InterestinhowmicroRNAscanbeusedinthepreventionand treatment of cardiovascular disease (CVD) 4 has increased over

Transcription

1 MicroRNA-410 regulated lipoprotein lipase variant rs13702 is associated with stroke incidence and modulated by diet in the randomized controlled PREDIMED trial 1 3 Dolores Corella, Jose V Sorlí, Ramon Estruch, Oscar Coltell, Carolina Ortega-Azorín, Olga Portolés, Miguel A ngel Martínez-González, Mónica Bulló, Montserrat Fitó, Fernando Arós, José Lapetra, Eva M Asensio, Guillermo T Sáez, Lluís Serra-Majem, Carlos Muñoz-Bravo, Valentina Ruiz-Gutiérrez, Miquel Fiol, Ernest Vinyoles, Xavier Pintó, Kris Richardson, Emilio Ros, and Jose M Ordovás ABSTRACT Background: MicroRNAs have emerged as important epigenetic regulators in cardiovascular diseases (CVDs). Using an observational meta-analysis design, we previously characterized a gain-offunction microrna-410 target site polymorphism (rs13702t.c) in the 3#untranslated region of the lipoprotein lipase (LPL) gene. The C allele was associated with lower triglycerides, and this association was modulated by fat intake. Objectives: We aimed to extend our findings by assessing the interaction between the rs13702 polymorphism and fat intake on triglycerides at baseline and longitudinally by using a dietary intervention design. We also examined as a primary outcome the association of this variant with CVD incidence and its modulation by the Mediterranean diet (MedDiet). Design: We studied 7187 participants in the PREDIMED (Prevención con Dieta Mediterránea) randomized trial that tested a MedDiet intervention compared with a control diet, with a median 4.8-y followup. LPL polymorphisms and triglycerides were determined and CVD assessed. Gene-diet interactions for triglycerides were analyzed at baseline (n = 6880) and after a 3-y intervention (n = 4131). Oxidative stress parameters were investigated in a subsample. Results: The rs13702t.c polymorphism was strongly associated with lower triglycerides in C allele carriers and interacted synergistically with dietary monounsaturated (P = 0.038) and unsaturated fat intake (P = 0.037), decreasing triglycerides at baseline. By 3 y, we observed a genediet interaction (P = 0.025) in which the C allele was associated with a greater reduction in triglycerides after intervention with MedDiet, high in unsaturated fat. Although the polymorphism was associated with lower stroke risk (HR: 0.74; 95% CI: 0.57, 0.97; P = per C allele), this association reached statistical significance only in the MedDiet intervention (HR: 0.58; 95% CI: 0.37, 0.91; P = in C compared with TT carriers), not in the control group (HR: 0.94; 95% CI: 0.55, 1.59; P = 0.805). Conclusion: We report a novel association between a microrna target site variant and stroke incidence, which is modulated by diet in terms of decreasing triglycerides and possibly stroke risk in rs13702 C allele carriers after a high-unsaturated fat MedDiet intervention. This trial was registered at controlled-trials.com as ISRCTN Am J Clin Nutr 2014;100: INTRODUCTION InterestinhowmicroRNAscanbeusedinthepreventionand treatment of cardiovascular disease (CVD) 4 has increased over 1 From the Department of Preventive Medicine and Public Health, School of Medicine, University of Valencia, Valencia, Spain (DC, JVS, CO-A, OP, and EMA); CIBER Fisiopatología delaobesidadynutrición, Instituto de Salud Carlos III, Madrid, Spain (DC, JVS, RE, OC, CO-A, OP, MÁM-G, MB, M Fitó, FA,JL,EMA,GTS,LS-M,CM-B,VR-G,MFiol,EV,XP,andER);Department of Internal Medicine, Hospital Clinic, IDIBAPS, Barcelona, Spain (RE); Department of Computer Languages and Systems, University Jaume I, Castellon, Spain (OC); Department of Preventive Medicine and Public Health, School of Medicine, University of Navarra, Pamplona, Spain (MÁM-G); Human Nutrition Unit, Faculty of Medicine, IISPV, University Rovira i Virgili, Reus, Spain (MB); Cardiovascular Epidemiology Unit, Municipal Institut for Medical Research, Barcelona, Spain (M Fitó); Department of Cardiology, Hospital Txagorritxu, Vitoria, Spain (FA); Department of Family Medicine, Primary Care Division of Sevilla, Sevilla, Spain (JL); Department of Biochemistry and Molecular Biology, Clinical Analysis Service-CDB, HGUV, University of Valencia, Valencia, Spain (GTS); Department of Clinical Sciences, University of Las Palmas de Gran Canaria, Las Palmas de Gran Canaria, Spain (LS-M); Department of Epidemiology, School of Medicine, University of Málaga, Málaga, Spain (CM-B); Instituto de la Grasa, Consejo Superior Investigaciones Cientificas, Seville (VR-G); University Institute for Health Sciences Investigation, Hospital Son Dureta, Palma de Mallorca, Spain (M Fiol); Primary Care Division, Catalan Institute of Health, Barcelona, Spain (EV); Lipids and Vascular Risk Unit, Internal Medicine, Hospital Universitario de Bellvitge, Hospitalet de Llobregat, Barcelona, Spain (XP); Nutrition and Genomics Laboratory, JM-USDA Human Nutrition Research Center on Aging at Tufts University, Boston, MA (KR, JMO); Lipid Clinic, Endocrinology and Nutrition Service, Institut d Investigacions Biomèdiques August Pi Sunyer, Hospital Clinic, Barcelona, Spain (ER); IMDEA Alimentación, Madrid, Spain (JMO); and Department of Cardiovascular Epidemiology and Population Genetics, Centro Nacional de Investigaciones Cardiovasculares, Madrid, Spain (JMO). 2 Supported by The Spanish Ministry of Economy and Innovation and the Fondo Europeo de Desarrollo Regional (projects CNIC-06/2007, RTIC G03/ 140, RTIC RD 06/0045, CIBER 06/03, PI , PI , PI , PI , PI11/02505, PI12/0153, PI13/00462, SAF , and AGL C03-03); contracts 53-K and from the US Department of Agriculture Research Service; the Generalitat Valenciana, Spain (ACOMP , AP111/10, AP-042/11, ACOM2011/145, ACOMP/2012/190, ACOMP/2013/159, and ACOMP/213/165); recercaixa (2013ACUP00194); and the Instituto de Salud Carlos III, Spain, and CP 06/00100 from the Health Department of the Catalan Government (Generalitat de Catalunya) (M Fitó). 3 Address correspondence to D Corella, PhD, Genetic and Molecular Epidemiology Unit, Valencia University, Blasco Ibañez, Valencia, Spain. dolores.corella@uv.es. Received September 30, Accepted for publication May 21, First published online June 25, 2014; doi: /ajcn Am J Clin Nutr 2014;100: Printed in USA. Ó 2014 American Society for Nutrition 719

