Genetic and Lifestyle Predictors of Nonfatal Myocardial Infarction in the Costa. Rica Study. A Dissertation Presented.

Transcription

1 Genetic and Lifestyle Predictors of Nonfatal Myocardial Infarction in the Costa Rica Study A Dissertation Presented By Stella Aslibekyan To The Department of Community Health In Partial Fulfillment of the Requirements For the Degree of Doctor of Philosophy In the Subject of Epidemiology Brown University Providence, Rhode Island May 2011

2 Copyright 2011 by Stella Aslibekyan

3 This dissertation by Stella Aslibekyan is accepted in its present form by the Department of Community Health as satisfying the dissertation requirement for the degree of Doctor of Philosophy (Epidemiology). Date: Ana Baylin, MD, DrPH, Advisor Recommended to the Graduate Council Date: Crystal D. Linkletter, PhD, Reader Date: Eric B. Loucks, PhD, Reader Date: Jose M. Ordovas, PhD, Reader Approved by the Graduate Council Date: Peter M. Weber, PhD, Dean of the Graduate School iii

4 Curriculum Vitae Stella Aslibekyan (b. 1984) graduated from Stanford University with a Bachelor of Arts with honors in Human Biology and a minor in Economics in She continued her education at the Harvard School of Public Health, where in 2008 she received a Master of Science degree in Epidemiology. Her research in chronic disease epidemiology has been accepted for publication in the American Journal of Cardiology, Journal of Nutrition, and Nutrition, Metabolism, and Cardiovascular Diseases. Additionally, she designed and taught several courses on gene-diet interactions in human disease pathogenesis. During her time at Brown University, Ms. Aslibekyan was awarded the Sidney Frank Graduate Fellowship and the Reginald D. Archambault Award for Teaching Excellence. She serves as the President-Elect of the Society for Epidemiologic Research Student Caucus. After graduation, she will pursue a post-doctoral fellowship in epidemiology under the mentorship of Dr. Donna Arnett at the University of Alabama at Birmingham. iv

5 Preface and Acknowledgments I would like to express my most profound gratitude to my dissertation advisor, Dr. Ana Baylin, for her caring and practical guidance, perpetually appropriate words of encouragement, patience with my bad SAS code and faulty Skype connections, and the inspiration to continue on the road of academic epidemiology. I would also like to thank my dissertation committee, whose feedback shaped this project and helped me grow as a scientist: Dr. Crystal Linkletter, for curing my phobia of multiple comparisons corrections once and for all; Dr. Eric Loucks, for his valuable expertise in social epidemiology; and Dr. Jose Ordovas, for being the god of gene-diet interactions. Finally, I would like to thank my outside reader, Dr. Hannia Campos, for the opportunity to work on the Costa Rica Study and all her support throughout this and other projects. This work is dedicated to the people without whom this Ph.D. saga would never have reached its happy ending: - To my father, whose own dissertation trials have inspired and informed my work (or at the very least scared me enough into finishing it); - To my mother, whose outlook on life makes the Yes, We Can crowds look like hopeless pessimists; - To Lan Ngo, whose friendship and support must be the biggest blessings to ever come out of Statistics office hours; - And most importantly, to the memory of my grandfather, Felix S. Aslibekyan, who taught me not to ask stupid questions. v

6 Table of Contents Signature Page....iii Curriculum Vitae....iv Preface and Acknowledgments....v Table of Contents....vi List of Tables....vii List of Illustrations......ix Introduction..1 Chapter 1 Development of a cardiovascular risk score for use in low- and middle-income countries...4 Chapter 2 Fatty acid desaturase gene variants, cardiovascular risk factors, and the risk of myocardial infarction in the Costa Rica Study..23 Chapter 3 Fatty acid elongase gene variants, cardiovascular risk factors, and the risk of myocardial infarction in the Costa Rica Study..42 Bibliography Supplemental Material vi

7 List of Tables Table Frequency of compliance with international guidelines for healthy lifestyle and income standards by case-control status Table Odds ratio estimates of nonfatal myocardial infarction associated with the value of the cardiovascular risk score in healthy participants Table General characteristics of the Costa Rica Study population included in the desaturase analysis Table Least square means of adipose fatty acids by quintile of genetic factor score among controls Table Least square means of blood lipids and inflammatory markers by quintile of genetic factor score among controls Table Association between genetic variation in the FADS cluster and the risk of myocardial infarction Table Least square means of LDL cholesterol by tertiles of the genetic factor score and adipose tissue ALA among controls Table vii

8 General characteristics of the Costa Rica Study population included in the elongase analysis Table Least square means of adipose fatty acids by genotype among controls Table Least square means of blood lipids and inflammatory markers by genotype among controls Table Risk of nonfatal myocardial infarction by genotype Supplemental Table Distribution of cardiovascular risk factors in the healthy population Supplemental Table Comparison of two cardiovascular risk score definitions by component Supplemental Table Single nucleotide polymorphism analysis: least square means of selected adipose fatty acids, blood lipids, and inflammatory markers by genotype among controls viii

9 List of Illustrations Figure Distribution of the study participants according to the number of cardiovascular risk recommendations followed Figure Receiver operating characteristic (ROC) curves evaluating performance of Score 1 (A) and Score 2 (B) models in the validation data set Supplemental Figure Pairwise standardized linkage disequilibrium coefficients (Lewontin s D') for the elongase family polymorphisms ix

10 INTRODUCTION Cardiovascular disease (CVD) is a leading cause of death, disability, and healthcare expenses throughout the world (1, 2). Over the past few decades, the burden of the CVD epidemic has been steadily shifting to middle- and low-income countries, with 80% of all chronic disease deaths and 63% of all CVD deaths now occurring in the developing world (3-7). As more countries across the globe undergo modernization and epidemiologic transition, the burden of CVD morbidity and mortality is likely to rise dramatically. Despite these projections, there remains a substantial research gap between the developed and the developing world, since most findings from industrialized countries are not necessarily applicable in the changing context of the global CVD epidemic (8). The need for epidemiologic studies investigating genetic and environmental determinants of CVD in the setting of epidemiologic and nutritional transition is evident and urgent. The role of environmental factors such as diet, physical activity, smoking, and alcohol intake in CVD etiology has been conclusively established in different populations, including developing countries (5). In Chapter 1, we developed and validated a novel score that is entirely comprised of such lifestyle factors and is the first chronic disease risk score derived in a non-western population undergoing nutritional transition. Our score presents a simple, low-cost alternative to the existing measures of cardiovascular risk stratification in resource-limited settings. 1

11 However, a growing body of evidence suggests that the effect of lifestyle factors on CVD risk is frequently modified by genetic variants within populations (9, 10). Furthermore, studies that consider genetic and environmental factors alone tend to consistently underestimate the population risk attributable to both exposures (11). To correct these estimates, many recent inquiries into CVD etiology have focused on geneenvironment, and particularly gene-diet interactions (11, 12). For example, several genetic loci (APOE, APOA1, APOC3, and others) in conjunction with dietary exposures have been shown to mediate CVD risk by altering blood lipid levels in different populations (13-15). The evidence for other gene-nutrient interactions is also emerging. In particular, much emphasis has been placed on the role of genes involved in polyunsaturated fatty acids (PUFAs) biosynthesis, and their interactions in the development of cardiovascular and metabolic outcomes. Despite the wealth of evidence, the effect of PUFAs on cardiovascular health has not been consistent across populations (16-20). Although such heterogeneity may arise from variation in dietary sources of PUFAs and/or ranges of intake, it has also been hypothesized that genetic polymorphisms may modify the cardiovascular effects of dietary PUFAs. Specifically, genetic variants that affect expression or efficiency of desaturases and elongases, which are the critical enzymes in the n-3 and n-6 pathways, may alter the effect of dietary PUFA intake on the risk of CVD. In Chapters 2 and 3, we investigated associations between common polymorphisms in the desaturase and elongase genes respectively, adipose tissue PUFAs, intermediate cardiovascular risk markers such as serum lipids and inflammatory markers, and the risk of myocardial infarction (MI). We also assessed potential interactions 2

12 between the elongase and desaturase genotypes and dietary PUFAs, namely alphalinolenic and linoleic acids. The overall goal of this study is to further our understanding of the role that genetic and environmental predictors play in the etiology of nonfatal MI in a population undergoing nutritional transition. It is our hope that this study will help bridge the gap in CVD research between the developed and developing countries, as well as identify better nutritional recommendations for Hispanic populations worldwide. 3

13 CHAPTER 1 Development of a cardiovascular risk score for use in low- and middle-income countries 1.1 Background Summary measures of cardiovascular risk such as the Framingham score have long been used in public health research and practice (21-24). These risk scores play an important role in screening programs, identifying susceptible individuals before the onset of clinical symptoms and facilitating primary prevention (25). Cardiovascular risk scores are also used in epidemiologic research, namely as measures of exposure, stratification variables, or measures of potential confounders (26, 27). The predictive ability of existing cardiovascular risk scores varies greatly between populations, and is particularly decreased in ethnic minority communities (28). Such variability merits further investigation because most cardiovascular risk measures in use were derived in high income Western cohorts (29, 30) and evidence of their generalizability to the context of low or middle income populations worldwide, or even to some ethnically and economically diverse communities of high income countries, is limited (31). Additionally, no commonly used cardiovascular risk score incorporates nutritional predictors despite overwhelming evidence, including studies from developing countries, linking diet to the risk of heart disease (32). Dietary risk factors are modifiable and less costly to measure than many of the intermediate risk factors (e.g. blood lipid levels), 4

14 which makes them an attractive option for use in resource-poor settings. As more populations begin the nutritional and epidemiologic transitions associated with modernization, new methods of risk stratification may be necessary to address the increasing global burden of cardiovascular disease. The purpose of our study is to derive and validate a novel cardiovascular risk score comprised of predictors such as diet, physical activity, smoking, alcohol consumption, waist-to-hip ratio, and socioeconomic status, in a population-based casecontrol study of MI in Costa Rican adults. Although an argument can be made that socioeconomic status is not truly a modifiable risk factor, it is included into the score as an important upstream determinant of cardiovascular risk that could be impacted by economic development and policy changes. 1.2 Methods Study Population The population of the Costa Rica Study included 4,547 Hispanics who resided in the Central Valley of Costa Rica between 1994 and 2004 (33-36). Cases of first nonfatal acute MI were ascertained by two independent cardiologists in the participating hospitals and deemed eligible if they met the World Health Organization criteria (37), survived hospitalization, were under 75 years of age on the day of their first MI, and able to answer the questionnaire. Eligible cases (n= 2273) were matched by 5-year age group, sex, and area of residence to population controls (n= 2274), identified randomly using data from the National Census and Statistics Bureau of Costa Rica. Women comprised 27% of all study participants (1209 total, 605 controls and 604 cases). After the cases 5

15 were discharged from the hospital, all cases and controls received home visits, during which trained study workers collected lifestyle and medical history data, anthropometric measurements, and biological specimens. Information on diet, physical activity, smoking, alcohol intake, socioeconomic status, and medical history was collected using questionnaires (38-40). Dietary exposures were ascertained both via food frequency questionnaires (FFQ) and biological markers, specifically adipose tissue concentrations of selected fatty acids (41, 42). To avoid reverse causation and recall bias, data on exposures among cases were recorded as close to the time of MI as possible. Participation was 98% for cases and 88% for controls. All participants provided written informed consent. The study has been approved by the Human Subjects Committee of the Harvard School of Public Health and the University of Costa Rica Descriptive Statistics All statistical analyses were conducted using Statistical Analysis Software version 9.2 (Cary, NC). To assess dietary habits in the Costa Rica Study population, the study used World Health Organization nutrient intake goals (43, 44). Participants were considered compliant with a specific guideline if their self-reported values fell within the recommended range of intake. Participants were excluded from this analysis if they were missing information on any of the covariates, yielding a sample size of 4091 (Figure 1.1). In an additional analysis, cases were re-matched to controls on age, sex, and area of residence to preserve the study design, resulting in a sample size of 3968 (Table 1.1). The distribution of risk factors by case-control status was compared using McNemar s test. Frequency tables and a histogram were constructed to ascertain the proportion of the 6

