Long-term exposure to fine particulate matter, residential proximity to major roads and measures of brain structure

Abstract

Background and Purpose

Long-term exposure to ambient air pollution is associated with cerebrovascular disease and cognitive impairment, but whether it is related to structural changes in the brain is not clear. We examined the associations between residential long-term exposure to ambient air pollution and markers of brain aging using magnetic resonance imaging (MRI).

Methods

Framingham Offspring Study participants who attended the seventh examination, were at least 60 years old and free of dementia and stroke were included. We evaluated associations between exposures (fine particulate matter (PM_2.5) and residential proximity to major roadways) and measures of total cerebral brain volume, hippocampal volume, white matter hyperintensity volume (log-transformed and extensive white matter hyperintensity volume for age) and covert brain infarcts. Models were adjusted for age, clinical covariates, indicators of socioeconomic position, and temporal trends.

Results

A 2 μg/m³ increase in PM_2.5 was associated with -0.32% (95%CI: -0.59, -0.05) smaller total cerebral brain volume and 1.46 (95%CI: 1.10, 1.94) higher odds of covert brain infarcts. Living further away from a major roadway was associated with 0.10 (95%CI: 0.01, 0.19) greater log-transformed white matter hyperintensity volume for an interquartile range difference in distance, but no clear pattern of association was observed for extensive white matter.

Conclusions

Exposure to elevated levels of PM_2.5 was associated with smaller total cerebral brain volume, a marker of age-associated brain atrophy, and with higher odds of covert brain infarcts. These findings suggest that air pollution is associated with insidious effects on structural brain aging even in dementia-and stroke-free persons.

Introduction

Long-term exposure to particulate air pollution has been associated with higher incidence of stroke¹ and impaired cognitive function in older adults.^{2, 3} Long-term exposures have also been associated with changes in cerebral hemodynamics,⁴ impaired microvascular reactivity⁵ and greater carotid atherosclerotic burden.⁶ Air pollution has been hypothesized to affect the central nervous system through activation of systemic inflammatory pathways and vascular dysfunction.⁷ Particulate air pollution is a pervasive component of urban and suburban ambient air pollution. Animal models have shown that particles can translocate from the nose via the olfactory nerve into the brain and evidence of these particles has been found in the striatum, frontal cortex, and cerebellum.⁸ However, it is not known whether long-term exposures to air pollution at urban background levels are related to measures of structural integrity and atrophy in the brains of older adults.

Magnetic Resonance Imaging (MRI) of the brain can detect early vascular impairment⁹ that is associated with subsequent risk of dementia and stroke.¹⁰ We therefore investigated the associations between exposure to fine particulate matter (PM_2.5) and residential proximity to major roadways with measures of total cerebral brain volume (TCBV), hippocampal volume (HV), white matter hyperintensity volume (WMHV), and covert brain infarcts (CBI) in the Framingham Offspring Study. We hypothesized that higher long-term exposure to ambient air pollution would be associated with subclinical damage as indicated by smaller TCBV and HV, larger WMHV and higher odds of CBI.

Materials and Methods

Study Participants

The design of the Framingham Offspring Study has been detailed previously.^{10, 11} Community dwelling participants living in the New England Region with no history of dementia, stroke, or transient ischemic attack who attended the seventh examination (1998–2001) and were aged ≥60 years at the time of MRI were eligible for inclusion in this study (n=943). All participants provided written informed consent, and the Institutional Review Boards at Beth Israel Deaconess Medical Center and Boston Medical Center approved the protocol.

Exposure Assessment

PM_2.5 satellite data

Participant primary addresses at the seventh examination were geocoded using ArcGIS 10 (ESRI, Redlands, CA) and census tract median household income was assigned (US Census 2000). Beginning in the year 2000, Moderate Resolution Imaging Spectroradiometer (MODIS) satellite-derived Aerosol Optical Density (AOD) measurements were used to predict daily PM_2.5 concentration at a 10×10 km spatial resolution across New England as previously described.^{12, 13}

Near roadway exposure

Residential proximity to the nearest A1, A2, or A3 roadway was determined by US Census Features Class in ArcGIS. We categorized proximity based on the following cutpoints: <50, 50 to <100, 100 to <200, 200 to <400, and 400 to <1,000 meters. We also evaluated the continuous association between the natural logarithm of proximity to a major roadway and neuroimaging outcomes because we have previously reported that this exposure and mortality were linearly associated.¹² Participants living further than 1,000 m from a major road in rural areas were excluded in primary analyses because the exposures of individuals living in exurb areas beyond 1,000 meters away from a major roadway are likely to be different from those of people living in urban and suburban areas.

