Subgroups at high risk for ischaemic heart disease:identification and validation in 67 000 individuals from the general population.

BACKGROUND: The aetiology of ischaemic heart disease (IHD) is complex and is influenced by a spectrum of environmental factors and susceptibility genes. Traditional statistical modelling considers such factors to act independently in an additive manner. The Patient Rule-Induction Method (PRIM) is a multi-model building strategy for evaluating risk attributable to context-dependent gene and environmental effects. METHODS: PRIM was applied to 9073 participants from the prospective Copenhagen City Heart Study (CCHS). Gender-specific cumulative incidences were estimated for subgroups defined by categories of age, smoking, hypertension, diabetes, body mass index, total cholesterol, high-density lipoprotein cholesterol and triglycerides and by 94 single nucleotide variants (SNVs).Cumulative incidences for subgroups were validated using an independently ascertained sample of 58 240 participants from the Copenhagen General Population Study (CGPS). RESULTS: In the CCHS the overall cumulative incidences were 0.17 in women and 0.21 in men. PRIM identified six and four mutually exclusive subgroups in women and men, respectively, with cumulative incidences of IHD ranging from 0.02 to 0.34. Cumulative incidences of IHD generated by PRIM in the CCHS were validated in four of the six subgroups of women and two of the four subgroups of men in the CGPS. CONCLUSIONS: PRIM identified high-risk subgroups characterized by specific contexts of selected values of traditional risk factors and genetic variants. These subgroups were validated in an independently ascertained cohort study. Thus, a multi-model strategy may identify groups of individuals with substantially higher risk of IHD than the overall risk for the general population.

Validated context-dependent associations of coronary heart disease risk with genotype variation in the chromosome 9p21 region: the Atherosclerosis Risk in Communities study.

Markers of the chromosome 9p21 region are regarded as the strongest and most reliably significant genome-wide association study (GWAS) signals for Coronary heart disease (CHD) risk; this was recently confirmed by the CARDIoGRAMplusC4D Consortium meta-analysis. However, while these associations are significant at the population level, they may not be clinically relevant predictors of risk for all individuals. We describe here the results of a study designed to address the question: What is the contribution of context defined by traditional risk factors in determining the utility of DNA sequence variations marking the 9p21 region for explaining variation in CHD risk? We analyzed a sample of 7,589 (3,869 females and 3,720 males) European American participants of the Atherosclerosis Risk in Communities study. We confirmed CHD-SNP genotype associations for two 9p21 region marker SNPs previously identified by the CARDIoGRAMplusC4D Consortium study, of which ARIC was a part. We then tested each marker SNP genotype effect on prediction of CHD within sub-groups of the ARIC sample defined by traditional CHD risk factors by applying a novel multi-model strategy, PRIM. We observed that the effects of SNP genotypes in the 9p21 region were strongest in a sub-group of hypertensives. We subsequently validated the effect of the region in an independent sample from the Copenhagen City Heart Study. Our study suggests that marker SNPs identified as predictors of CHD risk in large population based GWAS may have their greatest utility in explaining risk of disease in particular sub-groups characterized by biological and environmental effects measured by the traditional CHD risk factors.

Efficient identification of context dependent subgroups of risk from genome-wide association studies.

We have developed a modified Patient Rule-Induction Method (PRIM) as an alternative strategy for analyzing representative samples of non-experimental human data to estimate and test the role of genomic variations as predictors of disease risk in etiologically heterogeneous sub-samples. A computational limit of the proposed strategy is encountered when the number of genomic variations (predictor variables) under study is large (>500) because permutations are used to generate a null distribution to test the significance of a term (defined by values of particular variables) that characterizes a sub-sample of individuals through the peeling and pasting processes. As an alternative, in this paper we introduce a theoretical strategy that facilitates the quick calculation of Type I and Type II errors in the evaluation of terms in the peeling and pasting processes carried out in the execution of a PRIM analysis that are under-estimated and non-existent, respectively, when a permutation-based hypothesis test is employed. The resultant savings in computational time makes possible the consideration of larger numbers of genomic variations (an example genome-wide association study is given) in the selection of statistically significant terms in the formulation of PRIM prediction models.

Neutral genomic regions refine models of recent rapid human population growth.

