Common Genetic Variation, Residential Proximity to Traffic Exposure, and Left Ventricular Mass: The Multi-Ethnic Study of Atherosclerosis

Background Elevated left ventricular mass (LVM) is a strong predictor of negative cardiovascular outcomes, including heart failure, stroke, and sudden cardiac death. A relationship between close (< 50 m compared with > 150 m) residential proximity to major roadways and higher LVM has previously been described, but the mechanistic pathways that are involved in this relationship are not known. Understanding genetic factors that influence susceptibility to these effects may provide insight into relevant mechanistic pathways. Objective We set out to determine whether genetic polymorphisms in genes affecting vascular and autonomic function, blood pressure, or inflammation influence the relationship between traffic proximity and LVM. Methods This was a cross-sectional study of 1,376 genotyped participants in the Multi-Ethnic Study of Atherosclerosis, with cardiac magnetic resonance imaging performed between 2000 and 2002. The impact of tagged single-nucleotide polymorphisms (tagSNPs) and inferred haplotypes in 12 candidate genes (ACE, ADRB2, AGT, AGTR1, ALOX15, EDN1, GRK4, PTGS1, PTGS2, TLR4, VEGFA, and VEGFB) on the relationship between residential proximity to major roadways and LVM was analyzed using multiple linear regression, adjusting for multiple potential confounders. Results After accounting for multiple testing and comparing homozygotes, tagSNPs in the type 1 angiotensin II receptor (AGTR1, rs6801836) and arachidonate 15-lipoxygenase (ALOX15, rs2664593) genes were each significantly (q < 0.2) associated with a 9–10% difference in the association between residential proximity to major roadways and LVM. Participants with suboptimal blood pressure control demonstrated stronger interactions between AGTR1 and traffic proximity. Conclusions Common polymorphisms in genes responsible for vascular function, inflammation, and oxidative stress appear to modify associations between proximity to major roadways and LVM. Further understanding of how genes modify effects of air pollution on CVD may help guide research efforts into specific mechanistic pathways.

volume 118 | number 7 | July 2010 • Environmental Health Perspectives Research Longterm exposure to ambient air pollution has been associated with cardiovascular mor bidity and mortality in numerous epidemio logic studies (Dockery et al. 1993;Miller et al. 2007;Pope et al. 2004), but the mechanisms responsible for its observed effects remain poorly understood (Bhatnagar 2006). A vari ety of hypothesized pathways have been impli cated in human and animal studies, including an effect of air pollution on endothelial func tion and vascular tone, autonomic function, thrombosis, blood pressure, and oxidative stress. Recent reviews of the studies impli cating these pathways have been published (Brook 2007;Mills et al. 2009;Pope and Dockery 2006;Simkhovich et al. 2008).
Among studies that have demonstrated increased cardiovascular morbidity and mor tality associated with air pollution exposures, trafficrelated air pollution in particular has been increasingly recognized as an important source of cardiovascular disease (CVD) modifying exposures. Recent work in the MultiEthnic Study of Atherosclerosis (MESA) has demonstrated a link between close (< 50 m) residential major roadway proximity, an indicator of high traffic expo sure, and higher left ventricular mass (LVM) index (Van Hee et al. 2009). Increased LVM is associated with the subsequent development of depressed ejection fraction (Drazner et al. 2004), and left ventricular hypertrophy is associated with heart failure, arrhythmia, and sudden cardiac death Levy et al. 1990). Efforts to elucidate the ways in which air pollution might relate to higher LVM may provide not only greater under standing of the underlying CVD process itself but also potential strategies for intervention.
