Short-term exposure to air pollution and biomarkers of cardiovascular effect: a repeated measures study

Abstract

To help understand the pathophysiologic mechanisms linking air pollutants and cardiovascular disease (CVD), we employed a repeated measures design to investigate the associations of four short-term air pollution exposures – particulate matter less than 2.5 micrometers in diameter (PM_2.5), nitrogen dioxide (NO₂), ozone (O₃) and sulfur dioxide (SO₂), with two blood markers involved in vascular effects of oxidative stress, soluble lectin-like oxidized LDL receptor-1 (sLOX-1) and nitrite, using data from the Multi-Ethnic Study of Atherosclerosis (MESA). Seven hundred and forty participants with plasma sLOX-1 and nitrite measurements at three exams between 2002 and 2007 were included. Daily PM_2.5, NO₂, O₃ and SO₂ zero to seven days prior to blood draw were estimated from central monitors in six MESA regions, pre-adjusted using site-specific splines of meteorology and temporal trends, and an indicator for day of the week. Unconstrained distributed lag generalized estimating equations were used to estimate net effects over eight days with adjustment for sociodemographic and behavioral factors. The results showed that higher short-term concentrations of PM_2.5, but not other pollutants, were associated with increased sLOX-1 analyzed both as a continuous outcome (percent change per interquartile increase: 16.36%, 95%CI: 0.1-35.26%) and dichotomized at the median (odds ratio per interquartile increase: 1.21, 95%CI: 1.01-1.44). The findings were not meaningfully changed after adjustment for additional covariates or in several sensitivity analyses. Pollutant concentrations were not associated with nitrite levels. This study extends earlier experimental findings of increased sLOX-1 levels following PM inhalation to a much larger population and at ambient concentrations. In light of its known mechanistic role in promoting vascular disease, sLOX-1 may be a suitable translational biomarker linking air pollutant exposures and cardiovascular outcomes.

Graphical Abstract

Capsule summary

In an elderly population, higher short-term PM_2.5 concentrations were associated with increased sLOX-1, a translational biomarker linking air pollutant exposures and cardiovascular outcomes.

Introduction

Heart disease is the leading cause of death in the United States, responsible for one in four deaths each year (“Heart Disease Facts & Statistics ∣ cdc.gov,” 2018). Population studies have reported associations between ambient air pollutants and cardiovascular disease (CVD) (Beelen et al., 2014; Cosselman et al., 2015; Gill et al., 2011; Patel et al., 2016). Understanding the mechanisms by which air pollution affects the cardiovascular system could point the way to preventing adverse impacts. At least three different but potentially overlapping mechanisms of particulate matter (PM) cardiovascular effects have been proposed: 1) pulmonary inflammation leading to systemic inflammation, 2) modulation of autonomic influences through pulmonary irritant receptors, and 3) direct target organ effects of pollutants or their products that pass into the systemic circulation (Brook, 2007; Brook et al., 2010; Franklin et al., 2015; Rajagopalan et al., 2018).

The concept that pulmonary inflammation leads to systemic inflammation is more complex than previously appreciated, given recent findings of serum-borne inflammatory bioactivity. Pulmonary responses to PM and certain gases activate endogenous peptidases and promote oxidative modification of endogenous phospholipids and proteins (Aragon et al., 2017; Kampfrath et al., 2011; Mostovenko et al., 2019; Mumaw et al., 2016; Robertson et al., 2013). Reaction by-products (modified phospholipids, peptides) enter the circulation to drive endothelial cell activation via cell surface pattern recognition and scavenger receptors, such as lectin-like oxidized low density lipoprotein (oxLDL) receptor 1 (LOX-1), CD36 and toll-like receptors, ultimately promoting vascular inflammatory outcomes including recruitment of leukocytes, reduced endothelial barrier integrity, and release of chemokines (Aragon et al., 2017; Kampfrath et al., 2011; Lund et al., 2011; Mostovenko et al., 2019; Mumaw et al., 2016; Rao et al., 2014; Robertson et al., 2013). Plasma soluble LOX-1 (sLOX-1) reflects LOX-1 activation and cleavage by peptidases, such as ADAM10 (Mitsuoka et al., 2009). Previous studies have shown that sLOX-1 is a sensitive and specific biomarker for several CVDs (Jin and Cong, 2019; Kume et al., 2010; Pirillo and Catapano, 2013). Lund et al (2011) originally identified a putative role for LOX-1 in driving systemic vascular inflammation in response to inhaled emissions (Lund et al., 2011). Using controlled experimental inhalation exposures in humans and mice, vehicle emissions were found to induce LOX-1 and sLOX-1 in the aorta and circulation, while antibody-based inhibition of LOX-1 reduced vascular oxidative stress in response to those exposures (Lund et al., 2011).

