Metabolomic Consequences of Genetic Inhibition of PCSK9 Compared With Statin Treatment

Supplemental Digital Content is available in the text.

S tatins are first line therapy to lower blood levels of low-density lipoprotein cholesterol (LDL-C) and reduce the risk of cardiovascular events. [1][2][3] Treatment with proprotein convertase subtilisin/kexin type 9 (PCSK9) inhibitors has emerged as an additional effective therapy to lower LDL-C, resulting in reductions of ≈45% to 60%. 4,5 Large cardiovascular outcome trials have recently demonstrated that PCSK9 inhibitors reduce the risk of major cardiovascular events when added to statin treatment. 6,7 Based on the first major outcome trials, 6,7 there has been some suggestions that PCSK9 inhibitors may be slightly less efficacious than statins for equivalent LDL-C reductions; however, other reports suggest that this is not the case, with apparent differences in cardiovascular event reduction explained by the short duration of the PCSK9 trials. 8 Assessment of the detailed lipoprotein and other metabolic effects of statins and PCSK9 inhibitors could provide a more detailed understanding of these lipid-lowering therapies and shed light on potential differential effects on lipid metabolism.
The anticipated pharmacological effects of PCSK9 inhibitors may be assessed by LDL-C lowering alleles in the PCSK9 gene, which act as unconfounded proxies for the lifetime effects of treatments. [9][10][11] The observation of a prominent lower risk of coronary heart disease with LDL-C-lowering alleles in PCSK9 was pivotal for accelerating the development of anti-PCSK9 therapeutics. 10 Supporting the validity of using genetic proxies for molecular characterization of lipid-lowering targets, we have previously shown that LDL-C-lowering alleles in HMGCR (the gene encoding the target for statins) closely recapitulate the detailed metabolic changes associated with starting statin therapy in longitudinal cohorts, as assessed by nuclear magnetic resonance (NMR) metabolomics. 12 These detailed metabolic effects of statins were recently confirmed in PREVEND IT (Prevention of Renal and Vascular End-stage Disease Intervention Trial), a small randomized trial. 13 Other studies have assessed the associations of PCSK9 variants with lipoprotein subclass profiles, 14,15 and the treatment effects of PCSK9 inhibitors on lipoprotein particle concentrations and lipidomic measures have been examined in several small trials. [16][17][18] However, prior studies have had limited power to assess potential differences between PCSK9 inhibition and statin therapy for equivalent reductions in LDL-C, complicating direct comparisons of their impact on detailed lipid and metabolite measures.
In the present study, we examined the effects of statin therapy and genetic inhibition of PCSK9 on a circulating profile of 228 metabolic measures, quantified by NMR metabolomics, including lipoprotein subclasses, their lipid concentrations and composition, fatty acid balance, and several nonlipid pathways. The metabolic effects of statin treatment were assessed in a large randomized, placebo-controlled trial. In the absence of NMR metabolomics data from a large randomized trial of PCSK9 inhibitor therapy, the anticipated pharmacological effects were examined for a loss-offunction variant in the PCSK9 gene. 10,19 Comparing the metabolomic effects of genetic inhibition of PCSK9 to statin therapy provides an opportunity to examine possible discrepancies in many circulating biomarkers, and in turn elucidate potential therapeutic differences in the molecular mechanisms to reduce cardiovascular risk.

METHODS
The authors declare that the summary statistics are available within the article and its online-only Data Supplement. The individual patient data analyzed in this study are available by application to the respective cohort committees.

Study Design
An overview of the study design is shown in Figure 1 [24][25][26], and China (China Kadoorie Biobank 27 ). All study participants provided written informed consent, and study protocols were approved by the local ethics committees.
PROSPER is a double-blind, randomized, placebo-controlled trial investigating the benefit of pravastatin (40 mg/d) in elderly individuals at risk of cardiovascular disease, with 5804 participants (70-82 years old) from Scotland, Ireland, and The Netherlands enrolled between December 1997 and May 1999. 28 All participants had above average plasma total cholesterol concentration (4.0-9.0 mmol/L) at baseline, and 50% had prior vascular disease. For the present study, 5359