Several genes and genetic pathways have been noted to have important functions in the postulated mechanisms responsible for the impact of air pollution on CVD. Components of the renin-angiotensin system, including angiotensinogen (AGT), angiotensinconvert ing enzyme (ACE), and type 1 angiotensin II receptor (AGTR1) genes, have welldescribed impacts on inflammation, blood pressure, and vasoconstriction (Gradman 2009) that are mediated in part via G protein-coupled receptor kinases (GRKs), including GRK4 (Felder and Jose 2006). Prostaglandin endoperoxide synthase 1 and 2 genes (PTGS1 and PTGS2), as well as 15lipoxygenase gene (ALOX15), play important roles in vascular inflammation and oxidative stress that leads to CVD (Dogné et al. 2005; Mochizuki and Kwon 2008;Wittwer and Hersberger 2007). Endothelin1 (EDN1) is a potent vasocon strictor that activates signaling pathways lead ing to atherosclerosis (Ivey et al. 2008). The beta2 adrenergic receptor (ADRB2) affects cardiac autonomic function and the develop ment of heart failure (Triposkiadis et al. 2009). Together with ALOX15, vascular endothelial growth factors A and B (VEGFA and VEGFB) promote vascular growth, which contributes to the development of atherosclerotic plaque and plaque instability (Mochizuki and Kwon 2008;Sluimer and Daemen 2009). Tolllike recep tor 4 (TLR4) affects the inflammatory response that influences initiation and progression of atherosclerosis (Pasterkamp et al. 2004).
Background: Elevated left ventricular mass (LVM) is a strong predictor of negative cardiovascular outcomes, including heart failure, stroke, and sudden cardiac death. A relationship between close (< 50 m compared with > 150 m) residential proximity to major roadways and higher LVM has previously been described, but the mechanistic pathways that are involved in this relationship are not known. Understanding genetic factors that influence susceptibility to these effects may provide insight into relevant mechanistic pathways. oBjective: We set out to determine whether genetic polymorphisms in genes affecting vascular and autonomic function, blood pressure, or inflammation influence the relationship between traffic proximity and LVM. Methods: This was a crosssectional study of 1,376 genotyped participants in the MultiEthnic Study of Atherosclerosis, with cardiac magnetic resonance imaging performed between 2000 and 2002. The impact of tagged singlenucleotide polymorphisms (tagSNPs) and inferred haplotypes in 12 candidate genes (ACE, ADRB2, AGT, AGTR1, ALOX15, EDN1, GRK4, PTGS1, PTGS2, TLR4, VEGFA, and VEGFB) on the relationship between residential proximity to major roadways and LVM was analyzed using multiple linear regression, adjusting for multiple potential confounders. results: After accounting for multiple testing and comparing homozygotes, tagSNPs in the type 1 angiotensin II receptor (AGTR1, rs6801836) and arachidonate 15lipoxygenase (ALOX15, rs2664593) genes were each significantly (q < 0.2) associated with a 9-10% difference in the asso ciation between residential proximity to major roadways and LVM. Participants with suboptimal blood pressure control demonstrated stronger interactions between AGTR1 and traffic proximity. conclusions: Common polymorphisms in genes responsible for vascular function, inflamma tion, and oxidative stress appear to modify associations between proximity to major roadways and LVM. Further understanding of how genes modify effects of air pollution on CVD may help guide research efforts into specific mechanistic pathways. We conducted a study to determine whether polymorphisms in these prespecified genes involved in the regulation of vascular tone, blood pressure, autonomic function, and oxidative stress modify the previously observed relationship between proximity to major roadways and LVM in MESA.

Materials and Methods
Population and sampling. MESA is a prospec tive cohort study designed to examine the pro gression of subclinical CVD; it enrolled 6,814 men and women 44-85 years of age who were free of clinical CVD at entry. The partici pants were recruited from six U.S. communi ties: Baltimore, Maryland; Chicago, Illinois; Forsyth County, North Carolina; Los Angeles County, California; Northern Manhattan, New York; and St. Paul, Minnesota. Details of the sampling, recruitment, and data collec tion have been previously reported (Bild et al. 2002). A subcohort of 2,880 unrelated MESA subjects were selected for genetic studies from subjects who gave informed consent for DNA extraction and use in genetic studies and had sufficient DNA available for this study. Priority was given to subjects who were included in a 1,000person subset of participants with addi tional blood biomarker meas urements, supple mented by random selection from remaining participants to fulfill balanced ethnic group representation (720 African American, 720 Hispanic, 720 Chinese, and 720 European American) and equality by sex.