In another proposed mechanism of cardiovascular effects, air pollution exposure results in increased plasma nitrite (Langrish et al., 2013). A disturbance in either production or availability of nitric oxide (NO) is thought to be responsible for the functional alterations seen in endothelial dysfunction, and plays a key role in the development of atherosclerotic lesions (Cannon, 1998; Libby et al., 2002; Ross, 1999; Shimokawa, 1999; Toborek and Kaiser, 1999). Plasma nitrite concentrations reflect acute changes in endothelial NO synthase (eNOS) activity and NO production (Kleinbongard et al., 2006). In an intervention study, Laumbach et al. (2014) reported acute increases in plasma nitrite concentrations in healthy human subjects exposed to traffic emissions from the New Jersey Turnpike (Laumbach et al., 2014). The same group reported increases in plasma nitrite concentrations in college students coincident with short-term increases in ambient particulate matter smaller than 2.5 micrometers in diameter (PM_2.5) (Gandhi et al., 2014).

In the current study, we exploit the strengths of a repeated measures design in a population-based multi-ethnic study of older adults – the Multi-Ethnic Study of Atherosclerosis (MESA), and investigate the associations between short-term air pollution exposures and two blood markers of vascular effect. Each participant had repeated blood measurements of sLOX-1 and nitrite at three separate exams, and we assessed 24-hr averaged outdoor air pollutant concentrations at several lagged periods prior to each blood draw for four agents: PM_2.5, nitrogen dioxide (NO₂), ozone (O₃) and sulfur dioxide (SO₂). We hypothesize that higher pollutant concentrations concurrent with or prior to the blood collection would be associated with higher levels of plasma sLOX-1 and/or nitrite. Our findings may serve to enhance our understanding of the pathophysiologic mechanisms linking ambient air pollution and cardiovascular disease.

Methods

Study population

MESA is a prospective cohort study aiming to examine the progression of subclinical and clinical cardiovascular disease in an elderly population initially free of clinical CVD. The MESA Air Pollution Study (MESA Air) is an ancillary study to MESA that added additional study participants and obtained detailed air pollution data. A diverse, population-based sample of 6,814 asymptomatic men and women aged 45-84 were recruited at baseline from six U.S. communities, including Baltimore, MD, Chicago, IL, Winston-Salem, NC, Los Angeles, CA, New York, NY and St. Paul, MN. From 2002 through 2012, five follow-up exams were conducted, and each examination included a blood draw, anthropometric measurements, and collection of questionnaire data. Details of the sampling, recruitment and data collection have been described elsewhere (Bild et al., 2002). For this study, 750 participants were randomly selected from those who submitted blood samples at Exam 2 (September 2002-February 2004), Exam 3 (March 2004-September 2005) and Exam 4 (September 2005-May 2007). Those without a valid address history were excluded (N=10). Institutional review board approval was granted at each study site, and written informed consent was obtained from participants.

Outcomes

sLOX-1 and nitrite were measured in plasma samples at each of the three exams. Blood was drawn after 12 hours of fasting. Assays were performed in the Laboratory for Clinical Biochemistry Research at the University of Vermont. sLOX-1 was measured using a DuoSet ELISA kit (R&D Systems); two human plasma-based quality control samples run over the course of the study yielded coefficients of variation (CVs) of 11.5% (mean value: 110 pg/ml) and 18.5% (2,753 pg/ml). Most sLOX-1 samples were tested with a CV close to 11.5%. Nitrite was measured using the Cayman Chemical Nitrate/Nitrite Assay Kit in which Griess reagents were used to convert nitrite into a deep purple azo compound, whose absorbance was measured photometrically to determine nitrite concentration. Quality control samples run over the course of the study yielded CVs of 10.3% (23 umol/L) and 7.4% (106.1 umol/L).

