The Association of Long-Term Exposure to Particulate Matter Air Pollution with Brain MRI Findings: The ARIC Study

Common environmental pollutants may promote cognitive decline, cognitive impairment, and dementia. In particular, recent epidemiologic studies have reported that higher exposure to particulate air pollution is associated with increased risk of cognitive decline, cognitive impairment, and dementia (Power et al. 2016a; Tzivian et al. 2016; Xu et al. 2016). Although this body of work is highly suggestive, work linking air pollution to MRI markers of brain injury may provide mechanistic insight and would allay concerns about residual confounding by sociodemographic and socioeconomic characteristics that are common to studies of air pollution and cognition (Casanova et al. 2016; Chen et al. 2015; Wilker et al. 2015; Wilker et al. 2016). However, relatively little work has been done to examine the link between particulate air pollution and available markers of brain injury, and prior studies exclusively report on associations between recent air pollution exposures and markers of brain injury (Chen et al. 2015; Power et al. 2016a; Wilker et al. 2015). However, current brain health is a result of cumulative causes of brain injury that likely accumulate over decades, including aggregating proteins, ischemic injury, inflammation and oxidative stress, or exposure to toxins. As such, it is reasonable to expect that air pollution exposures over the prior years to decades may significantly contribute to current brain health. In addition, prior studies on air pollution and markers of brain injury are limited by lack of understanding of the selection process by which persons were selected for neuroimaging, which may lead to bias (Weuve et al. 2015).

To address these limitations, we conducted a study to quantify the association of long-term past exposure to particulate matter air pollution with MRI markers of neurodegeneration and subclinical cerebrovascular disease in older adults from the Atherosclerosis Risk in Communities Neurocognitive Study (ARIC-NCS). We hypothesized that long-term past exposure to particulate matter (PM) air pollution, specifically PM $<2.5\mathrm{\mu }\mathrm{m}$ in aerodynamic diameter (${\mathrm{PM}}_{2.5}$), would be associated with smaller total brain volumes, as atrophy is an etiologically nonspecific indicator of cumulative brain damage, and increased risk of subclinical cerebrovascular disease. We also considered associations with regional brain volumes, given focal atrophy may suggest that PM exposures contribute to the pathogenesis of specific neurodegenerative processes.

In total, 1,753 persons met our eligibility criteria and were included in the analyses. At the time of MRI, participants were on average 76 y old, 40% were male, and 45% had greater than a high school education. Table 1 provides demographic and clinical characteristics of the study participants, as well as information on our MRI outcomes by study site. Overall, the Minnesota site was the most affluent of the four sites, followed in order by North Carolina, Maryland, and Mississippi.

Of the four sites, Minnesota and Mississippi had the lowest PM exposures, while Maryland and North Carolina had the highest (Table 2). Variation in exposure to ${\mathrm{PM}}_{10}$ was generally larger than variation in exposure to ${\mathrm{PM}}_{2.5}$. Site-specific coefficients of variation for our exposure estimates ranged from 0.03 to $0.11{\mathrm{\mu }\mathrm{g}/\mathrm{m}}^3$ for ${\mathrm{PM}}_{10}$ and 0.02 to $0.10{\mathrm{\mu }\mathrm{g}/\mathrm{m}}^3$ for ${\mathrm{PM}}_{2.5}$.

As there was evidence of moderate to high heterogeneity ($I^2>40\%$) across sites when considering analyses of ${\mathrm{PM}}_{2.5}$ and brain volumes, we discuss both the site-specific and pooled analyses (Table 3). In the Minnesota site, higher ${\mathrm{PM}}_{2.5}$ exposures were generally associated with smaller total and regional brain volumes, with slightly stronger associations observed when considering the 1990–1998 exposure period compared to the 1999–2007 exposure period. This pattern was not observed in the other three sites. Results from the Maryland and North Carolina sites were consistently null. In the Mississippi site, there was some evidence to support a protective association between higher PM exposures and larger AD signature region, temporal lobe, and occipital lobe volumes, regardless of exposure period; associations with other regions were typically null. When site-specific associations were pooled via meta-analysis, the resulting effect estimates were generally null. However, consistently adverse associations between ${\mathrm{PM}}_{2.5}$ exposure from 1999–2007 and frontal lobe volumes across sites resulted in a small, marginally significant pooled association [beta: $-0.02$ SD units per $1\text{-}{\mathrm{\mu }\mathrm{g}/\mathrm{m}}^3$ higher exposure; 95% confidence interval (CI): $-0.04$, 0.00] Similarly, consistently adverse associations between higher ${\mathrm{PM}}_{2.5}$ exposures in all three time periods and smaller deep-gray volumes across the Minnesota, Maryland, and North Carolina sites resulted in small, marginally significant pooled associations (e.g., for mean ${\mathrm{PM}}_{2.5}$ from 1990–2007, beta: $-0.03$ SD units per $1\text{-}{\mathrm{\mu }\mathrm{g}/\mathrm{m}}^3$ higher exposure; 95% CI: $-0.08$, 0.00). The overall pattern of site-specific and combined results was similar across our sensitivity analyses, including analyses implementing inverse probability weighting (Tables S1 and S2) and those omitting use of sampling weights (Table S3).