Volumetric Brain MRI

TCBV, HV, and WMHV assessments and inter-rater reliability have been described previously.^14-17 Total cranial volume was determined by manual delineation of the intracranial vault and total brain parenchymal volume was determined by mathematical modeling. TCBV was then computed as a ratio of brain parenchymal volume to total cranial volume. The T2-weighted double spin-echo coronal sequences were acquired in 4-mm contiguous slices. Extensive WMHV (EXT-WMHV) was determined as a binary outcome by whether the log(WMHV/total cranial volume) was >1 SD above the age-adjusted mean in this cohort.^{10, 18} The presence of CBI was determined manually on the basis of size (≥3 mm), location, and characteristics of the lesions.¹⁹

Additional Covariates

History of cardiovascular disease was determined as previously described.²⁰ Prevalent diabetes was defined as a fasting glucose ≥126 mg/dL or oral hypoglycemic or insulin use at an examination or any previous history of diabetes (excluding gestational diabetes). Smoking status (never, current, former), pack-years smoked (<10 years, ≥10 years, missing), education (no high school, high school, some college, bachelor's or higher), and alcohol intake (0, 1-7 drinks per week, 7-14, 15 or more) were self-reported. Fasting homocysteine was measured in plasma. Systolic and diastolic seated blood pressures were calculated as the mean of two measurements taken during the clinical exam.

Statistical Methods

Linear and logistic regression models were used to evaluate continuous outcomes (TCBV, HV, and WMHV) and dichotomous outcomes (EXT-WMHV and CBI) respectively. We first adjusted for age at MRI, [age at MRI]², sex, time from exam 7 to MRI, median household income, date of MRI, smoking status, pack-years smoked, education, alcohol intake, and sine and cosine of MRI date to account for seasonal trends (Model 1). We then added covariates thought to be potential confounders that could also be mediators of the associations between particulate air pollution and brain structure including: natural logarithm of homocysteine, systolic blood pressure, diabetes, cardiovascular disease, history of atrial fibrillation, hypertension medications, and obesity (body mass index [BMI]≥30 kg/m²) (Model 2).

We tested whether observed associations differed by factors related to biological susceptibility and socioeconomic position as an evaluation of effect modification using cross-product terms for: sex, diabetes, obesity, current and former smoking (versus never smoking), and median household income < 25^th percentile (≤$44,901).

In sensitivity analyses, we accounted for clustering by census tract to further control confounding by socioeconomic position using generalized estimating equations with exchangeable working correlation matrix. Nonlinearity was evaluated using restricted cubic splines with knots at the 5^th, 27.5^th, 50^th, 72.5^th, and 95^th percentiles. We evaluated associations with PM_2.5 restricting to participants living within 1000 m of a major road. We also considered whether adjusting for CBI altered findings for associations between exposures of interest and TCBV.¹⁶

Models of PM_2.5 were scaled to 2 μg/m³, which approximates the interquartile range (IQR,1.7 μg/m³). We present the results of log-linear residential proximity to major roadway within 1000m analyses for an IQR. All analyses were performed in SAS version 9.3 or STATA Version 12. Plots were created using the POSTRCSPLINE package in Stata.²¹

Results

Table 1 shows population characteristics. The median (interquartile range (IQR)) of PM_2.5 exposure was 11.1 (1.7) μg/m³. Participants lived a median distance [25th to 75^th percentile] from a major road of 173 [48 to 415] meters (Table 2). The Spearman's rank correlation between the natural logarithm of residential distance from a major road and PM_2.5 was -0.15.

Table 1.

Population characteristics (n=943).