Air pollution exposure estimation

Daily mean concentrations of PM_2.5, NO₂, O₂ and SO₂ 0-7 days prior to blood draw at each visit were obtained from monitoring stations in central locations of the six study regions. As some study sites only had one monitor operational in certain time periods, to apply a universal definition of the exposure source, we chose the long-running monitoring sites which were the most central to the spatial distribution of study participants in each region and had the least missing data. The distances from each participant’s residence to the central monitoring site were all within 100 kilometers, with 80-97% within 25 kilometers depending on study sites.

Covariates

Individual-level information was obtained largely from self-reported questionnaires. Non-time varying covariates included sex, race/ethnicity (White/Caucasian, Chinese-American, African-American or Hispanic), site (Baltimore, Chicago, Winston-Salem, Los Angeles, New York or St. Paul), study exam (2, 3 or 4), household income (<$25,000, $25,000-$49,999, or ≥$50,000), educational level (high school or less, college attendance, graduate or professional schooling), smoking status (never, former or current smokers), second-hand smoking exposure (hours/week), medication use (e.g., anti-inflammatory medications such as aspirin or NSAIDs), and medical histories of diabetes (normal blood glucose, impaired fasting glucose or diabetes) and hypertension (yes/no). Time-varying covariates besides those used in the air pollutant pre-adjustment (see Statistical analysis, below) included age (years) and body mass index (BMI, kg/m²). Neighborhood socio-economic status (SES) was specified as a summary score of neighborhood disadvantage derived by principal component analysis in each visit, based on percentage with bachelor’s degree, managerial occupation, high school education, interest/dividend/rental income, and household income greater than $50,000, as well as median home value and median household income, as described elsewhere (Diez Roux et al., 2006). A higher score indicated lower neighborhood SES.

Statistical analysis

Daily air pollution data of PM_2.5, NO₂, O₃ and SO₂ were extracted for all of the MESA Air study regions over the entire 4.5-year period covering the three study exams. To control for temporal confounding, the short-term air pollutant concentrations were pre-adjusted using site-specific splines for calendar time (6 degrees of freedom [df]/year), temperature (4 df), relative humidity (4 df) and an indicator variable for day of the week. This approach makes use of all available air pollution data to increase precision for models assessing health effects of short-term pollutant exposures, rather than carrying out adjustment directly in the health models, which would only make use of the air pollution data shortly preceding each of the three exams for each subject (Szpiro et al., 2014). The daily air pollutant residuals from the pre-adjusted models 0-7 days prior to blood draw at each visit were assigned to each individual based on their home addresses and dates of exams. We set the undetectable values of both sLOX-1 and nitrite to be missing, and natural log transformed the detectable values because of right skewness. The respective air pollution effect estimates were scaled to their interquartile range (IQR): 7.60 μg/m³ for PM_2.5, 7.12 ppb for NO₂, 7.95 ppb for O₃ and 1.95 ppb for SO₂. As two thirds of the sLOX-1 samples were undetected, we further categorized all sLOX-1 measurements into two levels split at the median of detectable values at each exam.

Both descriptive and inferential analyses were performed. We used generalized estimating equations (GEE) as the primary statistical approach to account for within-subject correlation of the repeated measures (Diggle, 1994; LIANG and ZEGER, 1986). An exchangeable working correlation matrix was specified, assuming that the measurements had no logical ordering over time. In an exploratory analysis, we assessed the independent association of each air pollutant with plasma nitrite and sLOX-1 on the day of (lag 0) and on each of seven days before the blood draw. To estimate the cumulative or net effect of air pollution exposures across the week before blood measurements, we applied unconstrained distributed lag GEE (Schwartz, 2000). The mean model equation was:

$$\text{ln}(Y_{it})={\beta }_0+{\beta }^{\ast }\left(w_0{X}_{it}+..+w_7{X}_{i(t-7)}\right)+Z_{it}{\beta }_2+{\epsilon }_i,$$