2 720 CORELLA ET AL recent years, because it may be possible to exploit them in diagnostics, prognostics, and therapeutics (1, 2). micrornas are small noncoding RNAs (3) that have emerged as important epigenetic regulators of relevant functions related to CVD, and their dysregulation has been reported in cardiovascular events (2, 4 6). They act as posttranscriptional inhibitors of gene expression by binding to microrna recognition elements within the 3#untranslated region (UTR) of their target messenger RNAs (3, 7). The most critical region for this binding and repression is that referred to as the seed site (6, 7). Single-nucleotide polymorphisms (SNPs) can alter microrna binding by either creating a new site or destroying an existing target site (6 8). In previous work (9), we demonstrated that the minor allele of the rs13702 in the 3#UTR of the lipoprotein lipase (LPL) gene disrupts an microrna recognition element seed site for the microrna-410, resulting in a gain of function. This microrna- 410 target site SNP was consistently associated with lower circulating triglycerides and higher HDL cholesterol concentrations in our meta-analysis, including 27,756 participants (9). The LPL gene encodes a multifunctional enzyme that is a master regulator in the lipolytic processing of triglyceride-rich lipoproteins (10). Deficient LPL expression is associated with hypertriglyceridemia (10, 11). Plasma levels of the oxidation marker malondialdehyde were also significantly increased in LPL-deficient mice (12), suggesting a parallel enhancement of oxidative damage, but no such association has been reported with LPL variants in humans. In addition to lipids (13 17), LPL-SNPs have also been associated with type 2 diabetes (18), premature atherosclerosis (19), and the risk of CVD, mainly stroke (10, 11, 20 22). Although several LPL-SNPs have been related to these traits, the functional variants have not yet been identified unequivocally. This step is key for the development of more precise genomic tools for CVD prevention and treatment. In previous years, functional importance had largely been placed on the Ser447Stop (rs328) SNP in exon 9 (10, 11). However, emerging findings concerning epigenetic regulation by micrornas (3, 4) suggest that other functional variants in the 3#UTR (13, 17, 23, 24), mainly the microrna-410 target site rs13702 SNP, may be more relevant (9, 13, 23, 24). Our bioinformatics and experimental evidence supporting the functionality of this SNP (9) prompted us to investigate further into its inverse association with CVD. In prior work (9), we also found a gene-diet interaction of this SNP with PUFAs on fasting triglycerides. Although other gene-diet interactions of LPL-SNPs on lipids have been reported (25 27), no study has analyzed the dietary modulation of the association between LPL-SNPs or even microrna target site SNPs and CVD incidence in a randomized trial. Therefore, our goals were first to extend our initial feedings by analyzing the rs13702-by-fat interaction on triglycerides at baseline and longitudinally after intervention with the Mediterranean diet (Med- Diet), rich in unsaturated fat. Second, as a primary outcome, we examined the association of this microrna-410 target site SNP with CVD incidence and its modulation by the MedDiet (median follow-up: 4.8 y). In addition, we investigated the association of the rs13702 with oxidative stress in a subsample and whether its effects on lipids and CVD were independent from the rs328 SNP. 4 Abbreviations used: CVD, cardiovascular disease; LD, linkage disequilibrium; MedDiet, Mediterranean diet; PREDIMED, Prevención con Dieta Mediterránea; SNP, single-nucleotide polymorphism; UTR, untranslated region. SUBJECTS AND METHODS Participants We included 7187 participants (3063 men and 4124 women) in the PREDIMED (Prevención with Dieta Mediterránea) trial (28) for whom DNA was available and the LPL-rs13702 polymorphism determined. PREDIMED is a multicenter, randomized, controlled clinical trial ( ISRCTN ) aimed at assessing the effects of the MedDiet on the primary prevention of CVD (28, 29). The completion date of this study was 1 December 2010, and the total number of randomized participants was 7447 (see flowchart in Supplemental Figure 1 under Supplemental data in the online issue). The 7187 participants included did not differ in the main characteristics from those of the total cohort. Briefly, from 1 October 2003, physicians in primary care centers selected participants at high cardiovascular risk. Eligible were communitydwelling persons (55 80 y for men; y for women) who fulfilled at least one of 2 criteria: type 2 diabetes or 3 or more cardiovascular risk factors [hypertension, dyslipidemia, BMI (in kg/m 2 ) $25, current smoking, or a family history of premature CVD (28)]. Exclusion criteria included previous CVD, any severe chronic illness, and drug or alcohol addiction. Participants were randomly assigned to these interventions: a MedDiet (2 groups, one supplemented with extra-virgin olive oil and the other with nuts) and a control group (low-fat diet). Participants assigned to both MedDiet groups received intensive training to follow the MedDiet and allotments of either extra-virgin olive oil or mixed nuts (30 g/d), whereas those assigned to the control diet were instructed to reduce the intake of all types of fat (28). Median follow-up time was 4.8 y (IQR: y). Because both MedDiet intervention groups had a similar effect on reducing the incidence of total CVD and stroke, these groups were pooled and analyzed together. For budgetary reasons, triglycerides were determined in odd years (baseline, 1 y, 3 y, 5 y, etc). Considering that, in the PREDIMED study, recruitment was open from 2003 through 2009, not all participants completed the 5-y follow-up by 1 