and red a positive association with respect to the allele that increases the trait with the smallest P-value across all 216 metabolite traits. The units are standard deviations. A star (★) denotes genome-wide significant association (P-value < 5 × 10−8) and a dot (•) suggestive, i.e. nominally significant association (P<0.05). The APOB variants show heterogeneity in their effects on the lipoprotein subclasses. The variants with similar effect profiles cluster together in terms of genomic positions, thus likely affecting specific functional domains of the encoded protein.

Figure 2.

The associations of the independent and region's strongest variant in the APOB locus to the lipoprotein subclass measures. The effect estimates of the associations between the SNPs in the APOB locus and lipoprotein subclasses plotted as heat maps; each row represents a SNP and each column a metabolite. The grid on the right next to the heat map illustrates conditional model that led to the discovery of the SNP or indicates whether the SNP is the strongest variant in the region with symbols described in the figure legend. The strength and direction of the effect is illustrated with the colour scale, blue indicating a negative and red a positive association with respect to the allele that increases the trait with the smallest P-value across all 216 metabolite traits. The units are standard deviations. A star (★) denotes genome-wide significant association (P-value < 5 × 10−8) and a dot (•) suggestive, i.e. nominally significant association (P<0.05). The APOB variants show heterogeneity in their effects on the lipoprotein subclasses. The variants with similar effect profiles cluster together in terms of genomic positions, thus likely affecting specific functional domains of the encoded protein.

DISCUSSION

In this study, we characterized the 95 recently reported lipid loci by using serum NMR-based characterization of 95 lipoprotein and lipid measures along with 121 other metabolic variables, and a comprehensive set of 440 807 directly genotyped and imputed markers using the 1000 Genomes imputation reference panel. In total, for 30 loci we provided new association profiles to either new phenotypes or to a new lead SNP.

Our approach of characterizing the lipid regions both by phenotype and by genotype offered several benefits in providing insight into lipoprotein metabolism and the genetic architecture of the lipid loci. First, we showed that in the majority of the loci, the previously reported variants demonstrate stronger associations to the more detailed phenotypes than to the enzymatic lipid measures. The median estimated proportion of phenotype variance explained by the previously identified markers across metabolites was 69% higher than for the enzymatic lipids. As conventional laboratory measures of HDL-C, LDL-C, TC and TG sum up the lipids carried in lipoprotein particles of various sizes and composition, our results are likely revealing more specific metabolic functions of these loci.

Secondly, many loci were associated with a wide range of metabolite traits. These include the heterogeneous association patterns to lipoprotein subclasses (Fig. 3) and also associations to many small molecules not known to be directly related to lipid metabolism. While some loci showed consistent associations over certain types of lipoproteins, for example, SORT1 associated to all LDL particles and to IDL but showed little association with the other subclass measures, LPL, LIPC, PLTP and CETP demonstrated opposite associations between the lipid measures of larger and smaller HDL particles for the same allele. Although originally identified as lipid loci, our associations were not limited to lipoprotein measures, and a few genetic loci showed evidence of a role in other processes of energy metabolism. Expectedly, the SNP in the FADS1-2-3 gene cluster, the genes of which encode for FA desaturase enzymes, associated to multiple measures of FA levels and degree of unsaturation. However, somewhat surprisingly, GCKR, a gene encoding for glucokinase regulatory protein, which inhibits glucokinase in hepatocytes and often implicated in glucose (2) and TG metabolism (1), showed the strongest evidence of association with the ratio of alanine with glutamine and in total associated significantly with 70 metabolite measures.

Figure 3.