For the present study, we included all participants in the genetic study database (n = 2,847) who underwent cardiac magnetic resonance imaging (MRI) at examination 1 (n = 2,152), had accurate address information for exposure assignment, and represented the groups with the greatest contrast in traffic proximity exposure (living within 50 m com pared with ≥ 150 m from a major roadway) and the largest LVM association from our prior study (Van Hee et al. 2009), leaving 1,376 participants for analysis.
Study procedures. Medical history, address, anthropometric measurements, physi cal examination, and laboratory data for the present study were taken from the first exami nation of the MESA cohort, July 2000 to August 2002 (Bild et al. 2002). Information about age, sex, ethnicity, and medical history were obtained by questionnaires administered at the screening and the first examination. Institutional review boards at all participating centers approved the protocol, ensuring con formance with the Declaration of Helsinki, and all participants gave informed consent.
Geocoding and exposure assessment. Participants' residential addresses were assigned geographic coordinates using ArcGIS 9.1 software (ESRI, Redlands, CA) in conjunc tion with the Dynamap/2000 street network and geocoding database (Tele Atlas, Boston, MA). We calculated proximity to traffic by measuring the distance from geocoded home address to the nearest major roadway (inter state, state, or county highway or major arte rial), with a maximum search radius of 150 m. Exposure groups were divided into 0-50 m and > 150 m from the nearest major road way. These two groups were chosen a priori based upon the results of our prior study, which demonstrated the largest associations between traffic exposure and LVM compar ing the group residing within 50 m of a major roadway with the group living > 150 m away (Van Hee et al. 2009). This exposure contrast is consistent with the observed distribution of nearroadway air pollutants, with many pollut ants rapidly approaching background levels by 150-200 m from large roadways (Hoek et al. 2002;Smargiassi et al. 2005;Zhu et al. 2002).
Cardiac MRI imaging. LVM was obtained by cardiac MRI. Images were acquired by 1.5T MRI scanners (Signa LX and CVi, GE Healthcare, Waukesha, WI; and Somatom Vision and Sonata, Siemens Medical Solutions, Berlin, Germany) using a protocol previously described (Natori et al. 2006). All MRI data were submitted to the MESA MRI Reading Center at Johns Hopkins Hospital for central ized processing using MASS software, version 4.2 (Medis, Leiden, Netherlands). DNA extraction. DNA was extracted from peripheral leukocytes isolated from packed cells of anticoagulated blood by use of a commer cially available DNA isolation kit (Puregene; Gentra Systems, Minneapolis, MN). The DNA was quantified by determination of absorbance at 260 nm followed by PicoGreen analysis (Molecular Probes, Inc., Eugene, OR). Two vials of DNA were stored per participant at -70°C and subsequently aliquoted for use.
Selection of singlenucleotide poly morphisms. Singlenucleotide polymorphisms (SNPs) were selected in candidate gene loci according to the following criteria: a) within the proximal and distal 10kb regions 5´ and 3´ to the given candidate gene (NCBI Build 35; National Center for Biotechnology Information 2004); b) compatibility with the Illumina GoldenGate technology (Fan et al. 2006) as determined by the Assay Design Tool (TechSupport, Illumina, San Diego, CA); and c) minor allele frequency (MAF) > 0.05 or a tag (r 2 > 0.8) for another SNP with MAF > 0.05 as determined by applying the multi locus or "aggressive" "Tagger" option of Haploview version 3 (Barrett et al. 2005) using International HapMap project data for CEPH (Caucasian) and Yoruban populations (release 19;International HapMap Consortium 2003). In some cases a complete set of tagged SNPs (tagSNPs) for a given candidate gene was not possible because of these competing criteria. Additional SNPs were added from a) LDselect analysis of resequencing information from the SeattleSNPs project if available (Carlson et al. 2004;Nickerson 2003); b) nonsynonymous SNPs from dbSNP release 124 (National Center for Biotechnology Information 2005); and c) SNPs with prior reports of association proposed by a MESA investigator.
