Differentiating distinct and converging neural correlates of types of systemic environmental exposures

Abstract Systemic environmental disadvantage relates to a host of health and functional outcomes. Specific structural factors have seldom been linked to neural structure, however, clouding understanding of putative mechanisms. Examining relations during childhood/preadolescence, a dynamic period of neurodevelopment, could aid bridge this gap. A total of 10,213 youth were recruited from the Adolescent Brain and Cognitive Development study. Self‐report and objective measures (Census and Federal bureau of investigation metrics extracted using geocoding) of environmental exposures were used, including stimulation indexing lack of safety and high attentional demands, discrepancy indexing social exclusion/lack of belonging, and deprivation indexing lack of environmental enrichment. Environmental measures were related to cortical thickness, surface area, and subcortical volume regions, controlling for other environmental exposures and accounting for other brain regions. Self‐report (|β| = .04–.09) and objective (|β| = .02–.06) environmental domains related to area/thickness in overlapping (e.g., insula, caudal anterior cingulate), and unique regions (e.g., for discrepancy, rostral anterior and isthmus cingulate, implicated in socioemotional functions; for stimulation, precuneus, critical for cue reactivity and integration of environmental cues; and for deprivation, superior frontal, integral to executive functioning). For stimulation and discrepancy exposures, self‐report and objective measures showed similarities in correlate regions, while deprivation exposures evidenced distinct correlates for self‐report and objective measures. Results represent a necessary step toward broader work aimed at establishing mechanisms and correlates of structural disadvantage, highlighting the relevance of going beyond aggregate models by considering types of environmental factors, and the need to incorporate both subjective and objective measurements in these efforts.


