In this study composed of approximately 20,000 European American and 6,200 African American participants, we identified three new loci associated with multiple metabolic trait domains: APOC1, BRAP, and PLCG1. These loci were in or near genes previously associated with atherogenic dyslipidemia, vascular inflammation, type I diabetes, vascular dysfunction, and central adiposity. No previous genome-wide or gene-centric studies examining evidence for pleiotropy in metabolic domains has detected these loci at genome-wide significant levels.
The pathogenesis of the clustering of metabolic phenotypes remains poorly understood, although it is likely that a sedentary lifestyle, combined with dietary patterns and genetic susceptibility factors, contribute. Candidate genes associated with metabolic syndrome phenotypes largely reflect current knowledge of established pathways regulating obesity, free fatty acid metabolism, insulin sensitivity, lipid metabolism, and inflammation. Although candidate gene and GWA studies have successfully identified loci influencing variation in these pathways, studies examining genetic factors influencing the co-occurrence of metabolic phenotypes are limited. Additionally, those that examine the clustering of syndromic components using the pre-defined clinical cutpoints are largely inconsistent or inconclusive. This general lack of success may reflect ongoing controversy over metabolic syndrome definitions, leading to phenotypic heterogeneity and inconsistent genetic findings across studies 
. The utility of studying the syndrome as a binary entity as opposed to a series of component traits is also debated 
, especially since the dichotomization of interval scale traits will discard information.
Methods for examining evidence of pleiotropy remain uncommon in the GWA literature and most likely reflect the lack of methodologies and software that are scalable to GWA studies. In this paper, we present a statistically efficient and computational feasible approach to testing for pleiotropy on a genome-wide scale. Our method is applicable to population-based and family studies and identified several associations that would not have been identified through typical univariate analyses. The approach presented herein is also not limited to metabolic phenotypes. Instead, our method could be applied to any set of interval scale traits that are heritable and exhibit evidence of phenotypic clustering.
Although alternative analytic approaches were available, for example estimating principal components for all traits simultaneously, we focused on the phenotype clusters presented in . First, evaluating the nineteen phenotypes of interest as six domains of interest is biologically plausible given evidence of phenotypic clustering. It was also easier to interpret principal components that were derived in separate phenotype domains rather than components estimated simultaneously. Additionally, estimating principal components within each phenotype domains ensured that each domain was sufficiently represented in the analysis.
Challenges to the approach presented herein include careful phenotype curation, made more difficult by the inclusion of 19 traits across multiple cohorts that were not measured with a common protocol. Only the ARIC and CARDIA studies had full phenotype information on all 19 traits and CHS was the only study with all traits measured during a single visit. The use of a multivariate phenotype comprised of 19 variables also limited the number of contributing cohorts and the identification of replication cohorts, as few studies have such comprehensive phenotypic data. Nonetheless, we were able to identify approximately 25,000 participants from studies that used standardized, comparable protocols and many of the associations were consistent across cohorts.
Further challenges that are not unique to large scale genetic studies incorporating a phenomics approach include the consistency of results across populations defined by age, race, sex, or other demographic characteristics. For example, the three new loci identified in the European American population were not detected in the African American population. Given a modest sample size of 6,287 participants it is difficult to determine whether an inability to generalize results to the African American population reflects different patterns of LD, varying environmental contexts, or limited statistical power. Variation in mean age between contributing cohorts, which ranged from 25 years in the CARDIA study to 72 years in the CHS, could introduce additional heterogeneity, as associations between metabolic phenotypes have been shown to diminish with age 
. Finally, marked variation in the prevalence of the metabolic syndrome by gender, regardless of clinical definition, suggest the possibility of sex-specific metabolic syndrome effects 
. Analyses that examine modification by sex, age, and other important clinical covariates are therefore warranted.
Our use of the IBC array, which is composed of variants implicated in cardiovascular, inflammatory, hemostasis/coagulation, and metabolic pathways, was beneficial in that it allowed us to leverage the wealth of information on pathways implicated in metabolic disturbances while reducing multiple testing penalties. Admittedly this approach was limited in that it potentially excludes novel pathways not captured by the IBC chip. Although imputation allowed us to increase the number of variants, genome-wide approaches might identify additional pleiotropic loci.
In summary, our results support phenomics as a complementary approach that leverages phenotypic variation for the evaluation of pleiotropy, a clear limitation of existing studies examining the metabolic syndrome using clinical definitions. Our approach, which is applicable to studies of heritable, clustered interval scale outcomes, also takes advantage of the wealth of phenotype data available in longitudinal cohort studies as well as emerging analytical and bioinformatics approaches. Ultimately, these results support the presence of genetic variants with pleiotropic effects on adiposity, inflammation, glucose regulation, dyslipidemia, vascular dysfunction and thrombosis. Such loci may help characterize metabolic dysregulation and identify targets for intervention.