Characteristic	Median [IQR] or N(%)
Age at MRI (yr)	68 [9]
Systolic Blood Pressure (mmHg)	129 [25]
Diastolic Blood Pressure (mmHg)	72 [13]
Male	456 (48%)
Prevalent Cardiovascular Disease	130 (14%)
Smoking Status
Current	73 (8%)
Former	532 (56%)
Never	337 (36%)
Missing	1 (<1%)
Prevalent Heart Failure	46 (5%)
Diabetes	134 (14%)
Hypertension Medication Use	366 (39%)
Homocysteine	8.3 [3.1]
Education
Less than high school	51 (5%)
High school	311 (33%)
Some college /associate's degree	275 (29%)
Bachelor's or higher	298 (32%)
Missing Education	8 (<1%)
Median Household Income	63,479 [29,270]
Total Cerebral Brain Volume	78.41 [4.39]
Hippocampal Volume	0.33 [0.07]
Log White Matter Hyperintensity Volume	−2.69 [1.30]
Extensive White Matter Hyperintensity	135 (14%)
Covert Brain Infarcts	133 (14%)

Table 2.

Exposure characteristics

Exposure	Median [IQR] or n(%)	Range
PM2.5 (μg/m3)*	11.1 [1.7]	7.7–17.6
Distance to Major Road (m)†	173 [367]	0-993
Distance by category (m)
<50	226 (24%)
50 to <100	87 (9%)
100 to<200	149 (16%)
200 to<400	186 (20%)
400 to<1000	226 (24%)
≥1000	69 (7%)

^*Estimates unavailable on 13 participants.

^†69 participants living ≥1000 m from a major road excluded.

Higher PM_2.5 was associated with smaller TCBV and 1.46 times (95%CI: 1.10, 1.94) higher odds of CBI (Tables 3 and 4). A 2 μg/m³ increase in PM_2.5 was associated with a 0.32 unit difference in TCBV (95%CI: -0.59,-0.05). There was no clear pattern of association between PM_2.5 and HV, WMHV, or EXT-WMHV.

Table 3.

Associations between exposures and continuous volumetric outcomes.

	Model 1*				Model 2†
Outcome	Exposure	β	95%CI	n	β	95%CI	n
Hippocampal Volume	<50	0.004	(−0.005, 0.014)	865	0.005	(−0.004, 0.015)	853
	50 to <100	0.003	(−0.011, 0.016)		0.003	(−0.010, 0.017)
	100 to <200	0.005	(−0.006, 0.016)		0.005	(−0.006, 0.016)
	200 to <400	0.003	(−0.007, 0.013)		0.003	(−0.007, 0.013)
	400 to <1000	REF			REF
	Log(distance)‡	−0.002	(−0.007, 0.003)	865	0.003	(−0.008, 0.002)	853
	PM2.5§	0.0001	(−0.005, 0.005)	921	0.0004	(−0.005, 0.005)	909
Log (White Matter Hyperintensities)	<50	−0.17	(−0.35, 0.01)	873	−0.16	(−0.34, 0.02)	861
	50 to <100	−0.13	(−0.37, 0.11)		−0.12	(−0.36, 0.11
	100 to <200	−0.07	(−0.27, 0.13)		−0.06	(−0.26, 0.13)
	200 to <400	0.02	(−0.16, 0.21)		0.03	(−0.15, 0.22)
	400 to <1000	REF			REF
	Log(distance)‡	0.10	(0.010, 0.19)	873	0.09	(−0.0004, 0.18)	861
	PM2.5§	−0.06	(−0.16, 0.03)	929	−0.08	(−0.17, 0.01)	917
Total Cerebral Brain Volume	<50	0.23	(−0.29, 0.75)	873	0.09	(−0.42, 0.60)	861
	50 to <100	0.45	(−0.24, 1.14)		0.47	(−0.21, 1.16)
	100 to <200	−0.02	(−0.60, 0.56)		−0.14	(−0.71, 0.43)
	200 to <400	−0.19	(−0.73, 0.35)		−0.30	(−0.84, 0.23)
	400 to <1000	REF			REF
	Log(distance)‡	−0.15	(−0.41, 0.11)	873	−0.10	(−0.36, 0.16)	861
	PM2.5§	−0.32	(−0.59, −0.05)	929	−0.26	(−0.53, 0.004)	917

^*Model 1 adjusted for age, age , sex, time from exam 7 to MRI, median household income, date of MRI, smoking status, pack-years smoked, education (no high school, high school, some college, bachelor's or higher), drinking categories and sine and cosine of MRI date to account for seasonal trends.

