Outdoor Air Pollution Is Related to Amygdala Subregion Volume and Apportionment in Early Adolescence

Abstract

Background

Outdoor air pollution exposure is associated with structural and functional brain differences and an increased risk for psychopathology. Although the neural mechanisms remain unclear, air pollutants may impact mental health by altering brain regions implicated in psychopathology, such as the amygdala. Here, we examined the association between ambient air pollution exposure and amygdala subregion volumes in 9- to 10-year-olds.

Methods

Cross-sectional data from 4473 (55.4% male) Adolescent Brain Cognitive Development (ABCD) Study participants were leveraged. Air pollution exposure was estimated based on each participant’s primary residential address. Using the CIT168 atlas, we quantified total amygdala and 9 subregion volumes from T1- and T2-weighted images. We investigated associations between criteria pollutants (i.e., fine particulate matter [PM_2.5], nitrogen dioxide, and ground-level ozone), 15 PM_2.5 components, and amygdala subregion volumes and relative volume fractions using both single-pollutant linear mixed-effects regression and partial least squares correlation (PLSC) co-exposure modeling approaches.

Results

No significant associations were detected using single-pollutant models. Rather, in examining mixtures of exposures with PLSC, 1 latent dimension (52% variance explained) captured a positive association between calcium and several basolateral subregions. Latent dimensions were also identified for amygdala relative volume fractions (ranging from 30% to 82% variance explained), with PM_2.5 and component co-exposure being associated with increases in lateral, but decreases in medial and central, relative volume fractions.

Conclusions

PM_2.5 and its components are associated with distinct amygdala differences, potentially playing a role in risk for adolescent mental health problems.

Plain Language Summary

Air pollution is known to harm human physical and mental health. In this study, we explored how air pollution affects the amygdala, a brain region responsible for emotional processing, in 4473 9- to 10-year-olds. While no direct associations were identified between air pollutants and total amygdala size, co-exposure to PM_2.5 and its associated components was associated with specific patterns of differences in amygdala subregions. These differences may be related to emotional development and contribute to risk for mental health concerns in adolescents.

Plain Language Summary

Air pollution is known to harm human physical and mental health. In this study, we explored how air pollution affects the amygdala, a brain region responsible for emotional processing, in 4473 9- to 10-year-olds. While no direct associations were identified between air pollutants and total amygdala size, co-exposure to PM_2.5 and its associated components was associated with specific patterns of differences in amygdala subregions. These differences may be related to emotional development and contribute to risk for mental health concerns in adolescents.

Nearly 1 in 5 children in the United States experience mental health concerns (1), with the average age of onset in adolescence, peaking at 14.5 years (2). While social, environmental, and genetic risk factors have been identified, the etiology of most psychopathologies remains unclear (3). Understanding the risk factors associated with the emergence of these conditions is critical for prevention and early intervention. In this regard, mounting evidence supports an association between outdoor air pollution exposure and risk for psychopathology, albeit with mixed findings depending on pollutants, exposure timing, and outcomes (4, 5, 6, 7). Air pollution is a ubiquitous mixture of chemicals, which is increasingly recognized as a pervasive neurotoxicant (8,9), and has been associated with many neurodevelopmental (10, 11, 12) and mental health (13, 14, 15) conditions. Several criteria pollutants are routinely monitored, including ground-level ozone (O₃), nitrogen dioxide (NO₂), and particulate matter (PM) (16). O₃ is produced through chemical reactions between nitrogen oxides and volatile organic compounds in the presence of ultraviolet light (17), whereas NO₂ is emitted by vehicles and power plants when fuel is burned (18). PM is composed of many particles and liquid droplets through chemical reactions between pollutants and other sources (19). Fine particulate matter (PM_2.5; <2.5 μm) and smaller particles may be particularly harmful for human health due to their ability to penetrate deeper into the lungs (20,21). The components in PM_2.5 can be further classified and have different environmental origins and chemical features (22). Studying criteria air pollutants together with PM components allows for a more robust understanding of the contributions of different pollution sources and the implications of distinct pollutants (23).