December 2010, the study s closing date. Therefore, for the analysis of the interaction between the polymorphism and intervention with the MedDiet, 3-y follow-up data were used, because most participants completed this follow-up period and had their triglycerides determined as well as genetic and dietary data (n = 4131 participants). The institutional review board of each participating center approved the study protocol, and all participants provided written informed consent. For budgetary reasons, we also conducted additional genetic analyses of LPL-SNPs on participants (n = 1048) recruited in one of the field centers (Valencia region), where the laboratory for genetic analyses is located. Demographic, clinical, anthropometric, and dietary measurements The baseline examination included assessment of standard cardiovascular risk factors, medication use, sociodemographic factors, and lifestyle variables, as previously detailed (28). Food consumption was determined by a validated semiquantitative food-frequency questionnaire (30). The level of adherence to the MedDiet was measured by a validated 14-item questionnaire (31). Physical activity was estimated by the validated Minnesota Leisure-Time Physical Activity questionnaire (28). Weight and

3 microrna-410, LPL-rs13702 SNP, DIET, AND STROKE 721 height were measured with calibrated scales and a wall-mounted stadiometer, respectively (29). Biochemical data were obtained at baseline and in odd years. Biochemical determinations, DNA extraction, and genotyping Blood samples were obtained after an overnight fast and were frozen at 2808C. Fasting glucose, total cholesterol, triglycerides, HDL cholesterol, and LDL cholesterol were measured by using standard enzymatic methods (29). In participants whose triglyceride levels were,400 mg/dl (n = 6837), LDL cholesterol concentrations were estimated by using the Friedewald formula. Biochemical measures were available for nearly 7000 participants at baseline (n = 6880 for triglycerides). At 3 y, plasma triglycerides were available for 4131 participants. Genomic DNA was extracted from buffy coat. We genotyped the LPL-rs13702 polymorphism in all PREDIMED participants on a 7900HT Sequence Detection System (Applied Biosystems) by using a fluorescent allelic discrimination TaqMan assay. The LPLrs328 (Ser/Stop) SNP was also genotyped by a TaqMan assay in PREDIMED-Valencia participants (n = 1029 participants had both polymorphisms determined). Genotype frequencies did not deviate from Hardy-Weinberg equilibrium expectations. In addition, we genotyped another 22 LPL-SNPs by using the HumanOmniExpress BeadChip (Illumina) (see Supplemental Figure 2 under Supplemental data in the online issue) in a subsample consisting of 192 randomly selected PREDIMED-Valencia participants. Determination of oxidative stress parameters Oxidative stress markers malondialdehyde and 8-oxo-7#8#- dihydro-2#-deoxyguanosine in nuclear DNA were determined in circulating mononuclear blood cells as previously reported (32) in a subsample of 525 randomly selected participants recruited in the Valencia field center. Outcome ascertainment In the PREDIMED study, the primary end point was the occurrence of the first major CVD event and comprised stroke, myocardial infarction, or cardiovascular death (28). We used 4 sources of information to identify end points: 1) direct participant contact, 2) family physicians, 3) yearly review of medical records, and 4) consultation of the National Death Index. Medical records related to end points were examined by the end-point adjudication committee, whose members were blinded to treatment allocation. End points confirmed by the committee occurring between 1 October 2003 and 1 December 2010, were included in the analyses. The criteria for adjudicating primary outcomes are detailed elsewhere (28). Statistical analyses Linkage disequilibrium (LD) was assessed with the Haploview software package (version 4.2; Broad Institute). x 2 Tests were used to test differences in percentages. Triglycerides were logtransformed for statistical analyses. LPL-polymorphisms were first tested as additive with genotype coded by the number of minor alleles (0/1/2 copies). Furthermore, C allele carriers were grouped together and compared with TT individuals to increase statistical power when the number of participants in the CC group was small (ie, in the analyses of cardiovascular diseases as well as in the stratified analysis by diet and in the subsample analyses). We used ANOVA tests to compare the means of plasma lipids and oxidative stress parameters by LPL-polymorphisms. Multivariable adjustments for continuous variables were carried out by general linear model analysis. Models were sequentially adjusted for age, sex, field center, type 2 diabetes, BMI, adherence to the MedDiet at baseline, total energy intake, alcohol consumption, smoking, physical activity, medications (lipid-lowering, hypoglycemic, and antihypertensive drugs), hypertension, or plasma lipids as specifically indicated in the corresponding analyses. In subsample analyses, multivariable models also were adjusted for the LPL-rs328 (Ser/Stop) SNP. Gene-diet interaction analysis between the LPL-rs13702 (microrna-410) SNP and dietary fat intake in determining triglycerides at baseline was carried out as previously described (9). We used multivariable linear regression models, including main effects and interaction terms (genotype 3 dietary fat product). Dietary fats were considered continuous variables (expressed in percentage of energy intake) and the LPL-rs13702 SNP an additive variable while adjusting for age, sex, BMI, and field center. Additional adjustments were made as indicated. Selected gene-diet interactions were depicted by computing the predicted values for each individual from the adjusted regression model and plotting them against fat intake by LPL-rs13702 genotypes. Dichotomous variables for fatty acids and adherence to the MedDiet were created by using the sample means as cutoffs. Stratified analyses by dietary intake or medication use were also