ORIGINAL RESEARCH ARTICLE
samples (2659 on pravastatin) were measured by NMR metabolomics; all were previously unthawed 6-month postrandomization EDTA plasma samples stored at −80°C. 28 Metabolite data from baseline samples were not available, however the randomization should ensure that there are limited betweengroup differences at baseline. Replication of the metabolic effects of pravastatin in PROSPER was done by comparison with recent results from PREVEND-IT. 13 The metabolic effects of PCSK9 inhibition were assessed via the principle of Mendelian randomization using rs11591147-T (R46L), a loss-of-function allele robustly associated with lower LDL-C and decreased cardiovascular risk. 10,11 The frequency of carriers of this effect allele was 2.2% (N=3135 carriers); clinical characteristics of these individuals are specified in Table I in the online-only Data Supplement. Additional genetic variants in the PCSK9 locus, which have previously been used in Mendelian randomization studies on PCSK9, 11,29 and display low linkage disequilibrium with rs11591147 (R 2 <0.2), were assessed in sensitivity analyses. To complement the comparison of PCSK9 rs11591147-T effects against the statin trial, we further examined the metabolic effects of rs12916-T in HMGCR in the same study population, acting as a pseudotrial of a very small statin dose by naturally occurring randomization of HMG-CoA (HMG-coenzyme A) reductase inhibition. 12,30 Among single nucleotide polymorphisms in HMGCR, rs12916 exhibits the strongest association with LDL-C and has been shown to affect hepatic HMGCR expression as well as cardiovascular risk. 11,12,30 Last, to corroborate the validity of using genetic proxies to mimic the randomized trial effects, we compared metabolic effects of statin treatment in PROSPER with the corresponding effects of HMGCR rs12916-T. Pregnant women and individuals on lipid-lowering treatment were excluded from the analyses where information was available. Details of the cohorts are provided in Methods and Table I in the online-only Data Supplement.

Lipid and Metabolite Quantification
High-throughput NMR metabolomics was used to quantify 228 lipoprotein lipids and polar metabolite measures from serum or plasma samples in the PROSPER trial and eight cohorts by the Nightingale platform (Nightingale Health Ltd, Helsinki, Finland). This provides simultaneous quantification of routine lipids, particle concentration, and lipid composition of 14 lipoprotein subclasses, abundant fatty acids, amino acids, ketones, and glycolysis-related metabolites in absolute concentration units (Table II in the online-only Data Supplement). 31 The Nightingale NMR metabolomics platform has been widely used in epidemiological studies, 12,32,33 and the measurement method has been previously described. [31][32][33][34][35]

Statistical Analyses
The effects of statin therapy on the 228 metabolic measures in the PROSPER trial were assessed by comparing the mean metabolite concentrations in the treatment group with the placebo group at 6 months after randomization. The betweengroup difference in concentrations for each metabolic measure was quantified separately using linear regression with metabolite concentration as outcome and treatment status as predictor, adjusted for age and sex. All metabolite concentrations were scaled to standard deviation (SD) units before assessing the differences, to enable comparison of measures with different units and across wide ranges of concentration levels. Results in absolute units are presented in Table III in the online-only Data Supplement. The percentage differences in metabolite concentrations, relative to the placebo group, were examined as secondary analyses.
The effect of genetic inhibition of PCSK9 on each of the 228 metabolic measures was analyzed separately by fitting linear regression models with metabolite concentrations as outcome and rs11591147-T allele count as predictor, representing the number of LDL-C lowering alleles. For sensitivity analysis, we conducted equivalent tests of each metabolic measure with rs12916-T in HMGCR as the predictor. All genetic analyses assumed an additive effect and were adjusted for age, sex, and the first four genomic principal components. Effect sizes and standard errors from each cohort were combined using inverse variance-weighted fixed effect meta-analysis. All effect sizes were scaled to SD units of metabolite concentrations, as for analyses of PROSPER. The similarity between the overall patterns of metabolic effects attributable to PCSK9 inhibition and statin therapy was summarized using the linear fit of the effect estimates of 153 metabolic measures, 12 covering all assayed measures except lipoprotein lipid ratios and 5 polar metabolites that could not be reliably quantified in PROSPER.
To facilitate comparison between the substantial metabolic effects of statin therapy with the smaller effects from genetic inhibition of PCSK9, results are presented relative to an equivalent (1-SD) lowering of LDL-C within each study design (as quantified by NMR metabolomics). 12,35 For the statin trial, the estimates derived from comparing statin treatment with placebo were divided by 1.19 (because statins lowered LDL-C by 1.19 SD); for PCSK9 genetic associations, per-allele effect estimates were divided by 0.44; for sensitivity analyses using rs12916 in HMGCR, per-allele effect estimates were divided by 0.078. The scaling relative to LDL-C was used to interpret the reported effect sizes as a change in concentration in each metabolic measure (in SD units) that accompanies a 1-SD lowering of LDL-C by statin therapy and genetic inhibition of PCSK9. This scaling is in line with the principles of Mendelian randomization assuming that the genetic variants in PCSK9 and HMGCR serve as instruments for the LDL-C exposure.
Although 228 metabolic measures in total were examined, the number of independent tests performed is lower because of the correlated nature of the measures. 35 The number of independent tests was estimated by taking the number of principal components explaining 99% of the variation in the metabolic measures. 36 Thus, significance was considered at P<0.0003 to account for the testing of 54 independent metabolic measures and 3 sets of analyses conducted (main effects of statins, PCSK9, and differences in their effects). The significance of differences in the effect estimates was determined using the formula below, and the corresponding P values were derived from the normal distribution: To facilitate visualization of the results, we focused on 148 measures that cover all the metabolic pathways assayed; results for the remaining measures are shown in Figure I in the online-only Data Supplement and Tables IV through VI in the online-only Data Supplement. Statistical analyses were conducted using R3.2 (www.r-project.org). Figure 1 provides an overview of the study design. Characteristics of the study populations are shown in the Table. Characteristics of each of the eight cohorts used for the genetic analyses (N=72 185) are provided separately in Table I in the online-only Data Supplement. The metabolic effects of statin treatment in PROSPER (pravastatin 40 mg/daily) and genetic inhibition of PCSK9 are compared in Figure 2. Overall, there was a high concordance of association of statin treatment and genetic inhibition of PCSK9 across the detailed metabolic profile (R 2 =0.88). Nonetheless, some discrepancies in effect sizes between statin treatment and PCSK9 rs11591147 were evident, primarily for very-low-density lipoprotein (VLDL) lipids.