Similarly, there was some evidence to suggest heterogeneity of association across sites when considering analyses of ${\mathrm{PM}}_{10}$ and brain volumes (Table 4). Site-specific analyses suggested adverse associations between higher mean ${\mathrm{PM}}_{10}$ over 1999–2007 or 1990–2007 and smaller total brain volumes, occipital lobe volumes, and deep-gray volumes in Minnesota (Table 4). As with the ${\mathrm{PM}}_{2.5}$ analyses, we also observed protective associations between higher long-term ${\mathrm{PM}}_{10}$ exposure and larger occipital lobe, temporal lobe, and AD signature region volumes in Mississippi. As with ${\mathrm{PM}}_{2.5}$, there was little evidence of an association between long-term ${\mathrm{PM}}_{10}$ exposure and total or regional brain volumes in pooled analyses, with the exception of an adverse association between higher mean ${\mathrm{PM}}_{10}$ in all three time periods and smaller deep-gray-region volumes (e.g., the ${\mathrm{PM}}_{10}$ 1990–2007 time period, beta: $-0.02$; 95% CI: $-0.04$, 0.00). As above, the overall pattern of site-specific and combined results was similar across sensitivity analyses, including analyses implementing inverse probability weighting (Tables S4 and S5) and those omitting use of sampling weights (Table S6).

When considering the relation between ${\mathrm{PM}}_{2.5}$ or ${\mathrm{PM}}_{10}$ and the presence of MRI markers of cerebrovascular disease, there was little statistical evidence of heterogeneity of association across the four sites; thus, we focused on the analyses pooling estimates from all four sites via meta-analysis. Overall, there was little conclusive evidence to support an association between higher exposure to ${\mathrm{PM}}_{2.5}$ or ${\mathrm{PM}}_{10}$ in any time period and the presence of MRI markers of cerebrovascular disease in pooled analyses combining all four sites (Tables S7 and S8). However, the odds ratios (ORs) for the pooled associations between a $1\text{-}{\mathrm{\mu }\mathrm{g}/\mathrm{m}}^3$-higher mean ${\mathrm{PM}}_{2.5}$ exposure and either lacunes or subcortical microbleeds were consistently in the range of 1.04 to 1.10, although these associations were not statistically significant. Similarly, although the ORs for the pooled associations between a $1\text{-}{\mathrm{\mu }\mathrm{g}/\mathrm{m}}^3$-higher mean ${\mathrm{PM}}_{10}$ exposure and microbleeds were consistently in the range of 1.04 to 1.05; these associations were also not statistically significant. Results from our sensitivity analyses were broadly consistent with our primary analysis findings, including analyses implementing inverse probability weighting or omitting use of sampling weights (Tables S9 to S14).

In pooled analyses combining all four sites, higher mean ${\mathrm{PM}}_{2.5}$ and ${\mathrm{PM}}_{10}$ exposures in the 5 to 20 y prior were associated with smaller deep-gray regional brain volumes and higher ${\mathrm{PM}}_{2.5}$ exposures 5–14 y prior were marginally associated with smaller frontal lobe volumes. We found little evidence in support of an association between higher long-term exposure to ${\mathrm{PM}}_{2.5}$ or ${\mathrm{PM}}_{10}$ over our three time periods of exposure and other brain volume measures or markers of cerebrovascular and small vessel disease in pooled analyses. However, there was evidence of significant heterogeneity in associations between PM and brain volumes by study site. When considering site-specific associations, we consistently observed smaller total and regional brain volumes with greater long-term exposure to ${\mathrm{PM}}_{2.5}$ in the Minnesota site, but not the other three sites. Throughout, where there was evidence of an association between PM exposure and brain volumes, the magnitude of these associations was similar to that seen in prior analyses in this sample, considering the association between midlife blood pressure and brain volumes. (Power et al. 2016b) For reference, the $-0.05$ to $-0.1$ SD unit effect size observed in the Minnesota site can be interpreted as loss of approximately 0.5% to 1% of regional brain volume.