carried out. We also analyzed the effects of LPLrs13702 (additive) and diet intervention (MedDiet compared with control diet) on 3-y changes of triglycerides by using general linear models in 4131 participants with available data (because the recruitment lasted from 2003 through 2009, many participants had not completed 5 y of follow-up in December 2010). The change in triglycerides (3 y compared with baseline) was predicted in a multivariable regression model adjusted for sex, age, field center, type 2 diabetes, BMI, medications, incidence of CVD, and baseline triglycerides. In addition, an interaction term between the polymorphism and dietary intervention was included in the model. The 2 MedDiet intervention groups were pooled together after the homogeneity of the effect was checked in each group (P-interaction = 0.468). To examine the association between the LPL-rs13702 SNP and incidence of major CVD (total, stroke, and myocardial infarction) including all participants (n = 7187), we used Cox regression models with length of follow-up as the primary time variable. The exposure time was calculated as the time between randomization and the date of a major CVD; the date of the last interview, 1 December 2010, or the date at death, whichever came first. We first estimated the incidence rate for the 3 genotypes and fitted additive models. HRs with 95% CIs for the LPL-rs13702 SNP were estimated. Models were multivariable adjusted for covariates as indicated. Finally, we evaluated the modulation by diet of the associations between the SNP and CVD by estimating the HRs for the LPL-rs13702, stratified by dietary intervention groups (MedDiet compared with control diet). Analyses were based on the intent-to-treat principle. Unadjusted (model 1) and multivariable-adjusted Cox regression models were fitted as detailed in the Results. Kaplan-Meier survival curves were plotted to estimate the probability of remaining free of total CVD or stroke

4 722 CORELLA ET AL during follow-up. Formal tests for the interaction between the LPL-rs13702 SNP and the dietary intervention in determining stroke were also carried out by analyzing the product term of these variables in the Cox regression models. Testing this interaction in a Cox model estimates the departure from multiplicativity instead of the departure from additivity that is tested for the product term in a linear regression model (33). Therefore, taking into account the power limitations to detect a multiplicative interaction, we considered P, 0.10 as suggestive of a significant interaction term to be confirmed in the stratified analysis (34). Statistical analyses were performed with the SPSS Statistics version 21.0 (SPSS Inc). All tests were 2-tailed, and P, 0.05 was considered statistically significant, except for the tests of interaction in determining CVD outcomes; in this case, P, 0.10 was retained for significance. RESULTS Baseline demographic, biochemical, clinical, and lifestyle characteristics of the 7187 participants according to the randomly assigned dietary intervention groups (MedDiet groups or control) are shown in Table 1. Baseline characteristics for these participants according to the microrna-410 target site SNP LPL-rs13702 are shown in Supplemental Table 1 (under Supplemental data in the online issue). There were no significant differences in age, BMI, smoking, drinking, physical activity, hypertension, or dietary intake among LPL-rs13702 genotypes (minor allele frequency: 0.34). However, we detected statistically significant differences in the prevalence of type 2 diabetes and treatment with fibrates (lower in C allele carriers). Baseline association between the microrna-410 target site SNP LPL-rs13702 and plasma lipid concentrations in the PREDIMED participants To verify that the associations observed in our previous report (9) were evident in the PREDIMED participants, we first conducted baseline associations between this SNP and plasma lipid concentrations. As expected, we observed (Table 2) a strong inverse association between this polymorphism and triglycerides (B: mmol/l, or 211 mg/dl, per C allele; 95% CI: 20.20, 20.05; P = ) in the entire population. This association remained statistically significant after multivariable adjustment (P = ). We also observed an association TABLE 1 Baseline demographic, clinical, lifestyle, and genetic characteristics of the study participants by intervention group 1 Total (n = 7187) MedDiet with EVOO (n = 2479) MedDiet with nuts (n = 2361) Control group (n = 2347) Age (y) Sex, F 4124 (57.4) 1454 (58.7) 1271 (53.8) 1399 (59.6) BMI (kg/m 2 ) Current smokers 1007 (14.0) 346 (14.0) 339 (14.4) 322 (13.7) Family history of premature CVD 1614 (22.5) 564 (22.8) 511 (21.6) 539 (23.0) Waist circumference (cm) Systolic blood pressure (mm Hg) Diastolic blood pressure (mm Hg) Type 2 diabetes 3473 (48.3) 1241 (50.1) 1095 (46.4) 1137 (48.4) Hypertension 5953 (82.8) 2035 (82.1) 1951 (82.6) 1967 (83.8) Dyslipidemia 5198 (72.3) 1775 (71.6) 1729 (73.2) 1694 (72.2) Lipid-lowering drugs, statins 2888 (40.2) 1016 (41.0) 925 (39.2) 947 (40.3) Lipid-lowering drugs, fibrates 380 (5.3) 120 (4.8) 140 (5.9) 120 (5.1) Hypoglycemic drugs, insulin 374 (5.2) 122 (4.9) 121 (5.1) 131 (5.6) Hypoglycemic drugs, oral 2116 (29.4) 744 (30.0) 654 (27.7) 718 (30.6) Antihypertensive drugs 5228 (72.7) 1797 (72.5) 1699 (72.0) 1732 (73.8) LPL-rs13702 TT 3218 (44.8) 1082 (43.6) 1068 (45.2) 1068 (45.5) TC 3156 (43.9) 1097 (44.3) 1046 (44.3) 1013 (43.2) CC 813 (11.3) 300 (12.1) 247 (10.5) 266 (11.3) Adherence to the MedDiet Energy intake (kcal/d) Carbohydrate (% of energy) Protein (% of energy) Total fat (% of energy) Saturated fat (% of energy) MUFA (% of energy) PUFA (% of energy) n26 (% of energy) n23 (% of energy) Alcohol consumption (g/d) Physical activity (METs $ min 21 $ d 21 ) Values are means 6 SDs for continuous variables and n (%) for categorical variables. CVD, cardiovascular disease; EVOO, extra virgin olive oil; MedDiet, Mediterranean diet; MET, metabolic equivalent task. 2 Based on a 14-point screener of adherence.