A summary of the gene effects on lipoprotein subclasses with the primary pathways of endogenous lipoprotein metabolism. From top to bottom: VLDL particles are assembled in the liver and secreted into bloodstream where the particles are converted through a number of steps into IDL and subsequently to LDL particles. LDL and/or IDL particles are taken up by the liver and peripheral tissues, where the particles donate the lipids they carry and the particles are subsequently degraded. The nascent HDL produced in liver and peripheral tissues are precursors of the HDL particles that transport lipids from the periphery to the liver. A number of genes are involved in these processes; the genes encoding for the most relevant transfer proteins, enzymes and receptors as well as the genes affecting the lipoprotein measures genome-wide significantly in the present study are illustrated. In case functional evidence exists, genes from the latter category are also placed close to the expected site of action. White bars indicate the range of lipoprotein subclasses affected by each gene in the present study. Green and red dots illustrate that the gene only has significant associations with HDL particles or with apoB-containing lipoproteins, respectively. The IDL fraction is rarely included in the conventional laboratory measurements of blood lipids as an entity, yet our results underscore the prominent role of this lipoprotein class, which in many cases is distinct from LDLs; IDLs in concert with very small VLDLs associated with majority of the genes, including LIPC that showed specific associations to these particles, thus emphasizing the role of these particles in the lipoprotein cascade in the borderline of TG-dominated VLDL particles and cholesterol-carrying LDLs. In addition, PPP1R3B strongly affected IDL particles and the cholesterol content of large LDLs.

Figure 3.

A summary of the gene effects on lipoprotein subclasses with the primary pathways of endogenous lipoprotein metabolism. From top to bottom: VLDL particles are assembled in the liver and secreted into bloodstream where the particles are converted through a number of steps into IDL and subsequently to LDL particles. LDL and/or IDL particles are taken up by the liver and peripheral tissues, where the particles donate the lipids they carry and the particles are subsequently degraded. The nascent HDL produced in liver and peripheral tissues are precursors of the HDL particles that transport lipids from the periphery to the liver. A number of genes are involved in these processes; the genes encoding for the most relevant transfer proteins, enzymes and receptors as well as the genes affecting the lipoprotein measures genome-wide significantly in the present study are illustrated. In case functional evidence exists, genes from the latter category are also placed close to the expected site of action. White bars indicate the range of lipoprotein subclasses affected by each gene in the present study. Green and red dots illustrate that the gene only has significant associations with HDL particles or with apoB-containing lipoproteins, respectively. The IDL fraction is rarely included in the conventional laboratory measurements of blood lipids as an entity, yet our results underscore the prominent role of this lipoprotein class, which in many cases is distinct from LDLs; IDLs in concert with very small VLDLs associated with majority of the genes, including LIPC that showed specific associations to these particles, thus emphasizing the role of these particles in the lipoprotein cascade in the borderline of TG-dominated VLDL particles and cholesterol-carrying LDLs. In addition, PPP1R3B strongly affected IDL particles and the cholesterol content of large LDLs.

Thirdly, examination of the dense set of variants around the lead SNPs in the lipid loci led to the discovery of additional variants that showed stronger associations to the metabolites in 27 loci and 2 or more statistically independent variants for 12 loci. This demonstrates the power of utilizing a dense marker set and a homogeneous population for further characterization of the genetic association signals. For some of the loci with multiple independent associations, the trait variance explained by the locus was considerably larger. For example, the four independent LIPC variants explained on average 2.7 times more of the variance in very large and large HDL particles and up to 9.86% of the variance of a single trait (the TG content of very large HDL particles). Similarly, the two distinct association profiles across lipoprotein subclasses were captured by four SNPs in the APOB locus. The two variants in the 5′ end of APOB at the domain of apoB that binds microsomal transfer protein (MTP) (14) affected IDL and LDL particles as MTP–apoB interaction is essential for VLDL particle assembly (16,17). The two other APOB variants associated with particles from medium-sized VLDLs to IDLs reside near the region encoding for the LDLR-binding domain of apoB and therefore the SNPs could disrupt the clearance of LDL from circulation. Our findings extend the previous observations by Chasman et al. (7), who studied a set of 22 lipoprotein measures and identified 3 classes of associations in the APOB region with the associations to VLDL and LDL measures residing in separate groups.

Our study sample included >8000 Finnish individuals from 5 population-based cohorts. Since Finns are a genetic isolate within Europe, the associations found here may require verification in other populations. However, our focus on analysing the Finnish population allowed us to minimize heterogeneity previously observed in genetic marker frequencies (18,19), thus increasing our statistical power. We used an NMR-based metabolomics approach to dissect a serum metabolome comprising of 216 metabolite measures, a priori defined physiologically relevant ratios and other derived variables. Alternative screening techniques such as liquid chromatography coupled with mass spectrometry may offer a better and more detailed coverage of individual lipid molecules and other circulating metabolites. However, the metabolomics platform applied allowed the measurement of the most comprehensive lipoprotein subclass distribution to date in a high-throughput manner, enabling a precise characterization of the lipid gene effects on the lipoprotein cascade.