^†Model 2 adjusted for model 1 covariates + log (homocysteine), systolic blood pressure, diabetes, cardiovascular disease, history of AF, hypertension medications, and obesity.

^‡Scaled to difference between 25^th and 75^th percentile of distance (367 m).

^§Scaled to 2μg/m³ difference in PM_2.5.

Table 4.

Associations between exposures and binary outcome

	Model 1*				Model 2†
Outcome	Exposure	OR	95%CI	n	OR	95%CI	n
Extensive White Matter Hyperintensity Volume for Age	<50	0.88	(0.51, 1.52)	873	0.94	(0.53, 1.67)	861
	50 to <100	0.56	(0.25, 1.29)		0.59	(0.25, 1.38)
	100 to <200	0.86	(0.46, 1.60)		0.96	(0.51, 1.83)
	200 to <400	1.05	(0.60, 1.84)		1.16	(0.65, 2.06)
	400 to <1000	REF				REF
	Log(distance)‡	1.11	(0.84, 1.48)	873	1.09	(0.81, 1.47)	861
	PM2.5§	1.00	(0.76, 1.32)	929	0.94	(0.70, 1.26)	917
Covert Brain Infarcts	<50	1.21	(0.67, 2.17)	870	1.29	(0.70, 2.36)	861
	50 to <100	1.25	(0.58, 2.67)		1.16	(0.53, 2.56)
	100 to <200	1.17	(0.61, 2.23)		1.10	(0.56, 2.15)
	200 to <400	1.69	(0.95, 3.00)		1.72	(0.95, 3.11)
	400 to <1000	REF			REF
	Log(distance)‡	1.05	(0.79, 1.40)	870	1.02	(0.75, 1.37)	861
	PM2.5§	1.46	(1.10, 1.94)	926	1.37	(1.02, 1.85)	917

^*Model 1 adjusted for age, age , sex, time from exam 7 to MRI, median household income, date of MRI, smoking status, pack-years smoked, education (no high school, high school, some college, bachelor's or higher), drinking categories and sine and cosine of MRI date to account for seasonal trends.

^†Model 2 adjusted for model 1 covariates + log (homocysteine), systolic blood pressure, diabetes, cardiovascular disease, history of AF, hypertension medications, and obesity.

^‡Scaled to difference between 25^th and 75^th percentile of distance (367 m).

^§Scaled to 2μg/m³ difference in PM_2.5.

An IQR difference in residential proximity to a major road was associated with 0.10 (95%CI: 0.01, 0.19) higher WMHV. A similar pattern was observed with categories of distance. However, there was no evidence of an association between proximity to a major road and EXT-WMHV, nor was it associated with TCBV, HV, or CBI.

There was no evidence of effect modification for the observed associations by sex, diabetes diagnosis, obesity, smoking, or median income below the 25^th percentile.

Only the association between PM_2.5 and TCBV met criteria for significant deviation from linearity, suggesting a stronger association at lower exposure levels and wide confidence intervals at high levels (Supplemental Figure I). We found no material differences when we considered restricting PM_2.5 analyses to regions within 1000 m of a major road or clustering by socioeconomic position. Adjustment for covert brain injury did not alter the association between PM_2.5 and TCBV.