Effects on Lipoprotein Lipids
The specific effects of statin therapy and genetic inhibition of PCSK9 on lipid fractions and 14 lipoprotein subclasses are shown in Figure 3. Scaled to the same lowering of LDL-C, PCSK9 rs11591147 displayed similar effects as statin therapy for total cholesterol and intermediate-density lipoprotein cholesterol, with no effect on high-density lipoprotein cholesterol. However, PCSK9 rs11591147 had a weaker effect on lowering VLDL cholesterol compared with statins (54% versus 77%, relative to the lowering effect on LDL-C [% LDL-C ]; P het =2×10 -7 ). These results were substantiated by the pattern of reduction in lipoprotein subclass particles: although the effects were similar for lowering particle concentrations in all 3 (small, medium, and large) LDL subclasses, the extent of lowering of small, medium-sized, and large VLDL particle concentrations was smaller for PCSK9 rs11591147 com-pared with statin therapy. A similar discrepancy was observed for cholesterol concentrations within the 6 VLDL subclasses (Figure 4).
Both for statin therapy and genetic inhibition of PCSK9, the effects on triglyceride measures were modest compared with those observed for cholesterol levels in the same lipoprotein subfractions ( Figure 3). The most pronounced lowering of triglycerides was seen for intermediate-density lipoprotein and LDL particles. For the equivalent reductions in LDL-C, PCSK9 rs11591147 displayed a weaker effect than statin therapy on lowering total plasma triglycerides (16% LDL-C versus 37% LDL-C ; P het =3×10 -6 ). Similar differences were seen for VLDL and high-density lipoprotein triglycerides. Consistent with the observed discrepancies for lowering of medium and large VLDL particles, genetic inhibition of PCSK9 resulted in modestly larger VLDL size, whereas statin therapy had no effect on this measure. The effects on apolipoprotein concentrations were broadly similar, albeit a larger decrease was observed with statins for the ratio of apolipoprotein B to A-I.