Our findings highlight the importance of biochemically accurate trait measures (20) and support the findings that refined phenotyping of lipoproteins often results in stronger associations than are achieved by studying the conventional lipoprotein measures (7,21). As lipoprotein particle size and composition are known to be important for disease susceptibility and risk (22,23), our findings provide a foundation for further mechanistic studies to identify potential intervention targets. In particular, the HDL subclasses seem to differ in their relationship to coronary heart disease (CHD); the larger HDL particles are generally considered more anti-atherogenic than the smaller ones (23–26). Therefore, the observed heterogeneity in the biological effects may mask the detection of an association between CHD and the genes found to associate with HDL-C.

To conclude, our study demonstrates the advantage of the combination of detailed genotyping and a wide metabolite panel. The study reveals a broader set of metabolic processes governed by genes originally identified using a limited set of clinical lipid measurements and shows that in a homogeneous population the variants best tagging the variation in the metabolite traits are often different from those first reported. With further understanding of these processes, possible novel interventions for metabolic diseases may be revealed.

MATERIALS AND METHODS

Study cohorts

We used genetic and metabolomic data from five population-based Finnish cohorts (n = 8330): NFBC1966, YF, HBCS, H2000 and DILGOM. The cohorts have been described in detail elsewhere (27–31). In brief, NFBC1966 is a birth cohort and a follow-up study of children born in 1966 in the two northernmost provinces of Finland; YF is a longitudinal follow-up study of children and adolescents from all around Finland; HBCS comprises of men and women born 1934–1944 in Helsinki, Finland; DILGOM is an extension to the National FINRISK Study 2007 survey and includes Finnish individuals 25–74 years of age; H2000 is a subset from the Health 2000 survey collected in 2000 and includes metabolic syndrome cases and their matched controls. All participants provided informed consent, and local ethical committees at participating institutions approved individual study protocols.

Genotyping and imputation

NFBC1966 was genotyped using Illumina HumanHap CNV 370k array, DILGOM and H2000 with Illumina HumanHap 610k array, and YF and HBCS using custom generated Illumina HumanHap 670k array. Quality control was performed independently for each study prior to imputation. Low quality markers (locus missingness > 5%) and poor DNA samples (individual missingness >5%) were removed. In addition, individuals with high genomic heterozygosity (indicating sample contamination), gender discrepancies or closely related individuals were removed from the data. The cleaned genome-wide data were imputed using IMPUTE software (32). Imputation included a 1000 Genomes reference (33), HapMap3 reference (34), which included an additional Finnish reference in HapMap3 depth (18), and HapMap2 reference (35). The benefit of the additional Finnish reference has been discussed in detail by Surakka et al. (18). The imputation references for the variants which were not directly genotyped among the 102 reported lead SNPs (1) are listed in Supplementary Material, Table S9.

Cis-eQTL analysis

Whole blood expression data were available for a subset of the DILGOM cohort (n = 585), described previously by Inouye et al. (30). In brief, to obtain stabilized total RNA in study III, the PAXgene Blood RNA System (PreAnalytiX GMbH, Hombrechtikon, Switzerland) was used. Biotinylated cRNA (750 ng) was hybridized onto Illumina HumanHT-12 Expression BeadChips (Illumina Inc., San Diego, CA, USA), according to the manufacturer's protocol. The correlation of the allele dosages of the reported lead SNPs and the SNPs identified in the combined metabolic and genetic characterization and all expression probes within 1 Mb of the SNPs was tested using Spearman rank correlation in R. All significant eQTLs (P<6 × 10−5) are summarized in Supplementary Material, Table S7.