Strengths of this study include the relatively large number of participants with MRI, our ability to use weighting to account for selection into the MRI subcohort in primary analyses and attrition from the baseline ARIC visit in sensitivity analyses, and consideration of long-term, cumulative past exposures. Our coefficients of variation for air pollution exposure estimates are similar to those calculated from other studies based in geographically constrained locations (albeit typically using shorter averaging periods) (Power et al. 2016a), suggesting that the variation in exposure at our four sites is similar to that found in other locations. However, the relatively small number of persons in each center limits our power to detect small true effects, systematically evaluate the potential for nonlinear associations, or assess effect modification by age or other personal factors. In addition, mild brain atrophy may have several root causes. Heterogeneity in the causes of neurodegeneration in our sample may contribute to the muted dose–response, especially if only a subset of the potential causes of neurodegeneration, including both neurodegenerative diseases and other sources of brain injury, are related to air pollution exposure. Similarly, our findings do not preclude the possibility of neurotoxic effects on the brain that are not captured by the considered MRI markers of brain injury; studies considering alternate markers (e.g., cortical thickness) may be useful. We did not consider associations with more recent exposures, or with cumulative exposures that include recent exposures. As such, we cannot comment on the relative importance of recent versus past exposures or whether recent exposures are an acceptable surrogate for long-term cumulative exposures. As with many recent studies of the health effects of air pollution, we used modeled exposure measures using residential address rather than personal exposure metrics, and we were unable to address the issue of indoor air pollution. Moreover, we cannot discount the possibility that regional variation in predictive accuracy of our model may complicate or invalidate comparison of site-specific effect estimates. Finally, we cannot exclude the possibility of chance findings.

There are a small number of reports considering the association between PM exposures and MRI-based measurements of brain structure or subclinical cerebrovascular disease. Collectively, including the current study, this body of literature fails to identify a consistent pattern of associations, as results are frequently null, with the few positive findings differing across studies. In a study nested within the Women’s Health Initiative Memory Study (WHIMS), ${\mathrm{PM}}_{2.5}$ air pollution exposures in 1999–2006 were not associated with gray-matter brain volumes assessed in 2005–2006 (Chen et al. 2015). However, higher ${\mathrm{PM}}_{2.5}$ exposures were associated with smaller, normal-appearing white matter brain volumes, with magnitudes of association of roughly 0.01-SD units volume per interquartile range increase in exposure. Additional analyses in WHIMS using a voxel-based approach found ${\mathrm{PM}}_{2.5}$ exposures in the 3 y prior to MRI were associated with areas of smaller cortical gray-matter and subcortical white-matter volumes (Casanova et al. 2016). Notably, the authors also report significant clusters of association whereby higher ${\mathrm{PM}}_{2.5}$ was associated with larger deep gray–matter nuclei volumes in WHIMS participants (Casanova et al. 2016), opposite to our own observations of associations between higher PM and smaller regional gray-matter volumes in ARIC participants. In participants from the Framingham Offspring Study who lived in the New England region, higher past-year ${\mathrm{PM}}_{2.5}$ exposure was associated with smaller total cerebral brain volumes and greater risk of covert brain infarcts, but not with WMH volumes, age-adjusted extensive WMH volumes, or hippocampal volumes (Wilker et al. 2015). Finally, in a study of participants from the Massachusetts Alzheimer’s Disease Research Center Longitudinal Cohort, there was no association between higher ${\mathrm{PM}}_{2.5}$ exposures in 2003 and either brain parenchymal fraction (a measure of brain atrophy) or the presence of microbleeds at an MRI between 2004 and 2010, while there was a protective association between higher ${\mathrm{PM}}_{2.5}$ exposure and smaller WMH volumes (Wilker et al. 2016).