| INTRODUCTION
Beyond the individual, the larger environmental and social context (i.e., systemic-and structural-level environmental factors, including local, neighborhood, regional, or even country-level characteristics) has been shown to impact physical and mental health, among other critical lifestyle outcomes (Arcaya et al., 2016;Laraia et al., 2012;Ludwig et al., 2012). Yet, most investigations have focused on adult populations (Arcaya et al., 2016). Much less is known about the effect of exposures during the dynamic developmental period of late childhood and preadolescence (Leventhal & Dupéré, 2019). Further, relative to individual-level exposures (such as childhood trauma, life events, and bullying exposure), structural or systems level environmental factors have received relatively less attention in the literature-this is especially the case with regards to identifying putative biological or developmental mechanisms related to these factors.
The existing literature suggests different dimensions of environmental exposures could relate to both convergent and distinct neural structures across neurodevelopment (McLaughlin, Sheridan, Humphreys, Belsky, & Ellis, 2021;Vargas, Conley, & Mittal, 2020). Neural correlates, while theorized, have yet to be tested and thus remain poorly understood. Further, understanding neural correlates of different structural exposures is ultimately crucial from an epidemiological and etiological standpoint (Minh, Muhajarine, Janus, Brownell, & Guhn, 2017). Improving existing conceptualizations of systemic barriers to healthy development stands to inform health policy, as well as prevention and intervention efforts at the societal level.
Chronic stress has long been identified as a central vulnerability factor toward a host of mental, physical health, and life outcomes (Bauer, 2008;Juster, McEwen, & Lupien, 2010;Lupien, Juster, Raymond, & Marin, 2018;McEwen, 2017). Classically, individual-level stressors including childhood trauma, bullying, and parental conflict have been extensively studied with regards to underlying neural and biological mechanisms (Ellis, Boyce, Belsky, Bakermans-Kranenburg, & Van IJzendoorn, 2011;McLaughlin, Sheridan, & Lambert, 2014). The literature on individual stressors and neural correlates has allowed for a more nuanced understanding of mechanisms of influence, along with pinpointing possible intervention and prevention targets. In contrast to the established research on individual stressors, the broader environmental and social context (i.e., systemic or structural level factors) has been relatively understudied with regards to neural correlates (Bronfenbrenner & Morris, 1998;Glass & McAtee, 2006).
Characteristics of the broader environment, particularly factors such as neighborhood poverty, exposure to crime, population density, and crime exposure, could be disadvantageous at a systemic level and have downstream impacts on the individual. Indeed, neighborhood and structural characteristics have been reliably associated with adverse health outcomes and alterations in physical development (Arcaya et al., 2016;Leventhal & Dupéré, 2019;Ludwig et al., 2012).  (Jeon, Mishra, Ouyang, Chen, & Huang, 2015;Lyall et al., 2015;Tamnes et al., 2017;Wierenga, Langen, Oranje, & Durston, 2014). During late childhood, gray matter features are undergoing foundational developmental processes (Jeon et al., 2015;Wierenga et al., 2014). Widespread gray matter volume decreases are taking place; in addition, the developmental timing of volume, thickness, and surface area varies by cortical region, and cortical thickness and surface area develop independently of one another (Wierenga et al., 2014). Along with total gray matter volume decreases, cortical thinning and pruning processes are particularly active during adolescence (Norbom et al., 2021).
The marked neural reorganization occurring at this age yields greater plasticity, or sensitivity to environmental influences (Nelson III & Gabard-Durnam, 2020;Pechtel & Pizzagalli, 2011). As such, these gray matter metrics could provide unique insights into developmental processes and underlying biological mechanisms that are influenced by environmental factors. Cortical thickness may index synaptic pruning, cell shrinkage, apoptosis, and dendritic arborization (Jeon et al., 2015;Tamnes et al., 2017). Surface area could reflect processes related to cortical folding and gyrification (Garcia, Kroenke, & Bayly, 2018). Collectively these metrics could provide unique insights for emerging types of environmental vulnerability in the years prior to adolescence and young adulthood, when other contextual, interpersonal, and neurodevelopmental vulnerabilities could compound preexisting risk factors. Different types of exposures could require different considerations with regards to prevention and intervention efforts for mental and physical health vulnerability. As such, understanding neural correlates of types of environmental exposures is a crucial first step.
As the largest study to date on adolescent development, the adolescent brain and cognitive development (ABCD) study ® provides an excellent opportunity to further understand these questions (Casey et al., 2018). Indeed, existing literature has already yielded insights into the subject, relating environmental factors to neural structure and function. Existing studies have found relations between broader neighborhood disadvantage, resting state and structural features (Hackman et al., 2021;Rakesh, Seguin, Zalesky, Cropley, & Whittle, 2021). Other work harnessing ABCD has focused on neural correlates of single features, that is, neighborhood deprivation (Mullins, Campbell, & Hogeveen, 2020;Taylor, Cooper, Jackson, & Barch, 2020;Vargas, Damme, & Mittal, 2020). Studies distinguishing types of systemic environmental exposures are sparser. To our knowledge, one study explored distinct types of environmental exposures (using neighborhood features acquired through geocoding participant addresses), along with relations to brain structure (Karcher, Schiffman, & Barch, 2021). However, this study examined aggregate brain metrics of cortical thickness, volume and surface area, curtailing the ability to make inferences with regards to specific neural correlates and mechanisms underlying each environmental exposure. Investigations seeking to establish specificity of systemic environmental exposure on neurodevelopment are thus needed to enrich current conceptualizations of environmental vulnerability and putative neural correlates, as a first step toward understanding possible mechanisms.
To this end, a recent review delineated three systemic environmental exposure dimensions based on available evidence from epidemiological and neuroscience literature (Vargas, Conley, & Mittal, 2020).
The resulting stimulation, deprivation, and discrepancy (SDD) model posed three environmental exposure dimensions that are theorized to confer both converging and distinguishable effects on neural structure (figure 1 in the study by Vargas, Damme, Osborne, & Mittal, 2021). The environmental dimensions include stimulation exposures, with intermediary mechanisms of high sensory demands and lack of safety (e.g., high neighborhood crime and population density), discrepancy exposures, with intermediary mechanisms of social exclusion, low social capital and lack of belonging (e.g., high neighborhood income inequality), and deprivation exposures with intermediary mechanisms of lack of environmental enrichment (e.g., neighborhood median family income). In an earlier study, the specificity of the environmental domains in the SDD theory was tested through exploratory and confirmatory factor analyses. Environmental dimensions were distinguishable and related to vulnerability to psychopathology (Vargas et al., 2021). As such, some support has been found in the ABCD data for the distinctness of the domains.
Although the review outlined theorized neural regions that could be specific to each domain, these hypotheses have yet to be directly tested. As mentioned earlier, childhood and preadolescence are prime periods of marked environmental sensitivity, characterized by widespread neural plasticity and gray matter development. As such, efforts to understand these environmental factors and neural correlates during the childhood and preadolescence developmental period could contribute to crucial intervention and prevention efforts. These systemic exposures could only confer a generalized effect on neural structure, as found in the study by . Or, the environmental dimensions could relate specifically to certain regions, while also exhibiting broader effects at the whole brain level (as hypothesized by the SDD theory). Limited research testing these questions with different environmental exposures in the same sample hinder ability to further clarify these matters. As such, the current study marks the first explicit test of neural correlates of these three domains together, accounting for unique influences over and above other domains.
The present study sought to explore gray matter neural correlates for environmental dimensions of stimulation, discrepancy, and deprivation. First, self-report and subjective measures were identified, consistent with the SDD model. Then, subsamples were created based on the top 25 percentile of exposure to the environmental factors.
Finally, to establish specificity over and above general neighborhood disadvantage, cortical thickness and surface area, and subcortical volumes were used to predict exposure to the self-report and objective environmental dimensions, while accounting for other cortical and subcortical regions, and for exposure to other domains. Controlling for all cortical/subcortical regions within each analysis allowed for exploring SDD theory specificity predictions of certain regions over and above other regions (see supplementary material). Taken together, the current analyses allow for an opportunity to understand theorized environmental dimensions and neural correlates.
2 | MATERIALS AND METHODS 2.1 | Self-report questionnaires Self-report scales relevant to the three domains were chosen across administered scales . Self-report measures were developed by the ABCD team to index environmental and cultural factors that could be relevant to development (Alegria et al., 2004;Zucker et al., 2018). As such, these measures index structural factors/ exposures that occur at the systems level (Table 1)  .