Serum NMR metabolomics

All samples were analysed using the same high-throughput serum NMR metabolomics platform (8). This methodology provided information on 117 serum metabolites and 99 derived variables. The directly measured metabolites include lipoprotein subclass distribution and lipoprotein particle concentrations, low-molecular-weight metabolites such as amino acids, ketone bodies and creatinine, and detailed molecular information on serum lipid extracts including free and esterified cholesterol, sphingomyelin, degree of saturation and ω-3 FAs. Derived variables include selected ratios of metabolites implicated in lipolysis, proteolysis, ketogenesis and glycolysis as well as reagents and products of enzymatic reactions and measures obtained with the extended-Friedewald formula (36). Further details of the NMR spectroscopy, quantification data analyses as well as the full metabolite identifications have been described previously (6,8). A full list of the studied metabolites and derived variables is provided in Supplementary Material, Table S10.

In short, the NMR metabolomics platform included the measurement of total lipid and particle concentrations in 14 lipoprotein subclasses. The measurements of these subclasses have been calibrated using high performance liquid chromatography methods (37). The subclasses were as follows: chylomicron and largest VLDL particles (particle diameters from ∼75 nm upwards), five different VLDL subclasses: very large VLDL (average particle diameter 64.0 nm), large VLDL (53.6 nm), medium-size VLDL (44.5 nm), small VLDL (36.8 nm) and very small VLDL (31.3 nm); IDL (28.6 nm); three LDL subclasses: large LDL (25.5 nm), medium-size LDL (23.0 nm) and small LDL (18.7 nm); and four HDL subclasses: very large HDL (14.3 nm), large HDL (12.1 nm), medium-size HDL (10.9 nm) and small HDL (8.7 nm). The following components of the lipoprotein particles were also quantified: phospholipids, TG, TC, free cholesterol and cholesterol esters; due to resolution and concentration issues, all of these components are not available for every subclass. In addition, mean particle diameters of VLDL, LDL and HDL fractions were calculated on the basis of the corresponding subclass distributions (IDL particles were included in the LDL fraction).

Statistical analysis

The association between 216 metabolic measures and the 102 reported lead SNPs representing the 95 known lipid loci was tested in each cohort separately with a linear additive model. Individuals on lipid-lowering medication, pregnant and those not fasted were excluded. The metabolic traits were adjusted for age, sex and 10 principal components and inverse normal transformed prior association analyses. The cohort specific association results were combined using an inverse variance weighted fixed effects meta-analysis. An association was considered significant if it reached genome-wide significance (P<5 × 10−8) and suggestive if the P-value was <0.05 and ≥5 × 10−8. The results of the association analysis are given in Supplementary Material, Table S1.

To quantify the gain from using the more detailed phenotype measures over the enzymatic lipids, we calculated a P-gain statistic for each SNP where a metabolomics trait showed a smaller P-value than the best enzymatic trait. P-gain is the ratio of the P-values of two associations, in this case the ratio of the P-value of the association to the best enzymatic trait and the P-value of a metabolomics trait. To correct for multiple testing of 216 metabolite traits, the threshold for significant P-gain was set to 47, which is the number of principal components explaining >99% of the variation in the metabolomics data.

To evaluate whether alternative variants within the lipid loci tag the causative variants better in the Finnish population or whether there are other stronger signals in the loci, we performed association analyses for all the variants within the lipid loci using 1 Mb analysis windows around each of the 102 lead SNPs. Additionally, in order to identify further independent signals within the lipid loci, we performed association analyses conditioning each locus with the genotypes at the reported lead SNP (or two lead SNPs for seven loci). Analysis windows spanning 2Mb around the lead SNP (or a window including an overlap of 1 Mb flanking regions of both lead SNPs) were used. For those loci where genome-wide significant associations (P<5 × 10−8) to any of the 216 metabolic traits remained after conditioning, a further round of conditional analyses was performed adding the most significantly associated independent SNP of the region across all the traits to the covariates using analysis windows including an overlap of 1 Mb flanking regions of the lead SNP (or lead SNPs) and the independent variant. On each round of conditional analyses, the cohorts were analysed separately and combined in a meta-analysis as previously.

R statistics software was used for phenotype adjustments. The association analyses were performed using SNPTEST and R and meta-analyses with META.