Human populations have experienced dramatic growth since the Neolithic revolution. Recent studies that sequenced a very large number of individuals observed an extreme excess of rare variants and provided clear evidence of recent rapid growth in effective population size, although estimates have varied greatly among studies. All these studies were based on protein-coding genes, in which variants are also impacted by natural selection. In this study, we introduce targeted sequencing data for studying recent human history with minimal confounding by natural selection. We sequenced loci far from genes that meet a wide array of additional criteria such that mutations in these loci are putatively neutral. As population structure also skews allele frequencies, we sequenced 500 individuals of relatively homogeneous ancestry by first analyzing the population structure of 9,716 European Americans. We used very high coverage sequencing to reliably call rare variants and fit an extensive array of models of recent European demographic history to the site frequency spectrum. The best-fit model estimates ∼ 3.4% growth per generation during the last ∼ 140 generations, resulting in a population size increase of two orders of magnitude. This model fits the data very well, largely due to our observation that assumptions of more ancient demography can impact estimates of recent growth. This observation and results also shed light on the discrepancy in demographic estimates among recent studies.

Variation in the maternal corticotrophin releasing hormone-binding protein (CRH-BP) gene and birth weight in Blacks, Hispanics and Whites.

BACKGROUND: Given the unique role of the corticotrophin-releasing hormone (CRH) system in human fetal development, the aim of our study was to estimate the association of birth weight with DNA sequence variation in three maternal genes involved in regulating CRH production, bioavailability and action: CRH, CRH-Binding Protein (CRH-BP), and CRH type 1 receptor (CRH-R1), respectively, in three racial groups (African-Americans, Hispanics, and non-Hispanic Whites). METHODS: Our study was carried out on a population-based sample of 575 mother-child dyads. We resequenced the three genes in mouse-human hybrid somatic cell lines and selected SNPs for genotyping. RESULTS: A significant association was observed in each race between birth weight and maternal CRH-BP SNP genotypes. Estimates of linkage disequilibrium and haplotypes established three common haplotypes marked by the rs1053989 SNP in all three races. This SNP predicted significant birth weight variation after adjustment for gestational age, maternal BMI, parity, and smoking. African American and Hispanic mothers carrying the A allele had infants whose birth weight was on average 254 and 302 grams, respectively, less than infants having C/C mothers. Non-Hispanic White mothers homozygous for the A allele had infants who were on average 148 grams less than those infants having A/C and C/C mothers. CONCLUSIONS: The magnitudes of the estimates of the birth weight effects are comparable to the combined effects of multiple SNPs reported in a recent meta-analysis of 6 GWAS studies and is quantitatively larger than that associated with maternal cigarette smoking. This effect was persistent across subpopulations that vary with respect to ancestry and environment.

Knowledge-driven analysis identifies a gene-gene interaction affecting high-density lipoprotein cholesterol levels in multi-ethnic populations.

Total cholesterol, low-density lipoprotein cholesterol, triglyceride, and high-density lipoprotein cholesterol (HDL-C) levels are among the most important risk factors for coronary artery disease. We tested for gene-gene interactions affecting the level of these four lipids based on prior knowledge of established genome-wide association study (GWAS) hits, protein-protein interactions, and pathway information. Using genotype data from 9,713 European Americans from the Atherosclerosis Risk in Communities (ARIC) study, we identified an interaction between HMGCR and a locus near LIPC in their effect on HDL-C levels (Bonferroni corrected P(c) = 0.002). Using an adaptive locus-based validation procedure, we successfully validated this gene-gene interaction in the European American cohorts from the Framingham Heart Study (P(c) = 0.002) and the Multi-Ethnic Study of Atherosclerosis (MESA; P(c) = 0.006). The interaction between these two loci is also significant in the African American sample from ARIC (P(c) = 0.004) and in the Hispanic American sample from MESA (P(c) = 0.04). Both HMGCR and LIPC are involved in the metabolism of lipids, and genome-wide association studies have previously identified LIPC as associated with levels of HDL-C. However, the effect on HDL-C of the novel gene-gene interaction reported here is twice as pronounced as that predicted by the sum of the marginal effects of the two loci. In conclusion, based on a knowledge-driven analysis of epistasis, together with a new locus-based validation method, we successfully identified and validated an interaction affecting a complex trait in multi-ethnic populations.

Long-range autocorrelations of CpG islands in the human genome.