16 study sample that reported adherence to World Health Organization guidelines on nutrient intake and healthy lifestyle, as well as fell above the national poverty line threshold Measurement of the Cardiovascular Risk Score Components Two versions of the cardiovascular risk score were developed, one based on World Health Organization nutrient intake and physical activity recommendations and poverty line standards outlined above (Score 1, n= 839 cases and 839 controls) and one incorporating biomarker measures and more refined socioeconomic status measures available in the Costa Rica Study (Score 2, n=696 cases and 696 controls). Cardiovascular risk score components were selected based on a prior analysis of modifiable MI risk factors in our study population as well as international guidelines for healthy lifestyle (34, 43). The selected risk score components showed expected associations with the risk of MI in our study population (Supplemental Table 1.1). Score 1 was derived to ensure a simple, low-cost risk estimation algorithm that could be adapted to a variety of populations, while Score 2 was designed to include the most reliable measures of risk factors available for the Costa Rica Study population (Supplemental Table 1.2). The healthy diet score used in Score 2 was derived as a composite measure of total dietary intake of saturated fats, cholesterol, polyunsaturated fats, fiber, folate, and adipose tissue -linolenic acid (ALA) and total trans fats. Adipose tissue ALA was chosen for inclusion in Score 2 due to its importance as a cardioprotective factor in the study population, characterized by low ALA intake (20). Although adipose tissue was used in this population as the most reliable measure of long- 7

17 term intake, ALA intake as estimated by FFQ could also constitute a valid measure. Physical activity information was collected using a questionnaire described in more detail in previous publications (38). Briefly, participants reported the average frequency and time spent on several occupational and leisure time activities during the last year. Energy expenditure for each activity was calculated as the product of frequency, time, and intensity measured in METS (metabolic equivalents, defined as the energy expenditure for sitting quietly or approximately kj. kg body -1. h -1 ) (38). The physical activity questionnaire was validated by its ability to predict fitness level measured by the Harvard Step test, plasma lipids, and obesity, in our previous studies in Puriscal, Costa Rica (39, 40). Smoking and alcohol intake were measured using questionnaires. Anthropometric measurements, including waist-to-hip ratio, were collected the morning after an overnight fast by trained fieldworkers while subjects wore light clothing and no shoes. Measurements were performed in duplicate, with the average used in the analysis. Finally, income was measured by showing participants index cards with ranges of income (in US dollars per month) and asking them to select the appropriate index card Derivation of the Cardiovascular Risk Score The derivation data sets were developed using the complete case method, i.e. participants with missing values for any of the covariates were excluded from the analysis. Additionally, participants with a self-reported history of diabetes, hypertension, or current use of medication for chronic conditions were excluded from the derivation data set to avoid reverse causation (34). After restriction, remaining cases were re-matched to controls on age, sex, and area of residence to preserve the study design. 8

18 Participants were categorized according to each risk score component as follows. All dietary variables included in Score 1 (trans fats, polyunsaturated fats, saturated fats, cholesterol fiber, and folate) were included as categorical variables to model the relation between each dietary factor and MI. For each component of the healthy diet measure used in Score 2 (polyunsaturated fats, saturated fats, cholesterol fiber, folate, and adipose tissue trans fats and -linolenic acid), the participants were assigned scores from 0 to 4 corresponding to the quintile of intake, with 4 representing the lowest risk quintile (Supplemental Table 1.1). Quintiles of the healthy diet score were based on all participants. The assigned quintile values were then summed to produce the dietary score for each participant (45). Therefore, the resulting dietary score ranged from 0, which indicated the lowest possible adherence to dietary guidelines, to 28, which represented the highest possible adherence to dietary guidelines. For Score 1, being physically active was defined as expending more than 10% of daily energy in the performance of moderate- and high-intensity physical activities (at least four times the basal metabolism rate) (46). For Score 2, physical activity was included as a continuous variable, defined as total METS expended over a 24-hour period. For both Score 1 and Score 2, smoking was defined as a dichotomous variable (currently smoking vs. not), while alcohol intake was measured in grams per day and defined as a categorical variable with the following cutoffs: 0 (not drinkers), , , over 10. For both scores, participants were classified as healthy if their waist-to-hip ratio value lies below the cutoff of 0.90 for men and 0.85 for women as per World Health Organization guidelines (47). For Score 1, socioeconomic status was classified as low if a participant s self-reported annual income fell below the threshold of twice the national poverty line for the year of 9

19 recruitment into the study (48, 49). For Score 2, we used the socioeconomic status index (continuous variable), described in previous publications from our group as a comprehensive measure of education, occupation, income and household possessions (50). For each version of the score, conditional logistic regression models were fit with MI as the outcome and cardiovascular risk score components as predictors, while matching on age, sex, and area of residence to control for potential confounding by these demographic factors. The obtained regression coefficients for each score component were then multiplied by the values of cardiovascular risk score components, and summed across components to produce the final value of the cardiovascular risk score. Thus, the final cardiovascular risk score value represents a weighted sum of individual risk score components (51). Two regression models, each adjusted for age, sex, and area of residence, were fit to assess the discriminatory ability of Score 1 and Score 2: one with each score as a continuous variable and one with indicator variables corresponding to quintiles of each score s distribution. A test for linear trend was performed on categorical models, using the median value of each quintile as a continuous predictor. Sensitivity analyses were conducted using different subsets of components of the cardiovascular risk scores. For example, a conditional logistic regression model was fit including foods as continuous variables (using type of oil used in the household as proxy for fat intake and number of servings of fruits, vegetables, and beans as proxies for fiber and folate) in an effort to make cardiovascular risk assessment more accessible in resource-limited settings. In another sensitivity analysis, a conditional logistic regression model was fit replacing waist-to-hip ratio with waist circumference as a continuous 10

20 variable. Because the models were not nested, we compared their fit using the Akaike Information Criterion (AIC) (52) Validation of the Cardiovascular Risk Score The validation data set comprised all study participants excluded from the derivation data set, i.e. participants with a self-reported history of hypertension, diabetes, and/or hypercholesterolemia. The models for both Score 1 and Score 2, derived in healthy participants, were used to predict the probability of MI in the validation data set, and ROC curves were constructed from all combinations of sensitivity and (1 specificity) characterizing Score 1 and Score 2 respectively. Areas under the ROC curve (c-statistic) were used to assess the models predictive ability in the validation data set and compared to similar measures of performance in currently used cardiovascular risk stratification models (28, 30). For a comprehensive validation of the score models, the validation and derivation data sets were then switched, with Scores 1 and 2 now derived in the subpopulation with a history of chronic disease and validated in the healthy subpopulation using the ROC method described above. 1.3 Results Descriptive Statistics We estimated the proportion of the entire study population by case/control status (n=1984 for cases, 1984 for controls after restriction to participants with complete information on covariates) that reported adherence to international guidelines for healthy lifestyle and income standards (Table 1.1). Compared to controls, cases were less likely to follow the 11

21 recommendations regarding intake of saturated fats, polyunsaturated fats, trans fats and cholesterol, as well as smoking, but were more likely to follow guidelines on fiber and folate. Additionally, cases were more likely to exhibit higher waist-to-hip ratios, and to be below the income threshold of twice the national poverty line. Physical activity and alcohol consumption patterns did not significantly differ by case/control status. Overall, compliance was lowest for saturated fat intake (23% for cases, 30% for controls) and moderate alcohol consumption (16% for both). The majority of participants (total n=4091) complied with approximately half of all lifestyle recommendations (Figure 1.1). The mean age of the study population was 58 years (range: years) Score Derivation Of all risk score components, current smoking was associated with the biggest increase in the risk of MI [multivariate-adjusted OR= 3.17 (95% CI = )] (n=1386) (Supplemental Table 1.1). Values of the healthy diet score in this population ranged from 0 to 28. For most factors, a gradient of risk was observed between the categories of the variable. Higher values of cardiovascular risk scores (n=1678 for Score 1 and n=1392 for Score 2) were associated with a significant increase in the risk of MI in both the continuous and categorical models [Score 1: OR for the continuous variable= 2.72 (95% CI = , two-sided P < ), for the categorical model, two-sided P-trend < Score 2: OR for the continuous variable = 2.71 (95% CI = , two-sided P <0.0001), for the categorical model, two-sided P-trend <0.0001] (Table 1.2). The estimated odds ratios represent the exponentiated slope of the corresponding regression 12

22 line. The fit of the continuous models was preferred to the categorical ones due to their lower AIC values. The score was robust to the substitution of type of oil for fat intake and number of servings of fruits, vegetables, and beans for fiber and folate [OR for the continuous model= 2.72 (95% CI= )]. Finally, replacing waist-hip-ratio with waist circumference slightly increased the estimates for both scores [OR for Score 1 (continuous)= 2.75 (95% CI= ); OR for Score 2 (continuous)= 2.73 (95% CI= )] Score Derivation ROC curves were constructed to evaluate the performance of both scores in the validation data set (Figure 1.2, panels A and B). The area under the curve was estimated at 0.63 and 0.64 for Score 1 and Score 2 respectively. When the derivation and validation data sets were switched, the c-statistics for the corresponding ROC curves (not shown) were estimated at 0.65 for Score 1 and 0.67 for Score 2. In the sensitivity analysis replacing nutrients in Score 1 with foods, c-statistics were estimated at 0.63 and 0.65 (ROC curves not shown). 1.4 Discussion We developed and validated a new cardiovascular risk score in a population from Costa Rica, a middle income country undergoing nutritional and epidemiologic transition. An increase of one unit in the score value was associated with a more than two-fold increase in the risk of first nonfatal MI in the study population. The results were robust to inclusion of participants with known risk factors for cardiovascular disease, and the 13

23 discrimination ability of the score as evaluated by the area under the ROC curve was typically somewhat lower than that of other established risk measures. For example, the c-statistic for the Framingham risk score in diverse ethnic subgroups ranges from 0.63 to 0.79 in men and 0.66 to 0.83 in women (28). It is also important to note that the reported c-statistic estimates are conservative, as the predictive quality of our models may have been adversely affected by differences in characteristics (specifically, chronic disease history) between the derivation and validation data sets. Additionally, some of the components of the Framingham score (e.g. blood lipids) are considerably more difficult and expensive to measure than lifestyle variables. Other proposed cardiovascular risk measures, such as novel biomarkers, are even more costly to implement and offer only modest improvements in predictive ability beyond the Framingham risk score (53). In a limited-resource setting, our proposed score provides a feasible and effective alternative to currently available risk measures. Our new risk score is exclusively comprised of lifestyle risk factors, namely diet, physical activity, smoking, alcohol consumption, socioeconomic status, and waist-to-hip ratio. The modifiable nature of this score has several important implications. First, it empowers clinical and public health practice, because it clearly illustrates how the risk of coronary heart disease in the study population can be reduced by addressing each of the score components. Second, it presents a convenient summary estimate of cardiovascular risk due to lifestyle factors, facilitating future epidemiologic research on heart disease in Costa Rica. For example, the new score could be used as a risk stratification variable, a measure of lifestyle confounders, or a convenient summary of environmental factors in studies investigating gene-environment interactions. Third, it enables risk estimation for 14