The brain mechanisms that link air pollution and risk for psychopathologies remain unclear. Human research has demonstrated that inhaled pollutants deposit in the nasal cavity and lungs where, depending on the pollutant, they can rapidly enter the bloodstream and circulate throughout the body (24). A study using a 3-dimensional cultured human brain model has shown that PM_2.5 can directly access the brain by crossing the blood-brain barrier (BBB), inducing BBB and neuronal damage, cell death, and neuroinflammation (25). Among PM components, heavy metals and organic compounds are emerging as particularly toxic for human health, as they have been shown to induce neuronal cell death, oxidative stress, and metal dyshomeostasis in the brain [(26, 27, 28); see reviews (29,30)]. Considering cellular neurotoxicity and epidemiological evidence that links air pollution to risk for mental health conditions, such as depression and anxiety (14,15,31, 32, 33, 34), investigation of whether exposure to air pollution may manifest in changes to brain regions involved in these conditions is warranted. In particular, the amygdala is a limbic structure that plays a role in emotion regulation, fear, and social behavior (35). Early adolescence marks a critical period of amygdala development, as it reaches peak total volume at 9 to 11 years (36), while the development of corticolimbic circuitry continues through adolescence (37,38). While neuroimaging research has typically investigated the amygdala as a singular structure, it is composed of functionally and cytoarchitecturally distinct subregions (39), which have been associated with emotional processing (40), anxiety (41, 42, 43), neurodevelopmental (41,44), and mood (45, 46, 47) disorders. However, few studies have examined associations between air pollution and amygdala structure during development, and they have yielded mixed findings. While prenatal PM exposure has been linked with smaller amygdala volumes in infants (48) and 9- to 12-year-olds (49), prenatal silicon exposure (i.e., PM component) was associated with larger amygdalae, while 2 other PM components (i.e., polycyclic aromatic hydrocarbons; organic carbon) were associated with smaller amygdalae at 9 to 12 years (50). Several other studies have failed to detect an association between PM exposure and amygdala volumes in youths (50, 51, 52, 53). These inconsistences could be due to differences in magnetic resonance imaging (MRI) acquisition and the specific pollutants assessed. No study has assessed the association between air pollution exposure and amygdala subregions or explored the impact of co-exposures to PM_2.5 components on amygdalae subregions.

The aim of the current cross-sectional study was to fill these gaps by examining associations between ambient air pollutants and amygdala subregion volumes in 9- to 10-year-olds. To accomplish this, we took a 2-pronged approach. First, we used single-pollutant linear mixed-effects (LME) regression to investigate whether each pollutant related to total amygdala volume. Next, considering that individuals are not exposed to pollutants in isolation (54), and individual amygdala subregion volumes are highly correlated, we implemented a multivariate modeling approach, known as partial least squares correlation (PLSC), to identify latent associations between co-exposure to pollutants and 9 amygdala subregion outcomes per hemisphere. Given the number of studies that have failed to find an association between air quality and amygdala volumes, we hypothesized that air pollutants might not be associated with total amygdala volume but rather that co-exposure to PM_2.5 components would be associated with distinct amygdala subregion differences at 9 to 10 years of age.

Methods and Materials

Study Design

The current study utilized a subsample of cross-sectional baseline data from the National Institute of Mental Health (NIMH) Data Archive annual 3.0 (raw imaging data) and 5.0 (all other data) releases (55,56) collected from the larger ABCD (Adolescent Brain Cognitive Development) Study. The ABCD Study implemented identical recruitment protocols to enroll 11,880 9- and 10-year-old children (mean age = 9.49 years; 48% female) from 21 sites between October 2016 and October 2018 across the United States in a 10-year longitudinal study (57, 58, 59). Participants were eligible if they were 9.0 to 10.99 years of age at the baseline visit and were fluent in English. Centralized institutional review board (IRB) approval was obtained from the University of California San Diego, and study sites obtained approval from their local IRB. Each child’s parent or legal guardian provided written consent, and each child provided written assent. A subset of 4473 participants with air pollution data, high-quality amygdala data from Siemens 3T scanners, and all covariates were included in the current study (see Figure S1 and Neuroimaging).