where ln(Y_it) was the natural log transformed biomarker, X_i(t–n) was the time-varying short-term air pollution exposure for subject i, Z_it was the vector of covariates measured at either baseline (time-invariant) or time-varying for subject i at exam t, and ε_i was the error term. β* was the sum of the coefficients of air pollutants on days 0-7 (w₀-w₇), and its exponential form minus one [(e^β* – 1) * 100%] was interpreted as the net effect of one IQR increase in the air pollutant concentration over day lags 0-7 on percentage change in sLOX-1 or nitrite. For binary sLOX-1, GEE with the logit link function was performed to estimate the associations between air pollution exposures over lagged days and sLOX-1 levels above the median of detectable values. We developed a directed acyclic graph (DAG) to inform our thinking about the role of the various covariates, based partly on existing literature on CVD risk factors. A hierarchical adjustment approach of four models was used to assess sensitivity of the estimates of effect to covariate selection. Model 1 was adjusted for only study site. Model 2 additionally included age, sex and race/ethnicity. Model 3 was considered the full model, with additions to Model 2 of individual SES variables of household income and educational level, neighborhood SES, behavioral factors of second-hand smoking, smoking status and use of NSAIDs or aspirin. Model 4 was an extended model and further controlled for BMI and history of diabetes and hypertension.

Several sensitivity analyses were performed. First, the degrees of freedom were doubled for the site-specific time trend in the air pollutant pre-adjustment models from six per year to twelve per year. Second, we included PM_2.5 with either NO₂ or O₃ in separate two-pollutant models. Third, we repeated the analysis with O₃ after applying an additional exclusion to remove the three study sites (Winston-Salem, Baltimore and St. Paul) with only seasonal operation of O₃ monitors. Fourth, we excluded from the analysis those who lived more than 25 kilometers away from the central monitors. Fifth, to evaluate potential confounding by long-term air pollution exposures, we further controlled for the one-year averages before each visit in the fully adjusted models (Model 3) using central monitor measurements at each site that had at least 200 days of daily measurements. In addition, we performed mixed effects tobit regression with a random intercept to handle the censoring in sLOX-1 measurements and further verify our findings. Lastly, an interaction term between the neighborhood SES disadvantage summary score and study site was added to the full models to handle the potential non-linearity induced by neighborhood SES. The analyses were conducted in STATA 15 (StataCorp, Texas, the U.S.) and RStudio (R Cord Team, Vienna, Austria); a two-tailed p-value of ≤ 0.05 was considered statistically significant.

Results

The study sample of 740 MESA participants had an average age at baseline of 63.3 years (range from 46 to 89) and an approximately equal number of men and women (Table 1). As with the overall MESA sample, these participants were racially diverse, with more White (39.6%) and fewer Chinese American (12.0%) participants. They were relatively well-educated with approximately 40% having bachelor’s degree or above. More than half were current or former smokers. One-third had impaired fasting glucose or diabetes and approximately half had a history of hypertension.

Table 1.

Baseline characteristics of participants (n=740)

Variables	N	Mean (SD) / %
Study site
Winston-Salem	121	16.4%
New York	130	17.6%
Baltimore	90	12.2%
St. Paul	116	15.7%
Chicago	148	20%
Los Angeles	135	18.2%
Socio-Demographics
Female	361	48.8%
Age at Exam 2 in years	740	63.3 (10.1)
Race/ethnicity
White, Caucasian	291	39.3%
Chinese American	89	12.0%
African American	204	27.6%
Hispanic	156	21.1%
Annual household income
<$25,000	203	27.9%
$25,000-$49,999	233	32%
>$50,000	292	40.1%
Highest Educational Levels
High school or less	238	32.2%
College attendance	356	48.1%
Graduate or professional schooling	146	19.7%
Behavioral Risk Factors
Smoking history
Never	330	44.8%
Former	340	46.2%
Current	66	9%
Second-hand smoking exposure (hours/week)	690	3.1 (10.3)
Anthropometric Measurements
BMI at Exam 2, kg2/m	740	28.4 (5.6)
Medical History
Diabetes
Impaired fasting glucose	147	19.9%
Diabetes	131	17.7%
Hypertension	382	51.6%
Medicine Use
Ever on regular aspirin medication	364	49.2%
Ever on regular NSAID medication other than aspirin	193	26.1%