| Objective neighborhood features
Residential history was collected through addresses where participants had lived across their lifetime. Addresses were used to determine census tracts corresponding to each location. Each tract T A B L E 1 Self-report scales used for subjective measures of environmental exposures, along with domains each measure represents Bureau of Investigation (FBI) data were used to calculate neighborhood population density, total crimes occurring in certain neighborhood, average neighborhood income inequality (i.e., higher neighborhood income inequality meaning that higher income individuals receive much larger percentages of the total income in a given neighborhood), and median family income. Since these metrics are compiled based on government data, they will be referred to as "objective neighborhood features," drawing a contrast from neighborhood features of interest that are also assessed through self-report, such as the ABCD Parent neighborhood safety/crime survey (NSC).
See with evidence that a lack of sense of belonging within one's culture, along with lack of participation and engagement with the majority culture and with one's culture, are cultural/systems level factors that can confer a lack of social capital and social exclusion (Emerson, Minh, & Guhn, 2018;Veling et al., 2008;Yang, Lei, & Kurtulus, 2018). For objective measures within the discrepancy domain, neighborhood income inequality was chosen, given evidence of high income inequality being linked to lack of belonging and feeling of social exclusion, consistent with the discrepancy domain (Vargas et al., 2021;. For the deprivation domain, the ABCD parent's demographic survey was used to index lack of access to environmental enrichment (with questions probing for access to resources such as access to doctors if needed, food, and utilities; Table S1). For objective measures, neighborhood median family income was used as a measure of neighborhood deprivation.

| Structural MRI
Participants completed a high-resolution T1-weighted structural MRI scan (1-mm isotropic voxels) using scanners from GE Healthcare Structural MRI data were processed using FreeSurfer version 5.3.0 (http://surfer.nmr.mgh.harvard.edu/) (Fischl, Sereno, Tootell, & Dale, 1999) according to the standard processing pipelines (Casey et al., 2018). Processing included removal of nonbrain tissue, segmentation of gray and white matter structures (Fischl et al., 2002), and cortical parcellation. All scan sessions underwent radiological review whereby scans with incidental findings were identified and excluded.
Quality control for the structural images comprised visual inspection of T1 images and Free-Surfer outputs for quality (Hagler et al., 2019).

