Air pollution is a complex mixture of particulate matter, volatile organic compounds, and gaseous pollutants such as oxides of nitrogen. Epidemiological analyses have demonstrated an association between short- and long-term exposures to air pollution, especially particulate matter <2.5 μm in aerodynamic diameter (PM_2.5) and increased cardiovascular morbidity and mortality ((1),2). It is postulated that this adverse cardiovascular effect is related to systemic inflammation, oxidative stress, or autonomic nervous system imbalance effects in the artery wall (3). Hence, one of the intermediate steps by which PM_2.5 exposure increases cardiovascular morbidity and mortality may be via functional changes in the endothelium–smooth muscle complex.

Previous studies have shown associations between short-term exposure to PM_2.5 and nitric oxide (NO)-mediated endothelial dysfunction measured by using forearm plethysmography (4) or flow-mediated dilation (FMD) using brachial ultrasound ((5),6). In a study of healthy volunteers situated for 2 h at different urban bus stops, a 30-μg/m³ increase in PM_2.5 exposure corresponded to a 0.5% reduction in FMD (7). In contrast, other experimental studies have shown associations with vasoconstriction measured as a decrease in baseline arterial diameter (BAD) but not FMD ((8),9). These studies provide different conclusions: PM_2.5 exposure may primarily affect NO-mediated endothelial dysfunction or it may alter elaboration of vasoconstrictors.

Observations from short-term exposure studies provide only limited insight into effects of pollutants and mechanisms. Are the vascular responses transient or do lasting effects accrue after long-term exposure? We hypothesized that repeated short-term insults to the vasculature result in persistent endothelial dysfunction related to long-term exposures. To test this hypothesis, we investigated the relationship between PM_2.5 exposures and changes in the brachial artery in the MESA (Multi-Ethnic Study of Atherosclerosis) cohort. Leveraging the detailed exposure assignment of the MESA Air Pollution (MESA Air) study, we examined whether long- and short-term exposure to PM_2.5 is associated with decreased FMD and/or decreased BAD in this cohort.

MESA is a prospective study designed to investigate the prevalence and progression of subclinical atherosclerosis in participants aged 45 to 84 years who were free of clinical cardiovascular disease at the time of enrollment (10). The study includes 6,814 participants of white, African American, Hispanic, and Chinese descent from 6 US communities. The institutional review boards of each study site approved this study, and all participants gave written informed consent. Demographic characteristics, medical history, anthropometry, laboratory data, and brachial ultrasound measurements for the current analysis were taken at the first examination (July 2000 to August 2002). For this study, we included all participants with reliable brachial ultrasound measurements for whom complete covariate information and exposure estimates were available (N = 3,040).

The brachial examination procedure, inclusion and exclusion criteria, and reproducibility measures are reported in detail elsewhere (11). Briefly, participants were examined in the supine position after at least a 6-h fast and 15 min of rest. A standard blood pressure cuff was positioned around the right arm, 2 inches below the antecubital fossa, and the brachial artery was imaged 5 to 9 cm above the antecubital fossa by using a linear-array multifrequency transducer operating at 9 MHz and the LOGIQ 700 ultrasound (General Electric Logic 700 Device General Electric Medical Systems, Waukesha, Wisconsin). After the baseline images were obtained continuously for 30 s to obtain the BAD, the cuff was inflated to 50 mm Hg for 5 min above the participant's systolic blood pressure (SBP). After cuff deflation, the maximum arterial diameter (MAD) was measured. Image analysis was performed at the Wake Forest University Cardiology Image Processing Laboratory by using a previously validated semiautomated system that uses media-adventitial interfaces (12).

BAD (in millimeters) and FMD% were our main outcome measures and are based on recent guidelines (13). The number of sonographers acquiring images varied between the cities, ranging from 1 in St. Paul to 9 in Chicago. Each sonographer underwent central training, performed a supervised examination, and acquired at least 5 examinations acceptable to the core laboratory before certification. Intrareader reproducibility for BAD and FMD, evaluated by comparing an original and a blinded quality control reread of ultrasounds from 40 participants, was reported previously (11).

Estimates of long-term average ambient PM_2.5 concentrations were computed for each participant by using a hierarchical spatio-temporal model fit with an estimation procedure described elsewhere ((14),(15),16). Monitoring data came from the U.S. Environmental Protection Agency's Air Quality System regulatory monitoring stations supplemented by several fixed-site monitoring stations in each MESA community and home-monitoring at 10% of MESA participant homes, in campaigns specifically deployed for the MESA Air cohort (17). The hierarchical model decomposed the space-time field of concentrations into 3 components: 1) spatially varying long-term averages; 2) spatially varying seasonal and long-term trends; and 3) spatially correlated but temporally independent residuals. The spatially varying long-term averages and seasonal trends were modeled by using land-use regression and spatial smoothing of residuals using universal kriging. A large suite of spatial covariates such as proximity to major roadways and local land-use were used as predictors in the universal kriging models, after dimension reduction by partial least-squares. The spatially correlated but temporally independent residuals were predicted by using ordinary kriging. These components were combined to predict concentrations at each subject's home location for every 2-week period starting in January 1999. An average of exposures for the calendar year 2000 were used for this analysis. We used ArcGIS 9.1 software (Esri, Redlands, California); Dynamap/2000 street network and geocoding database (TeleAtlas, Boston, Massachusetts) were used to determine participants' residential addresses.