5 microrna-410, LPL-rs13702 SNP, DIET, AND STROKE 723 TABLE 2 Association between the LPL-rs13702 polymorphism and plasma lipids: crude means, ORs for hypertriglyceridemia, and unadjusted and adjusted linear regression coefficients (B) 1 Crude Adjusted LPL-rs13702 genotypes 2 Model 1 3 Model 2 4 Model 3 5 TT (n = 3073) TC (n = 3035) CC (n = 772) P 6 B P-additive effects B P-additive effects P Total cholesterol (mmol/l) (20.029, 0.040) (20.035, 0.030) LDL cholesterol (mmol/l) (20.007, 0.055) (20.016, 0.042) HDL cholesterol (mmol/l) (0.016, 0.041) (0.018, 0.042) Triglycerides (mmol/l) (20.196, ) (20.125, ) Hypertriglyceridemia (%) Hypertriglyceridemia 1 (reference) 0.76 (0.68, 0.84) 0.72 (0.60, 0.85) (0.75, 0.88) (0.76, 0.89) Hypertriglyceridemia was defined as fasting triglycerides.1.7 mmol/l. P values for triglycerides were obtained after log transformation. Regression coefficients (B) per-variant allele effects (genotypes coded as 0, 1, and 2 according to the number of minor alleles; additive). 2 Biochemical data were available for total cholesterol (n = 6919 participants), HDL cholesterol (n = 6836 participants), LDL cholesterol (n = 6781 participants), and triglycerides (n = 6880 participants). 3 Estimates were derived from linear or logistic regression models for continuous and categorical variables, respectively. 4 Adjusted for sex, age, center, type 2 diabetes, and lipid-lowering and hypoglycemic drugs. Estimates were derived from linear or logistic models adjusted for covariates. 5 Adjusted as for model 2 and for BMI, smoking, drinking, physical activity, and total energy intake. 6 Unadjusted P values for the polymorphism for the comparison of means or ORs as categorical (3 categories). ANOVA tests, x 2 test, and logistic regression analysis, respectively. 7 Mean 6 SD (all such values). 8 B or OR; 95% CI in parentheses (all such values).

6 724 CORELLA ET AL with higher HDL cholesterol concentrations (B: mmol/l per C allele; 95% CI: 0.02, 0.04; P = ). No significant associations were detected for total cholesterol or LDL cholesterol. Considering that many PREDIMED participants were taking lipid-lowering drugs and that in the previous study (9), these participants were excluded, we analyzed the influence of medication use on the polymorphism effects on plasma triglycerides. Because we found no heterogeneity as a result of statin use (see Supplemental Figure 3 under Supplemental data in the online issue), we did not exclude participants taking statins from the analyses. Regarding fibrates (see Supplemental Figure 4 under Supplemental data in the online issue), we detected a significant interaction because the magnitude of the polymorphism effects was stronger in participants taking fibrates (5.3% of the population) than in the others. However, considering that the genotype effect was not abolished by fibrate treatment, we decided to include all participants in the analyses, adjusting or stratifying by fibrate treatment to account for this heterogeneity. Influence of the LPL-rs328 (Ser/Stop) SNP on the association between the LPL-rs13702 (microrna-410) and baseline plasma lipids To elucidate whether the associations with plasma lipids observed for the LPL-rs13702 SNP were independent of the LPLrs328 (Ser/Stop), we analyzed this SNP in the PREDIMED- Valencia participants (n = 1048) (see Supplemental Table 2 under Supplemental data in the online issue). Both polymorphisms were in partial LD (D#: 0.80, r 2 = 0.22; P, 0.001). Additional LD parameters with other LPL-SNPs in this Mediterranean population are shown in Supplemental Table 3 (under Supplemental data in the online issue). We observed a strong inverse association between the LPL-rs13702 SNP and triglycerides in the PREDIMED-Valencia participants (P = 0.002), and this association remained statistically significant after additional adjustment for the LPL-rs328 SNP (P = 0.004) (see Supplemental Table 4 under Supplemental data in the online issue). Conversely, the LPL-rs328 SNP tended to have weaker associations with triglycerides (P = 0.185), and the inverse trend totally disappeared (P = 0.841) after adjustment for the LPLrs13702 SNP (see Supplemental Table 5 under Supplemental data in the online issue). Baseline association between the microrna-410 target site SNP LPL-rs13702 and oxidative stress parameters To extend our findings, we analyzed whether the LPL-rs13702 SNP was also associated with oxidative stress parameters in a subsample. We found that this SNP was strongly associated (P = 0.004) with lower levels of malondialdehyde, an indicator of lipid oxidation, in C allele carriers (P for additive effects in a model adjusted for sex, age, field center, type 2 diabetes, BMI, and medications) (see Supplemental Figure 5 under Supplemental data in the online issue). This association remained statistically significant even after further adjustment for triglycerides (P = 0.003), suggesting additional effects of this SNP on oxidative stress parameters. Likewise, we also observed a significant association (P = 0.033) with lower levels of an indicator of oxidative DNA damage, the 8-oxo-7#8#-dihydro-2#- deoxyguanosine in nuclear DNA (see Supplemental Figure 6 under Supplemental data in the online issue), which reinforced an additional association of the LPL-rs13702 SNP with oxidative stress. Baseline gene-diet interactions in the association between the microrna-410 target site SNP LPL-rs13702 and fasting triglycerides We used the same statistical model (model 1) that we fitted in the previous meta-analysis (9) in which dietary fatty acids (MUFA, PUFA, or saturated fat) were considered continuous variables and the polymorphism an additive variable, as well as further adjustment for additional covariates (model 2), and obtained regression coefficients for the interaction terms between the LPL-rs13701 and fat intake at baseline before randomization (Table 3). We obtained a statistically significant