Air Pollution Exposure

Annual average concentrations of ambient air pollution exposure were estimated at the primary residential address of each child, as previously described (60). Daily estimates of PM_2.5 (μg/m³) and NO₂ (parts per billion [ppb]) and daily 8-hour maximums of O₃ (ppb) were derived at a 1-km² resolution using hybrid spatiotemporal models, which utilize satellite-based aerosol optical depth models, land-use regression, weather data, and chemical transport models (61, 62, 63). Monthly estimates of 15 PM_2.5 components (bromine [Br], calcium [Ca], copper [Cu], elemental carbon [EC], iron [Fe], potassium [K], ammonium [NH₄⁺], nickel [Ni], nitrate [NO₃⁻], organic carbon [OC], lead [Pb], silicon [Si], sulfate [SO₄²⁻], vanadium [V], and zinc [Zn]) were derived at a 50-m spatial resolution using hybrid spatiotemporal models, as previously described (64,65). Exposure estimates for criteria pollutants and PM_2.5 components were then averaged for the 2016 calendar year to correspond to the baseline enrollment period of the ABCD Study and assigned to the primary residential address of each child, provided by the caregiver at the baseline study visit.

Neuroimaging

As previously published, a harmonized data protocol was utilized across all ABCD Study sites (66). Raw T1-weighted (T1w) and T2-weighted (T2w) images were downloaded from the ABCD 3.0 release (NDA 3.0 data release 2023) (55); CIT168 atlas construction, validation, comparison with other atlases, and individual difference estimates are described elsewhere (67,68), as are descriptions of each subregion along with its successful application to estimate amygdala subregions in children and adolescents (69, 70, 71, 72). For each participant, we quantified in vivo probabilistic volumes for 9 amygdala subregions of interest per hemisphere (69,73) (Figure 1). T1w and T2w images were registered via the Human Connectome Project (HCP) minimal preprocessing pipeline (74). The CIT168 atlas, a high-resolution in vivo probabilistic atlas of human amygdala nuclear subregions, was implemented to segment each individual’s amygdala into 9 subregions (67,68). Probabilistic volumes were calculated for the left and right hemisphere total amygdala and 9 subregions: the lateral nucleus (LA), dorsal and intermediate divisions of the basolateral nucleus (BLDI), ventral division of the basolateral nucleus and paralaminar nucleus (BLVPL), basomedial nucleus (BM), central nucleus (CEN), cortical and medial nuclei (CMN), amygdala transition areas (ATA), amygdalostriatal transition area (ASTA), and anterior amygdala area (AAA).

Figure 1.

(A) Schematic of CIT168 amygdala subregion labels. Note: The anterior amygdala area (AAA) is not visible in this view. (B) Representative Adolescent Brain Cognitive Development (ABCD) Study participant segmentations using the CIT168 atlas. Three-dimensional image created from in vivo segmentation of a representative participant using the Quantitative Imaging Toolbox (111). ASTA, amygdalostriatal transition area; ATA, amygdala transition area; BLDI, basolateral dorsal and intermediate subdivision; BLVPL, basolateral ventral and paralaminar subdivision; BM, basomedial nucleus; CEN, central nucleus; CMN, cortical and medial nuclei; LA, lateral nucleus.

In cross-sectional analyses, brain volumes were adjusted using total intracranial volume (ICV) to compensate for differences in head size, as this enables comparisons of brain structures across individuals with different cranial sizes. Therefore, we adjusted for ICV in the total amygdala volume analyses and the amygdala subregion volume analyses. We also calculated amygdala subregion volumes normalized to the total amygdala volume, known as relative volume fractions (RVFs), by dividing each region’s probabilistic volume by the total probabilistic volume of the hemispheric amygdala (69, 70, 71, 72). RVFs aim not to correct for differences in head size but rather to understand whether there are relative size differences in amygdala subregions. By doing so, we aimed to address complementary questions. Analyses that use probabilistic volumes test whether air pollution exposure is associated with amygdala subregion volumes, whereas analyses that use amygdala RVFs investigate whether air pollution is related to differences in the relative subregion composition, or apportionment, of the amygdala (69) (see Supplemental Methods).