In this paper, we use a statistical estimator developed in astrophysics to study the distribution and organization of features of the human genome. Using the human reference sequence we quantify the global distribution of CpG islands (CGI) in each chromosome and demonstrate that the organization of the CGI across a chromosome is non-random, exhibits surprisingly long range correlations (10 Mb) and varies significantly among chromosomes. These correlations of CGI summarize functional properties of the genome that are not captured when considering variation in any particular separate (and local) feature. The demonstration of the proposed methods to quantify the organization of CGI in the human genome forms the basis of future studies. The most illuminating of these will assess the potential impact on phenotypic variation of inter-individual variation in the organization of the functional features of the genome within and among chromosomes, and among individuals for particular chromosomes.

Deep resequencing reveals excess rare recent variants consistent with explosive population growth.

Accurately determining the distribution of rare variants is an important goal of human genetics, but resequencing of a sample large enough for this purpose has been unfeasible until now. Here, we applied Sanger sequencing of genomic PCR amplicons to resequence the diabetes-associated genes KCNJ11 and HHEX in 13,715 people (10,422 European Americans and 3,293 African Americans) and validated amplicons potentially harbouring rare variants using 454 pyrosequencing. We observed far more variation (expected variant-site count ∼578) than would have been predicted on the basis of earlier surveys, which could only capture the distribution of common variants. By comparison with earlier estimates based on common variants, our model shows a clear genetic signal of accelerating population growth, suggesting that humanity harbours a myriad of rare, deleterious variants, and that disease risk and the burden of disease in contemporary populations may be heavily influenced by the distribution of rare variants.

Impact of direct soil exposures from airborne dust and geophagy on human health.

Over evolutionary time humans have developed a complex biological relationship with soils. Here we describe modes of soil exposure and their biological implications. We consider two types of soil exposure, the first being the continuous exposure to airborne soil, and the second being dietary ingestion of soils, or geophagy. It may be assumed that airborne dust and ingestion of soil have influenced the evolution of particular DNA sequences which control biological systems that enable individual organisms to take advantage of, adapt to and/or protect against exposures to soil materials. We review the potential for soil exposure as an environmental source of epigenetic signals which may influence the function of our genome in determining health and disease.

The effects of maternal weight gain patterns on term birth weight in African-American women.

OBJECTIVE: The goals of our study were (1) to estimate the trends in maternal weight gain patterns and (2) to estimate the influence of variation in maternal weight and rate of weight gain over different time periods in gestation on variation in birth weight in African-American and non-African-American gravidas. STUDY DESIGN AND SETTING: Data from a prospective cohort study in which pregnant women were monitored at multiple time points during pregnancy were analysed. Maternal weight was measured at three times during pregnancy: preconception (W(0)); 16-20 weeks gestation (W(1)); 30-36 weeks gestation (W(2)), in a cohort of 435 women with full-term singleton pregnancies. The relationship between gestational age-adjusted birth weight (aBW) and measures of maternal weight and rate of weight gain across pregnancy was estimated using a multivariable longitudinal regression analysis stratified on African-American race. RESULTS: The aBW was significantly associated with maternal weight measured at any visit in both strata. For African-American women, variation in aBW was significantly associated with variation in the rate of maternal weight gain in the first half of pregnancy (W(01)) but not the rate of maternal weight gain in the second half of pregnancy (W(12)); while for non-African-American women, variation in aBW was significantly associated with W(12) but not W(01). CONCLUSION: Factors influencing the relationship between aBW and maternal weight gain patterns depend on the context of the pregnancy defined by race. Clinical decisions and recommendations about maternal weight and weight gain during pregnancy may need to account for such heterogeneity.

Context-dependent associations between variation in risk of ischemic heart disease and variation in the 5' promoter region of the apolipoprotein E gene in Danish women.

OBJECTIVE: Variations in the noncoding single-nucleotide polymorphisms (SNPs) at positions 560 and 832 in the 5' promoter region of the apolipoprotein E gene define genotypes that distinguish between high and low concentrations of plasma total and high-density lipoprotein cholesterol and triglycerides. We addressed whether these genotypes improve the prediction of ischemic heart disease (IHD) in subsamples of individuals defined by traditional risk factors and the genotypes defined by the epsilon(2), epsilon(3), and epsilon(4) alleles in exon 4 of the apolipoprotein E gene. METHODS AND RESULTS: In a sample of 3686 female and 2772 male participants of the Copenhagen City Heart Study who were free of IHD events, 576 individuals (257 women, 7.0% and 319 men, 11.5%) were diagnosed as having developed IHD in 6.5 years of follow-up. Using a stepwise Patient Rule-Induction Method modeling strategy that acknowledges the complex pathobiology of IHD, we identified a subsample of 764 elderly women (> or =65 years) with hypertriglyceridemia who had a history of smoking, a history of hypertension, or a history of both in which the A(560)T(832)/A(560)T(832) and A(560)T(832)/A(560)G(832) 5' 2-SNP genotypes had a higher cumulative incidence of IHD (172/1000) compared to the incidence of 70/1000 in the total sample of women. CONCLUSIONS: Our study validates that 5' apolipoprotein E genotypes improve the prediction of IHD and documents that the improvement is greatest in a subset defined by a particular combination of traditional risk factors in Copenhagen City Heart Study female participants. We discuss the use of these genotypes in medical risk assessment of IHD in the population represented by the Copenhagen City Heart Study.