24 the entire lifestyle pattern, from diet and health behaviors to socioeconomic status. Fourth, the choice of total physical activity rather than solely recreational actvity provides a more reliable estimate of caloric expenditure in the setting of middle or low income countries, where recreational physical activity is less common (54). Finally, the score can be easily adapted to incorporate risk factors important to other populations, or to include FFQ measures of dietary variables when biomarker data are not available. Depending on specific study designs and objectives, new adaptations of our score can also include variables like age and sex, which are not modifiable but highly predictive of cardiovascular risk across populations. The limitations of the study include the use of the complete case method to construct the validation and derivation data sets. Participants with missing values for any of the covariates were dropped from the analysis, which requires the assumption that the data are missing at random. Although there is no strong evidence to suggest otherwise, this assumption has not been tested in this study. Another limitation are possible inaccuracies in self-report of nutritional intake, as all dietary covariates could not be ascertained using biomarker data. In addition to non-differential misclassification associated with the use of an FFQ, under-reporting of caloric intake and adverse health habits has been well documented in a variety of populations and is especially prevalent among high risk participants, potentially resulting in a risk estimate that is biased towards the null (55). Additionally, our definition of the outcome as nonfatal MI did not include other adverse coronary heart disease events such as fatal MI or stroke, limiting the scope of our risk score. Further, since the blood lipid measures among cases in the Costa Rica Study were taken post-mi, we could not directly compare the performance of our score 15

25 with that of the Framingham risk score in our study population. Moreover, the estimates provided by the two versions of the score are also not directly comparable, as the analyses were based on slightly different study populations due to availability of covariate information. Finally, as this was a case-control study with participants matched on age and sex, these findings apply primarily to the age groups investigated in this study (ages 18-86). In order to be incorporated for use in the general population, these findings should be replicated, and researchers should consider providing age- and sex-specific cardiovascular risk score algorithms. While the burden of cardiovascular disease in Costa Rica is substantially lower than the average for the Americas (188 deaths /100,000 population vs. 281 deaths/100,000 population, age-standardized) (56), compliance with international healthy lifestyle guidelines in our study population is similar to or worse than other middle income countries. A recent study of modifiable cardiovascular risk factors in Peru reported the prevalence of non-smokers at 64% and of regular physical activity at 42%, although in contrast to our data, approximately two thirds of the study sample reported healthy dietary habits (57). The reasons for the discrepancy between higher risk factor prevalence and lower cardiovascular mortality rates are unknown but may include improved access to health services in Costa Rica, differences in risk factor assessment, or other unmeasured factors. The new cardiovascular risk score, derived and validated using data from the Costa Rica Study, presents a quantitative summary of modifiable coronary heart disease predictors in this population. Our score represents a novel approach to cardiovascular risk assessment in middle- and low-income countries as a simple, low-cost alternative to the 16

26 Framingham measure. Furthermore, this score offers a unique framework for assessing the impact of lifestyle factors on the risk of chronic disease in the context of nutritional transition. Future directions in validation of the cardiovascular risk score should entail a comparison of this method with established scores such as the Framingham or SCORE equations, as well as applications of the score to other populations, especially in developing countries. 17

27 Figure 1.1 Distribution of the study participants (n=4091) according to the number of cardiovascular risk recommendations followed. 18

28 Figure 1.2 Receiver operating characteristic (ROC) curves evaluating performance of Score 1 (A) and Score 2 (B) models in the validation data set of the Costa Rica Study. A B 19

29 Table 1.1 Frequency of compliance with international guidelines for healthy lifestyle and income standards in the Costa Rica Study population (n=1984 cases, 1984 controls) by case-control status. Cases n (%) Controls n (%) Healthy diet Saturated fats < 10% of total energy 450 (23) * 593 (30) Polyunsaturated fats = 6-10% of total energy 1044 (53) * 1114 (56) Trans fats < 1% of total energy 679 (34) * 744 (38) Cholesterol < 300 mg/d 1014 (51) * 1252 (63) Fiber > 25 g/d 951 (48) * 862 (41) Folate > 400 g/d 944 (56) * 818 (51) Physically active ** 702 (35) 751 (38) Waist-to-hip ratio 0.90 for men, 0.85 for women 202 (10) * 327 (16) Currently non-smoking 1182 (60) * 1557 (78) Alcohol intake = g/d 323 (16) 327 (16) Income > twice the national poverty line 1734 (87) * 1798 (91) * P <0.05 (McNemar s test) 20

30 ** Physically active defined as expending more than 10% of daily energy in the performance of moderate- and high-intensity physical activities (at least four times the basal metabolism rate) 21

31 Table 1.2 Odds ratio estimates of nonfatal myocardial infarction associated with the value of the cardiovascular risk score in healthy participants of the Costa Rica Study (n=1678 for Score 1 and n=1392 for Score 2). Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 Model * OR (95% CI) AIC P Score 1, categorical 1.00 (referent) 1.25 (0.90, 1.74) 1.91 (1.37, 2.66) 3.52 (2.51, 4.93) 6.26 (4.33, 9.05) < (trend) Score 2, categorical 1.00 (referent) 1.78 (1.21, 2.62) 2.25 (1.52, 3.32) 5.10 (3.37, 7.72) 8.11 (5.25, 12.55) < (trend) Score 1, continuous 2.72 (2.28, 3.24) Score 2, continuous 2.71 (2.26, 3.26) < < * All models were adjusted for age, sex, and residence (by matching) 22

32 CHAPTER 2 Fatty acid desaturase gene variants, cardiovascular risk factors, and the risk of myocardial infarction in the Costa Rica Study 2.1 Background Long-chain omega-3 polyunsaturated fatty acids (n-3 PUFAs), obtained either through dietary intake or synthesized endogenously from ALA, have long been known to be protective against heart disease, diabetes, and other chronic outcomes (58, 59). Previous studies showed that in populations with low intake of marine fatty acids, dietary intake of ALA also confers cardioprotective benefits, either by itself or through conversion to long-chain PUFAs (60). The delta5- and delta6-desaturases, encoded by FADS1 and FADS2 genes respectively, play a critical role in the conversion pathway and could have implications for chronic disease risk, especially in the context of a long-chain PUFA deficient diet (61, 62). Several studies, including a recent genome-wide association scan (GWAS), linked polymorphisms in the FADS gene cluster to PUFA concentrations in serum phospholipids and erythrocyte cell membranes (63-69). A previous analysis of our data from the Costa Rican population, characterized by low fish intake, examined the effect of a common single nucleotide polymorphism (SNP) in the FADS2 promoter region on the risk of nonfatal myocardial infarction (MI) (70). Although a consistent decrease in adipose and plasma PUFA concentrations was observed with an increase in number of copies of the minor allele, the association with MI was not significant (70). 23

33 The mechanisms by which genetic variation in desaturases impacts cardiovascular health are unclear and likely to involve multiple pathways. A study conducted in a population with high dietary intake of n-6 and low intake of n-3 fatty acids has linked the number of risk alleles in the FADS cluster to high-sensitivity C-reactive protein (hscrp) concentrations and prevalence of coronary artery disease, suggesting inflammation as a mechanism of interest (65). Another potential pathway supported by extensive evidence from genome-wide studies links desaturase polymorphisms to changes in serum cholesterol and triglycerides in a variety of populations (71-74). While most published studies have considered FADS gene variants individually, the complex nature of biological mechanisms underlying the association between desaturases and cardiovascular risk warrants a more comprehensive approach. The objective of this study is to evaluate the association between genetic variation in the entire FADS cluster and the risk of MI, and to examine potential gene-diet interactions with the intake of ALA and LA in the fish-deficient Costa Rican population. Additionally, we investigate the association between FADS cluster variants and intermediate outcomes such as serum lipids and inflammation, providing novel mechanistic insights into the relation between desaturases, dietary fatty acids, and cardiovascular disease. 2.2 Methods Study Population The population of the Costa Rica Study, described in detail in prior publications, included 4548 Hispanics who resided in the Central Valley of Costa Rica between 1994 and

34 (33-35) Cases of first nonfatal acute MI were ascertained by two independent cardiologists in the participating hospitals and deemed eligible if they met the World Health Organization criteria, survived hospitalization, were under 75 years of age on the day of their first MI, and able to answer the questionnaire (37). Eligible cases were matched by 5-year age group, sex, and area of residence to population controls, identified randomly using data from the National Census and Statistics Bureau of Costa Rica. After the cases were discharged from the hospital, all cases and controls received home visits, during which trained study workers collected lifestyle and medical history data, anthropometric measurements, and biological specimens. Participation was 98% for cases and 88% for controls. The study population is appropriate for investigating genetic markers of disease due to its origin in a small number of founders and low rates of migration (70). The original sample size was 2274 cases and 2274 controls. Participants missing information on outcomes, exposure, or covariates were excluded from the analysis. Excluded participants did not differ (p>0.05) from the included participants for demographic (age, sex, and area of residence), dietary (LA and ALA) or genetic covariates (individual ancestral proportions), reducing the possibility of bias due to the complete case approach. Of the original sample, information on adipose tissue LA and ALA was obtained on 1858 cases and 1956 controls. Of these, genotype information was available on 953 cases and 1039 controls. For the analyses of adipose tissue fatty acids, inflammatory markers, and blood lipids, 7 additional participants were excluded due to missing information on ancestry, yielding a sample size of 1032 controls. For the MI analysis, 6 additional controls and 44 cases were excluded due to missing information on 25

35 covariates such as smoking, physical activity, adipose tissue trans fats, waist circumference, or saturated fat intake, resulting a sample size of 909 cases and 1026 controls. Additionally, cases were re-matched to controls on age, sex, and area of residence to preserve the gains in efficiency and validity due to matched study design; 148 controls and 31 cases were lost during the re-matching process, resulting in a final sample size of 878 controls and 878 cases. All participants provided written informed consent. The study was approved by the Human Subjects Committee of the Harvard School of Public Health and the University of Costa Rica Exposure Ascertainment Exposures were ascertained via adipose tissue biomarkers for the following fatty acids: 18:3n-3 (ALA), 18:2n-6 (LA), 18:3n-6 (gamma-linolenic, GLA), 20:3n-3 (eicosatrienoic, ETA), 20:2n-6 (eicosadienoic, EDA), 20:3n-6 (dihomo-gamma-linolenic, DGA), 20:4n-6 (arachidonic, AA), 20:5n-3 (eicosapentaenoic, EPA), and 22:6n-3 (docosahexaenoic, DHA). Advantages of using adipose tissue biomarkers to characterize long-term nutritional intake include slow turnover, absence of recall bias, and lack of response to conditions of acute disease (42). However, previous analyses of our data showed that adipose tissue concentrations of AA are poorly correlated with dietary intake due to high metabolic regulation; additionally, little is known regarding correlations of dietary intake with adipose tissue ETA and EDA (42). Therefore, we used adipose tissue concentrations of these fatty acids as metabolic markers representing the endogenous component explained by genetic variation, adjusted for the exogenous component (dietary intake) as measured by the previously validated FFQ (41). 26

36 All biological samples were collected following an overnight fast. Subcutaneous adipose biopsies, collected following an overnight fast, were performed with a 16-gauge needle using a modification of the method proposed by Beynen and Katan (75). Fatty acids from adipose tissue were quantified by gas-liquid chromatography (42). Peak retention times and area percentages of total fatty acids were identified with the use of known standards (NuCheck Prep) and were analyzed with the ChemStation A software (Agilent Technologies) (76). Average coefficients of variation for twelve blind duplicates were 3.87% for ALA and 14.2% for EPA (76). Biomarker analyses of inflammation and serum lipids were restricted to control subjects to preclude reverse causation. Blood tubes were centrifuged within 6 hours at 2,500 rpm for 20 min to separate plasma. Plasma triglyceride, low-density lipoprotein (LDL), high-density lipoprotein (HDL), and total cholesterol were assayed with enzymatic reagents (Boehringer-Mannheim) (77). Cholesterol measurements were standardized according to the program specified by the Centers for Disease Control and the National Heart, Lung and Blood Institute (77). To ascertain the presence of inflammation, the study used hscrp and vascular cell adhesion molecule-1 (VCAM-1), two plasma biomarkers that have been extensively validated and linked to cardiovascular risk in large-scale prospective studies (78-80). Concentrations of hscrp were measured using immunoturbidometry on Roche Modular P chemistry autoanalyzer (Hoffman La Roche). The average coefficient of variation for 12 blind duplicates was 1.42% and ranged from 0% to 5%. The Quantikine Human svcam-1 assay (R&D Systems, kit lot # ) was used to measure svcam-1. The average coefficient of variation for 12 blind duplicates was 9.58%. All assays were performed with the same batch of reagents. 27