Effects on Lipoprotein Composition
In addition to affecting the absolute lipid concentrations, both statin therapy and genetic inhibition of PCSK9 had prominent effects on the relative abundance of lipid types (free and esterified cholesterol,

ORIGINAL RESEARCH ARTICLE
triglycerides, and phospholipids) in differently sized lipoprotein subclasses (Figure 4). The most pronounced lipoprotein composition effects were observed within LDL subclasses, with substantial lowering in the relative abundance of cholesteryl esters in LDL particles, alongside increases in the abundance of free cholesterol and phospholipids. These effects were very similar for statin treatment compared with PCSK9 rs11591147 for the equivalent reductions in LDL-C.
Subtle discrepancies between statin and genetic inhibition of PCSK9 were observed (eg, for the extent of lowering the fraction of free cholesterol in VLDL particles). The relative fraction of triglycerides in LDL and other apolipoprotein B-carrying particles increased similarly for both statins and PCSK9 rs11591147, whereas statin therapy caused larger decreases in the relative abundance of triglycerides within high-density lipoproteins.

Effects on Fatty Acids and Polar Metabolites
The effects of statin therapy and genetic inhibition of PCSK9 on fatty acid concentrations and the balance of fatty acid ratios are shown in Figure 5. Absolute concentrations of all fatty acids were lowered, with the most pronounced lowering for concentrations of linoleic acid, an omega-6 fatty acid commonly bound to cholesteryl esters in LDL particles. For the same lowering of LDL-C, the effects of statins and PCSK9 rs11591147 were broadly similar, albeit with the lowering of total fatty acids being stronger in the case of statins (67% LDL-C versus 50% LDL-C ; P het =2×10 -4 ). The effects on the fatty acid ratios were generally modest, both for statin therapy and PCSK9 rs11591147. A pronounced discrepancy between these was observed for the overall degree of fatty acid unsaturation (16% LDL-C reduction for PCSK9 versus 26% LDL-C increase for statin; P het =4×10 -20 ). We further assessed the effects of statin therapy and PCSK9 rs11591147 on polar metabolites and other metabolic measures quantified simultaneously in the metabolomics assay, including circulating amino acids, glycolysis metabolites, ketone bodies, and GlycA, a marker of chronic inflammation 37 ( Figure 6). Statin therapy caused only minor effects on these metabolic measures; the strongest lowering effects were observed for GlycA (17% LDL-C ) and isoleucine (7% LDL-C ). The effects of PCSK9 rs11591147 were also very close to null for these measures, including for glycolysis related metabolites and markers of insulin resistance. Of note,

ORIGINAL RESEARCH ARTICLE
information on glucose, lactate, and pyruvate were not available in the PROSPER trial because of glycolysis progression after sample collection.

Comparison With PREVEND-IT Trial and Mendelian Randomization
To replicate the detailed metabolic effects of statins observed in PROSPER, we compared them with recent results from the PREVEND-IT trial obtained using the same NMR metabolomics platform. 13  We further compared the metabolic effects of PCSK9 rs11591147 to those caused by rs12916 in the HMGCR gene, hereby using the genetic variants to ef-fectively act as 2 naturally occurring trials in the same study population. The overall pattern of metabolic effects was highly similar for PCSK9 rs11591147 and HMGCR rs12916 (R 2 =0.92; Figure IIIA in the online-only Data Supplement). Nonetheless, scaled to the equivalent LDL-C reductions, similar deviations were observed for VLDL lipids as when comparing PCSK9 rs11591147 with the statin trial results ( Figure IV in the online-only Data Supplement). Specifically, the lowering effect of HMGCR rs12916-T on particle concentrations of all VLDL subclasses was more similar to the effects of statin treatment than to those of PCSK9 rs11591147-T, except in the case of very small VLDL. Differences were also observed for cholesterol and triglyceride concentrations in VLDL subclasses, whereas the lowering of total and saturated fatty acids was similar for the HMGCR and PCSK9 variants. However, power to detect statistical differences on individual measures was modest, because of the much weaker LDL-C lowering effect of HMGCR rs12916. Last, the overall pattern of metabolic effects of statin therapy in PROSPER was highly concordant to effects of HMGCR rs12916 (R 2 =0.95; Figure IIIB in the online-only Data Supplement), signifying pharmacological and genetic inhibition of HMG-CoA reductase, respectively.
In sensitivity analyses, the pattern of metabolic effects from PCSK9 rs11591147 was consistent across the cohorts ( Figure V in the online-only Data Supplement). We also observed similar detailed patterns of metabolic