We assessed each participant's short-term exposure to PM_2.5 concentrations on the day of the examination (day 0), 1 day prior, 2 days prior, and the average of days 0 to 2. Daily average PM_2.5 concentrations were derived from an air quality system central-site monitoring station with complete daily average information available for the study city during the examination period (14).

A number of covariates were considered in the multivariate analysis. These included age, gender, ethnicity, education, income, body surface area, smoking status, alcohol consumption, dietary fat intake, emotional distress derived from a standardized anxiety scale, and physical activity. We also adjusted for waist to hip ratio, blood glucose, SBP, diastolic blood pressure, high-density lipoprotein, total cholesterol, triglycerides, homocysteine, fibrinogen, and C-reactive protein. We also adjusted for specific medications, including anti-inflammatory agents, antihypertensives, lipid-lowering medications, and antioxidants (vitamin C).

Using Stata version 10.1 (Stata Corp., College Station, Texas) and SAS version 9.1.3 (SAS Institute, Inc., Cary, North Carolina), summary statistics were calculated and analyses performed. We tested for differences between the full cohort and those with complete data by using a t test for continuous covariates and chi-square test for the categorical covariates. Linear regression modeling was performed to examine associations between PM_2.5 estimates and the main vascular outcomes. Because the sonographers vary by study location and can influence these outcomes, the analyses were adjusted by using an indicator variable for sonographer. Each sonographer worked in only 1 study site; in effect, our analysis controls for study site as well.

The ratio outcome (FMD%) requires careful statistical consideration because the denominator (BAD) may itself be associated with the exposure. In our primary approach, we included 1/BAD as a covariate in regression models to obtain unbiased effect estimates and to increase precision (18). We also evaluated the FMD% outcome without adjusting for 1/BAD and have referred to it as “FMD% without adjustment for 1/BAD.” In a sensitivity analysis, we also used simple extent of dilation (“FMDmm”), calculated as MAD-BAD, as reported in the Framingham Heart Study ((13),19). Results are reported per interquartile range (IQR) change in PM_2.5.

The role of all the aforementioned covariates, including secondhand smoke exposures, family history of myocardial infarction, serum cotinine, and forced expiratory volume in 1 s, were examined. However, these were not included in the final model due to a large amount of missing data.

To control for temporal and meteorological confounding in short-term analyses, B-splines were used for city-specific trends in calendar time (12 df/year), temperature (6 df), and relative humidity (6 df) and included a city-specific day of the week indicator. In addition, long- and short-term exposures were evaluated jointly, including the temporal and meteorological confounders.

In another sensitivity analysis, the final model was not controlled for sonographer (or city) to determine between-city estimates. In separate city-specific analyses, the influence of within-city exposure contrasts was estimated in the final model controlled for sonographer. Differential susceptibility was investigated by stratifying for age categories, gender, ethnicity, diabetes mellitus status, hypertension categories (using the criteria of the Seventh Report of the Joint National Committee on Prevention, Detection, Evaluation, and Treatment of High Blood Pressure, as derived from the blood pressure variables from examination 1) (20), antihypertensive use, obesity (obese if body mass index [BMI] ≥30 kg/m²), smoking status, and residential stability (people who lived in the same address for ≥5 versus <5 years' duration). In addition, we tested for interaction for age, BMI, SBP, and diastolic blood pressure as continuous variables.

The effect of lipid-lowering medications and the antihypertensive medications classified as angiotensin-converting enzyme (ACE) inhibitors, angiotensin receptor blockers, beta-blockers, calcium channel blockers, diuretics, and vasodilators was investigated. Analysis was conducted by using stratification and by evaluating effect modification through interaction terms for the respective categories.