37 Ancestry was estimated using a set of 39 informative markers with allele frequencies from Amerindian, European, and West African samples (81). Expected variance of estimated individual ancestry proportions for any particular set of loci was calculated using a maximum likelihood approach. Based on the set of selected markers, individual ancestral proportions were estimated with a high degree of precision (standard error 0.15) (81) SNP Selection and Genotyping Fifteen SNPs were selected as tagging SNPs using information from the HapMap Project ( and the National Center of Biotechnology Information ( The tagging variants were selected so that even if they do not have a functional role, they would be in linkage disequilibrium with the true causative polymorphism and thus serve as its efficient surrogates (82). Of those, seven SNPs located on the FADS gene cluster were selected for analysis based on previously published evidence of their common availability and role in FA metabolism: rs (C/T), rs (T/deletion), rs (C/T), rs (A/T), rs (C/G), rs (T/C), rs (C/T). (64, 65, 70, 83) A DNA stock sample of approximately 200 mg was collected from all study participants and stored at 80 C in Costa Rica. This sample was extracted from frozen buffy coats using the Qiagen QIAamp DNA Blood Kit. Purity was determined by the ratio of absorbence at 260 to 280 (A 260 /A 280) to be between for all samples. For high-throughput genotyping 15ng/ml working solutions were prepared and aliquoted into 96-well plates. Genotyping was performed at the University of Cincinnati using the 28

38 SNPlex Genotyping System (Applied Biosystems). Fragmented genomic DNA (50 ng) was dried into each well of a 384-well plate (approximately 1 ng DNA per genotype). After phosphorylation of Oligonucleotide Ligation Assay (OLA) probes and universal linkers, allele-specific ligation and enzymatic purification were performed. Polymerase chain reaction (PCR) utilized universal biotinylated primers, so amplicons could be captured on streptavidin-coated plates. Single-strand PCR products were hybridized with a universal set of fluorescently dye-labeled mobility modifiers, the ZipChute probes that have a unique sequence corresponding to each SNP. ZipChute probes were eluted and separated for detection by capillary electrophoresis on ABI PRISM 3130XL DNA Analyzer (Applied Biosystems). Data were collected, formatted, processed, and analyzed using the GeneMapper Analysis Software (Version 4.0), which assigned individual genotypes Statistical Analysis Data from the Costa Rica Study were analyzed using the SAS software package (Version 9.2; SAS Institute Inc, Cary, NC). To assess the significance of differences in general characteristics and potential confounders, we used paired t-tests for continuous variables, McNemar s tests for categorical variables, and Fisher s exact test for minor allele frequencies. The ALLELE procedure was used to test for deviations from Hardy- Weinberg equilibrium among controls. Of all SNPs, only rs was found to be in violation of the Hardy-Weinberg equilibrium and thus removed from all subsequent analyses. 29

39 Of the six remaining SNPs, a score were constructed to capture genetic variation in the FADS cluster. Linear regression models were fit among the controls with AA as the outcome and all six SNPs as predictors, adjusting for age, sex, and ancestry. AA was chosen as the outcome due to its highly regulated biosynthesis. The obtained regression coefficients for each SNP were then multiplied by 0, 1, or 2 for wild type homozygotes, heterozygotes, and mutant homozygotes respectively, and summed across SNPs to produce the value of the genetic score. Because the number of minor alleles is inversely associated with adipose tissue AA, the obtained regression coefficients were negative; therefore, the sign of the final value was reversed to ensure that the genetic score accurately captures the relation between individual SNPs and outcomes. The final genetic score value represents a weighted sum of individual genetic variants in the FADS cluster. Linear regression models were fit among controls to evaluate the association between the genetic score and the adipose tissue concentrations of long chain PUFAs as well as plasma concentrations of hscrp, VCAM-1, and serum lipids. Least square means and 95% confidence intervals were used to report the relation between the outcomes and FADS genetic variants. Log-transformations were carried out for non-normally distributed variables (GLA, hscrp, and triglycerides) and geometric means were reported. The intermediate risk factors models were adjusted for age, sex, and residence area (by matching) and ancestry, while the PUFA models were additionally adjusted for dietary intake of fatty acids. The relation between genetic variation in the FADS cluster and the MI outcome was modeled using conditional logistic regression, adjusted for age, sex, residence area (by matching), and ancestry with the genetic score as the explanatory 30

40 variable. Additional analyses were conducted using individual SNPs as predictors and adipose tissue fatty acids, inflammation markers, blood lipids, and MI as outcomes. Finally, departures from additivity were considered for the genetic score and the precursor fatty acids (ALA and LA). For the outcomes that showed a statistically significant relationship (p-value for linear trend<0.20) with both fatty acids and the genetic score, interaction terms were added to the linear regression models, which were further adjusted for dietary and demographic confounders. Again, participants were excluded from the analyses if they were missing information on exposures, outcomes, or any of the covariates, with the final sample size of 945. Homogeneity across tertiles of the genetic risk score was assessed using partial F-tests. 2.3 Results The general characteristics of the study population are summarized by case/control status in Table 2.1. None of the selected SNPs or individual ancestral proportions differed significantly by disease status. Cases were significantly more likely to report MI risk factors, specifically smoking and lower household income. Additionally, cases had significantly lower adipose tissue concentrations of ALA and LA. As shown in Table 2.2, least square means of adipose tissue ALA and LA did not vary significantly by quintile of the genetic score. Adipose tissue GLA, AA, and EPA showed a statistically significant decrease across genetic score quintiles; a borderline significant linear decrease was also observed for DHA. On the other hand, EDA, ETA, and DGA showed a significant linear increase across quintiles of the genetic score. The results were consistent with single SNP analyses, summarized in Supplemental Table

41 Table 2.3 shows the least square means of blood lipids and inflammatory biomarkers by quintile of the genetic score. VCAM-1, HDL-, LDL-, and total cholesterol were not significantly associated with genetic variation in the desaturase cluster. A significant linear increase was observed for triglycerides, while hscrp showed a borderline significant decreasing linear trend across quintiles of the genetic score. Genetic variation in the FADS cluster was not associated with the risk of first nonfatal MI in the Costa Rica study (Table 2.4). Additive interactions between the genetic score and LA/ALA were evaluated for hscrp, triglycerides, and LDL cholesterol. No statistically significant interactions were observed for triglycerides or hscrp, or for LA (data not shown). The association of ALA with LDL varied by tertile of the genetic score (p for interaction= 0.02), with the lipid least square means increasing in the first and second tertiles but decreasing in the third tertile (Table 2.5). 2.4 Discussion Our findings show that genetic variation in the FADS cluster is associated with adipose tissue fatty acids, serum triglycerides, and hscrp in the Costa Rica Study. Specifically, we demonstrated linear associations of a genetic score composed of six desaturase SNPs with decreases in adipose AA, GLA, and EPA as well as with increases in adipose EDA, ETA, and DGA. Our results contribute to the body of evidence in support of robust associations between long-chain PUFAs and FADS cluster polymorphisms, including a recent GWAS conducted in a European population (63). Specifically, in concordance with our findings, the strongest signal in the FADS cluster reported by Tanaka et al. (rs174537) was similarly associated with a decrease in long-chain PUFA (namely AA 32

42 and EPA) concentrations (63). Additionally, the observed increase in serum triglycerides is consistent with the known effects of long-chain n-3 fatty acids on blood lipids (84, 85). On the other hand, the decrease in hscrp, a marker of systemic inflammation, may be explained by the decrease in pro-inflammatory eicosanoids at lower concentrations of AA (86). The observed associations of desaturase polymorphisms with intermediate cardiovascular risk factors, however, do not translate into a change in the risk of MI in our study population. Given the evidence of manifold physiological effects of PUFAs, is likely that the overall effect of FADS polymorphisms on cardiovascular health is determined by a combination of mechanisms such as inflammation, hyperlipidemia, endothelial function, cardiac rhythm, thrombosis, and others. As such, the cardioprotective effect of lowered inflammation may be counterbalanced by the effect of increased triglycerides or other unknown mechanisms, leading to a null association overall. The strengths of our study include its large size, high response rates, the representativeness of the sample of the Costa Rican population, and extensive information on genetic and dietary covariates including biomarker measures. Additionally, the information on blood lipids and inflammatory markers among controls was available to elucidate the mechanisms linking dietary intake of PUFAs, genetic variation in desaturases, and the risk of MI. Furthermore, the genetic score used to capture variation in desaturases has a distinct biological meaning, because its weights were derived from the observed relation between the selected SNPs and adipose tissue AA, the synthesis of which is highly regulated by the n-6 pathway. Finally, our gene-diet 33

43 interaction analyses estimate departure from additivity and thus represent true underlying biological interactions rather than statistical artifacts. The results of this study should be interpreted in light of several important limitations. First, the observational nature of the Costa Rica Study precludes from establishing any causal relations between the genetic and dietary exposures and the outcomes. Second, the selected SNPs may merely be in linkage disequilibrium with the true causal variant and not have any physiologic effects of their own. Third, although VCAM-1 is a valid marker of systemic inflammation, there is evidence that it is not associated with heart disease and thus may not be an appropriate intermediate cardiovascular risk factor (87). In conclusion, we established associations between a novel measure of genetic variation in the desaturase cluster, adipose tissue fatty acids, and intermediate cardiovascular risk factors such as serum triglycerides and systemic inflammation in the Costa Rica Study. Future studies investigating multiple mechanisms of PUFA action will be important to understanding the biological interplay between genetic variation in desaturases and cardiovascular health. 34

44 Table 2.1 General characteristics of the Costa Rica Study population. Variable Cases (n=878) Controls (n=878) Age, years 58.2± ±11.0 % Female Monthly household income, USD * 470± ±420 % Current smokers * Waist circumference, cm 90.8± ±9.9 Physical activity, MET 35.1± ±15.3 Adipose tissue fatty acids, % of total Alpha-linolenic * 0.64± ±0.21 Linoleic * 15.4± ±3.8 Minor allele frequency, % rs (C/T) rs (T/deletion) rs (C/T) rs (A/T) rs (G/C)

45 rs174627(c/t) Individual admixture, % European 57.5± ±8.2 Amerindian 38.4± ±7.6 West African 4.2± ±3.6 * P<0.05 for paired t-test (continuous variables), McNemar s test (for sex and smoking), or Fisher s exact test (for minor allele frequencies) 36

46 Table 2.2 Least square means * (+/- standard errors) of adipose fatty acids by quintile of genetic factor score among controls in the Costa Rica Study (n=1032 unless indicated otherwise). Q1 Q2 Q3 Q4 Q5 P for trend LA / / / / / GLA (n=983) / / / / / < ALA / / / / / EDA / / / / / DGA (n=1024) / / / / / ETA / / / / / < AA (n=1031) / / / / / < EPA / / / / / <

47 DHA / / / / / * Models adjusted for age/sex/residence (by matching), ancestry, and dietary fatty acids 38

48 Table 2.3 Lease square means * (+/- standard errors) of blood lipids and inflammatory markers by quintile of genetic factor score among controls in the Costa Rica Study (n=1026 unless indicated otherwise). Q1 Q2 Q3 Q4 Q5 P for trend VCAM-1 (n=838) / / / / / hscrp (n=836) 2.43+/ / / / / HDL Cholesterol (n=1021) / / / / / LDL Cholesterol (n=951) / / / / / Total Cholesterol / / / / / Triglycerides / / / / / * Models adjusted for age/sex/residence (by matching) and ancestry 39

49 Table 2.4 Association between genetic variation in the FADS cluster and the risk of myocardial infarction in the Costa Rica Study, adjusted for age, sex, residence, and ancestry (n=1756). Genetic Factor Quintile n OR (95% CI) (referent) (0.90, 1.62) (0.85, 1.55) (0.88, 1.57) (0.73, 1.34) P-value for trend=