DISCUSSION
This study elucidates the comprehensive metabolic effects associated with statin therapy and PCSK9 inhibition. The results demonstrate that, in comparison with statin therapy, genetic inhibition of PCSK9 yields com-parable changes across many different markers of lipid metabolism when scaled to the equivalent lowering of LDL-C. However, our results also suggest that PCSK9 inhibitors may be somewhat less efficacious at lowering VLDL particles. This could potentially contribute to subtle differences in potency for lowering cardiovascular disease for the equivalent reductions in LDL-C, 6 because recent evidence suggests that VLDL cholesterol and other triglyceride-rich lipoprotein measures may causally contribute to the development of coronary heart disease independent of LDL-C. [38][39][40] Moreover, trial data suggest that VLDL cholesterol is a stronger predictor of cardiovascular event risk than LDL-C among patients on statin therapy. 41,42 Statins and PCSK9 inhibitors both lower circulating LDL-C levels via upregulation of LDL receptors on cell surfaces. Consistent with this shared mechanism for clearance of LDL particles, we found that statins and genetic inhibition of PCSK9 caused a highly consistent pattern of change across the detailed metabolic profile.

ORIGINAL RESEARCH ARTICLE
The metabolomic profiling of the PROSPER trial corroborates previous studies with detailed measurements of metabolic effects of statin therapy, both as assessed in longitudinal cohorts and in a small randomized trial. 12,13 By profiling a large number of individuals from multiple cohorts, our results also validate and extend previous studies examining the detailed metabolic effects of PCSK9 rs11591147. 14,15 Importantly, in relation to assessment of potential side-effects of PCSK9 inhibition, we did not observe effects on amino acids or other nonlipid metabolites, many of which are associated with risk of incident diabetes and cardiovascular events. 32,43,44 Our results on the pattern of lowering VLDL particles attributable to genetic inhibition of PCSK9 are consistent with 2 small trials assessing the effects of the PCSK9 inhibitors alirocumab and evolocumab on lipoprotein particle concentrations; both trials showed substantial reductions in small and medium-sized VLDL particles, whereas the particle concentration of the large VLDL fraction was not affected. 16,17 Similar results were also found in a small PCSK9-inhibitor trial using separation of VLDL subfractions and other lipid measures by ultracentrifugation, which also corroborate our results on a stronger effect on lowering of VLDL cholesterol in comparison with total plasma triglycerides. 18 However, differences in assay methods complicate direct comparison of these trials to our results. Overall, these results provide orthogonal evidence for diverse lipoprotein lipid alterations by PCSK9 inhibitors, coherent with the comprehensive metabolic effects of statins. Currently licensed PCSK9 inhibitors are given either instead of statins-when there is strong evidence of statin intolerance in those with familial hypercholesterolemia-or in addition to maximally tolerated statins in patients with existing vascular disease. 45 Such treatment with PCSK9 inhibitors has been shown to be more efficacious in lowering LDL-C than the most potent statins. [4][5][6]8 Mendelian randomization studies comparing PCSK9 and HMGCR gene scores on cardiovascular outcomes have indicated nearly identical protective effects for equivalent reductions in LDL-C. 11,46 However, when scaling the metabolic effects to an equivalent lowering in LDL-C, the results of the present study indicate subtle differences on multiple lipoprotein lipid measures. The most notable discrepancy was for VLDL lipids, suggesting weaker potency of PCSK9 inhibitors in clearance of these triglyceride-rich lipoproteins as compared with statins. These findings are supported by a recent study providing evidence that statins, but not PCSK9 inhibitors, improve triglyceride-rich lipoprotein metabolism after an oral fat load in normolipidemic men. 47 The causal consequences of these differences in medium-sized and large VLDL particles, that are rich in triglycerides, remain unclear and warrant further investigation; whereas intermediate-density lipoprotein and the smallest VLDL particles can penetrate the arterial wall to cause atherosclerosis, it is commonly perceived not be to the case for larger VLDL particles. 38,48 We also observed a difference in lowering of VLDL cholesterol levels; the cholesterol concentrations of VLDL particles are strongly associated with risk of myocardial infarction, 44 and some studies have suggested that VLDL cholesterol could underpin the link between triglycerides and cardiovascular risk. 38,41 If these VLDL particles do play a causal role in vascular disease, the discrepancy between statin therapy and PCSK9 inhibition could translate into slightly more potent cardiovascular risk reduction for the same LDL-C lowering for statins as compared with PCSK9 inhibition. We acknowledge that the present comparison of detailed metabolic effects of statin therapy and PCSK9 inhibition does not directly inform on the cardiovascular benefits of anti-PCSK9 therapies above current optimal care, but potentially in keeping with our findings, the cardiovascular outcome trials on PCSK9 inhibition have demonstrated slightly weaker cardiovascular event lowering compared with meta-analysis of statin trials per mmol/L reduction in LDL-C. 6 Although potential explanations for this discrepancy include the short trial duration and choice of primary end point, other explanations, such as differences in anti-inflammatory effects, have also been suggested. 