Heat map visualization

To enable simultaneous detection and comparison of the multiple associations and to illustrate the metabolic continuum of the lipoprotein subclasses, the effect estimates for all of the tests were plotted on colour-coded 2D maps, i.e. heat maps, each line representing a SNP and each column a metabolite and the colour range from blue to red a negative and a positive association, respectively. In each heat map, the effect allele of each SNP was selected to be the one that increased the trait with the strongest association in the original study. In order to group genes with similar effects together, the SNPs were ordered with multidimensional scaling according to their effect profiles on the subclass variables, the apolipoprotein B (apoB) pathway (VLDL, IDL and LDL variables) and the variables related to HDL metabolism contributing with equal weight to the similarity measures. For the visualization of the loci with multiple signals, SNPs of the same locus were clustered together, and the colouring for each SNP was determined by the allele that increased the trait with the most significant association.

Clinical lipid measurements

For comparison with the subclass data and the original study, also the associations between the clinical measurements of TC, TG, LDL-C and HDL-C and the lead SNPs of the 95 lipid loci were tested. The association analysis was performed as described above using the same cohorts and individuals. The laboratory procedures for the clinical lipoprotein measurements have been described previously (1). The results of the association analysis are given in Supplementary Material, Table S1 with the associations to the metabolite traits.

SUPPLEMENTARY MATERIAL

Supplementary Material is available at HMG online.

FUNDING

This work was supported by the Academy of Finland [grant numbers 137870 (to P.S.), 132629 (to M.J.), 139635 (to V.S.), 126925, 121584, 124282, 129378, 117787 and 41071 (to The Young Finns Study)]; the Responding to Public Health Challenges Research Programme of the Academy of Finland [grant numbers 129269 (to M.J.S.) and 129429 (to M.A.-K.)]; The European Community's Seventh Framework Programme (FP7/2007-2013), BioSHaRE Consortium (grant agreement 261433); the Finnish Cardiovascular Research Foundation (to M.J.S., M.A.-K. and S.R.); the Social Insurance Institution of Finland (to The Young Finns Study); Kuopio, Tampere and Turku University Hospital Medical Funds (to The Young Finns Study); Turku University Foundation (to The Young Finns Study); Juho Vainio Foundation (to The Young Finns Study); Paavo Nurmi Foundation (to The Young Finns Study); Finnish Foundation for Cardiovascular Research (to The Young Finns Study); Finnish Cultural Foundation (to The Young Finns Study and T.T.); Yrjö Jahnsson Foundation (to The Young Finns Study); Emil Aaltonen Foundation (to T.L.); Tampere Tuberculosis Foundation (to T.L. and M.K.); the Jenny and Antti Wihuri Foundation (to A.J.K.); Instrumentarium Science Foundation (to T.T.) and Aalto University School of Science and Technology researcher training scholarship (to T.T.).

ACKNOWLEDGEMENTS

The authors would like to acknowledge the substantial contribution of the late Academian of Science Leena Peltonen. The expert technical assistance in the statistical analyses by Irina Lisinen and Ville Aalto are gratefully acknowledged. This research was conducted using the resources of CSC—IT Center for Science Ltd., Finland, and FiMM Technology Centre, University of Helsinki, Finland.

Conflict of Interest statement. None declared.