50 Genetic Score Tertiles Table 2.5 Least square means (+/- standard errors) of LDL cholesterol by tertiles of the genetic factor score and adipose tissue ALA among controls in the Costa Rica Study. LDL cholesterol * (n=945) ALA Tertiles / / / / / / / / /-3.99 P for interaction 0.02 * Model adjusted for age/sex/residence (by matching), smoking, income, physical activity, waist circumference, intake of saturated fatty acids, adipose tissue trans 18:2 FAs, LA, and ancestry 41

51 CHAPTER 3 Fatty acid elongase gene variants, cardiovascular risk factors, and the risk of myocardial infarction in the Costa Rica Study 3.1 Background Elongases of very long chain fatty acids 2, 4, and 5, encoded by genes ELOVL2, ELOVL4, and ELOVL5 respectively, play a key role in the biosynthetic pathway of PUFAs (88, 89). As numerous epidemiologic and laboratory studies have demonstrated associations between PUFA metabolism and the risk of several complex diseases, it is likely that genetic variants in the ELOVL family that alter expression or efficiency of elongases have implications for chronic disease development (59). Because the ELOVL genes were cloned recently, the evidence for their role in human disease pathogenesis is extremely limited. A recent genome-wide association study demonstrated associations of SNPs in the ELOVL2 gene to plasma and erythrocyte concentrations of several n-3 and n-6 very long-chain PUFAs in two population-based cohorts (90). Another genome-wide scan identified ELOVL5 as a susceptibility gene for normal tension glaucoma (91), while other reports linked changes in expression of elongase 5 to increased risk of depressive disorders (92, 93). The mechanisms underlying the observed associations between elongases and disease risk are unclear but probably involve changes in conversion of precursor fatty acids, namely ALA and LA into very long-chain fatty acids (63, 91, 92). 42

52 To date, no epidemiologic studies to our knowledge have linked genetic variation in the ELOVL genes to the risk of cardiovascular outcomes. We hypothesize that polymorphisms in ELOVL2, ELOVL4, and ELOVL5 genes that affect enzyme expression or efficiency are associated with changes in very long-chain PUFA biosynthesis, and therefore with changes in intermediate cardiovascular risk factors and ultimately in the risk of nonfatal MI. Furthermore, our study aims to evaluate whether elongase polymorphisms modify the established associations of dietary precursor PUFAs with cardiovascular outcomes in a population characterized by low intake of n-3 fatty acids. 3.2 Methods Study Population The population of the Costa Rica Study, described in detail in prior publications, included 4548 Hispanics who resided in the Central Valley of Costa Rica between 1994 and 2004 (33-35). Cases of first nonfatal acute MI were ascertained by two independent cardiologists in the participating hospitals and deemed eligible if they met the World Health Organization criteria, survived hospitalization, were under 75 years of age on the day of their first MI, and able to answer the questionnaire (37). Eligible cases were matched by 5-year age group, sex, and area of residence to population controls, identified randomly using data from the National Census and Statistics Bureau of Costa Rica. After the cases were discharged from the hospital, all cases and controls received home visits, during which trained study workers collected lifestyle and medical history data, anthropometric measurements, and biological specimens. Participation was 98% for cases 43

53 and 88% for controls. The study population is appropriate for investigating genetic markers of disease due to its origin in a small number of founders and low rates of migration (70). The original sample size was 2274 cases and 2274 controls. Participants missing information on outcomes, exposure, or covariates were excluded from the analysis. Excluded participants did not differ (p>0.05) from the included participants for demographic (age, sex, and area of residence), dietary (LA and ALA) or genetic covariates (individual ancestral proportions), reducing the possibility of bias due to the complete case approach. Of the original sample, information on adipose tissue LA and ALA was obtained on 1858 cases and 1956 controls. Of these, ELOVL genotype information was available on 1057 cases and 1127 controls. For the analyses of adipose tissue fatty acids, inflammatory markers, and serum lipids, 8 additional participants were excluded due to missing information on ancestry. One additional participant was excluded from the analyses of adipose tissue fatty acids due to missing information on dietary exposures, yielding a final sample size of 1118 controls. For the MI analysis, 7 additional controls and 42 cases were excluded due to missing information on covariates such as smoking, physical activity, adipose tissue trans fats, waist circumference, or saturated fat intake, resulting a sample size of 1015 cases and 1111 controls. Additionally, cases were re-matched to controls on age, sex, and area of residence to preserve the gains in efficiency and validity due to matched study design; 132 controls and 36 cases were lost during the re-matching process, resulting in a final sample size of 979 controls and 979 cases. 44

54 All participants provided written informed consent. The study was approved by the Human Subjects Committee of the Harvard School of Public Health and the University of Costa Rica Measurement of Outcomes and Covariates Exposures were ascertained via adipose tissue biomarkers for the following fatty acids: 18:3n-3 (ALA), 18:2n-6 (LA), 18:3n-6 (GLA), 20:3n-3 (ETA), 20:2n-6 (EDA), 20:3n-6 (DGA), 20:4n-6 (AA), 20:5n-3 (EPA), 22:5n-3 (docosapentaenoic, DPA), 22:6n-3 (DHA), and 22:4n-6 (adrenic, ADA). Advantages of using adipose tissue biomarkers to characterize long-term nutritional intake include slow turnover, absence of recall bias, and lack of response to conditions of acute disease. (42) However, previous analyses of our data showed that adipose tissue concentrations of AA and DPA are poorly correlated with dietary intake due to high degree of metabolic regulation of both fatty acids; additionally, little is known regarding correlations of dietary intake with adipose tissue ETA, EDA, and ADA. (42) Therefore, we used adipose tissue concentrations of these fatty acids as metabolic markers representing the endogenous component explained by genetic variation, adjusted for the exogenous component (dietary intake) as measured by the previously validated FFQ. (41) Collection of all biological samples, analyses of adipose tissue fatty acids and serum inflammatory markers and lipids, and ancestry ascertainment were conducted as described in Section SNP Selection and Genotyping 45

55 Thirty-one (31) SNPs were identified in the ELOVL2, ELOVL4, and ELOVL5 genes using information from the HapMap Project ( and the National Center of Biotechnology Information ( To increase efficiency, 10 of 31 SNPs were selected as the tagging SNPs using linkage disequilibrium block information obtained from HaploView software. (Supplemental Figure 3.1) The tagging variants were selected so that even if they do not have a functional role, they would be in linkage disequilibrium with the true causative polymorphism and thus serve as its efficient surrogates.(82) Additionally, SNPs were removed from the analysis if the frequency of the minor allele in the study population was under 10%, leaving 8 SNPs: rs (ELOVL2), rs (ELOVL2), rs (ELOVL2), rs (ELOVL4), rs (ELOVL5), rs (ELOVL5), rs (ELOVL5) and rs (ELOVL5). Genotyping was performed as described in Section Statistical Analysis Data from the Costa Rica Study were analyzed using the SAS software package (Version 9.2; SAS Institute Inc, Cary, NC). To assess the significance of differences in general characteristics and potential confounders, we used paired t-tests for continuous variables, McNemar s tests for categorical variables, and Fisher s exact test for minor allele frequencies. The ALLELE procedure was used to test for deviations from Hardy- Weinberg equilibrium among controls. Of all SNPs, one (rs ) was found to be in violation of the Hardy-Weinberg equilibrium and thus removed from all subsequent analyses. 46

56 Linear regression models, adjusted for dietary and demographic covariates, were fit among controls to evaluate the association between the ELOVL SNPs and the adipose tissue, plasma, and erythrocyte concentrations of very long chain PUFAs as well as on plasma concentrations of hscrp, VCAM-1, and serum lipids. Least square means and 95% confidence intervals were used to report the relation between the outcomes and ELOVL genetic variants. Log-transformations were carried out for non-normally distributed variables (GLA, hscrp, and triglycerides) and geometric means were reported. The intermediate risk factors models were adjusted for age, sex, and residence area (by matching) and ancestry, while the PUFA models were additionally adjusted for dietary intake of fatty acids. The relation between ELOVL SNPs and the MI outcome was modeled using conditional logistic regression, adjusted for age, sex, residence area (by matching), and ancestry; both additive and codominant models were considered. To address the problem of multiple comparisons, the significance level was adjusted using the false discovery rate approach, and corrected p-values were reported in addition to nominal p-values. (94, 95) Finally, departures from additivity were considered for SNPs rs and rs761179, and the precursor fatty acids (ALA and LA). For the outcomes that showed a robust statistically significant relationship with both fatty acids and SNPs (total and LDL cholesterol), interaction terms were added to the linear regression models, which were further adjusted for dietary and demographic confounders. Again, participants were excluded from the analyses if they were missing information on exposures, outcomes, or any of the covariates, with the final sample size of 1104 for total cholesterol and 1026 for LDL cholesterol. Homogeneity across genotypes was assessed using partial F-tests. 47

57 3.3 Results The general characteristics of the study population by case/control status are presented in Table 3.1. Cases were significantly more likely to report MI risk factors such as lower household income, smoking, and lower physical activity. Cases also had significantly lower adipose tissue concentrations of ALA and LA. None of the selected SNPs or individual ancestral proportions differed significantly by disease status. In multivariate-adjusted models corrected for multiple comparisons, none of the adipose tissue PUFAs were significantly associated with the number of minor allele copies in 7 ELOVL cluster SNPs (Table 3.2). Similarly, serum inflammatory markers (VCAM-1 and hscrp), HDL cholesterol, and triglycerides did not vary significantly by ELOVL genotypes (Table 3.3). However, LDL- and total cholesterol showed robust linear decreases as the number of copies of the minor alleles in two ELOVL5 SNPs increased (multiple comparison adjusted p-values= and 0.01 respectively for both rs and rs761179) (Table 3.3). In both additive and codominant models, the risk of first nonfatal myocardial infarction was not significantly associated with genetic variation in elongases (Table 3.4). Additive interactions between the precursor fatty acids (ALA and LA) and SNPs rs and rs were evaluated for the LDL- and total cholesterol outcomes. Our data present no evidence that these ELOVL5 polymorphisms modify the relation between dietary intake of ALA or LA and serum LDL- and total cholesterol (data not shown). 48

58 3.4 Discussion Findings from this study suggested a novel association of SNPs in the ELOVL5 gene with serum LDL- and total cholesterol in a large population-based case-control study of nonfatal MI. Additionally, this study demonstrated null associations of ELOVL polymorphisms with adipose tissue PUFAs, selected markers of systemic inflammation, HDL cholesterol and triglycerides, and nonfatal MI. We also evaluated whether the rs and rs ELOVL5 polymorphisms modified the relations between precursor PUFAs and LDL- and total cholesterol, and found no evidence in support of such gene-diet interactions. The null fatty acid findings from our study are in contrast to the genome-wide scan conducted by Tanaka et al., which reported an association between an ELOVL2 polymorphism and an increase in plasma EPA in a European population, as well as with an increase in DPA and a decrease in DHA in a replication cohort of American adults. (63) Although the differences in PUFA ascertainment (adipose tissue versus plasma and erythrocytes) or elongase genes (ELOVL5 versus ELOVL2) may be relevant to this discrepancy, the lack of statistical power in the Costa Rica Study is also a possible explanation. Consistent with the results reported by Tanaka et al., we observed increases in adipose tissue EPA as the number of minor allele copies increased for both rs and rs761179; however, the observed increases were not statistically significant once adjusted for multiple comparisons. The mechanisms underlying the observed associations between the rs and rs polymorphisms and serum cholesterol are likely to involve increased elongase 5 transcription, which could affect the conversion of n-3 and n-6 precursor fatty acids 49