Of the 6,814 MESA participants, only a subset of images from the brachial ultrasound examinations was analyzed due to quality control and funding reasons and also only in 5 of the 6 MESA study sites. Similar to the main cohort, our study population of 3,040 subjects included 50% female patients and 18% with less than a high school education. The exclusion of 1 site resulted in a lower proportion of African Americans and higher proportion of Chinese Americans than in the overall MESA cohort. Fifteen percent of the study population used lipid-lowering drugs, and 34% were treated for hypertension (Table 1). Approximately 21% of the participants had moved residential location within 5 years before their brachial examination. Estimated long-term PM_2.5 concentrations ranged from 10.6 to 24.7 μg/m³, with an IQR of 3 μg/m³. Short-term PM_2.5 concentrations ranged from 1 to 74 μg/m³, with an IQR of 12 μg/m³ (Figure 73_gr1). The mean BAD and FMD% in this cohort were 4.3 ± 0.8 mm and 4.4 ± 2.8%, respectively (Table 2).

We found a significant inverse association between long-term PM_2.5 concentrations and FMD% but not with BAD. For every 3-μg/m³ increase in the annual average of PM_2.5, FMD% decreased by 0.3% (95% CI: −0.6 to −0.03; p = 0.03) independent of cardiovascular risk factors ((Table 3),Figure 73_gr2). A weaker but highly statistically significant relationship was observed between long-term PM_2.5 concentrations and FMD% (−0.1% [95% CI: −0.2 to −0.04]; p = 0.005) without adjustment for sonographer or city (6). Exclusion of sonographers who performed 1 to 2 examinations did not affect these estimates. Restricting analysis to those with a stable residential address for ≥5 years did not affect the association (6). In sensitivity analyses using the FMD% without adjustment for 1/BAD (−0.3% [95% CI: −0.5 to 0.01]; p = 0.06) or the simple FMDmm (−0.01 mm [95% CI: −0.02 to 0.001]; p = 0.07), this negative association persisted.

Short-term PM_2.5 concentrations were associated with a small but not statistically significant reduction in FMD% (−0.1% [95% CI: −0.2 to 0.04]; p = 0.4) and BAD (−0.01 mm [95% CI: −0.05 to 0.01]; p = 0.4) for the day 1 before examination when adjusted for risk factors, seasonality, and meteorology (Table 4). Associations with long-term PM_2.5 exposure did not significantly change when short-term exposure estimates or temperature or season were included in the model (6).

Models stratified according to city demonstrated that the relationship between long-term PM_2.5 concentration and reduced FMD% was consistently negative across the 5 communities, except for St. Paul, which had the lowest mean PM_2.5 concentrations. Chicago participants showed the largest decrease in FMD% per IQR increase in PM_2.5 (Table 3). Similar effect estimates were noted when sonographer was not included as a covariate.

A significant interaction was found between age and the association between long-term PM_2.5 and the FMD% (p < 0.001) but not for BMI or blood pressure. Younger age was associated with a larger magnitude of effect of PM_2.5 on reduced endothelial function (FMD% −0.5% [95% CI: −0.8 to −0.2%]; p = 0.001). Stratified analyses suggest that women, never smokers, participants with stages 1 and 2 hypertension, and those not taking antihypertensive medications tend to have a slightly greater negative association of PM_2.5 with FMD% (6). Although no category of medication usage demonstrated statistically significant effect modification, the association between long-term PM_2.5 and FMD% seemed to be ameliorated by ACE inhibitor use (n = 163); there was a 0.3% increase (95% CI: −1.04 to 1.6) in FMD% per 3-μg/m³ annual increase in PM_2.5 (6).

In this large, multisite, multiethnic cohort, a significant association was observed between long-term residential PM_2.5 concentrations and NO-mediated endothelial dysfunction as assessed by FMD, independent of major cardiovascular risk factors. Although previous studies ((5),(6),21) have assessed the relation between short-term exposure to PM_2.5 and endothelial function, this is the first investigation of the relation between long-term PM_2.5 exposure and endothelial function. Our study provides unique insights into the mechanisms of long-term PM_2.5 exposure and increased cardiovascular mortality.

Increasing age, SBP, BMI, and smoking are inversely related to FMD (19). The magnitude of the long-term effect of an IQR (difference between the 25th and 75th percentile) increase in PM_2.5 on FMD% is comparable to the effect of 5 years' increase in age, or of active tobacco smoking, in this population. A 3-μg/m³ increase in PM_2.5 concentration occurs as a contrast in residential exposure between less polluted and more polluted areas in most major U.S. metropolitan areas. Taken together, these data suggest that PM_2.5 exerts a clinically relevant degree of effect on endothelial dysfunction.