Selection of ancestry informative markers. Ancestry informative markers (AIMs) were selected from an Illumina proprietary SNP database to maximize the difference in allele frequencies between any pair of ethnic groups: European American versus African American, European American versus Chinese American, and African American versus Chinese American. Additional markers infor mative for MexicanAmerican ancestry were selected from published lists (Choudhry et al. 2006;CollinsSchramm et al. 2004). All sub jects were genotyped for the 199 total AIMs selected for inclusion. Principal components of AIMs were calculated using Stata (version 10.1; StataCorp LP, College Station, TX).
Genotyping. Genotyping was performed by Illumina Genotyping Services (Illumina Inc.) using their proprietary GoldenGate assay. Of 150 SNPs initially typed, 11 failed quality control measures and were excluded. After excluding 24 SNPs with an overall MAF < 2%, 115 common tagSNPs in the 12 genes under study were chosen for the present analysis. We inferred haplotypes using the expectationmaximization algorithm of PHASE 2.1 (Stephens and Scheet 2005). Haplotype uncertainty was estimated using the diplotype probability. Within a gene, haplotypes with probabilityweighted fre quencies < 2.5% were grouped together in a genespecific haplotype bin assigned "others." A total of 85 common haplotypes were inves tigated. Supplemental Material, Table 1 (doi:10.1289/ehp.0901535.), summarizes the tagSNPs and haplotypes examined.
Data quality control. Illumina performed initial quality control to identify samples and SNPs that failed genotyping according to proprietary protocols and sporadic failed genotypes with GenCall (Illumina) quality score < 99.99%. After removal of failed SNPs and samples, the genotype calling rate was 99.93%, with maximum missing data rate per sample of 2.1% and maximum missing per SNP of 4.98%. The cohort genetic data was checked for cryptic sample duplicates and dis crepancies in genetically predicted sex (using X markers) versus study database reported sex.
volume 118 | number 7 | July 2010 • Environmental Health Perspectives Samples with unresolved duplicate and sex discrepancies were removed from the genetic study database.
Statistical methods. We produced descrip tive statistics using frequencies and percent ages for categorical variables and means and SDs for continuous variables. The distribu tions of covariates were compared across proximity to major roadway categories and logtransformed LVM using analysis of vari ance methods for continuous variables and chisquare tests for categorical variables. SNP genotype frequencies were tested for racestratified Hardy-Weinberg equilib rium using exact tests. Racestratified linkage disequilibrium was assessed for each of the 12 genes under analysis.
A total of 1,139 participants had no miss ing data on any outcomes or covariates. An initial analysis was performed without impu tation and using models with reduced sets of covariates that required no imputation. Subsequently, iterative multiple imputation procedures were performed using switching regression and chained equations for the full 1,376 participants, including data on coti nine levels for 1,082 of 1,376 participants to impute secondhand smoking data (Royston 2007). This process resulted in the full sample of 1,376 adults with some imputed values.
The regression analysis was repeated on the imputed sample, yielding overall results simi lar to those estimated initially. All reported findings are from the imputed data set because the precision of the estimates is improved by the increased sample size, and the full data set is less likely to be subject to bias (Greenland and Finkle 1995;van der Heijden et al. 2006).
We first performed SNPspecific analyses to examine interactions between SNPs and very close (< 50 m) residential proximity to major roadway for logtransformed LVM. We then performed haplotype analyses to also examine interactions between haplotypes and very close proximity to major roadways for LVM. We fitted linear, additive (1 degree of freedom) models separately for each individual SNP (coded 0, 1, and 2 for minor allele cop ies) and for each haplotype (coded 0, 1, and 2 for number of haplotype copies). Diplotype probabilities were used as weights in the haplo type analyses clustered on participant identi fier. Individual Ftests for the product term of SNP × proximity < 50 m and haplotype × proximity < 50 m were calculated to obtain pvalues for interactions. The false discovery rate (FDR) correction (qvalue) was used to account for multiple testing (115 tests in each SNP model and 85 haplotypes in each haplo type model) (Hochberg and Benjamini 1990;Storey 2002). A qvalue of 0.2, corresponding to a 20% FDR, was selected as a threshold for acceptable level of significance (Smith et al. 2007). To account for potential confounding by population stratification, we adjusted for the first five principal components of AIMs in all analyses. All models were additionally adjusted for age, sex, height, and weight.