59 into EPA and AA respectively. The hypolipidemic effect of very long-chain PUFAs has been well-documented in epidemiologic studies. (96) Specifically, very long chain n-3 PUFAs lower the rate of LDL cholesterol synthesis, as well as reduce the size and increase the density of lipoproteins, thus diminishing their atherogenicity. (97) Because total cholesterol concentrations in serum are highly correlated with LDL cholesterol, a decrease in LDL synthesis also explains the observed decline in total cholesterol. Despite robust decreases in LDL- and total cholesterol, genetic variation in ELOVL5 was not associated with a decrease in MI risk in our study population. Because of the variety of physiologic pathways relevant to PUFAs, the overall effect of elongase polymorphisms on cardiovascular health involves a combination of mechanisms including but not limited to inflammation, changes in serum lipids and/or blood pressure, endothelial function, cardiac rhythm, and thrombosis. As such, it is possible that the cardioprotective effects of reduced serum LDL cholesterol are counterbalanced by yet unknown negative effects of elongase polymorphisms, leading to a null association overall. It is also possible that in the MI analyses, the true effect size was too small to detect given the statistical power of our study. To our knowledge, this is the first study to examine the relationship between elongase polymorphisms and cardiovascular disease outcomes, including inflammation, serum lipids, and myocardial infarction. In addition to the novelty of the question, the strengths of our study include its large size, high response rates, the representativeness of the sample of the Costa Rican population, and extensive information on genetic and dietary covariates including biomarker measures. Furthermore, we have addressed the multiple testing problem using the false discovery rate method; it is noteworthy that the 50

60 LDL-cholesterol results were robust to even more conservative forms of adjustment for multiple comparisons such as the Bonferroni correction. The results of this study should be interpreted in light of several important limitations. First, the reported association between ELOVL5 polymorphisms and LDLand total cholesterol needs to be replicated in an independent population to establish its validity. Second, the observational nature of the Costa Rica Study precludes from establishing any causal relations between the genetic and dietary exposures and the outcomes. Third, the selected SNPs may merely be in linkage disequilibrium with the true causal variant and not have any physiologic effects of their own. In conclusion, we found that an increase in the number of copies of the minor allele in two ELOVL5 polymorphisms (rs and rs761179) is associated with decreased serum LDL- and total cholesterol, but the observed improvement in the lipid profile does not translate into an attenuated risk of MI in our study population. Future studies are warranted to replicate the observed associations as well as to elucidate the biological mechanisms underlying the effect of genetic variation in elongases on cardiovascular health. 51

61 Table 3.1 General characteristics of the Costa Rica Study population. Variable Cases (n=979) Controls (n=979) Age, years 58.4± ±11.0 % Female Monthly household income, USD * 474± ±420 % Current smokers * Waist circumference, cm 90.8± ±9.7 Physical activity, MET * 35.1± ±16.1 Adipose tissue fatty acids, % of total Alpha-linolenic * 0.64± ±0.22 Linoleic * 15.4± ±3.9 Minor allele frequency, % rs (A/G) rs (A/G) rs (C/G) rs (C/T) rs (A/C)

62 rs (A/C) rs (C/T) Individual admixture, % European 57.6± ±8.2 Amerindian 38.2± ±7.6 West African 4.2± ±3.6 * P<0.05 for paired t-test (continuous variables), McNemar s test (for sex and smoking), or Fisher s exact test (for minor allele frequencies) 53

63 Table 3.2 Least square means * (+/- standard errors) of adipose fatty acids by genotype among controls in the Costa Rica Study (n=1118 unless otherwise specified). rs (ELOVL2) AA AG GG Nominal P value Corrected P value ALA 0.65± ± ± LA 15.79± ± ± GLA 0.060± ± ± (n=1067) EDA 0.213± ± ± ETA ± ± ± DGA 0.300± ± ± (n=1105) AA 0.47± ± ± (n=1117) EPA 0.043± ± ± DPA 0.178± ± ± DHA 0.134± ± ± ADA 0.197± ± ± rs (ELOVL2) AA AG GG Nominal P value Corrected P value ALA 0.65± ± ±

64 LA 15.82± ± ± GLA 0.063± ± ± (n=1067) EDA 0.219± ± ± ETA ± ± ± DGA 0.313± ± ± (n=1105) AA 0.48± ± ± (n=1117) EPA 0.043± ± ± DPA 0.184± ± ± DHA 0.139± ± ± ADA 0.201± ± ± rs (ELOVL4) CC CG GG Nominal P value Corrected P value ALA 0.70± ± ± LA 17.40± ± ± GLA 0.047± ± ± (n=1067) EDA 0.241± ± ± ETA ± ± ± DGA 0.314± ± ±

65 (n=1105) AA 0.44± ± ± (n=1117) EPA 0.040± ± ± DPA 0.182± ± ± DHA 0.130± ± ± ADA 0.197± ± ± rs (ELOVL5) CC CT TT Nominal P value Corrected P value ALA 0.67± ± ± LA 15.89± ± ± GLA 0.065± ± ± (n=1067) EDA 0.229± ± ± ETA ± ± ± DGA 0.329± ± ± (n=1105) AA 0.49± ± ± (n=1117) EPA 0.045± ± ± DPA 0.195± ± ± DHA 0.146± ± ±

66 ADA 0.212± ± ± rs (ELOVL5) AA AC CC Nominal P value Corrected P value ALA 0.65± ± ± LA 15.71± ± ± GLA 0.062± ± ± (n=1067) EDA 0.218± ± ± ETA ± ± ± DGA 0.315± ± ± (n=1105) AA 0.48± ± ± (n=1117) EPA 0.043± ± ± DPA 0.185± ± ± DHA 0.140± ± ± ADA 0.200± ± ± rs (ELOVL5) AA AC CC Nominal P value Corrected P value ALA 0.65± ± ± LA 15.73± ± ±

67 GLA 0.061± ± ± (n=1067) EDA 0.223± ± ± ETA ± ± ± DGA 0.318± ± ± (n=1105) AA 0.48± ± ± (n=1117) EPA 0.041± ± ± DPA 0.182± ± ± DHA 0.137± ± ± ADA 0.198± ± ± rs (ELOVL5) CC CT TT Nominal P value Corrected P value ALA 0.62± ± ± LA 15.29± ± ± GLA 0.059± ± ± (n=1067) EDA 0.224± ± ± ETA ± ± ± DGA 0.308± ± ± (n=1105) 58

68 AA 0.47± ± ± (n=1117) EPA 0.040± ± ± DPA 0.184± ± ± DHA 0.144± ± ± ADA 0.198± ± ± * Models adjusted for age/sex/residence (by matching), ancestry, and dietary fatty acids 59

69 Table 3.3 Least square means * (+/- standard errors) of serum lipids and inflammatory markers by genotype among controls in the Costa Rica Study (n=1112 unless otherwise specified). rs (ELOVL2) AA AG GG Nominal P value Corrected P value VCAM ± ± ± (n=916) hscrp 2.52± ± ± (n=913) HDL-Chol ± ± ± (n=1104) LDL-Chol ± ± ± (n=1034) Total Chol ± ± ± TG ** ± ± ± rs (ELOVL2) AA AG GG Nominal P value Corrected P value VCAM ± ± ± (n=916) hscrp 2.26± ± ± (n=913) 60

70 HDL-Chol ± ± ± (n=1104) LDL-Chol ± ± ± (n=1034) Total Chol ± ± ± TG ** ± ± ± rs (ELOVL4) CC CG GG Nominal P value Corrected P value VCAM ± ± ± (n=916) hscrp 2.86± ± ± (n=913) HDL-Chol ± ± ± (n=1104) LDL-Chol ± ± ± (n=1034) Total Chol ± ± ± TG ** ± ± ± rs (ELOVL5) CC CT TT Nominal P value Corrected P value VCAM ± ± ±

71 (n=916) hscrp 2.09± ± ± (n=913) HDL-Chol ± ± ± (n=1104) LDL-Chol ± ± ± (n=1034) Total Chol ± ± ± TG ** ± ± ± rs (ELOVL5) AA AC CC Nominal P value Corrected P value VCAM ± ± ± (n=916) hscrp 2.32± ± ± (n=913) HDL-Chol ± ± ± (n=1104) LDL-Chol ± ± ± (n=1034) Total Chol ± ± ± TG ** ± ± ± rs (ELOVL5) 62

72 AA AC CC Nominal P value Corrected P value VCAM ± ± ± (n=916) hscrp 2.31± ± ± (n=913) HDL-Chol ± ± ± (n=1104) LDL-Chol ± ± ± (n=1034) Total Chol ± ± ± TG ** ± ± ± rs (ELOVL5) CC CT TT Nominal P value Corrected P value VCAM ± ± ± (n=916) hscrp 2.74± ± ± (n=913) HDL-Chol ± ± ± (n=1104) LDL-Chol ± ± ± (n=1034) 63

73 Total Chol ± ± ± TG ** ± ± ± * Models adjusted for age/sex/residence (by matching) and ancestry ** Triglycerides 64

74 Table 3.4 The risk of nonfatal myocardial infarction by genotype in the Costa Rica Study (n=1958). SNP OR * (95% CI) Nominal P value Corrected P value rs (ELOVL2) Additive 0.99 (0.86, 1.15) Codominant (AG vs AA) 0.94 (0.77, 1.13) Codominant (GG vs AA) 1.08 (0.76, 1.55) rs (ELOVL2) Additive 1.06 (0.91, 1.24) Codominant (AG vs AA) 1.07 (0.89, 1.30) Codominant (GG vs AA) 1.12 (0.72, 1.74) rs (ELOVL4) Additive 0.98 (0.80, 1.20) Codominant (CG vs CC) 0.79 (0.35, 1.79) Codominant (GG vs CC) 0.81 (0.37, 1.78) rs (ELOVL5) Additive 0.82 (0.71, 0.96)

75 Codominant (CT vs CC) 0.98 (0.65, 1.48) Codominant (TT vs CC) 0.76 (0.51, 1.14) rs (ELOVL5) Additive 0.93 (0.82, 1.06) Codominant (AC vs AA) 0.98 (0.80, 1.20) Codominant (CC vs AA) 0.87 (0.68, 1.11) rs (ELOVL5) Additive 0.91 (0.80, 1.03) Codominant (AC vs AA) 0.90 (0.72, 1.13) Codominant (CC vs AA) 0.83 (0.65, 1.06) rs (ELOVL5) Additive 0.96 (0.83, 1.11) Codominant (CT vs CC) 0.82 (0.57, 1.18) Codominant (TT vs CC) 0.83 (0.58, 1.19) * Models adjusted for age/sex/residence (by matching) and ancestry 66

76 BIBLIOGRAPHY 1. Bonow RO, Smaha LA, Smith SC, Mensah GA, L'Enfant C. World Heart Day The international burden of cardiovascular disease: responding to the emerging global epidemic. Circulation 2002;106: McKee M, Leon D. Global epidemic of cardiovascular disease. Lancet 1999;353: Abegunde DO, Mathers CD, Adam T, Ortegon M, Strong K. Chronic diseases 1 - The burden and costs of chronic diseases in low-income and middle-income countries. Lancet 2007;370: Husten L. Global epidemic of cardiovascular disease predicted. Lancet 1998;352: Miranda JJ, Kinra S, Casas JP, Smith GD, Ebrahim S. Non-communicable diseases in low- and middle-income countries: context, determinants and health policy. Trop Med Int Health 2008;13: Murray CJL, Lopez AD. Global and regional cause-of-death patterns in Bull World Health Organ 1994;72: Reddy KS, Yusuf S. Emerging epidemic of cardiovascular disease in developing countries. Circulation 1998;97: Mendis S, Yach D, Bengoa R, Narvaez D, Zhang X. Research gap in cardiovascular disease in developing countries. Lancet 2003;361: Talmud PJ. Gene-environment interaction and its impact on coronary heart disease risk. Nutr Metab Cardiovasc Dis 2007;17:

77 10. Tonstad S. Do genetic determinates of ischaemic heart disease imply that prevention may become more difficult? J Inter Med 1998;243: Ordovas JM, Tai ES. Why study gene-environment interactions? Curr Opin Lipidol 2008;19: Williams CM, Ordovas JM, Lairon D, Hesketh J, Lietz G, Gibney M, et al. The challenges for molecular nutrition research 1: linking genotype to healthy nutrition. Genes Nutr 2008;3: Corella D, Ordovas JM. Single nucleotide polymorphisms that influence lipid metabolism: Interaction with dietary factors. Ann Rev Nutr 2005;25: Ruiz-Narvaez EA, Yang YD, Nakanishi Y, Kirchdorfer J, Campos H. APOC3/A5 haplotypes, lipid levels, and risk of myocardial infarction in the Central Valley of Costa Rica. J Lipid Res 2005;46: Yang YD, Ruiz-Narvaez E, Kraft P, Campos H. Effect of apolipoprotein E genotype and saturated fat intake on plasma lipids and myocardial infarction in the Central Valley of Costa Rica. Hum Biol 2007;79: Guallar E, Aro A, Jimenez FJ, Martin-Moreno JM, Salminen I, van't Veer P, et al. Omega-3 fatty acids in adipose tissue and risk of myocardial infarction - The EURAMIC study. Arterioscl Throm Vasc Biol 1999;19: Oomen CM, Ocke MC, Feskens EJM, Kok FJ, Kromhout D. -Linolenic acid intake is not beneficially associated with 10-y risk of coronary artery disease incidence: the Zutphen Elderly Study. Am J Clin Nutr 2001;74:

78 18. Pedersen JI, Ringstad J, Almendingen K, Haugen TS, Stensvold I, Thelle DS. Adipose tissue fatty acids and risk of myocardial infarction - a case-control study. Eur J Clin Nutr 2000;54: Pietinen P, Ascherio A, Korhonen P, Hartman AM, Willett WC, Albanes D, et al. Intake of fatty acids and risk of coronary heart disease in a cohort of Finnish men - The alpha-tocopherol, beta-carotene cancer prevention study. Am J Epidemiol 1997;145: Campos H, Baylin A, Willett WC. -linolenic acid and risk of nonfatal acute myocardial infarction. Circulation 2008;118: Brindle P, Beswick A, Fahey T, Ebrahim S. Accuracy and impact of risk assessment in the primary prevention of cardiovascular disease: a systematic review. Heart 2006;92: Eichler K, Puhan MA, Steurer J, Bachmann L. Prediction of first coronary events with the Framingham score: A systematic review. Am Heart J 2007;153: Tzoulaki I, Liberopoulos G, Ioannidis JP. Assessment of claims of improved prediction beyond the Framingham risk score. JAMA 2009;302: Wilson PW. Challenges to improve coronary heart disease risk assessment. JAMA 2009;302: World Health Organization. Prevention of Cardiovascular Disease: Guidelines for Assessment and Management of Cardiovascular Risk. Geneva: World Health Organization; Arbogast PG, Kaltenbach L, Ding H, Ray WA. Adjustment for multiple cardiovascular risk factors using a summary risk score. Epidemiology 2008;19:

79 27. Grewal J, Chan S, Frohlich J, Mancini GB. Assessment of novel risk factors in patients at low risk for cardiovascular events based on Framingham risk stratification. Clin Invest Med 2003;26: D'Agostino RB, Grundy S, Sullivan LM, Wilson P. Validation of the Framingham Coronary Heart Disease prediction scores - Results of a multiple ethnic groups investigation. JAMA 2001;286: Wilson PWF, D'Agostino RB, Levy D, Belanger AM, Silbershatz H, Kannel WB. Prediction of coronary heart disease using risk factor categories. Circulation 1998;97: Conroy RM, Pyorala K, Fitzgerald AP, Sans S, Menotti A, De Backer G, et al. Estimation of ten-year risk of fatal cardiovascular disease in Europe: the SCORE project. Eur Heart J 2003;24: Abegunde DO, Shengelia B, Luyten A, Cameron A, Celletti F, Nishtar S, Pandurangi V, Mendis S. Can non-physician health-care workers assess and manage cardiovascular risk in primary care? Bull World Health Organ 2007;85: Iqbal R, Anand S, Ounpuu S, Islam S, Zhang X, Rangajaran S, Chifamba J, Al- Hinai A, Keltai M, et al. Dietary patterns and the risk of acute myocardial infarction in 52 countries: results of the INTERHEART study. Circulation 2008;118(19) Kabagambe EK, Baylin A, Ascherio A, Campos H. The type of oil used for cooking is associated with the risk of nonfatal acute myocardial infarction in Costa Rica. J Nutr 2005;135:

80 34. Kabagambe EK, Baylin A, Campos H. Nonfatal acute myocardial infarction in Costa Rica - Modifiable risk factors, population-attributable risks, and adherence to dietary guidelines. Circulation 2007;115: Kabagambe EK, Baylin A, Ruiz-Narvaez E, Rimm EB, Campos H. Alcohol intake, drinking patterns, and risk of nonfatal acute myocardial infarction in Costa Rica. Am J Clin Nutr 2005;82: Kabagambe EK, Furtado J, Baylin A, Campos H. Some dietary and adipose tissue carotenoids are associated with the risk of nonfatal acute myocardial infarction in Costa Rica. J Nutr 2005;135: Tunstallpedoe H, Kuulasmaa K, Amouyel P, Arveiler D, Rajakangas AM, Pajak A. Myocardial infarction and coronary deaths in the World Health Organization MONICA Project - Registration procedures, event rates, and case fatality rates in 38 populations from 21 countries in 4 continents. Circulation 1994;90: Campos H, Siles X. Siesta and the risk of coronary heart disease: results from a population-based, case-control study in Costa Rica. Int J Epidemiol 2000;29: Campos H, Bailey SM, Gussak LS, Siles X, Ordovas JM, Schaefer EJ. Relationships of body habitus, fitness level, and cardiovascular risk factors including lipoproteins and apolipoproteins in a rural and urban Costa Rican population. Arterioscl Throm 1991;11: Campos H, Mata L, Siles X, Vives M, Ordovas JM, Schaefer EJ. Prevalence of cardiovascular risk factors in rural and urban Puriscal, Costa Rica. Circulation 1992;85:

81 41. Kabagambe EK, Baylin A, Allan DA, Siles X, Spiegelman D, Campos H. Application of the method of triads to evaluate the performance of food frequency questionnaires and biomarkers as indicators of long-term dietary intake. Am J Epidemiol 2001;154: Baylin A, Kabagambe EK, Siles X, Campos H. Adipose tissue biomarkers of fatty acid intake. Am J Clin Nutr 2002;76: World Health Organization. Introduction. In: Diet, Nutrition and the Prevention of Chronic Diseases. World Health Organization; 2003: World Health Organization. Folate and folic acid. In: Vitamin and Mineral Requirements in Human Nutrition. World Health Organization and Food and Agriculture Organization; Stampfer MJ, Hu FB, Manson JE, Rimm EB, Willett WC. Primary prevention of coronary heart disease in women through diet and lifestyle. N Engl J Med 2000;343: Bernstein MS, Morabia A, Sloutskis D. Definition and prevalence of sedentarism in an urban population. Am J Public Health 1999;89: Alberti K, Zimmet PZ. Definition, diagnosis and classification of diabetes mellitus and its complications, part 1. Diagnosis and classification of diabetes mellitus - Provisional report of a WHO consultation. Diab Med 1998;15: Cifras Basicas sobre Pobreza e Ingresos. San Jose, Costa Rica: El Instituto Nacional de Estadistica y Censos, Panorama social [Social panorama]. Santiago, Chile: Comision Economica de las Naciones Unidas para America Latina (CEPAL),

82 50. Colon-Ramos U, Kabagambe EK, Baylin A, Ascherio A, Campos H, Peterson KE. Socio-economic status and health awareness are associated with choice of cooking oil in Costa Rica. Pub Health Nutr 2007;10: Hippisley-Cox J, Coupland C, Vinogradova Y, Robson J, May M, Brindle P. Derivation and validation of QRISK, a new cardiovascular disease risk score for the United Kingdom: prospective open cohort study. Br Med J 2007;335: Akaike H. New look at statistical model identification. IEEE Transactions on Automatic Control 1974;19: Wang TJ, Gona P, Larson MG, Tofler GH, Levy D, Newton-Cheh C, Jacques PF, Rifai N, Selhub J, et al. Multiple biomarkers for the prediction of first major cardiovascular events and death. N Engl J Med 2006;355: Shilton T, Bauman A, Bull F, Sarmiento O. Effectiveness and Challenges for Promoting Physical Activity Globally. New York: Springer; Lichtman SW, Pisarska K, Berman ER, Pestone M, Dowling H, Offenbacher E, Weisel H, Heshka S, Matthews DE, et al. Discrepancy between self-reported and actual caloric intake and exercise in obese subjects. N Engl J Med 1992;327: World Health Organization. WHOSIS: WHO Statistical Information System. ( (Accessed January 7, 2010). 57. Medina-Lezama J, Morey-Vargas OL, Zea-Diaz H, et al. Prevalence of lifestylerelated cardiovascular risk factors in Peru: the PREVENCION study. Pan Am J Public Health 2008;24: Harris W. Omega-6 and omega-3 fatty acids: partners in prevention. Curr Opin Clin Nutr 2010;13:

83 59. Lecerf JM. Fatty acids and cardiovascular disease. Nutr Rev 2009;67: Baylin A, Kabagambe EK, Ascherio A, Spiegelman D, Campos H. Adipose tissue -linolenic acid and nonfatal acute myocardial infarction in Costa Rica. Circulation 2003;107: Lattka E, Illig T, Heinrich J, Koletzko B. FADS gene cluster polymorphisms: important modulators of fatty acid levels and their impact on atopic diseases. J Nutrigenet Nutrigenomics 2009;2: Marquardt A, Stohr H, White K, Weber BH. cdna cloning, genomic structure, and chromosomal localization of three members of the human fatty acid desaturase family. Genomics 2000;66: Tanaka T, Shen J, Abecasis GR, Kisialiou A, Ordovas JM, Guralnik JM, et al. Genome-wide association study of plasma polyunsaturated fatty acids in the InCHIANTI study. PLoS Genet 2009;5: Malerba G, Schaeffer L, Xumerle L, Klopp N, Trabetti E, Biscuola M, et al. SNPs of the FADS gene cluster are associated with polyunsaturated fatty acids in a cohort of patients with cardiovascular disease. Lipids 2008;43: Martinelli N, Girelli D, Malerba G, Guarini P, Illig T, Trabetti E, et al. FADS genotypes and desaturase activity estimated by the ratio of arachidonic acid to linoleic acid are associated with inflammation and coronary artery disease. Am J Clin Nutr 2008;88: Rzehak P, Heinrich J, Klopp N, Schaeffer L, Hoff S, Wolfram G, et al. Evidence for an association between genetic variants of the fatty acid desaturase 1 fatty acid 74

84 desaturase 2 (FADS1 FADS2) gene cluster and the fatty acid composition of erythrocyte membranes. Br J Nutr 2009;101: Schaeffer L, Gohlke H, Muller M, Heid IM, Palmer LJ, Kompauer I, et al. Common genetic variants of the FADS1 FADS2 gene cluster and their reconstructed haplotypes are associated with the fatty acid composition in phospholipids. Hum Mol Gen 2006;15: Xie L, Innis SM. Genetic variants of the FADS1 FADS2 gene cluster are associated with altered n-6 and n-3 essential fatty acids in plasma and erythrocyte phospholipids in women during pregnancy and in breast milk during lactation. J Nutr 2008;138: Koletzko B, Lattka E, Zeilinger S, Illig T, Steer C. Genetic variants of the fatty acid desaturase gene cluster predict amounts of red blood cell docosahexaenoic and other polyunsaturated fatty acids in pregnant women: findings from the Avon Longitudinal Study of Parents and Children. Am J Clin Nutr 2010;93: Baylin A, Ruiz-Narvaez E, Kraft P, Campos H. alpha-linolenic acid, Delta(6)- desaturase gene polymorphism, and the risk of nonfatal myocardial infarction. Am J Clin Nutr 2007;85: Sabatti C, Service SK, Hartikainen A-L, Pouta A, Ripatti S, Brodsky J, et al. Genome-wide association analysis of metabolic traits in a birth cohort from a founder population. Nat Genet 2009;41: Aulchenko YS, Ripatti S, Lindqvist I, Boomsma D, Heid IM, Pramstaller PP, et al. Loci influencing lipid levels and coronary heart disease risk in 16 European population cohorts. Nat Genet 2009;41:

85 73. Kathiresan S, Willer CJ, Peloso GM, Demissie S, Musunuru K, Schadt EE, et al. Common variants at 30 loci contribute to polygenic dyslipidemia. Nat Genet 2009;41: Zabaneh D, Balding DJ. A genome-wide association study of the metabolic syndrome in Indian Asian men. PLoS ONE 2010;5(8):e Beynen A, Katan M. Rapid sampling and long-term storage of subcutaneous adipose-tissue biopsies for determination of fatty acid composition. Am J Clin Nutr 1985;42: Truong H, DiBello JR, Ruiz-Narvaez E, Kraft P, Campos H, Baylin A. Does genetic variation in the Δ6-desaturase promoter modify the association between α- linolenic acid and the prevalence of metabolic syndrome? Am J Clin Nutr 2009;89: Williams ES, Baylin A, Campos H. Adipose tissue arachidonic acid and the metabolic syndrome in Costa Rican adults. Clin Nutr 2007;26: Blankenberg S, Rupprecht HJ, Bickel C, Peetz D, Hafner G, Tiret L, et al. Circulating cell adhesion molecules and death in patients with coronary artery disease. Circulation 2001;104: Koenig W, Sund M, Frohlich M, Fischer HG, Lowel H, Doring A, et al. C- reactive protein, a sensitive marker of inflammation, predicts future risk of coronary heart disease in initially healthy middle-aged men - Results from the MONICA (Monitoring Trends and Determinants in Cardiovascular Disease) Augsburg Cohort Study, 1984 to Circulation 1999;99:

86 80. Ridker PM, Hennekens CH, Buring JE, Rifai N. C-reactive protein and other markers of inflammation in the prediction of cardiovascular disease in women. N Engl J Med 2000;342: Ruiz-Narváez E, Bare L, Arellano A, Catanese J, Campos H. West African and Amerindian ancestry and risk of myocardial infarction and metabolic syndrome in the Central Valley population of Costa Rica. Hum Genet 2010;127: Johnson GCL, Esposito L, Barratt BJ, Smith AN, Heward J, Di Genova G, et al. Haplotype tagging for the identification of common disease genes. Nat Genet 2001;29: Truong H, DiBello JR, Ruiz-Narvaez E, Kraft P, Campos H, Baylin A. Does genetic variation in the Delta(6)-desaturase promoter modify the association between alpha-linolenic acid and the prevalence of metabolic syndrome? Am J Clin Nutr 2009;89: Skulas-Ray AC, Kris-Etherton PM, Harris WS, Vanden Heuvel JP, Wagner PR, West SG. Dose-response effects of omega-3 fatty acids on triglycerides, inflammation, and endothelial function in healthy persons with moderate hypertriglyceridemia. Am J Clin Nutr 2011;93: Makhoul Z, Kristal AR, Gulati R, Luick B, Bersamin A, Boyer B, et al. Associations of very high intakes of eicosapentaenoic and docosahexaenoic acids with biomarkers of chronic disease risk among Yup'ik Eskimos. Am J Clin Nutr 2010;91(3): Wall R, Ross RP, Fitzgerald GF, Stanton C. Fatty acids from fish: the antiinflammatory potential of long-chain omega-3 fatty acids. Nutr Rev 2010;68:

87 87. de Lemos JA, Hennekens CH, Ridker PM. Plasma concentration of soluble vascular cell adhesion molecule-1 and subsequent cardiovascular risk. J Am Coll Cardiol 2000;36: Jakobsson A, Westerberg R, Jacobsson A. Fatty acid elongases in mammals: their regulation and roles in metabolism. Prog Lipid Res 2006;45: Matsuzaka T, Shimano H. Elovl6: a new player in fatty acid metabolism and insulin sensitivity. J Mol Med. 2009;87: Tanaka T, Shen J, Abecasis GR, Kisialiou A, Ordovas JM, Guralnik JM, et al. Genome-wide association study of plasma polyunsaturated fatty acids in the InCHIANTI study. PLoS Genet 2009;5: Writing Committee for the Normal Tension Glaucoma Genetic Study Group of Japan Glaucoma Society, Meguro A, Inoko H, Ota M, Mizuki N, Bahram S. Genomewide association study of normal tension glaucoma: common variants in SRBD1 and ELOVL5 contribute to disease susceptibility. Ophthalmology 2010;117: e McNamara RK, Liu Y. Reduced expression of fatty acid biosynthesis genes in the prefrontal cortex of patients with major depressive disorder. J Affect Disord 2011;129: Mamalakis G, Kiriakakis M, Tsibinos G, Kafatos A. Depression and adipose polyunsaturated fatty acids in an adolescent group. Prostaglandins Leukot Essent Fatty Acids 2004;71: Benjamini Y, Hochberg Y. Controlling the false discovery rate-- a practical and powerful approach to multiple testing. J Royal Stat Soc, Ser B 1995;57:

88 95. Benjamini Y, Drai D, Elmer G, Kafkafi N, Golani I. Controlling the false discovery rate in behavior genetics research. Behav Brain Res 2001;125: Lopez-Huertas E. Health effects of oleic acid and long chain omega-3 fatty acids (EPA and DHA) enriched milks: a review of intervention studies. Pharmacol Res 2010;61: Leaf A, Weber PC. Cardiovascular effects of n-3 fatty acids. New Engl J Med 1988;318:

89 SUPPLEMENTAL MATERIALS Supplemental Table 1.1 Distribution of cardiovascular risk factors in the healthy population of the Costa Rica Study (n= 1386). Factor n Crude OR of MI (95% CI) Multivariate-adjusted * OR of MI (95% CI) Dietary score Quintile (referent) 1.00 (referent) Quintile (0.68, 1.39) 0.95 (0.64, 1.41) Quintile (0.45, 0.85) 0.68 (0.48, 0.97) Quintile (0.46, 0.97) 0.78 (0.52, 1.19) Quintile (0.38, 0.80) 0.59 (0.39, 0.90) Physical activity Quintile (referent) 1.00 (referent) Quintile (0.54, 1.08) 0.81 (0.55, 1.18) Quintile (0.52, 1.05) 0.71 (0.48, 1.05) Quintile (0.52, 1.05) 0.82 (0.56, 1.21) 80

90 Quintile (0.56, 1.15) 0.63 (0.42, 0.95) Socioeconomic status index Quintile (referent) 1.00 (referent) Quintile (0.39, 0.79) 0.60 (0.41, 0.88) Quintile (0.47, 1.00) 0.72 (0.48, 1.09) Quintile (0.38, 0.79) 0.60 (0.40, 0.89) Quintile (0.26, 0.60) 0.45 (0.28, 0.71) Waist-to-hip ratio 0.90 for men, (referent) 1.00 (referent) for women >0.90 for men, > (1.61, 3.32) 2.35 (1.59, 3.47) for women Smoking Current non-smoker (referent) 1.00 (referent) Current smoker (2.51, 4.17) 3.17 (2.43, 4.13) Alcohol consumption, g/day 81

91 (referent) 1.00 (referent) (0.56, 1.13) 0.77 (0.52, 1.14) (0.48, 1.06) 0.60 (0.39, 0.93) > (0.81, 1.40) 0.75 (0.55, 1.03) * Adjusted for age/sex/area of residence (by matching), as well as for all other cardiovascular risk factors as listed in the table 82

92 Supplemental Table 1.2 Comparison of two cardiovascular risk score definitions by component. Score 1 Score 2 Diet Trans fats Saturated fats Polyunsaturated fats Cholesterol Fiber Folate Other components Categorical by quintile of daily intake * (FFQ) Categorical by quintile of daily intake * (FFQ) Categorical by quintile of daily intake * (FFQ) Categorical by quintile of daily intake * (FFQ) Categorical by quintile of daily intake * (FFQ) Categorical by quintile of daily intake * (FFQ) --- Continuous; Adipose biomarker quintile Continuous; Daily intake * quintile (FFQ) Continuous; Daily intake * quintile (FFQ) Continuous; Daily intake * quintile (FFQ) Continuous; Daily intake * quintile (FFQ) Continuous; Daily intake quintile * (FFQ) Continuous; Adipose -linolenic acid quintile Physical activity Binary; active if moderate- and high-intensity activity > than 10% of expended daily energy Continuous; total METs expended over a 83

93 24-hour period Smoking Binary; current smoker Binary; current smoker Alcohol consumption Categorical; 0 g/day, g/day, g/day, g/day (FFQ) Categorical; 0 g/day, g/day, g/day, g/day (FFQ) Socioeconomic status Binary; poor if self-reported annual income < twice the national poverty line in the year of recruitment Continuous; Composite index of income, education, occupation, and household possessions Waist-to-hip ratio Binary; cutoff= 0.90 for men, 0.85 for women Binary; cutoff = 0.90 for men, 0.85 for women * Adjusted for total energy intake 84

94 Supplemental Table 2.1 Single SNP analysis: least square means * (+/- standard errors) of selected adipose fatty acids, blood lipids, and inflammatory markers by genotype among controls in the Costa Rica Study (n=1032 unless indicated otherwise). rs (FADS1) CC CT TT Nominal P value GLA (n=983) 0.069± ± ±0.002 < ETA ± ± ± < AA (n=1031) 0.54± ± ±0.01 < EPA 0.045± ± ±0.002 < hscrp (n=836) 2.54± ± ± Triglycerides ± ± ± (n=1026) rs (FADS1/FADS2 intergenic) TT T- -- Nominal P value GLA (n=983) 0.067± ± ± ETA ± ± ± < AA (n=1031) 0.54± ± ±0.01 < EPA 0.045± ± ±0.001 < hscrp (n=836) 2.32± ± ± Triglycerides ± ± ± (n=1026) 85

95 rs (FADS2) CC CT TT Nominal P value GLA (n=983) 0.068± ± ±0.002 < ETA ± ± ± < AA (n=1031) 0.53± ± ±0.01 < EPA 0.047± ± ±0.002 < hscrp (n=836) 2.43± ± ± Triglycerides ± ± ± (n=1026) rs (FADS1) AA AT TT Nominal P value GLA (n=983) 0.066± ± ±0.003 < ETA ± ± ± AA (n=1031) 0.51± ± ±0.02 < EPA 0.044± ± ± hscrp (n=836) 2.20± ± ± Triglycerides ± ± ± (n=1026) rs (FADS2) CC CG GG Nominal P value 86

96 GLA (n=983) 0.063± ± ± ETA ± ± ± AA (n=1031) 0.51± ± ±0.01 < EPA 0.043± ± ± hscrp (n=836) 2.46± ± ± Triglycerides ± ± ± (n=1026) rs (FADS2/FADS3 intergenic) CC CT TT Nominal P value GLA (n=983) 0.062± ± ± ETA ± ± ± AA (n=1031) 0.49± ± ± EPA 0.043± ± ± hscrp (n=836) 2.25± ± ± Triglycerides ± ± ± (n=1026) * Models with fatty acids as the outcome were adjusted for age/sex/residence (by matching), ancestry, and dietary fatty acids; models with inflammatory markers or blood lipids as outcomes were adjusted for age/sex/residence (by matching) and ancestry 87

97 Supplemental Figure 3.1 Pairwise standardized linkage disequilibrium (LD) coefficients (Lewontin s D') for the elongase family SNPs used in the study. LD blocks are defined by a black border. The intensity of the color corresponds to the value of the D' statistic. Panels A-C represent SNPs in genes ELOVL2, ELOVL4, and ELOVL5 respectively. A) B) 88

98 C) 89