| Sample selection
The ABCD study was designed to recruit a nationally representative  Higher values indicate higher feelings of safety within one's neighborhood (3 items ranging from 1 to 5, with 1 meaning strongly disagree and 5 meaning strongly agree), scores range from 3 to 15.
c Higher values indicate lower sense of belonging with ethnic group (6 items ranging from 1 to 5, with 1 meaning strongly disagree and 5 meaning strongly agree), scores range from 6 to 30.
d Higher values indicate higher participation in American culture (8 items ranging from 1 to 9, with 9 meaning completely agree), scores range from 8 to 72.
e Higher scores indicate greater degrees of deprivation (7 items ranging from 0 to 1, with 0 meaning no and 1 meaning yes), scores range from 0 to 7.
f Values indicate number of total crimes recorded within the Census-delineated neighborhood.
g Values indicate number of people per square mile.
h Values for income inequality represent the log of 100 Â ratio of the number of households with <10,000 annual income to the number of households with >50,000 annual income (Singh, 2003), higher values represent higher income inequality at the neighborhood level. Values range from À1.13 to 8.16. i The median of the yearly household income in U.S. dollars for households in a given neighborhood. stimulation/population density (n = 2,553), 2.95 for discrepancy/ income inequality (n = 2,566), and 50,357 for deprivation/median family income (n = 2,554; Table 2, Tables S2-S10). Analyses on selfreport features from our prior work accounted for exposures to other self-report domains. Similarly, analyses on objective features accounted for exposures to other objective domains. Analyses were run on the entire sample as well, and while not the focus of the study, these are presented in the supplementary material (Tables S11-S19).
2.6 | Cortical area/thickness and subcortical volumes as predictors of self-report and objective environmental domain exposures As such, results presented accounted for and corrected for all other brain regions within the same model. Prior to analyses, variables were converted to standardized units (z scores). Standardized brain metrics/ predictors did not correlate highly with each other (rs were below .5, in a vast majority of cases under .1); as such, collinearity due to brain metric predictors was not a concern in models that were run. The car package, version 3.0.11 was used to calculate variance inflation factor to further assess for multicollinearity; values were below 5, meeting conventional thresholds (Fox, 2015).

| Subcortical volumes as predictors of selfreport and objective environmental domain exposures
As described above, aseg Atlas subcortical regions (seven total) were used as predictors in a single varying-intercepts mixed effect model accounting for age, sex and other self-report/objective domains as fixed effects, and family and site as random effects, with stimulation/discrepancy/deprivation subjective and objective measures as outcome variables.

| Data analytic strategy
Prior to analyses, variables were converted to standardized units (z scores). Results were visualized using r packages ggseg version 1.6.3, and ggseg3d version 1.6.3 (Mowinckel & Vidal-Piñeiro, 2020 (Armstrong, 2014;Cabin & Mitchell, 2000;Fiedler, Kutzner, & Krueger, 2012): there were three distinct theoretically grounded tests (for stimulation, discrepancy, and deprivation), which were separately predicted for self-report and objective measures. The measures used were grounded in previous research that identified domains using factor analyses and related them to vulnerability for psychopathology (Vargas et al., 2021). In addition to correcting for all cortical and subcortical regions within the same model, further correction for comparisons was conducted for each gray matter metric tested (cortical thickness, surface area, and subcortical volume) using Bonferroni thresholds for three tests of one similar hypothesis, resulting in a threshold of 0.016 (see Table 3) (Bonferroni, 1936;Shaffer, 1995).

| Stimulation exposures
In addition to originally hypothesized regions, stimulation and discrepancy exposures were predicted by regions implicated in sensory processing, including auditory/language processing regions (transverse temporal), primary visual processing regions (lateral occipital), visual and semantic attention/integration (temporal pole), as well as some primary motor regions (precentral and paracentral). Consistent with stimulation, it is possible that high environmental attentional demands result in greater activity in regions involved in sensory processing and cue reactivity, altering neural structure during developmental sensitive periods (Ellis et al., 2011;Petanjek et al., 2011).
Given that the current study is cross sectional, future investigations are needed to further test the impact of environmental complexity on sensory processing. It could be that some discrepancy exposures partially engage some intermediary mechanisms from stimulation exposures. For example, exposures related to low social capital, low sense of belonging, and social exclusion could engage threat circuitry and cue reactivity .

| Discrepancy exposures
In tandem, Discrepancy exposures related to regions implicated in theorized intermediary mechanisms of social exclusion, lack of belonging and low social capital. Notably, isthmus cingulate, rostral and caudal anterior cingulate area related to self-report American culture participation and objective neighborhood income inequality (with rostral anterior cingulate not surviving Bonferroni correction). Cingulate regions have been related to processing of social threat, rejection, and lack of belonging, along with pain related processing (Adolphs, 2009;Eisenberger & Cole, 2012;Eisenberger, Lieberman, & Williams, 2003).