Given that the CIT168 atlas was developed and validated using Siemens MRI data (67,68), and large differences have been reported in the ABCD Study due to the scanner manufacturer (75,76), 7273 participants with Siemens MRI data collected from 13 ABCD sites were eligible for the current study (Table S1). Of these, 6525 met quality control criteria for image inclusion by ABCD, and 6449 successfully passed CIT168 amygdala segmentation. To ensure that estimates were reliable within individual amygdalae, we required each participant to have an intra-amygdala contrast-to-noise ratio (CNR) > 1.0 for T1w and T2w images [see (49) supplemental data]. A total of 4754 participants had a CNR > 1.0 in T1w or T2w images and therefore were considered to have high-quality amygdala segmentations (67,69,71). Due to the nature of PLSC, we were only able to include participants with complete listwise data (i.e., exposures, covariates, and brain data). From the 4754 participants with usable amygdala data, 281 participants were removed due to missingness (see Figure S1 for flow chart of participant selection and Table S2 for comparisons between the study sample, excluded participants, and the whole ABCD cohort).

Demographic Variables and Covariates

Covariates included in our analyses were selected based on previous literature and the construction of a directed acyclic graph (Figure S2) (77). Sociodemographic variables (see Supplemental Methods) such as race/ethnicity, total household income, highest household education, urbanicity, and parent-report perceived neighborhood safety scores were included due to known disparities in air pollution exposure and associations between these variables and amygdala size (78, 79, 80, 81). Measures of weekly physical activity and average daily screen time were included, as previously published (82), because they may relate to various neighborhood factors and influence time outdoors and thus exposure levels. We also included demographic factors and precision imaging variables such as the child’s age at scan, sex, body mass index (BMI), and handedness, as well as MRI scanner head coil and serial number, which account for site and scanner differences (83). Finally, ICV was included in all models using total amygdala or amygdala subregion volumes to account for variation in total cranial size.

Statistical Analyses

Linear Mixed Effects Modeling

Analyses were conducted using R version 4.3.2 (84). Thirty-six single-pollutant models were run to test the independent associations between air pollution and total amygdala volume (18 outdoor air pollutants for each hemisphere of total amygdala volume) (Supplemental Methods). Models included the same covariates but implemented site as a random effect in place of MRI serial number. False discovery rate correction was implemented to correct for multiple comparisons.

As part of the revision process, we conducted 2 sets of sensitivity analyses to test the robustness of our results (Supplemental Methods). First, because age and sex may not fully capture developmental status, we ran single-pollutant models identical to our original models except that puberty was substituted for age and sex. Next, to determine whether time spent outdoors (i.e., screen time and physical activity) related to other key variables, we ran 2 models testing their associations with all demographic variables (Tables S3 and S4) and assessed their correlations with all exposures (Figures S3 and S4). We performed a final set of sensitivity analyses excluding screen time and physical activity as covariates.

Partial Least Squares Correlation

PLSC analyses were conducted as previously described (82,85,86) (Supplemental Methods). PLSC is a multivariate statistical method that compares 2 multidimensional datasets with cross-correlated features (87) to identify patterns of shared covariance or latent dimensions. PLSC uses singular value decomposition on the correlation matrix of these data matrices to identify latent dimensions that capture the maximum shared covariance and the corresponding variables that contribute to these latent dimensions, explained by their loadings. Considering the multidimensionality and multicollinearity within both data matrices (i.e., air pollutants [Figures S5 and S6] and amygdala subregion volumes [Figure S7]), PLSC is an optimal statistical approach due to its ability to handle highly cross-correlated data. We ran 4 separate PLSC models, including 1) criteria pollutants and subregion absolute volumes, 2) PM_2.5 components and subregion absolute volumes, 3) criteria pollutants and amygdala RVFs, and 4) PM_2.5 components and amygdala RVFs. Before running PLSC, all components were converted to ng/m³ so that units were identical. Linear regression was applied to residualize the same set of covariates from both exposure and brain matrices for each model, and PLSC analysis was performed on the residuals.