Endothelial dysfunction measured by using FMD is a precursor for atherogenesis (22) and is associated with hypertension (23), passive smoking (24), and cardiovascular events ((11),25). Although the role of FMD as a predictor is still controversial, it is consistently associated with future cardiovascular events (26). In a nested case-cohort study using MESA data, abnormal FMD was predictive of incident cardiovascular events independent of other major cardiovascular risk factors (11). Furthermore, a recent meta-analysis using pooled data from 14 studies (including 5,447 participants) found that a 1% decrease in FMD% was associated with an 8% increase in cardiovascular mortality independent of risk factors (26). In our study, a 10 μg/m³ annual increase in PM_2.5 concentration was associated with a 1% decrease in FMD% (regardless of the adjustment for 1/BAD). Although we report our outcomes for an IQR increase in the PM_2.5 contrast, the equivalent of a 10-μg/m³ exposure contrast has previously been associated with a 9% increase in cardiopulmonary mortality in the American Cancer Society Cancer Prevention Study II (27) and a 24% increase in cardiovascular events in the Women's Health Initiative Observational Study (28).

Our finding regarding short-term exposures, although not statistically significant, is still notable given the prior association with endothelial dysfunction in both experimental ((8),9) and observational ((5),21) studies. Our results of larger associations with long-term compared with short-term associations (one-tenth of the effect estimate for short-term compared with long-term) parallel the larger magnitude of effect on cardiovascular mortality noted in previous population-based studies ((27),(28),29). For example, a 10-μg/m³ increase in short-term exposure to particulate matter is typically associated with a 0.1% to 0.5% mortality increase in the large multicity studies (30) compared with 10% or greater mortality increases in long-term exposure cohort studies ((1),(27),28). The lack of more robust findings for short-term analysis may be due to limited statistical power or the simplified approach to short-term exposure estimation compared with long-term exposures.

Several mechanisms have been proposed to explain the potential cardiovascular risk associated with pollutants, involving oxidative stress, inflammation, and autonomic imbalance, each of which can affect the endothelium directly or indirectly. A recent study found an association between arterial stiffness and annual average concentrations of nitrogen dioxide and sulfur dioxide but not PM_2.5 (31). Reduced production and efficacy of NO in the vasculature is a hallmark of endothelial dysfunction. Previous studies in humans ((8),9) have shown vasoconstriction and altered FMD with PM_2.5 but typically on an acute time scale. Stimulation of the angiotensin-1 receptor (32) or uncoupling of endothelial NO synthase (33) or generation of highly reactive oxygen species via nicotinamide adenine dinucleotide phosphate (reduced) oxidase and Toll-like receptor pathways (34) and Rho kinase activation (35) have been proposed as the relevant vascular mechanisms of PM_2.5. We found that long-term exposure to PM_2.5 might produce chronic changes in the brachial artery that negatively affects its ability to react to shear stress, primarily from NO-mediated endothelial dysfunction.

Insights into potential mechanisms of PM_2.5 exposure on endothelial function are further suggested by a subgroup analysis analyzing antihypertensive medication use. Although limited by subgroup sizes, we found that the use of ACE inhibitors, but not other classes of antihypertensive agents (including angiotensin receptor blockers) or lipid-lowering drugs, seemed to abrogate the association between long-term PM_2.5 concentrations and FMD%. Elevated endothelial kinins and NO from ACE inhibition could play a role in mitigating PM_2.5-induced endothelial dysfunction ((36),37). Recently, we found that the genetic variation in the renin-angiotensin-aldosterone pathway seemed to modify the effect of traffic-related air pollution on increased left ventricular mass in the MESA cohort (38).

Major strengths of our study include availability of vascular functional measurements in a large multiethnic cohort with high-quality control standards and excellent data on covariates. We took advantage of the specialized monitoring and sophisticated exposure models from MESA Air to predict spatially resolved, individual-specific estimates at each participant's home (15).

Our study has several limitations. Most importantly, this is a cross-sectional evaluation of images collected on 1 occasion; important information might be gained by a longitudinal study of vascular function. Our findings also may not be generalizable to either normal younger individuals or those with recognized clinical disease. It should be noted that although our long-term exposure estimates are derived from sophisticated exposure modeling methods, they are subject to measurement error and may not fully reflect an individual's long-term exposure because information on specific microenvironments, time-activity patterns, or periods of exposure >1 year before testing is lacking. Future MESA Air exposure metrics will be able to incorporate individual-level data on pollutant infiltration efficiencies and time spent indoors, which will improve these estimates.

FMD is a commonly applied marker for NO-mediated endothelial dysfunction (39). We found consistent associations between PM_2.5 and reduced NO-mediated endothelial dysfunction, whether expressed as FMD% (with or without adjustment for 1/BAD) or the simple change in FMDmm. Because MESA did not include nitroglycerin administration, we could not rule out NO-independent endothelial dysfunction in this analysis.

This is the first epidemiological study to suggest that long-term exposure to PM_2.5 is associated with decreased endothelial function in a conduit artery independent of cardiovascular risk factors. This finding provides a clue that long-term perturbations of the endothelium–smooth muscle complex from PM_2.5 exposure lead to chronic functional changes. These functional changes may partly explain the risk of cardiovascular events previously associated with these environmental exposures.