Interestingly, studies of the relationship between air pollution and cognitive or related outcomes (e.g., MRI markers or neuropathology) that consider geographically localized samples are more likely to report null associations. In contrast, studies considering participants spread over larger geographic regions have been more likely to report adverse associations (Power et al. 2016a). We suggest several potential explanations. First, studies in geographically constrained locations are typically small, and the range of exposures tends to be smaller. Thus, such studies are likely underpowered to detect small effects. A meta-analytic approach, such as demonstrated here, for combining information about multiple small, geographically constrained studies in different locations can overcome this limitation without inducing concerns about strong or intractable confounding that may arise in pooled analyses. Second, studies with wider geographic distribution may be more susceptible to confounding by characteristics that vary regionally. As we have previously demonstrated elsewhere (Power et al. 2016a), it appears unlikely that residual confounding may fully account for the adverse findings in more geographically dispersed settings, given the characteristics such a confounder would have to have in order to fully account for previously observed associations. However, this possibility cannot be fully discounted, especially given evidence in this study that exposure and confounder distributions across sites do not always overlap. Thus, residual confounding may still lead to a biased estimate of the true association in more expansive settings when spatial confounding is strong and meta-analysis of site-specific associations is not used. Finally, it is possible that a focus on quantifying exposure based on particulate mass is contributing to this heterogeneity of findings. If specific PM species or other physical characteristics such as surface area confer the relevant toxic effect, geographically constrained studies may be studying the impact of less toxic exposures, while geographically broad studies may be capturing mixtures of these effects due to their larger study area. Our finding of heterogeneity in association across sites would support this hypothesis, and the finding of adverse associations in Minnesota but not the other three sites may be attributable not to chance, but to the relative toxicity of exposures. Future work will be needed to understand the drivers of the divide in findings between these two study types in order to establish a causal effect of air pollution on late-life brain health.

Another potential explanation for the finding of adverse associations between ${\mathrm{PM}}_{2.5}$ and brain volumes in the Minnesota site, but not the others, lies in the potential nonlinearity of the association. Minnesota had the lowest air pollution levels of the four sites, and previous studies have suggested a nonlinear relationship between PM and both total cerebral brain volume (Wilker et al. 2015) and cognitive function (Ailshire and Crimmins 2014; Oudin et al. 2015; Power et al. 2011), whereby the strongest associations were observed at the lowest levels of exposure. Given relatively small samples per site, we were not able to assess nonlinearity of exposure within site, but hope others may be able to follow up on this possibility in the future.

In conclusion, we found no associations between cumulative past PM exposure and MRI-based markers of cerebrovascular disease. Combining data across sites, higher past PM exposures were associated with smaller deep-gray volumes across sites, and higher ${\mathrm{PM}}_{2.5}$ in 1999–2007 was marginally associated with smaller frontal lobe volumes. When considering individual sites, higher ${\mathrm{PM}}_{2.5}$ exposures were associated with smaller brain volumes in the Minnesota site. Further work will be needed to replicate these findings and understand the sources of heterogeneity across sites, and will require consideration of a broader number of sites.

The ARIC study is carried out as a collaborative study supported by National Heart, Lung, and Blood Institute contracts (HHSN268201100005C, HHSN268201100006C, HHSN268201100007C, HHSN268201100008C, HHSN268201100009C, HHSN268201100010C, HHSN268201100011C, and HHSN268201100012C). Neurocognitive data is collected by U01 HL096812, HL096814, HL096899, HL096902, and HL096917, with previous brain MRI examinations funded by R01-HL70825. The sponsors had no role in the design and conduct of the study; collection management, analysis, and interpretation of the data; preparation review; or approval of the manuscript. The authors thank the staff and participants of the ARIC study for their important contributions.

Supplemental Material is available online (https://doi.org/10.1289/EHP2152).

R.F.G. is Associate Editor for Neurology. C.R.J. serves on a scientific advisory board for Eli Lilly & Company and receives research support from the NIH/NIA (R01-AG011378, U01-HL096917, U01-AG024904, RO1 AG041851, R01 AG37551, R01AG043392, U01-AG06786) and the Alexander Family Alzheimer’s Disease Research Professorship of the Mayo Foundation. All other authors declare they have no actual or potential competing financial interests.

Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.