To test the significant number of latent dimensions for each model, permutation testing was performed where data were resampled 10,000 times without replacement. The probability of significance was determined based on the number of times permuted singular values exceeded the observed singular value, and the percentage of variance explained was assessed visually using scree plots. The percentage of variance explained was used to interpret the strength of the association between the 2 matrices across each latent dimension (88). To determine which variables were important in driving these latent dimensions, bootstrap testing evaluated the robustness of salience loadings onto significant latent dimensions by resampling the data 10,000 times, leaving out one sample each time. Confidence bootstrap ratios were derived by dividing the mean of a variable’s bootstrapped distribution by its standard deviation. Bootstrap ratios exceeding 2.5 (p < .01) were considered statistically reliable and significant (89).

Results

Sociodemographic characteristics of the final study sample are included in Table 1 and Tables S1 and S2. The final sample had more male White participants who were older, had lower BMI z scores, were more physically active, and were from households with higher income and education and safer neighborhoods than the full ABCD Study and excluded sample (Table S2). Annual average exposures are presented in Table S5. PM_2.5, NO₂, and O₃ exposure levels were significantly lower than current Environmental Protection Agency (EPA) standards (PM_2.5: 7.47 μg/m³ vs. EPA standard 9 μg/m³, t_4,472 = −69.59, p < .001; NO₂: 19.3 ppb vs. EPA standard 53 ppb, t_4,472 = −363.19, p < .001; O₃ 8-hour maximum: 42.1 vs. EPA standard 70, t_4,472 = −418.23, p < .001). However, these concentrations are significantly higher than the World Health Organization (WHO) 2021 Air Quality Guidelines of 5 μg/m³ for PM_2.5 (t_4,472 = 111.86, p < .001) and 10 ppb for NO₂ (t_4,472 = 100.14, p < .001) and significantly lower than the 60-ppb peak season 8-hour maximum for O₃ (t_4,472 = −268.53, p < .001) (90). Descriptive statistics for total amygdala and subregion volumes and RVFs are provided in Table S6.

Table 1.

Demographics for the Study Sample (N = 4473).

	Values
Age, Years	9.98 (0.63)
Sex
Female	1996 (44.6%)
Male	2477 (55.4%)
Race/Ethnicity
Asian	63 (1.3%)
American Indian/Native American/Native Hawaiian	34 (0.8%)
Hispanic	807 (18.0%)
Multiracial	340 (7.6%)
Non-Hispanic Black	586 (13.1%)
Non-Hispanic White	2622 (58.6%)
Other Pacific Islander/other	–
Missing/refused	19 (0.4%)
Do not know	2
Household Income
<$50,000	1067 (23.9%)
≥$50,000 and <$100,000	1241 (27.7%)
≥$100,000	1847 (41.3%)
Do not know/refused	318 (7.1%)
Highest Parental Education
<High school diploma	129 (2.9%)
High school diploma/general educational development	390 (8.7%)
Some college/associate’s degree	1111 (24.8%)
Bachelor’s degree	1256 (28.1%)
Postgraduate degree	1581 (35.3%)
Missing/refused	6 (0.1%)
Urbanicity
Rural area	335 (7.5%)
Urban clusters	156 (3.5%)
Urbanized area	3982 (89.0%)
Pubertal Stage
Prepuberty	2259 (50.5%)
Early puberty	1032 (23.1%)
Midpuberty	972 (21.7%)
Late puberty	57 (1.3%)
Postpuberty	3
Missing	150 (3.4%)
BMI z Score	0.3 (1.2)
Weekly Physical Activity, Days	3.6 (2.3)
Average Daily Screen Time, Hours	3.0 (2.4)
Neighborhood Safety	3.9 (0.9)

Values are presented as n, n (%), or mean (SD). The parent-report Neighborhood Safety/Crime Survey is on a scale of 1 to 5, with 5 indicating a higher degree of perceived neighborhood safety. Pubertal stage is calculated by converting the caregiver-report Pubertal Development Scale to a Tanner-like pubertal category score (see Supplemental Methods). Race/ethnicity categories reported in the table were collapsed into smaller categories in the final analytic models; see Table S2 for categories used in analyses. The BMI z score is used here as a measure for childhood body weight status normalized for sex and age.