Abbreviations: Body Mass Index, BMI; Nonsteroidal anti-inflammatory drug, NSAID

Of the 2,220 nitrite measurements (median 24.7 umol/L), four were outside the range of quantification (three below the minimum detection of 6 umol/L and one above the maximum detection of 400 umol/L) (Table 2). For sLOX-1 (median 94.5 pg/ml), on the other hand, 60% (n=1,322) were below the minimum detection of 7 pg/ml and 5% (n=108) above the maximum detection of 5000 pg/ml. At each exam, 22-23% of the participants had sLOX-1 measurements above the median of detectable values. Compared to participants with at least one undetectable sLOX-1 measurement, those with all three detectable sLOX-1 measurements were more likely to be white and non-smokers, and less likely to have diabetes or hypertension (Appendix Table 1). The two groups were otherwise similar with respect to other characteristics. The distributions of all unadjusted air pollutant concentrations are shown in Figure 1; all were right-skewed. Residuals for each pollutant from the pre-adjustment models controlling for meteorology, day of week and temporal trend, however, were distributed normally with mean zero.

Table 2.

Distribution of Biomarkers by Exam (Total 2220 tests in 740 participants)

Biomarkers	Exam	< Min detectable value (N)	> Max detectable value (N)	N	Mean (SD)	Min	Median	Max
Nitrite (umol/L)	Overall	3	1	2,216	31.6 (23.6)	6.6	24.7	347
	2	1	1	738	30.9 (24.4)	6.8	23.3	269.1
	3	1	0	739	31.7 (21.6)	6.6	24.9	162.4
	4	1	0	739	32.3 (24.9)	7.01	25.78	347
sLOX-1 (pg/ml)	Overall	1322	108	790	264.0 (399.8)	7.9	94.5	1993.7
	2	437	38	265	264.0 (395.2)	8.0	91.7	1896.4
	3	436	35	269	261.0 (405.4)	7.9	94.7	1972.9
	4	449	35	256	267.1 (400.1)	7.9	97.1	1993.7

Figure 1. Boxplots of Pollutant Concentrations before Pre-adjustment by Exam.

Shown are boxplots of the distribution of combined pollutant concentrations 0-7 days prior to blood draw at each visit before pre-adjustment in each of the six MESA regions. Note that each distribution is right-skewed.

A DAG showing the exposure-outcome associations is presented in Appendix Figure 1. Effect estimates of each individual day (from 0-7 days before blood draw) from GEE for both sLOX-1 and nitrite, unadjusted for effects at other lags, are shown in Appendix Figure 2. Results of the primary analysis from the unconstrained distributed lag GEE of the net effects over day lags 0-7 are shown in Table 3; effect estimates over all lagged days are shown in Appendix Figure 3. Among participants with detectable sLOX-1, estimates of net effect were increased for all four pollutants, but were larger and only evident with PM_2.5. With the fully adjusted model (Model 3, see Statistical analysis), there was an estimated 16.36% (percent change: 16.36%, 95%CI: 0.1-35.26%) net increase in sLOX-1 for each one IQR increase in PM_2.5 over day lags 0-7. When sLOX-1 was analyzed as a binary variable, each one-IQR increase of PM_2.5 over day lags 0-7 was associated with a 21% (odds ratio: 1.21, 95%CI: 1.01-1.44) net increase in the odds of sLOX-1 above the median of detectable values. There was little evidence of increased net effect of the other three air pollutants on sLOX-1. Effect estimates with nitrite were uniformly null for all four pollutants.

Table 3.