gene-diet interaction term (P = 0.038) between the LPL-rs13702 SNP and MUFA intake (as % of energy) in determining baseline triglycerides (B: mmol/l per variant C allele and % energy P = 0.038). This suggested an interactive effect between the C allele and MUFA intake, in the sense that the decrease in plasma triglycerides associated with each additional C allele became stronger as MUFA intake increased (resulting in steeper slopes in the regression lines for each genotype). We also analyzed n26, n23, and unsaturated fat, which were not explored in our previous work. Although we did not find any significant gene-diet interaction with PUFA on triglycerides, as we had observed in the meta-analysis (9), we detected a statistically significant interaction with unsaturated fat intake (P = 0.037), which could reconcile these and the previous results. Depicted in Figure 1 are the interaction effects for MUFA (A), PUFA (B), and unsaturated fatty acid intake (C) and the LPLrs13702 SNP on triglycerides. The association of the C allele with lower triglycerides was stronger when MUFA or unsaturated fat intake was high. When participants taking fibrates were excluded, we achieved a higher level of statistical significance for these gene-diet interactions at baseline (P = for TABLE 3 Regression coefficients (B) and SEs for the interaction terms between the LPL-rs13702 polymorphism and fat intake in determining baseline triglycerides 1 Fat intake (% of energy) B 6 SE 2 P 2 B 6 SE 3 P 3 mmol/l mmol/l MUFA PUFA n n Unsaturated fat SFA Total fat B for the interaction terms (6SE) and P were obtained in separated multivariable linear regression analyses implementing an additive genetic model with fat intake as continuous (% of energy) with adjustment for the indicated covariates. Triglycerides were log-transformed for P values. n = Models were adjusted for age, sex, field center, and BMI. 3 Models were also adjusted for type 2 diabetes, medications, Mediterranean diet adherence, total energy intake, smoking, drinking, and physical activity.

7 MUFA, P = for PUFA, and P = for unsaturated fat for model 2 in Table 3). In line with the meta-analysis (9), we too did not find any significant interaction of the polymorphism with PUFA or MUFA in determining HDL or LDL cholesterol concentrations (see Supplemental Table 6 under Supplemental data in the online issue). Moreover, we analyzed the interaction between MUFA intake and the LPL-rs13702 SNP in determining oxidative stress parameters in the subsample of participants at baseline. Our data also suggested a modulation of the effects of the polymorphism on malondialdehyde levels by MUFA intake (see Supplemental Figure 7 under Supplemental data in the online issue). Thus, when MUFA intake was high (above the sample mean, 19.5% energy), carriers of the C allele had significantly lower (P = 0.002) malondialdehyde concentrations than did noncarriers. This inverse association was not observed (P = 0.536) in the context of a low MUFA intake (,19.5% energy). microrna-410, LPL-rs13702 SNP, DIET, AND STROKE 725 Interaction between the microrna-410 target site SNP LPL-rs13702 and intervention with MedDiet on 3-y changes in triglycerides The LPL-rs13702 SNP was also strongly associated with triglycerides at 3 y (P-main effects = ). We detected a statistically significant gene-diet interaction (P = 0.025) between this SNP and the dietary intervention (MedDiet groups pooled together compared with the control group) for changes in triglycerides at 3 y in the randomized trial (Figure 2). These results were consistent with the gene-diet interaction observed for triglycerides at baseline. Thus, among participants allocated to the MedDiet intervention groups [in which an increase in unsaturated fat intake was observed (28)], carriers of the C allele (additive effects) had greater reductions in triglycerides than did noncarriers (P = 0.002). However, in the control group (low-fat diet group), no significant decreases in triglycerides in carriers of the C allele were observed (P = 0.387). When participants taking fibrates were excluded, the above-mentioned gene-diet interaction was also detected (P = 0.010). No heterogeneity of the interaction effects was observed for the 2 MedDiet intervention groups (P = 0.468). In both, a significant decrease in triglycerides in participants carrying the C allele was observed (not shown). Effects of the microrna-410 target site SNP LPL-rs13702 on the incidence of CVD and modulation by diet After a median follow-up of 4.8 y (IQR: y), 268 major cardiovascular events occurred among the 7187 participants (30,685 person-years of observation). Of these, 135 were strokes, 100 were myocardial infarctions, and the rest were other cardiovascular deaths. In the entire population, we observed no significant association between the LPL-rs13702 SNP and total CVD (HR: 0.95; 95% CI: 0.79, 1.14; P = per C allele) in the additive multivariable-adjusted model. Likewise, no significant association was obtained for myocardial infarction (HR: 0.99; 95% CI: 0.74, 1.34; P = per C allele in the multivariable-adjusted model). However, the SNP was significantly associated with a reduced incidence of stroke (HR: 0.74; 95% CI: 0.57, 0.97; P = per C allele in the additive model). Kaplan-Meier curves are shown in Figure 3 for total CVD-free FIGURE 1. Predicted values of baseline triglycerides (Ln) by the LPLrs13702 single-nucleotide polymorphism (additive) plotted against the MUFA (A), PUFA (B), or unsaturated fat (C) intake in the entire population (all participants with triglyceride data). The P values for the interaction term between the corresponding fat and the polymorphism were adjusted for sex, age, field center, type 2 diabetes, BMI, medications, smoking, drinking, physical activity, adherence to the Mediterranean diet, and total energy intake in the multivariable linear regression model (n = 3073 TT, n = 3035 TC, and n = 772 CC).