BMI, body mass index.

Linear Mixed Effects Modeling

Using LME modeling, no significant associations were identified between pollutants and total hemispheric amygdala volumes (Table S7). Sensitivity analyses that substituted pubertal development for age and sex (Table S8) and excluded physical activity and screen time (Table S9) revealed similarly null findings.

Partial Least Squares Correlation

Using PLSC, no significant latent dimensions were identified between exposure to criteria pollutants and amygdala subregion volumes (Figure S8). In contrast, PLSC identified one significant latent dimension between exposure to 15 PM_2.5 components and amygdala subregion volumes, explaining 52% of the variance (Figure S9). The most stable and reliable contributions to this latent variable were Ca and the bilateral LA, BLDI, and BLVPL and left BM and AAA (Figure 2), with higher Ca exposure being associated with larger volumes of these subregions.

Figure 2.

Variable loadings for the association between fine particulate matter components and amygdala subregion volumes. Variables passing bootstrap ratio threshold (2.5, p < .01) are displayed in color. Significant subregions mapped into 3-dimensional brain space using in vivo segmentation from a representative participant using the Quantitative Imaging Toolbox (111); red denotes subregions with positive loadings (enlargement). AAA, anterior amygdala area; ASTA, amygdalostriatal transition area; ATA, amygdala transition area; BLDI, basolateral dorsal and intermediate subdivision; BLVPL, basolateral ventral and paralaminar subdivision; BM, basomedial nucleus; Br, bromine; Ca, calcium; CEN, central nucleus; CMN, cortical and medial nuclei; Cu, copper; EC, elemental carbon; Fe, iron; K, potassium; L, left; LA, lateral nucleus; NH₄⁺, ammonium; Ni, nickel; NO₃⁻, nitrate; OC, organic carbon; Pb, lead; R, right; Si, silicon; SO₄²⁻, sulfate; V, vanadium; Zn, zinc.

PLSC analyses of amygdala apportionment revealed one significant latent dimension between criteria pollutants and amygdala subregion RVFs, which explained 82% of the variance (Figure S10). Within this significant dimension, PM_2.5 most strongly contributed to the association (Figure 3A). PLSC between PM_2.5 components and subregion RVFs identified 2 significant latent dimensions, which explained 39% and 30% of the variance, respectively (Figure S11). The strongest and most stable contributions to the first latent dimension were larger bilateral LA and smaller bilateral BM, right CEN, and left CMN RVFs (Figure 3B), suggesting that these amygdala subregions are central to the observed association between PM_2.5 component exposure and amygdala apportionment. The strongest contributions to the second latent dimension were K and OC, indicating that these 2 components drive the second latent association between exposure to the 15 PM_2.5 components and the relative composition of the amygdala (Figure 3B).

Figure 3.

Variable loadings for criteria air pollutants, fine particulate matter (PM_2.5) components, and amygdala subregion relative volume fractions. (A) Criteria pollutants; (B) PM_2.5 components. Variables passing bootstrap threshold (p < .01) are displayed in color. Significant subregions mapped into 3-dimensional brain space for visualization purposes using the Quantitative Imaging Toolbox (111); red denotes subregions with positive loadings (i.e., larger relative proportion with exposure), and blue denotes subregions with negative loadings (i.e., smaller relative proportion with exposure). AAA, anterior amygdala area; ASTA, amygdalostriatal transition area; ATA, amygdala transition area; BLDI, basolateral dorsal and intermediate subdivision; BLVPL, basolateral ventral and paralaminar subdivision; BM, basomedial nucleus; Br, bromine; Ca, calcium; CEN, central nucleus; CMN, cortical and medial nuclei; Cu, copper; EC, elemental carbon; Fe, iron; K, potassium; L, left; LA, lateral nucleus; NH₄⁺, ammonium; Ni, nickel; NO₂, nitrogen dioxide; NO₃⁻, nitrate; O₃, ground-level ozone; OC, organic carbon; Pb, lead; R, right; Si, silicon; SO₄²⁻, sulfate; V, vanadium; Zn, zinc.