Modifications to the Patient Rule-Induction Method that utilize non-additive combinations of genetic and environmental effects to define partitions that predict ischemic heart disease.

This article extends the Patient Rule-Induction Method (PRIM) for modeling cumulative incidence of disease developed by Dyson et al. (Genet Epidemiol 31:515-527) to include the simultaneous consideration of non-additive combinations of predictor variables, a significance test of each combination, an adjustment for multiple testing and a confidence interval for the estimate of the cumulative incidence of disease in each partition. We employ the partitioning algorithm component of the Combinatorial Partitioning Method to construct combinations of predictors, permutation testing to assess the significance of each combination, theoretical arguments for incorporating a multiple testing adjustment and bootstrap resampling to produce the confidence intervals. An illustration of this revised PRIM utilizing a sample of 2,258 European male participants from the Copenhagen City Heart Study is presented that assesses the utility of genetic variants in predicting the presence of ischemic heart disease beyond the established risk factors.

Ethnic heterogeneity in the longitudinal effects of placental vascular blood flow on birthweight.

OBJECTIVE: Our goals were: (1) to estimate the longitudinal trends in uterine artery (UtArt) and umbilical artery (UmArt) resistance indices (RIs) in different ethnic strata; (2) to estimate time-dependent changes across gestation in the influence of variation in UtArt and UmArt RI on variation in birthweight in different ethnic strata; and (3) to determine the optimum set of UtArt and UmArt RIs for predicting birthweight in different ethnic strata. STUDY DESIGN: Analyses were carried out on data collected in a prospective study of 535 multiethnic gravidas recruited from the Cedars-Sinai Medical Center (Los Angeles, CA). Baseline maternal characteristics were recorded at time of entry into the study. UtArt and UmArt RIs were measured on 3 occasions during pregnancy (visit 1, 16-20 weeks' gestation; visit 2, 21-29 weeks' gestation; and visit 3, 30-36 weeks' gestation). The outcome for this study was gestational age-adjusted birthweight (aBW). RESULTS: The average UtArt and UmArt RI decreased steadily across gestation for all ethnicities. The average UtArt RI at each visit and the average rate of change between visits were not significantly different among the ethnicities. However, the UmArt RI measured at visit 3 and its rate of change in the last trimester were significantly different among the ethnic groups (P < .02). After adjustment for traditional risk factors for fetal growth restriction, the magnitude and rate of change of UtArt RI significantly predicted aBW only in Hispanic women, whereas the magnitude and rate of change of UmArt RI predicted aBW only in African American women. The most parsimonious combination of UtArt and UmArt RI measurements at visits 1, 2, and 3 that predicted statistically significant variation in aBW differed by ethnicity. CONCLUSION: The relationships between aBW and longitudinally collected measures of UtArt and UmArt RI depend on the context defined by ethnicity and time of measurement after adjusting for a parsimoniously selected subset of traditional risk factors.

Understanding the accuracy of statistical haplotype inference with sequence data of known phase.

Statistical methods for haplotype inference from multi-site genotypes of unrelated individuals have important application in association studies and population genetics. Understanding the factors that affect the accuracy of this inference is important, but their assessment has been restricted by the limited availability of biological data with known phase. We created hybrid cell lines monosomic for human chromosome 19 and produced single-chromosome complete sequences of a 48 kb genomic region in 39 individuals of African American (AA) and European American (EA) origin. We employ these phase-known genotypes and coalescent simulations to assess the accuracy of statistical haplotype reconstruction by several algorithms. Accuracy of phase inference was considerably low in our biological data even for regions as short as 25-50 kb, suggesting that caution is needed when analyzing reconstructed haplotypes. Moreover, the reliability of estimated confidence in phase inference is not high enough to allow for a reliable incorporation of site-specific uncertainty information in subsequent analyses. We show that, in samples of certain mixed ancestry (AA and EA populations), the most accurate haplotypes are probably obtained when increasing sample size by considering the largest, pooled sample, despite the hypothetical problems associated with pooling across those heterogeneous samples. Strategies to improve confidence in reconstructed haplotypes, and realistic alternatives to the analysis of inferred haplotypes, are discussed.