8 726 CORELLA ET AL FIGURE 2. Effects of LPL-rs13702 genotype and intervention with the MedDiet (2 groups pooled) compared with the control diet on mean (6SEM) changes in triglycerides at 3 y. Data included 4131 participants with triglycerides available (n = 2928 MedDiet and n = 1203 control group). Means were adjusted for age, sex, field center, type 2 diabetes, BMI, cardiovascular disease events, and baseline triglycerides. P-interaction was obtained in the multivariable regression model adjusted for these covariates. Further adjustment for medications did not change significance (P = 0.042). P 1 : P for the polymorphism (additive) for the difference of adjusted means between genotypes in the MedDiet intervention groups. P 2 : P for the polymorphism (additive) comparing the genotypes in the control group. n = 1273 TT, n = 1312 TC, and n = 343 CC in the MedDiet intervention and n = 548 TT, n = 525 TC, and n = 130 CC in the control group. MedDiet, Mediterranean diet. (A) and stroke-free (B) survival by the LPL-rs13702 genotypes for the population as a whole. The significant association between LPL-rs13702 SNP and stroke seemed independent of the LPL-rs328 (Ser/Stop) because in estimating the incidence and stroke risk for both polymorphisms in the PREDIMED-Valencia population (see Supplemental Tables 7 and 8 under Supplemental data in the online issue), a stronger association was detected for the microrna-410 target site SNP rs13703 (P = 0.043) than for the rs328 (Ser/Stop) (P = 0.249). Moreover, when analyzing the modulation of the effects of the polymorphism on stroke by the dietary intervention in the entire population (Table 4), we found that the microrna target site rs13703 SNP was significantly associated with lower stroke risk only among participants randomly assigned to the MedDiet intervention groups (HR: 0.63; 95% CI: 0.43, 0.91; P = per C allele in the multivariable-adjusted model 3). However, in the control group, the polymorphism lost its inverse association with stroke (HR: 0.87; 95% CI: 0.58, 1.31; P = per C allele). When carriers of the variant allele were grouped in model 3, we observed a more significant interaction term between the SNP and the dietary intervention (P = 0.053), and we selected this model for further analyses. We did not find heterogeneity for this gene-diet interaction between the 2 MedDiet interaction groups, and results were presented together. Kaplan-Meier curves for stroke-free survival stratified by the dietary intervention are shown in Figure 4: (A) MedDiet and (B) control group by the LPL-rs13703 SNP (C compared with TT carriers). In this model, further adjustment for triglycerides and HDL cholesterol concentrations maintained the protective association of the C allele with stroke in participants in the MedDiet intervention groups. Moreover, the interaction term between the SNP and the intervention on stroke was more significant (P = 0.051). Taking into account that the mean follow-up was higher in the MedDiet groups than in the control group, we excluded 9 participants with late follow-up (.6 y). After this exclusion, we obtained a statistically significant interaction term between the LPLrs13703 SNP and dietary intervention for stroke at P, 0.05 (P-interaction = 0.044). This result added more evidence to the LPL-rs13703 MedDiet interaction. Also supporting the modulation by the MedDiet of the protective effects of the LPLrs13703 SNP on stroke risk are our results analyzing the effect of adherence to the MedDiet at baseline. We also observed that higher baseline adherence to the MedDiet was associated with significantly lower stroke risk in C allele carriers (see Supplemental Table 9 under Supplemental data in the online issue). DISCUSSION Here we provide initial evidence that a microrna target site SNP is associated with the incidence of stroke and modulated by a diet in a population at high cardiovascular risk. Although microrna research is growing exponentially and solid evidence supports the role of micrornas in CVD in animal models (4 6, 35, 36), few reports show associations between microrna target site SNPs and stroke (37 39). Recent case-control studies (37 39) provided first evidence that functional SNPs in micro- RNA target sites specifically in the 3#UTR of the angiopoietin 1 gene (rs ) (37), matrix metalloproteinase 9 gene (rs ) (38), and paraoxonase gene (rs ) (39) might be significantly associated with stroke risk by interfering with a microrna binding (microrna-211, microrna-491 5p, and microrna-616, respectively) site. Here we augment this

9 microrna-410, LPL-rs13702 SNP, DIET, AND STROKE 727 FIGURE 3. Cumulative total cardiovascular disease free survival (A) or stroke-free survival (B) by LPL-rs13702 genotypes (additive model) in the entire population (n = 7817). Cox regression models were adjusted for sex, age, field center, type 2 diabetes, hypertension, BMI, intervention group, medications, physical activity, alcohol, smoking, total energy intake, and baseline adherence to Mediterranean diet baseline. HRs were estimated in multivariable-adjusted Cox regression models per C allele. Both for A and B: n = 3218 TT, n = 3156 TC, and n = 813 CC. evidence showing that the rs13702t.c in the 3#UTR of the LPL gene, in which the C allele disrupts a microrna recognition element seed site for the microrna-410, was associated with lower stroke incidence in C allele carriers. Although significant associations between LPL-SNPs and stroke have been reported in previous works (20 22), none has specifically analyzed the rs13702 SNP, focusing on the traditional rs328 (Ser/ Stop). Our results suggest a higher relevance of the LPL-rs13702 variant on this association. Moreover, we report that the impact of the microrna-410 target site SNP LPL-rs13702 on stroke incidence was modulated by diet in a randomized nutritional intervention trial (28). Thus, only among participants assigned to a MedDiet intervention, supplemented with extra-virgin olive oil or nuts, did C allele carriers have a statistically significant lower stroke