REFERENCES

1
Teslovich
T.M.
Musunuru
K.
Smith
A.V.
Edmondson
A.C.
Stylianou
I.M.
Koseki
M.
Pirruccello
J.P.
Ripatti
S.
Chasman
D.I.
Willer
C.J.
, et al.  . 
Biological, clinical and population relevance of 95 loci for blood lipids
Nature
 , 
2010
, vol. 
466
 (pg. 
707
-
713
)
2
Dupuis
J.
Langenberg
C.
Prokopenko
I.
Saxena
R.
Soranzo
N.
Jackson
A.U.
Wheeler
E.
Glazer
N.L.
Bouatia-Naji
N.
Gloyn
A.L.
, et al.  . 
New genetic loci implicated in fasting glucose homeostasis and their impact on type 2 diabetes risk
Nat. Genet.
 , 
2010
, vol. 
42
 (pg. 
105
-
116
)
3
Dehghan
A.
Dupuis
J.
Barbalic
M.
Bis
J.C.
Eiriksdottir
G.
Lu
C.
Pellikka
N.
Wallaschofski
H.
Kettunen
J.
Henneman
P.
, et al.  . 
Meta-analysis of genome-wide association studies in >80 000 subjects identifies multiple loci for C-reactive protein levels
Circulation
 , 
2011
, vol. 
123
 (pg. 
731
-
738
)
4
Illig
T.
Gieger
C.
Zhai
G.
Romisch-Margl
W.
Wang-Sattler
R.
Prehn
C.
Altmaier
E.
Kastenmuller
G.
Kato
B.S.
Mewes
H.W.
, et al.  . 
A genome-wide perspective of genetic variation in human metabolism
Nat. Genet.
 , 
2010
, vol. 
42
 (pg. 
137
-
141
)
5
Hicks
A.A.
Pramstaller
P.P.
Johansson
A.
Vitart
V.
Rudan
I.
Ugocsai
P.
Aulchenko
Y.
Franklin
C.S.
Liebisch
G.
Erdmann
J.
, et al.  . 
Genetic determinants of circulating sphingolipid concentrations in European populations
PLoS Genet.
 , 
2009
, vol. 
5
 pg. 
e1000672
 
6
Inouye
M.
Kettunen
J.
Soininen
P.
Silander
K.
Ripatti
S.
Kumpula
L.S.
Hämäläinen
E.
Jousilahti
P.
Kangas
A.J.
Männistö
S.
, et al.  . 
Metabonomic, transcriptomic, and genomic variation of a population cohort
Mol. Syst. Biol.
 , 
2010
, vol. 
6
 pg. 
441
 
7
Chasman
D.I.
Pare
G.
Mora
S.
Hopewell
J.C.
Peloso
G.
Clarke
R.
Cupples
L.A.
Hamsten
A.
Kathiresan
S.
Malarstig
A.
, et al.  . 
Forty-three loci associated with plasma lipoprotein size, concentration, and cholesterol content in genome-wide analysis
PLoS Genet.
 , 
2009
, vol. 
5
 pg. 
e1000730
 
8
Soininen
P.
Kangas
A.J.
Würtz
P.
Tukiainen
T.
Tynkkynen
T.
Laatikainen
R.
Järvelin
M.R.
Kähönen
M.
Lehtimäki
T.
Viikari
J.
, et al.  . 
High-throughput serum NMR metabonomics for cost-effective holistic studies on systemic metabolism
Analyst
 , 
2009
, vol. 
134
 (pg. 
1781
-
1785
)
9
Lee-Rueckert
M.
Vikstedt
R.
Metso
J.
Ehnholm
C.
Kovanen
P.T.
Jauhiainen
M.
Absence of endogenous phospholipid transfer protein impairs ABCA1-dependent efflux of cholesterol from macrophage foam cells
J. Lipid Res.
 , 
2006
, vol. 
47
 (pg. 
1725
-
1732
)
10
Vikstedt
R.
Metso
J.
Hakala
J.
Olkkonen
V.M.
Ehnholm
C.
Jauhiainen
M.
Cholesterol efflux from macrophage foam cells is enhanced by active phospholipid transfer protein through generation of two types of acceptor particles
Biochemistry
 , 
2007
, vol. 
46
 (pg. 
11979
-
11986
)
11
Zambon
A.
Bertocco
S.
Vitturi
N.
Polentarutti
V.
Vianello
D.
Crepaldi
G.
Relevance of hepatic lipase to the metabolism of triacylglycerol-rich lipoproteins
Biochem. Soc. Trans.
 , 
2003
, vol. 
31
 (pg. 
1070
-
1074
)
12
Annema
W.
Tietge
U.J.
Role of hepatic lipase and endothelial lipase in high-density lipoprotein-mediated reverse cholesterol transport
Curr. Atheroscler. Rep.
 , 
2011
, vol. 
13
 (pg. 
257
-
265
)
13
Suhre
K.
Shin
S.Y.
Petersen
A.K.
Mohney
R.P.
Meredith
D.
Wagele
B.
Altmaier
E.
Deloukas
P.
Erdmann
J.
, et al.  . 
CARDIoGRAM
Human metabolic individuality in biomedical and pharmaceutical research
Nature
 , 
2011
, vol. 
477
 (pg. 
54
-
60
)
14
Segrest
J.P.
Jones
M.K.
De Loof
H.
Dashti
N.
Structure of apolipoprotein B-100 in low density lipoproteins
J. Lipid Res.
 , 
2001
, vol. 
42
 (pg. 
1346
-
1367
)
15
Kumar
V.
Butcher
S.J.
Öörni
K.
Engelhardt
P.
Heikkonen
J.
Kaski
K.
Ala-Korpela
M.
Kovanen
P.T.
Three-dimensional cryoEM reconstruction of native LDL particles to 16A resolution at physiological body temperature
PLoS ONE
 , 
2011
, vol. 
6
 pg. 
e18841
 