Estimated net effects over eight days of short-term increases in air pollutant concentrations on the biomarkers from fully adjusted unconstrained distributed lag GEE

	Nitrite (Continuous)		sLOX-1 (Continuous)		sLOX-1 (Binary)
Air pollutiona	% Change	95%CI	% Change	95%CI	Odds ratio	95%CI
PM2.5 0-7 day
Model 1	−0.53%	(−5.72%, 4.95%)	16.69%	(0.43%, 35.58%)	1.20	(1.02, 1.42)
Model 2	−0.33%	(−5.38%, 4.99%)	17.05%	(0.5%, 36.32%)	1.21	(1.02, 1.43)
Model 3	−1.41%	(−6.62%, 4.10%)	16.36%	(0.10%, 35.26%)	1.21	(1.01, 1.45)
Model 4	−1.38%	(−6.64%, 4.18%)	17.14%	(0.84%, 36.09%)	1.23	(1.02, 1.47)
NO2 0-7 day
Model 1	1.54%	(−3.31%, 6.63%)	11.61%	(−0.63%, 25.35%)	1.04	(0.95, 1.14)
Model 2	1.53%	(−3.20%, 6.48%)	11.54%	(−0.68%, 25.26%)	1.04	(0.95, 1.15)
Model 3	1.14%	(−3.78%, 6.30%)	10.49%	(−1.71%, 24.20%)	1.05	(0.95, 1.16)
Model 4	0.75%	(−4.15%, 5.90%)	10.78%	(−1.44%, 24.50%)	1.06	(0.95, 1.17)
O3 0-7 day
Model 1	−2.79%	(−8.82%, 3.63%)	6.40%	(−10.58%, 26.62%)	1.04	(0.91, 1.19)
Model 2	−3.90%	(−9.76%, 2.34%)	6.55%	(−10.27%, 26.52%)	1.04	(0.90, 1.20)
Model 3	−3.26%	(−9.56%, 3.48%)	8.62%	(−9.29%, 30.06%)	1.05	(0.91, 1.23)
Model 4	−3.52%	(−9.74%, 3.13%)	8.58%	(−9.31%, 30.00%)	1.05	(0.90, 1.22)
SO2 0-7 day
Model 1	1.62%	(−1.55%, 4.89%)	4.33%	(−2.94%, 12.14%)	1.02	(0.94, 1.10)
Model 2	1.81%	(−1.35%, 5.07%)	4.67%	(−2.56%, 12.43%)	1.02	(0.94, 1.10)
Model 3	2.18%	(−1.08%, 5.54%)	3.38%	(−4.46%, 11.86%)	1.00	(0.92, 1.08)
Model 4	1.93%	(−1.35%, 5.32%)	3.58%	(−4.21%, 12.00%)	1.00	(0.92, 1.09)

^a.Air pollutants were scaled to their respective interquartile range.

Model 1 adjusted for study site.

Model 2 additionally adjusted for age, sex and race/ethnicity.

Model 3 additionally adjusted for household income, educational level, neighborhood SES, second-hand smoking, smoking status and use of NSAIDs or aspirin

Model 4 additionally adjusted for BMI and history of diabetes and hypertension.

In sensitivity analyses, doubling the degrees of freedom for the site-specific time trends from 6 to 12 per year in the air pollutant pre-adjustment models did not attenuate the effect estimates for sLOX-1 and PM_2.5 appreciably (Appendix Table 2), nor when including PM_2.5 with either NO₂ or O₃ in the two-pollutant models (Appendix Figure 3). The estimated effects on both biomarkers remained largely null when the three study sites with only seasonal operation of O₃ monitors were excluded from the analysis, except that the O₃ point estimates approximately doubled for continuous sLOX-1 and remained non-significant (Appendix Table 3). Aside from the wider confidence intervals for the associations with O₃, no meaningful change in effect sizes and precision was shown in other pollutants after removing the participants living more than 25 kilometers away from the central monitors from the analysis (Appendix Table 4). Except for the inflated point estimate of O₃ with continuous sLOX-1, there were no appreciable changes in the fully adjusted model estimates for the other pollutants with a further control for one-year average air pollution concentrations (Appendix Table 5). The results of PM_2.5 and O₃ from the mixed effects tobit regressions with sLOX-1 were consistent with the findings from the primary analysis when censored data were set as missing (Appendix Table 6). However, the point estimates were greatly attenuated for NO₂, and the direction of association changed for SO₂, although their associations with sLOX-1 were not precise. The interaction between neighborhood SES and the study site was not statistically significant for most of the exposure-outcome associations except for NO₂ and nitrite (p interaction: 0.03), and including the term in the models did not impact the pollutant effect estimates (Appendix Table 7).