Discussion

In this study, we observed that an increase in PM_2.5 within the range of exposures observed at urban and suburban background levels in New England was associated with smaller -0.32 (95%CI: -0.59, -0.05) TCBV and with 1.46 times (95%CI: 1.10, 1.94) higher odds of CBI. These findings support the hypothesis that higher long-term exposures to ambient air pollution are associated with structural changes in the brain that could precede cognitive impairment and overt cerebrovascular damage.^22-24

To our knowledge, there are no published studies of associations between air pollution and brain volume or CBI in older adults. Though CBI may appear asymptomatic, these small infarcts typically located in deep regions of the brain have been associated with neurological abnormalities, poorer cognitive function²⁵ onset of dementia,²⁴ and are thought to reflect small vessel disease.²⁶ Smaller TCBV has been associated subsequent stroke among Framingham Offspring participants,²⁷ and also with poorer performance on tests of attention, executive and visuospatial function.¹⁶ The magnitude of association that we observed for a 2 μg/m³ increase in PM_2.5 was similar to approximately one year of brain aging computed as the ratio of the coefficients for PM_2.5 and age in the model. Adjustment for CBI did not alter this association, suggesting that atrophy was independent of the presence of asymptomatic injury and not merely a direct result of the presence of cerebral infarction.

The mechanisms through which air pollution may affect brain aging remain unclear, but systemic inflammation resulting from deposition of fine particles in alveoli is likely important. Upregulation of a pro-inflammatory state has been associated both with elevated risk of stroke^{7, 28} and cognitive decline.²⁹ Circulating levels of biomarkers indicative of systemic inflammation have been associated with lower brain volume.³⁰

Our findings are largely consistent with previous studies showing that long-term exposure to ambient air pollution is associated with vascular impairment.^{4, 5, 12} Several prior studies have reported associations between long-term pollution exposure and living close to major roads with incident stroke^{1, 31, 32} and poorer cognitive function in older adults.^{2, 3, 33} Living in a high air pollution region in Mexico City was associated with greater accumulation of β-amyloid-42 in the frontal cortex and hippocampus than living in a non-polluted area.³⁴

Though evidence regarding the associations between long-term air pollution exposures and white matter damage is limited, an ecological study in Mexico reported associations between higher levels of air pollution and white matter damage in children and dogs.³⁵ Our findings of a positive association between WMHV and living further from a major road but no association with EXTWMHV were unexpected. However, among Framingham Offspring Study participants, EXTWMHV was associated with poorer cognitive function¹⁸ and elevated risk of stroke¹⁰ but WMHV was not, suggesting a threshold for these associations. Different findings may have distinct underlying pathophysiologic mechanisms. To evaluate this will require additional studies in experimental models designed to address these questions.

There were some differences in associations we observed for PM_2.5 and residential proximity to a major roadway. While both capture features of long-term exposure to ambient air pollution, proximity is an integrated measure of exposure to traffic which includes vehicle emissions, noise, ultrafine particles, road dust, and gaseous pollutants such as nitrogen dioxide, carbon monoxide, and volatile organic compounds but does not specifically account for the intensity of traffic or meteorological conditions at a given location. In contrast, modeled PM_2.5 incorporates both locally and regionally generated air pollution. Hence they represent different aspects of ambient pollutant exposures.

Our study is not without limitations. Although we accounted for individual-level and area-level characteristics of socioeconomic position, there may be residual confounding. However, the results from our analysis taking spatial clustering into account were similar to our primary results and previous literature using modeled PM_2.5 has also shown that correlations between socioeconomic position and PM_2.5 are low within urban areas.³⁶ The Framingham Offspring population is comprised of mostly white participants. Therefore, these results may not be generalizable to other populations. Recent addresses were stable, with 91% of participants reporting the same address at the 6^th and 7^th examination cycles. Our PM_2.5 data are based on an average for the year 2001 similar to other large epidemiologic studies.^{12, 31} This approach limits the influence of secular trends in exposure, while capturing the spatial distribution of average PM_2.5.This study also has several strengths, including a relatively large, community-based sample, inclusion of both men and women, quantitative brain MRI, and individual-level estimates of exposures.

Summary

In conclusion, we observed evidence suggesting that long-term exposure to PM_2.5 is associated with lower TCBV and more CBI among a community-based sample of participants free of dementia and stroke. These findings suggest that relatively low urban background levels of particulate air pollution may contribute to the acceleration of atrophic changes and small vessel disease in older adults. Future studies will be necessary to confirm or refute these findings, extend the work to include longitudinal assessments, and to determine factors that mediate this association.