Additional potential confounders for the full model, chosen a priori based upon a prior study of the relationship between traffic exposure and LVM (Van Hee et al. 2009), and results of the initial descriptive analyses included household income, highest educa tional attainment level, systolic and diastolic blood pressure, medication use, low (LDL) and high (HDL) densitylipoprotein choles terol, physical activity, alcohol use, smoking and packyear history of smoking, secondhand smoke exposure, and diabetes status by fast ing blood glucose criteria or medication for diabetes. Medication use was modeled using indicator variables for ACE inhibitors, angiotensinreceptor blockers (ARBs), diuret ics, betablockers, lipidlowering medications, aspirin, nonsteroidal antiinflammatory drugs (NSAIDs), and cyclooxygenase (COX) inhibi tors. Because several of these covariates (par ticularly blood pressure) may lie in the causal biological pathway between the exposure and outcome, and as a sensitivity analysis, we also examined several partially adjusted models with reduced sets of covariates. Confidence intervals (CIs) were calculated at an alpha value of 0.05. We then conducted several sensitivity analyses, including stratification by race, blood pressure, and medication use. Subpopulation effects in all interaction models were calculated using linear combinations of predictors in each model (lincom command in Stata). Data were analyzed using Stata and R (ver sion 2.9; R Development Core Team 2009). The authors had full access to the data and take responsibility for the integrity of the data.

Results
Descriptive statistics. Table 1 indicates the overall distribution of participant characteris tics. The 1,376 participants in this study ranged in age from 44 to 84 years. Although all MESA participants were free of clinical CVD at baseline, a substantial percentage had cardiac risk factors (Table 1). On bivariate analyses, race, sex, site, antihypertensive medication use, body mass index (BMI), secondhand smoke exposure, and income were at least moder ately associated (p < 0.2) with proximity to major roadway (data not shown). African Americans, Hispanics, women, participants in New York and Baltimore, those on antihy pertensive medication and specifically ACE inhibitor therapy, and those with higher BMIs, higher secondhand smoke exposures, and lower income tended to reside nearer to major road ways. Associations between LVM and nearly every covariate were strong. AfricanAmerican race, St. Paul study site, male sex, younger age, hypertension, antihypertensive medication use, BMI, blood pressure, cigarette smoke expo sure, fasting blood glucose, income, education, lower HDL cholesterol, and higher height and weight were all associated with higher LVM on bivariate analysis. After adjustment for height and weight, LVM was positively associated with age (data not shown).
We noted no significant deviations from racestratified Hardy-Weinberg equilibrium, such as those associated with geno typing error, on exact testing at an alpha level of 0.05 with Bonferroni correction for 115 tests. Supplemental Material, Table 1 (doi:10.1289/ ehp.0901535), lists the selected SNPs and inferred haplotypes for this study, with the frequencies (for tagSNPs) or probability weighted frequencies (for inferred haplotypes) of each. The principal components of AIMs showed generally good discrimination between selfcategorized races, with Hispanics demon strating the most diverse ancestry and show ing some ancestral overlap with other racial groups [Supplemental Material, Figure 2 (doi:10.1289/ehp.0901535)].