risk than did participants with the TT genotype. However, those carriers of the protective C allele who were assigned to the low-fat group did not experience the same level of protection from this allele, and the statistical association became nonsignificant. Although a limitation of our study may be that the statistical significance of the interaction term was borderline at P, 0.05, our findings meet most of the criteria of credibility for a subgroup analysis (34). Thus, our subgroup variable was measured at baseline and randomly distributed, we have an a priori hypothesis of the protective effect in the MedDiet intervention group, the direction of the subgroup effect was specified, a small number of hypotheses were tested, the size of the interaction effect was large, and the interaction in determining stroke was consistent across closely related outcomes within the study. The dietary modulation of the microrna-410 target site SNP on stroke incidence is in agreement with our previous results showing that effects of this SNP on triglycerides were also modulated by dietary intake at the meta-analysis level, including 27,756 participants from 10 cohorts (9). In that meta-analysis, significant interaction effects of PUFA intake were found. However, in our exclusively Mediterranean population, we observed statistical significance with MUFA rather than with PUFA. In the meta-analysis (9), populations of different origins (American, European, etc) were included. Of all, only the InCHIANTI (Invecchiare in Chianti, Italy) study corresponded with a Mediterranean population. In this cohort, the gene-pufa interaction was not detected (9), and the regression coefficient for MUFA interaction was the greatest in magnitude (KR, unpublished data, 8 August 2013). This observation is consistent with the results obtained here. Differences in fatty acid intake among populations (9) may have influenced these findings, given the high MUFA intake characteristic of Mediterranean populations (28). Moreover, the origin of these MUFAs is mainly olive oil (29), whereas in other populations of Northern Europe and North America, it comes mainly from meats. This could explain the difference in the results observed and may suggest that unsaturated fatty acids (both PUFAs and MUFAs) could be driving this gene-diet interaction and that their relative contribution would depend on the greater or lower intake in each population analyzed. Accordingly, we report statistically significant results for the interaction between the LPL-rs13702 SNP with unsaturated fat and found statistically significant results showing that the decreasing effect on triglycerides per C allele was increased with higher intake of both MUFAs and PUFAs. Similarly, a recent study reported that unsaturated fat interacted with another LPL-rs polymorphism in determining HDL cholesterol concentrations in Korean participants (26). The gene-diet interaction on baseline triglycerides was also supported by our findings following the 3-y intervention with the MedDiet in the randomized trial. C carriers had a greater decrease in triglycerides when following a MedDiet in comparison with a low-fat diet. In a previous work (28), we detailed the changes in macronutrients associated with the MedDiet intervention and demonstrated a significant increase of MUFAs (mainly from olive oil) and PUFAs (mainly from nuts) in the MedDiet intervention groups. Only one previous intervention study had examined the effect of the rs13702 SNP on triglyceride changes (40). In the Look AHEAD study (40), despite the strong association of this SNP with baseline triglycerides, a significant interaction with the behavioral intervention on triglyceride changes at 1 y was not found. These results are in agreement with our observation that

10 728 CORELLA ET AL TABLE 4 Incidence rates and HRs for stroke depending on the LPL-rs13702 (microrna-410) polymorphism in the entire population and stratified by the intervention groups in the randomized trial 1 Stroke incidence Whole population (n = 7187) MedDiet intervention (n = 4840) Control group (n = 2347) LPL-rs13702 Cases Person-years of follow-up Incidence rate HR (95% CI) P Cases Person-years of follow-up Incidence rate HR (95% CI) P Cases Person-years of follow-up Incidence rate HR (95% CI) P Model 1 TT 71 13, (reference) (reference) (reference) TC 55 13, (0.55, 1.11) (0.40, 1.03) (0.62, 1.79) CC (0.24, 0.97) (0.16, 1.04) (0.21, 1.73) Per C 0.73 (0.56, 0.96) (0.45, 0.92) (0.60, 1.31) allele Model 2 TT 1.00 (reference) 1.00 (reference) 1.00 (reference) TC + CC 0.71 (0.50, 0.99) (0.37, 0.93) (0.54, 1.52) Per C 0.72 (0.55, 0.95) (0.44, 0.92) (0.57, 1.27) allele Model 3 TT 1.00 (reference) 1.00 (reference) 1.00 (reference) TC + CC 0.72 (0.51, 1.01) (0.37, 0.91) (0.55, 1.59) Per C 0.74 (0.57, 0.97) (0.43, 0.91) (0.58, 1.31) allele 1 Incidence rates were expressed per 1000 person-years of follow-up. MedDiet intervention groups were pooled together. Model 1 was unadjusted. These results were presented as descriptive with crude values. P-interaction SNP 3 intervention in the Cox regression for model 1: P = in the model including 3 categories; P-interaction SNP 3 intervention: in the model considering additive effects (per C allele). Model 2 was adjusted for age, sex, field center, type 2 diabetes, BMI, and dietary intervention groups. Results for the SNP in the Cox regression were presented for both C allele carriers grouped compared with TT and per C allele. P-interaction SNP 3 intervention: for C allele carriers grouped; P-interaction SNP 3 intervention: for additive effects. Model 3 was also adjusted for medications, hypertension, smoking, drinking, physical activity, baseline adherence to MedDiet, and total energy intake. Results for the SNP in the Cox regression were presented for both C allele carriers grouped compared with TT and per C allele. P-interaction SNP 3 intervention: for C allele carriers grouped; P-interaction SNP 3 intervention: for additive effects. MedDiet, Mediterranean diet; SNP, single-nucleotide polymorphism.