11,49 Our results provide an additional hypothesis for exploration: the apparent weaker cardioprotective effects of PCSK9 inhibitors compared with statins per unit reduction in LDL-C may be attributable to weaker reductions in VLDL lipid concentrations by PCSK9 inhibition. This hypothesis warrants further investigation, including elucidation of the causal role of triglyceriderich VLDL particles in tandem with further examinations of the detailed lipid effect of PCSK9 inhibitors.
Strengths and limitations of our study warrant consideration. The lack of NMR metabolomics data for a PCSK9 inhibition trial motivated the use of a loss-offunction variant in PCSK9 as a proxy for the anticipated therapeutic effects. The close match in the detailed metabolic effects of statin therapy and HMGCR observed in this study substantiates the validity of using genetic variants to mimic lipid-lowering effects in randomized trial settings. Although we note that the metabolic profile of other statins may differ from that of pravastatin, the similarity between HMGCR and statin therapy that we identified provides reassurances about the generalizability of our findings to other statin types. To robustly assess the metabolic effects of genetic inhibition of PCSK9, we had >5 times the sample size of prior studies examining PCSK9 rs11591147 on lipoprotein subclass profiles. 14,15 Despite the large sample size, we had limited power to detect effects on glycolysis-related metabolites because of preanalytical effects causing depletion of glucose levels in the blood samples. Notwithstanding the increased risk for type 2 diabetes mellitus linked with PCSK9 rs11591147-T, our ORIGINAL RESEARCH ARTICLE results indicating minute effects on glycolysis traits are in line with larger studies reporting null effects on fasting glucose for this SNP. The divergency in VLDL lipidlowering effects between statins and genetic inhibition of PCSK9 could potentially be attributable to differences in the clinical characteristics of the older, highrisk patients of the PROSPER trial in comparison with the younger cohort participants included in the genetic analyses. However, similar VLDL lowering effects attributable to pravastatin as observed here in PROSPER were recently reported in PREVEND IT 13 with younger and lower-risk trial participants ( Figure I in the onlineonly Data Supplement). The differences in VLDL effects were also recapitulated when directly comparing the effects of genetic inhibition of PCSK9 to that of HMGCR in the same study population ( Figure IV in the onlineonly Data Supplement), providing reassurance that the observed VLDL differences are primarily due to the molecular mechanisms. Furthermore, the scaling of results to the LDL-C lowering magnitude enables comparison of the metabolic effects regardless the possible differences in the absolute lipid levels between the study populations. A strength of the metabolomics platform used is the ability to profile lipoprotein subclasses and their lipid composition at high-throughput, however we acknowledge that other assays may provide even deeper characterization of lipid metabolism and nonlipid pathways to further clarify the molecular effects of lipid-lowering therapies. 50 In conclusion, we found highly similar metabolic effects of statin therapy and genetic inhibition of PCSK9 across a comprehensive profile of lipids, lipoprotein subclasses, fatty acids, and polar metabolites. The detailed profiling of lipoprotein subclasses revealed weaker effects of PCSK9 inhibition on VLDL particles and their cholesterol concentrations in comparison with statins, when scaled to an equivalent lowering of LDL-C. If some of these VLDL lipids have independent causal effects on cardiovascular risk, this could contribute to subtle differences in cardiovascular event reduction between statins and PCSK9 inhibitors. More broadly, these results exemplify the utility of large-scale metabolomics in combination with randomized trials and genetics to uncover potential molecular differences between related therapeutics.  F2-2012-279233). Data collection and metabolic profiling in the ALSPAC mother's study were obtained from British Heart Foundation (SP/07/008/24066) and the Wellcome Trust (WT092830M). Genetic data in the ALSPAC mothers was obtained through funding from the Wellcome Trust (WT088806). ALSPAC offspring genetic data were obtained with support from 23andMe. The FINRISK studies have received financial support related to the present study from the National Institute for Health and Welfare, the Academy of Finland (139635)