Primary SNP interaction analyses. All SNP-traffic proximity interaction models tested demonstrated significant evidence of interactions for two of the 12 genes tested, ALOX15 and AGTR1 [ Table 2; see also Supplemental Material, Figure 1 (doi:10.1289/ ehp.0901535)]. The minimally adjusted model (model 1) containing only study site and principal components of AIMs, age, sex, height, and weight showed significant (q < 0.2) evidence of interactions for rs389566 and rs6801836 in AGTR1, with a 9% (95% CI, 2-16%) difference in the impact of traffic prox imity (< 50 m residential proximity compared with > 150 m residential proximity) on LVM comparing homozygotes for rs389566 and an 11% (95% CI, 3-19%) difference comparing homozygotes for rs6801836. In all cases of significant interactions, the minor allele represented the deleterious allele. The MAF of polymorphisms conferring appar ent susceptibility to traffic proximity ranged from 23% (rs6801836) to 33% (rs389566) for AGTR1 and from 19% (rs2664593) to 21% (rs7220870) for ALOX15. The top four SNPs representing greatest evidence for interaction overall in models 4 and 5 con sist of one group of two adjacent tagSNPs on ALOX15 (rs2664593 at position 7 and rs7220870 at position 8) and two nearby tagSNPs on AGTR1 (rs389566 at position 10 and rs6801836 at position 12). Figure 1 represents the relative location of each evalu ated tagSNP on the two genes with signifi cant findings. In the Supplemental Material, Figures 3 and 4  . Despite this reduced power to detect interactions for haplotypes com pared with individual SNPs, we found sig nificant (q < 0.2) interactions for AGTR1F after taking into account multiple testing. In model 5, proximity to roadway had no significant association with LVM in individu als without this haplotype, whereas the pres ence of a single haplotype conferred an 8.6% (95% CI, 3.0-14.6%) positive difference in LVM (pvalue for interaction = 0.002) associ ated with close (< 50 m) proximity to a major roadway compared with living farther away (> 150 m). All five models showed consistent impacts on susceptibility, with the presence of AGTR1F showing increased susceptibil ity to traffic proximity on LVM. Complete results of all haplotype-roadway proximity interaction tests for all genes and all models are included in Table 3 of the Supplemental Material (doi:10.1289/ehp.0901535). Sensitivity analyses. As described above and shown in Table 2 and Supplemental Material, Figure 1 (doi:10.1289/ehp.0901535), results were not particularly sensitive to model selection. To investigate the impact of racial structure on the findings, we performed stratification by race (using a threeway inter action among SNP, proximity to roadway, and race) for the two significant (q < 0.2) inter active SNPs in the full model, an analy sis that includes adjustment for the principal components of AIMs (Figure 2). Although we found no significant evidence of a threeway inter action between rs6801836 (pvalue for threeway interaction = 0.7) or rs2664593 (pvalue for threeway interaction = 0.6), road way proximity, and race, the overall results suggest less pronounced gene-environment interactions for African Americans compared with other racial groups.
We also explored stratification by blood pressure categories [Seventh Report of the Joint National Committee on Prevention, Detection, Evaluation, and Treatment of High Blood Pressure (JNC VII) categories for all participants, binary optimal/ suboptimal blood pressure, binary hypertensive/not hypertensive] (Chobanian et al. 2003) and use of medications important in the renin angiotensin system or inflammatory pathways (ACE inhibitors, ARBs, aspirin, and NSAIDs) to determine whether these were groups par ticularly susceptible to this interaction. Among these analyses, we found significant evidence of interaction among rs6801836, proximity to traffic, and optimal versus suboptimal blood pressure categories. The difference in the LVM change associated with roadway proximity for individuals with greater numbers of the delete rious (G) allele in rs6801836 was significantly larger among individuals with suboptimal blood pressure by JNC VII criteria (systolic blood pressure ≥ 120 mmHg or diastolic blood pressure ≥ 80 mmHg, n = 732 partici pants) than among those with optimal blood pressure (pvalue for threeway interaction = 0.05; Figure 3). We found no other significant threeway interactions for the medication use or blood pressure categories tested.