16
Gordon
D.A.
Jamil
H.
Progress towards understanding the role of microsomal triglyceride transfer protein in apolipoprotein-B lipoprotein assembly
Biochim. Biophys. Acta
 , 
2000
, vol. 
1486
 (pg. 
72
-
83
)
17
Verges
B.
Abnormal hepatic apolipoprotein B metabolism in type 2 diabetes
Atherosclerosis
 , 
2010
, vol. 
211
 (pg. 
353
-
360
)
18
Surakka
I.
Kristiansson
K.
Anttila
V.
Inouye
M.
Barnes
C.
Moutsianas
L.
Salomaa
V.
Daly
M.
Palotie
A.
Peltonen
L.
, et al.  . 
Founder population-specific HapMap panel increases power in GWA studies through improved imputation accuracy and CNV tagging
Genome Res.
 , 
2010
, vol. 
20
 (pg. 
1344
-
1351
)
19
Service
S.
DeYoung
J.
Karayiorgou
M.
Roos
J.L.
Pretorious
H.
Bedoya
G.
Ospina
J.
Ruiz-Linares
A.
Macedo
A.
Palha
J.A.
, et al.  . 
Magnitude and distribution of linkage disequilibrium in population isolates and implications for genome-wide association studies
Nat. Genet.
 , 
2006
, vol. 
38
 (pg. 
556
-
560
)
20
Ala-Korpela
M.
Kangas
A.J.
Inouye
M.
Genome-wide association studies and systems biology: together at last
Trends Genet
 , 
2011
, vol. 
12
 (pg. 
493
-
438
)
21
Kaess
B.M.
Tomaszewski
M.
Braund
P.S.
Stark
K.
Rafelt
S.
Fischer
M.
Hardwick
R.
Nelson
C.P.
Debiec
R.
Huber
F.
, et al.  . 
Large-scale candidate gene analysis of HDL particle features
PLoS ONE
 , 
2011
, vol. 
6
 pg. 
e14529
 
22
Otvos
J.D.
Mora
S.
Shalaurova
I.
Greenland
P.
Mackey
R.H.
Goff
D.C.
Jr.
Clinical implications of discordance between low-density lipoprotein cholesterol and particle number
J. Clin. Lipidol.
 , 
2011
, vol. 
5
 (pg. 
105
-
113
)
23
Krauss
R.M.
Lipoprotein subfractions and cardiovascular disease risk
Curr. Opin. Lipidol.
 , 
2010
, vol. 
21
 (pg. 
305
-
311
)
24
Morgan
J.
Carey
C.
Lincoff
A.
Capuzzi
D.
High-density lipoprotein subfractions and risk of coronary artery disease
Curr. Atheroscler. Rep.
 , 
2004
, vol. 
6
 (pg. 
359
-
365
)
25
Ala-Korpela
M.
Critical evaluation of 1H NMR metabonomics of serum as a methodology for disease risk assessment and diagnostics
Clin. Chem. Lab. Med.
 , 
2008
, vol. 
46
 (pg. 
27
-
42
)
26
Ala-Korpela
M.
Soininen
P.
Savolainen
M.J.
Letter by Ala-Korpela et al regarding article, ‘Lipoprotein particle profiles by nuclear magnetic resonance compared with standard lipids and apolipoproteins in predicting incident cardiovascular disease in women’
Circulation
 , 
2009
, vol. 
120
 pg. 
e149
  
author reply e150
27
Rantakallio
P.
Groups at risk in low birth weight infants and perinatal mortality
Acta Paediatr. Scand.
 , 
1969
, vol. 
193
 