An application of the patient rule-induction method for evaluating the contribution of the Apolipoprotein E and Lipoprotein Lipase genes to predicting ischemic heart disease.

Different combinations of genetic and environmental risk factors are known to contribute to the complex etiology of ischemic heart disease (IHD) in different subsets of individuals. We employed the Patient Rule-Induction Method (PRIM) to select the combination of risk factors and risk factor values that identified each of 16 mutually exclusive partitions of individuals having significantly different levels of risk of IHD. PRIM balances two competing objectives: (1) finding partitions where the risk of IHD is high and (2) maximizing the number of IHD cases explained by the partitions. A sequential PRIM analysis was applied to data on the incidence of IHD collected over 8 years for a sample of 5,455 unrelated individuals from the Copenhagen City Heart Study (CCHS) to assess the added value of variation in two candidate susceptibility genes beyond the traditional, lipid and body mass index risk factors for IHD. An independent sample of 362 unrelated individuals also from the city of Copenhagen was used to test the model obtained for each of the hypothesized partitions.

Subsets of SNPs define rare genotype classes that predict ischemic heart disease.

Single nucleotide polymorphisms (SNPs) are hypothesized to explain the genetic predisposition to ischemic heart disease (IHD) in the general population. Lack of evidence for a role of such variation is fostering pessimism about the utility of genetic information in the practice of medicine. In this study we determined the utility of exonic and 5' SNPs in apolipoprotein E (APOE) and lipoprotein lipase (LPL) when considered singly and in combination for predicting incidence of IHD in 8,456 individuals from the general population during 24 years of follow-up. In men, LPL D9N improved prediction of IHD (P = 0.03) beyond smoking, diabetes and hypertension. The group of men heterozygous and homozygous for the rare D9N variant had a hazard ratio (HR) of 1.69 (95% confidence interval = 1.10-2.58) relative to the most common genotype. Pairwise combinations of D9N with -219G > T in APOE and N291S and S447X in LPL significantly improved the prediction of IHD (P = 0.05 in women, P = 0.04 in men, P = 0.03 in men, respectively) beyond smoking, diabetes and hypertension, and identified subgroups of individuals (n = 6-94) with highly significant HRs of 1.92-4.35. These results were validated in a case-control study (n = 8,806). In conclusion, we present evidence that combinations of SNPs in APOE and LPL identify subgroups of individuals at substantially increased risk of IHD beyond that associated with smoking, diabetes and hypertension.

Complex adaptive system models and the genetic analysis of plasma HDL-cholesterol concentration.

Despite remarkable advances in diagnosis and therapy, ischemic heart disease (IHD) remains a leading cause of morbidity and mortality in industrialized countries. Recent efforts to estimate the influence of genetic variation on IHD risk have focused on predicting individual plasma high-density lipoprotein cholesterol (HDL-C) concentration. Plasma HDL-C concentration (mg/dl), a quantitative risk factor for IHD, has a complex multifactorial etiology that involves the actions of many genes. Single gene variations may be necessary but are not individually sufficient to predict a statistically significant increase in risk of disease. The complexity of phenotype-genotype-environment relationships involved in determining plasma HDL-C concentration has challenged commonly held assumptions about genetic causation and has led to the question of which combination of variations, in which subset of genes, in which environmental strata of a particular population significantly improves our ability to predict high or low risk phenotypes. We document the limitations of inferences from genetic research based on commonly accepted biological models, consider how evidence for real-world dynamical interactions between HDL-C determinants challenges the simplifying assumptions implicit in traditional linear statistical genetic models, and conclude by considering research options for evaluating the utility of genetic information in predicting traits with complex etiologies.

Consistent effects of genes involved in reverse cholesterol transport on plasma lipid and apolipoprotein levels in CARDIA participants.