Discussion

This study is the first to explore associations between ambient air pollution exposure and amygdala subregion morphology. We aimed to identify novel associations between exposure to criteria pollutants and PM_2.5 components and amygdala subregions in a large sample of preadolescents by implementing a 2-pronged approach: assessing individual exposures via univariate analyses and examining patterns of co-exposure using multivariate methods. Consistent with previous studies (50, 51, 52, 53), our results suggest that 1 year of exposure to any single pollutant in early adolescence does not relate to total amygdala volume at 9 to 10 years of age. Rather, co-exposure to PM_2.5 is associated with distinct amygdala subregion differences in early adolescence. Specifically, annual average exposure to PM_2.5 and 3 PM_2.5 components—Ca, K, and OC—is associated with distinct increases in basolateral volumes and differences in amygdala subregion apportionment at ages 9 to 10. This work highlights the importance of using multipollutant analyses to model co-exposure to air pollutants given that individuals are not exposed to any pollutant in isolation (54,91).

While total amygdala volumes increase through early adolescence (36,92), human postmortem (93) and neuroimaging studies (41,42,46) suggest heterogeneity in amygdala subregion development during childhood and adolescence. Previous research shows nuclei-specific changes in amygdala neuron numbers (93,94) and age-related differences in amygdala RVFs (69), which may explain why findings from previous studies investigating associations between air pollution and total amygdala volume have been mixed depending on the timing of exposure and MRI. This study, being the first to examine air pollution and amygdala subregions, supports a growing body of literature using multipollutant models to link outdoor air pollution exposure, specifically PM_2.5 and its components, with amygdala structure and function during development. Using multipollutant models, higher prenatal exposure to coarse PM and lower exposure to NO₂ have been associated with smaller total amygdala volumes in infants from the United Kingdom (48). In a second study, results from multipollutant models demonstrated that prenatal exposure to Si—a PM component in dust that often coincides with Ca—was related to larger total amygdala volumes, whereas prenatal polycyclic aromatic hydrocarbon and OC exposure was associated with smaller amygdala volumes in 9- to 12-year-olds from the Netherlands (50). Inconsistencies across these studies may stem from differences in the child's age at MRI, geography, demographic variables, and pollutants examined (95). The current study highlights the importance of studying amygdala subregions, revealing associations between childhood air pollution exposure (PM_2.5 and its components, specifically) and amygdala subregion volumes at ages 9 to 10 years. Accounting for co-exposure to NO₂ and O₃, PM_2.5 was associated with unique and robust subregional amygdala volume patterns. Notably, one robust and reliable relationship was that higher Ca exposure—a PM component largely found in dust—was associated with a larger basolateral complex of the amygdala (BLA) (i.e., LA, BLDI, BLVPL, BM) and the AAA, which borders the BLA along a thin band of projection fibers (67). Furthermore, exposure to PM_2.5 and its components was strongly related to patterns of amygdala subregion apportionment. Specifically, exposure to PM_2.5 components was associated with proportionally larger lateral LA but smaller medial (i.e., BM, CMN) and central subregions. A second association was noted between K and OC, attributes of biomass burning, and overall patterns of amygdala apportionment. These findings suggest that exposure to PM_2.5 components is associated with notable increases in the size of certain amygdala subregions and may also impact the relative composition, or apportionment, of amygdala nuclei. Taken together with our previous findings showing childhood PM_2.5 exposure and longitudinal changes in resting-state functional connectivity of the amygdala and large-scale networks from ages 9 to 13 years in the ABCD Study (96), the current study further suggests that outdoor air quality may impact amygdala neurocircuitry during adolescence.