Discussion

Using data from a population cohort of older adults, we found that short-term increases in PM_2.5 concentrations were associated with elevated blood concentrations of sLOX-1, but not with increased blood nitrite.

The present study includes a number of important strengths. First, a longitudinal design with repeated measures of pollutant exposures and blood biomarkers, particularly on a large number of study participants, improves statistical efficiency and reduces the variance of the estimates of exposure effects (Sullivan, 2008). We focused our attention on the estimated net effects of exposure over a range of day lags rather than those of individual day lags given that individual day effects are often difficult to interpret and are subject to statistical variability. Second, using the complete exposure time series of pollutant concentrations to carry out the pre-adjustment of concentrations for time-varying factors provided more stable adjustments than would have been possible using only concentrations from around the time of the MESA exams. This approach enhanced our confidence that temporal trends and time-varying factors did not bias the health effect estimates. Third, participants were randomly selected from those in the cohort who provided data at each of three follow-up exams. The study was largely balanced by design with an equal and large number of participants providing data at each survey time. Fourth, the study exploited the strengths of the MESA cohort itself with its rigorous quality control procedures and good compliance (Bild et al., 2002).

This study demonstrates, now in a general population and at ambient concentrations, that sLOX-1 is increased following short-term increases in ambient PM concentrations. In controlled human diesel emissions exposure studies, exposure caused an increase in sLOX-1, and preclinical studies mechanistically tied the presence of circulating LOX-1 to oxidative changes in aortas in a vulnerable mouse model (Lund et al., 2011). Increased LOX-1 expression in aortas after O₃ exposure, with or without diesel PM, has also been reported in rat models (Kodavanti et al., 2011). Surface chemistry of the PM core has been documented in rats to play a role in regulating in vivo pulmonary toxicity responses, including LOX-1 and inducible NO synthase (Snow et al., 2014). Upon binding to oxidized LDL, LOX-1 augments atherosclerotic disease by increasing expression of adhesion molecules, downregulating nitric oxide production, and activating inflammatory responses through NF-κB (Kattoor et al., 2019). Importantly, numerous other ligands of the promiscuous LOX-1 receptor have been identified, including C-reactive protein, advanced glycation end-products and oxidatively modified lipids (Tian et al., 2019). The circulating form of sLOX-1 arises from peptidase cleavage, largely by ADAM10, which releases the extracellular LOX-1 domain into the circulation (Mitsuoka et al., 2009). sLOX-1 has also been shown to be elevated in various cardiovascular diseases and by cigarette smoking (Jin and Cong, 2019; Kume et al., 2010; Pirillo and Catapano, 2013; Takanabe-Mori et al., 2013). LOX-1, together with related scavenger and pattern-recognition receptors such as TLR4 and CD36, appears mechanistically involved in the systemic vascular effects of inhaled pollutants (Aragon et al., 2017; Kampfrath et al., 2011; Rao et al., 2014). Studies in mice have suggested that alterations in brain microvascular structure and integrity observed with mixed vehicle emission exposure may be mediated, at least in part, via LOX-1 signaling (Lucero et al., 2017; Suwannasual et al., 2018). While the link between inflammatory lung responses, circulating products and generation of sLOX-1 remains unclear, the present study provides evidence in support of this pathway with ambient levels of PM in a general population.

The second biomarker, blood nitrite, was employed to assess the role of NO, which is involved in the endothelial dysfunction that is central to the development of atherosclerosis. Any of the more thoroughly explored mechanisms of air pollution effects could potentially affect the production or availability of NO, and could therefore involve this mechanism. Our results did not provide support for this hypothesis, at least as reflected by blood nitrite and at the pollutant concentrations under study here. Our findings are not consistent with the randomized crossover study by Laumbach et al. (2014) in which an acute increase in plasma nitrite was observed in healthy human subjects exposed to traffic emissions on the New Jersey Turnpike (Laumbach et al., 2014). There are several possible reasons contributing to this difference, including the observational nature of our study, the much older age of our study participants, and the relatively lower ambient air pollution concentrations in the MESA six cities. The same group more recently reported findings similar to those from their experimental trial in an observational study based on college students (Gandhi et al., 2014). One preclinical study directly tied circulating nitrite levels to inhalation of nitric oxide, which is found in higher concentrations near roadways (Knuckles et al., 2011). Although our study shares certain similarities in study design and exposure levels with the observational study of college students, the age difference of the two study populations may again have contributed to the difference in findings.