OBJECTIVE: To identify common variations in genes in the reverse cholesterol transport pathway with nongender-specific influence on plasma lipid and apolipoprotein levels. METHODS AND RESULTS: An average of 5 single nucleotide polymorphisms (SNPs) were genotyped within each of 45 genomic regions (54 genes) in blacks (1131 females and 812 males) and whites (1102 females and 954 males) from the Coronary Artery Risk Development in Young Adults (CARDIA) study. SNPs and gene-based 3-SNP haplotypes were evaluated for their ability to predict variation in plasma apolipoproteins (apo) A-I and apoB, total cholesterol (TC), high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, and triglycerides (TG). We identified 14 SNPs in 6 candidate gene regions that explained statistically significant variation in the same trait in both genders of at least one race and with evidence of consistent genotype mean trend across gender within race. Haplotype analyses identified 9 candidate gene regions that explained statistically significant variation in one or both races. CONCLUSIONS: Four gene regions, ABCA1, APOA1/C3/A4/A5, APOE/C1/C4/C2, and CETP, explained plasma lipoprotein variation most consistently across strata. Other gene regions that influence plasma lipid and apolipoprotein levels within race include CYP7A1, LPL, PPARA, SOAT1, and SREBF2.

Contrasting multi-site genotypic distributions among discordant quantitative phenotypes: the APOA1/C3/A4/A5 gene cluster and cardiovascular disease risk factors.

Most tests of association between DNA sequence variation and quantitative phenotypes in samples of randomly chosen individuals rely on specification of genotypic strata followed by comparison of phenotypes across these strata. This strategy often succeeds when phenotypic differences are caused by one or two single nucleotide polymorphisms (SNPs) among the surveyed markers. However, when multiple-SNP haplotypes account for observed phenotypic variation, identification of the best partitioning requires examination of an inordinate number of SNP combinations. An alternative approach is to rank individuals by their phenotypic measures and ask whether attributes of the genotypic variation show a non-random distribution along this phenotypic ranking. One simple version of this strategy selects the top and bottom tails of the distribution, and then tests whether genotypes from these two samples are drawn from a single population. This framework does not require the recovery of phased haplotypes and allows contrasts between large numbers of sites at once. We use a method based on this approach to identify associations between plasma triglyceride level, a risk factor for cardiovascular disease, and multi-site genotypes located in the APOA1/C3/A4/A5 cluster of apolipoprotein genes in unrelated individuals (1,071 African-American females, 780 African-American males, 1,036 European-American females, and 930 European-American males) sampled from four US cities as part of the Coronary Artery Risk Development in Young Adults (CARDIA) study. Method performance is investigated using simulations that model genealogical variation and different genetic architectures. Results indicate that this multi-site test can identify genotype-phenotype associations with reasonable power, including those generated by some simple epistatic models.

Evidence for consistent intragenic and intergenic interactions between SNP effects in the APOA1/C3/A4/A5 gene cluster.

OBJECTIVE: Evaluate the consistency of the contribution of interactions between single nucleotide polymorphism (SNP) genotype effects to variation in measures of lipid metabolism across ethnic strata within gender. METHODS AND RESULTS: We considered 80 SNPs within the apolipoprotein (APO) A1/C3/A4/A5 gene cluster using an over-parameterized general linear model to identify SNPs whose genotype effects combine non-additively to influence plasma levels of high density lipoprotein cholesterol (HDL-C), total cholesterol (TC) and triglycerides (TG) in a consistent manner across ethnic strata. We analyzed population-based samples of unrelated 18 to 30 year old African-Americans (n = 1,858) and European-Americans (n = 1,973) ascertained without regard to health at four field centers (Birmingham, Ala.; Chicago, Ill.; Minneapolis, Minn. and Oakland, Calif., USA) by the Coronary Artery Risk Development in Young Adults (CARDIA) study. To identify which SNP genotype effects combine non-additively we used a two-tier analysis strategy. We first required that pairs of SNPs show statistically significant non-additivity in both ethnic strata within a gender, where experiment-wise significance was evaluated using a permutation test to determine the probability of observing the number of tests significant in both ethnic strata by chance alone. Second, we required no significant evidence of heterogeneity of the relationship between the phenotype and the two SNP genotypes across ethnic strata and across field centers within each ethnic group. From this strategy we identified ten pairs of SNPs, involving thirteen SNPs, that displayed statistically significant non-additivity of SNP genotype effects on TC. Only one of these thirteen SNPs had statistically significant genotype effects that were consistent across samples. CONCLUSION: Our analyses suggest that ignoring the contribution of interactions between SNP genotype effects when modeling multi-SNP genotype-phenotype relationships may result in an underestimate of the contribution of genetic variation to variation in quantitative cardiovascular disease (CVD) risk factor traits.