Discussion
We found significant evidence of gene-traffic interactions in MESA, with polymorphisms in AGTR1 and ALOX15, genes important in vascular function and inflammation/oxidative stress, associated with substantial alterations in the association between traffic proximity and LVM. Exploratory analyses additionally revealed that the interaction associated with   rs1076039  rs1871346  rs6502804  rs743646  rs7217186  rs7220870  rs2255888  rs9894225  rs748694  *rs2664593  rs916055  rs409742  rs422858  rs2638363  rs2131127  rs2638360  rs4681443  rs3772616  rs4488792  rs389566  rs385338  rs2320019 rs275646 rs275645 *rs6801836 rs718858 the common polymorphism in the AGTR1 gene (rs6801836) is larger for individuals with suboptimal blood pressure (> 120 mmHg systolic or > 80 mmHg diastolic) than for those with optimal blood pressure control. Haplotype analyses, although likely under powered in this setting, provided additional support for SNP findings in AGTR1.
AGTR1 is a wellknown regulator of blood pressure and common target of specific phar macologic intervention (ARB) in hypertension. The specific tagSNP locus (rs6801836) found to demonstrate interactive effects here has not yet been observed to show main effects in MESA or other studies (Bahrami et al. 2008). However, polymorphisms in AGTR1, and par ticularly the A1166C polymorphism (rs5186), have been associated with inflammation (Suchankova et al. 2009) and left ventricular hypertrophy (Smilde et al. 2007). Additionally, an association between the A1166C polymor phism and hypertension has been observed in a number of studies, although results have been somewhat inconsistent (Mottl et al. 2008). A possible explanation for the inconsistencies observed in these studies is the presence of unexamined interactive effects with environ mental factors, such as those observed in the present analysis. Our observation that blood pressure and the identified polymorphism in AGTR1 appear to interact to produce suscep tibility to the hypertrophic effects of traffic exposure is consistent with the wellknown role of AGTR1 in the regulation of blood pres sure and provides support to the hypothesis that the effect of air pollution is mediated by genes involved in blood pressure regulation. This observation also suggests an important role of both genetic and other environmental factors (in this case, factors that contribute to blood pressure) in modulating the effects of air pollution on CVD.
The protein encoded by ALOX15 is an oxidizing enzyme that can produce reactive lipid hydroperoxides within the vasculature.
Experimental and human epidemiologic stud ies have demonstrated a range of contrasting effects of ALOX15 on CVD, and its role at present is not fully understood (Mochizuki and Kwon 2008;Wittwer and Hersberger 2007). Variants specifically at the rs2664593 locus in the 5´ promoter region of the gene have previously (but inconsistently) been asso ciated with carotid atherosclerosis and have been shown to interact in the relationship between carotid wall thickness and insulin resistance (Bevan et al. 2009;McCaskie et al. 2008). In MESA, this specific SNP locus has not shown main effects on LVM (Bahrami H, unpublished observations).
Both ALOX15 and AGTR1 are likely to play important roles in inflammation, one of the primary putative mechanisms for the impact of air pollution on CVD and also an important mechanism involved in the devel opment of left ventricular hypertrophy. These genes' apparent interaction with air pollution to produce increases in LVM could suggest an important role of vascular inflammation in the pathogenesis of trafficrelated air pollu tion-induced CVD more generally.
Linkage disequilibrium mapping of AGTR1 and ALOX15 in these analyses sug gests that regions near to tagSNPs rs6801836 on AGTR1 and rs2664593 on ALOX15 are responsible for the interactions seen in this study. Finer mapping of ALOX15 and AGTR1 in these regions may help identify the specific polymorphism associated with these results.
Because the MESA cohort has four dis tinct racial groups, it represents a unique opportunity to examine gene-environment interactions in a multiethnic population. Analyses performed in this heterogeneous group must take into account the important issue of potential confounding by popula tion stratification. To account for this issue of multiple ethnic groups that also vary in structure by site, we adjusted for both study site and the principal components of AIMs.
We additionally examined threeway interac tions with race in an analysis that included both site and AIMs and observed no evidence of substantial confounding or effect modi fication by race. Although we found no sig nificant evidence overall of modification of the interactions by race, the finding that African Americans show apparently less pronounced interactions (Figure 2) bears further investiga tion. It is possible that additional relationships with other underlying genetic factors that pre dispose African Americans to known higher LVM (gene-gene interactions) may play a role in attenuating this particular interaction.