Suppl. 193)
pg. 
1+
 
28
Raitakari
O.T.
Juonala
M.
Rönnemaa
T.
Keltikangas-Järvinen
L.
Räsänen
L.
Pietikäinen
M.
Hutri-Kähönen
N.
Taittonen
L.
Jokinen
E.
Marniemi
J.
, et al.  . 
Cohort profile: the cardiovascular risk in Young Finns Study
Int. J. Epidemiol.
 , 
2008
, vol. 
37
 (pg. 
1220
-
1226
)
29
Eriksson
J.G.
Epidemiology, genes and the environment: lessons learned from the Helsinki Birth Cohort Study
J. Intern. Med.
 , 
2007
, vol. 
261
 (pg. 
418
-
425
)
30
Inouye
M.
Silander
K.
Hämäläinen
E.
Salomaa
V.
Harald
K.
Jousilahti
P.
Männistö
S.
Eriksson
J.G.
Saarela
J.
Ripatti
S.
, et al.  . 
An immune response network associated with blood lipid levels
PLoS Genet.
 , 
2010
, vol. 
6
 pg. 
e1001113
 
31
Perttilä
J.
Merikanto
K.
Naukkarinen
J.
Surakka
I.
Martin
N.W.
Tanhuanpää
K.
Grimard
V.
Taskinen
M.R.
Thiele
C.
Salomaa
V.
, et al.  . 
OSBPL10, a novel candidate gene for high triglyceride trait in dyslipidemic Finnish subjects, regulates cellular lipid metabolism
J. Mol. Med.
 , 
2009
, vol. 
87
 (pg. 
825
-
835
)
32
Howie
B.N.
Donnelly
P.
Marchini
J.
A flexible and accurate genotype imputation method for the next generation of genome-wide association studies
PLoS Genet.
 , 
2009
, vol. 
5
 pg. 
e1000529
 
33
1000 Genomes Project Consortium.
A map of human genome variation from population-scale sequencing
Nature
 , 
2010
, vol. 
467
 (pg. 
1061
-
1073
)
34
Altshuler
D.M.
Gibbs
R.A.
Peltonen
L.
Altshuler
D.M.
Gibbs
R.A.
Peltonen
L.
Dermitzakis
E.
Schaffner
S.F.
Yu
F.
, et al.  . 
International HapMap 3 Consortium
Integrating common and rare genetic variation in diverse human populations
Nature
 , 
2010
, vol. 
467
 (pg. 
52
-
58
)
35
Frazer
K.A.
Ballinger
D.G.
Cox
D.R.
Hinds
D.A.
Stuve
L.L.
Gibbs
R.A.
Belmont
J.W.
Boudreau
A.
Hardenbol
P.
, et al.  . 
International HapMap Consortium
A second generation human haplotype map of over 3.1 million SNPs
Nature
 , 
2007
, vol. 
449
 (pg. 
851
-
861
)
36
Niemi
J.
Mäkinen
V.P.
Heikkonen
J.
Tenkanen
L.
Hiltunen
Y.
Hannuksela
M.L.
Jauhiainen
M.
Forsblom
C.
Taskinen
M.R.
Kesäniemi
Y.A.
, et al.  . 
Estimation of VLDL, IDL, LDL, HDL2, apoA-I, and apoB from the friedewald inputs—apoB and IDL, but not LDL, are associated with mortality in type 1 diabetes
Ann. Med.
 , 
2009
, vol. 
41
 (pg. 
451
-
461
)
37
Okazaki
M.
Usui
S.
Ishigami
M.
Sakai
N.
Nakamura
T.
Matsuzawa
Y.
Yamashita
S.
Identification of unique lipoprotein subclasses for visceral obesity by component analysis of cholesterol profile in high-performance liquid chromatography
Arterioscler. Thromb. Vasc. Biol.
 , 
2005
, vol. 
25
 (pg. 
578
-
584
)

Author notes

These authors contributed equally.