Others have reported findings on biomarkers of air pollution cardiovascular effects in the MESA cohort. Hajat et al. (2015), also using a repeated measures design, investigated the associations of short and long-term air pollutant exposures with biomarkers of inflammation, endothelial activation and blood coagulation (Hajat et al., 2015). Increased PM_2.5 on the same day as the blood draw was associated with a small increase in E-selectin and possibly with increases in CRP and fibrinogen. That study, however, employed a more unbalanced design, with a larger number of study participants contributing data from only one exam than had data from multiple exams. Diez Roux et al. also assessed short-term PM_2.5 exposure associations on CRP using a cross-sectional design in MESA, but found no evidence of an association (Diez Roux et al., 2006). Again, as CRP is a confirmed ligand for LOX-1 and may actively increase vascular LOX-1 expression in a feed-forward manner, pollution-induced pathways may converge on endothelial cell scavenger and pattern recognition receptors to promote vascular disease (Kattoor et al., 2019).

Our study has some limitations. First, fully validated assays were not available when the biomarkers were measured. Due to the suboptimal sensitivity of the assays, analytical CVs were relatively high and about two-thirds of the measurements of sLOX-1 were below the limit of detection. This reduced our power to estimate the associations with continuous sLOX-1, and may have induced some degree of selection bias when the censored data in this measurement were set as missing in the primary analysis. To partly address this limitation, we first dichotomized sLOX-1 (including detectable and undetectable values) into high and low levels in the GEE models with the logit link function, and further performed mixed effects tobit regressions on a continuous scale to handle the censoring. The effects estimated from both analyses were generally in line with those from the primary analysis, providing some assurance that the results were not greatly biased. We also found no major differences between participants with at least one undetectable sLOX-1 measurement and those with all three detectable measurements. Nevertheless, the associations derived from the primary analysis need to be interpreted with caution. Second, the possibility of residual confounding persists in this observational study, although the within-person repeated measures analysis suggests that only confounders that vary over time are of concern. Pre-adjustment of air pollutant concentrations and inclusion of time-varying covariates further reduced time-varying sources of confounding. Previous studies also have reported individual and combined effects of road traffic noise and air pollution on increasing inflammatory biomarkers, but because we had no data on noise, we were unable to control for this potential confounder (Cai et al., 2016; Klompmaker et al., 2019). Third, we remain concerned about the potential for exposure measurement error, although the accuracy of shortterm, time-varying air pollution exposures using community-based regulatory monitors is relatively high for health analyses compared to long term exposures that rely on regulatory monitors and spatial contrasts (Zeger et al., 2000).

Conclusions

Using data from a medical research study of an elderly population in the U.S., the current analysis extends earlier experimental findings of increased circulating LOX-1 levels following PM inhalation to a much larger population and at ambient concentrations. In light of the known mechanistic role of sLOX-1 in promoting chronic vascular disease, the findings raise interest in sLOX-1 as a potential translational biomarker linking air pollution exposures and cardiovascular outcomes

Highlights.

1. Short-term increases in PM_2.5 were associated with increased sLOX-1 in an elderly population.

2. Nitrite was not associated with any of the four pollutants – PM_2.5, NO₂, O₃ and SO₂.

3. sLOX-1 has a known mechanistic role in promoting vascular inflammation and CVD.

4. The findings support the pathway linking PM_2.5 with vascular inflammation and CVD.

Abbreviations

CVD

cardiovascular disease

CV

coefficients of variation

DAG

directed acyclic graph

IQR

interquartile range

NO₂

nitrogen dioxide

O₃

ozone

PM_2.5

particulate matter less than 2.5 micrometers in diameter

SO₂

sulfur dioxide

SES

socio-economic status

sLOX-1

soluble lectin-like oxidized LDL receptor

MESA

the Multi-Ethnic Study of Atherosclerosis