Few studies to date have examined geneenvironment interactions in the cardiovascu lar health effects of air pollutants (Baccarelli et al. 2008;Chahine et al. 2007;Park et al. 2006;Schwartz et al. 2005). Of those stud ies, most have focused on individual functional    (Peters et al. 2007;Ren et al. 2010). This is the first study to explore multiple polymorphisms in pathways of inter est with the goal of furthering understanding of how air pollution affects cardiac structure specifically. This study has several limitations. We can not rule out the possibility that a portion of the findings may have arisen by chance. We chose an FDR of 0.2 as a measure of statistical significance in this study, as have other studies (Smith et al. 2007). By definition, the choice of an FDR threshold to account for multiple testing controls for an expected proportion of incorrectly rejected null hypotheses. Future studies examining these relationships in addi tional cohorts will be necessary to replicate and confirm these findings. As have other recent studies of the health impacts of air pol lution, we used a relatively coarse indicator of exposure to trafficrelated air pollution, advantages and disadvantages of which have been described in more detail previously (Van Hee et al. 2009). The exposure misclassifica tion produced by such methods may lead to bias in estimating the effects in subpopula tions and overall and may adversely affect standard error estimates (Gryparis et al. 2009;Kim et al. 2009;Szpiro et al. 2009). Because of the chosen exposure metric, this study is unable to distinguish between interactions produced by specific components of traffic related air pollutants or even nonair pollut ants related to traffic (e.g., noise; for more detail, see Van Hee et al. 2009). Ongoing work will help disentangle these trafficrelated influences on health.
Although the interactions observed in this study are relatively modest, in this same data set a 10mmHg increase in systolic blood pressure (one of the factors known to be most important in the development of increased LVM) is associated with only an adjusted 3% higher LVM. Given that individuals homozy gous for the deleterious alleles show a 7-8% difference in LVM associated with close prox imity to a major roadway, and given the high minor allele frequencies and high prevalence of close residential proximity to a major road way seen here, the public health impacts of these findings are potentially significant.
In our prior analysis of the relationship between proximity to roadways and VM (Van Hee et al. 2009), we reported results using an untransformed LVM index (rather than logtransformed LVM adjusted for height and weight) as the dependent variable in order to make the results more accessible to the pri mary audience. As we described in that report, the preferred statistical methods for regression analysis avoid the use of ratios such as the LVM index to prevent spurious correlation (Kronmal 1993). For this reason, and to meet the more stringent requirements of popula tion association studies, which require that the outcome be approximately normally distrib uted rather than only the residuals (Balding 2006), we have used log LVM adjusted for height and weight in this report. For compar ison purposes, we have reported the results of our prior study with the same parameteriza tion used here in Table 4 of the Supplemental Material (doi:10.1289/ehp.0901535).
In addition to replication of these find ings specifically, human studies such as phar macologic intervention trials may further understanding of these potentially modifiable genetic pathways. Although this study was not powered to detect threeway interactions among genes, air pollution, and medication use, the observation that the presence of opti mal blood pressure may attenuate the effect of genes and air pollution on cardiovascular outcomes is a provocative one.
As described above, the mechanisms responsible for the effects of air pollutants on CVD remain uncertain, although several stud ies have implicated inflammation and oxidative stress, as well as impaired vascular function. Modification of the effect of traffic exposures on LVM by genes involved in inflamma tory response and vascular function suggests that individuals with genetically determined impaired handling of inflammation and altered vascular responses have greater susceptibility to these effects. This finding lends further support to pathways involving these mechanisms.

Conclusions
SNPs in genes responsible for vascular func tion, inflammation, and oxidative stress (AGTR1 and ALOX15) modify associations between proximity to major roadways and LVM. Further understanding of how genes modify effects of air pollution on CVD may help guide research efforts into specific mecha nistic pathways.