Although additional research is needed, it is plausible that the observed associations between air pollution exposure and amygdala subregions may have long-term implications for emotional processing and risk for mental health concerns. The BLA is the main thalamic sensory and cortical input region of the amygdala, involved in emotional regulation and processing (97, 98, 99). The BLA plays a key role in conditioned fear and stress responses (97) and has been associated with an individual’s susceptibility to anxiety (100). The BM, which connects the LA and CEN, plays a significant role in the suppression of stress and fear responses, particularly in the context of social anxiety (101,102). The CEN receives intrinsic connections and is one of the major output nuclei of the amygdala; the CMN is another recipient of projections, particularly from the BLA, CEN, and olfactory bulb (67). Considering these roles in anxiety, fear conditioning, and social cognition (103), additional investigations are needed to determine whether our identified subregion patterns play a role in the underlying neural mechanisms that link air pollution to risk for psychopathologies. However, the causal delay between the neurotoxicant effects of air pollution and observable behavioral differences is an ongoing challenge. While studies have identified positive associations between exposure to PM_2.5 and anxiety and depression symptoms (14,15,31,32), some of them suggest a delayed onset between the timing of exposure and mental health concerns. A recent study showed that air pollution exposure at age 12 was not associated with concurrent mental health conditions but rather with higher incidence of depression at age 18 (33). Therefore, despite an absent association between annual PM_2.5 and internalizing or externalizing symptoms in 9- to 13-year-olds in the ABCD Study (104), the current differences in the amygdala, together with other notable outcomes (50,52,53,105, 106, 107), may reflect early biomarkers of neurotoxicity that ultimately contribute to increased risk for psychopathologies. Future longitudinal research is needed to determine how outdoor air pollution impacts trajectories of amygdala subregion development and apportionment and confirm its utility as a potential biomarker for later psychopathology.

Several strengths and limitations of the current study should be noted. We implemented the CIT168 atlas, which was created using in vivo Siemens MRI data from healthy young adult brains (67,68). While other amygdala segmentation approaches were created from postmortem samples from older male brains (108,109), the CIT168 atlas uses high-resolution HCP data (from which the ABCD Siemens protocol was derived) and probabilistic delineations to encode partial volume uncertainty in amygdala subregions. To improve individual-level volume estimates, we chose a priori to limit our analyses to Siemens data and use a stringent CNR criterion. While these strengthened the rigor of our estimates, it limited our sample size to 4473 participants. Our final sample included more male, White, and higher socioeconomic status participants, potentially limiting the generalizability of our findings. Furthermore, participants included in this study were exposed to air pollution levels below the EPA guidelines, though not below all WHO guidelines. Therefore, these findings do not necessarily translate to adolescents living in highly polluted countries. The data included in this study are cross-sectional as the ABCD Study does not have residentially mapped PM_2.5 component exposure data beyond the baseline visit. While we can assess how current levels of air pollution are related to amygdala volume at ages 9 to 10, we are unable to draw conclusions about the impact of chronic exposure on amygdala development. Lastly, future studies using various mixture modeling approaches (i.e., Bayesian Kernel Machine Regression, Weighted Quantile Sum Regression) or unsupervised learning approaches (i.e., principal component analysis to identify latent exposure profiles) may be useful for testing other interesting, yet complementary, questions. Given our study question and that many of these other mixture approaches require separate models per outcome, we chose to use PLSC, which allows for multidimensionality and multicollinearity of both exposures and outcomes, to identify novel associations between co-exposure to pollutants and amygdala subregions. However, one limitation of PLSC is that it does not produce interpretable effect sizes that allow for conclusions regarding amygdala differences per unit of exposure. Future studies may consider other methods to decipher whether there are overall cumulative effects, the magnitude of these effects, and identify the most toxic components on various amygdala nuclei during development.

Conclusions

Overall, the results of the current study show that co-exposures to various components—Ca, K, and OC—are related to differences in amygdala volumes and apportionment, with expansion in subregions involved in fear conditioning, and reduction in subregions responsible for anxiety and fear suppression. These findings support a growing body of literature identifying PM_2.5 as particularly harmful for human health (110) and suggest that air pollutant exposure may influence structural differences in amygdala subnuclei during a critical period of brain development.