| Deprivation exposures
Deprivation exposures related to thickness in regions implicated in theorized lack of access to environmental enrichment. Associations specific to thickness support theorized mechanisms of environmentally dependent pruning following synaptic proliferation during late childhood and preadolescence (Changeux & Danchin, 1976;Huttenlocher, de Courten, Garey, & Van der Loos, 1982;Petanjek et al., 2011). Superior frontal thickness predicted objective neighborhood median family income, consistent with the interpretation that exposure to deprivation could accelerate normative developmental synaptic pruning processes (Huttenlocher et al., 1982). In addition, as noted earlier, findings highlight the utility of having both self-report and objective measures across environmental dimensions (as self-report and objective deprivation related to distinct prefrontal regions). Perhaps self-report deprivation picks up on a specific sub-facet of deprivation that neighborhood median family income does not. Future studies are needed to explore this possibility. Even for discrepancy and stimulation domains, where there was some consistency in associated regions across self-report and objective measures, there were still regions that uniquely related to self-report or objective facets of the environmental domain. As such, the notion of subfacets within dimensions warrants further attention and study.

| Converging regions across exposures
Taken together, results suggest the three dimensions of systemic environmental exposures relate to specific neural structures, when controlling for other environmental exposures, and for other brain regions. Evidence was also found for common regions predicting multiple types of exposures-this was the case for insula and caudal anterior cingulate area/thickness, which predicted all three domain exposures (though insula did not pass correction for deprivation, and caudal anterior cingulate did not pass correction for discrepancy).
These regions could be implicated through more a general effect of chronic stress exposure (Bauer, 2008;Juster et al., 2010;McEwen, 2017;. As the insula and caudal anterior cingulate are both implicated in a host of cognitive, affective, and regulatory processes, perhaps exposure during pre-adolescence marks a sensitive period of neurodevelopment during which development of these regions is particularly malleable (Caruana, Jezzini, Sbriscia-Fioretti, Rizzolatti, & Gallese, 2011;Eisenberger & Cole, 2012;Uddin, Nomi, Hébert-Seropian, Ghaziri, & Boucher, 2017). Results support potential regional convergence in neural correlates and exposures. were also some results that were unique to analyses in the broader sample. Results unique to the broader sample were largely within selfreport measures. Amygdala volume related to both stimulation neighborhood safety and deprivation neighborhood median family income.
Rostral anterior cingulate thickness, which has been found to modulate amygdala-dependent fear learning, also related to self-report deprivation (Bissière et al., 2008). While the amygdala was a theorized region for the stimulation domain, it was not for deprivation, highlighting the need for future inquiry and studies examining different developmental periods of exposure.
For stimulation domain, neighborhood safety in the broader sample related to lingual thickness and pars orbitalis area, like other visuospatial and semantic processing regions found in the high exposure sample. The lateral orbitofrontal cortex, receiving inputs from visual processing regions, also related to neighborhood safety in the broader sample (Rolls, 2004). Subcortically, the thalamus, critical for perceptual processing, also related to neighborhood safety (Sherman & Guillery, 2006). Finally, while objective neighborhood crimes did not relate to brain morphometry in the high exposure sample, in the broader sample isthmus cingulate thickness, which has been related to stressful life event exposure, related to neighborhood total crimes (Calati et al., 2018).
For discrepancy exposures, results largely converged in the broader sample, though there was an association between pars opercularis thickness and discrepancy American culture participation, with relations to phonological processing, which had not been theorized.
Notably, deprivation exposures in the whole sample related to a wider range of regions, including several regions implicated in visual and sensorimotor processing (lateral occipital, precentral, superior parietal, inferior temporal); these results are consistent with theories of deprivation accelerating normative developmental synaptic pruning processes in the human visual cortex (Huttenlocher et al., 1982). Lastly, subcortically, hippocampal and caudate volume related to self-report deprivation, in line with experience-dependent plasticity conceptualizations (Kleber, Veit, Birbaumer, Gruzelier, & Lotze, 2010;Wenger & Lövdén, 2016).

| SDD theory, limitations, and future directions
The current investigation allowed for an initial test of an emergent conceptual framework, the SDD theory, finding evidence of distinct neural coordinates that remained even after accounting for exposure to other dimensions. The SDD theory proposes that exposure to types of systemic environmental factors can meaningfully aggregate over time and impact neural development across critical sensitive periods (Vargas et al., 2021;. Further, it poses the notion that teasing apart types, or dimensions, of environmental factors can yield insight to both converging and distinct underlying mechanisms. In a recent study, our group found support for the theory with regards to separating hypothesized stimulation, discrepancy and deprivation domains, and relating them to mental illness vulnerability (Vargas et al., 2021). The current study tested the neural structures hypothesized to relate to each domain, finding partial support for the theory.
While we expected specific neural regions to relate to all three domains, substantial convergence was found for only two out of three domains (for stimulation and discrepancy). Deprivation exposures could engage systems less related to conscious experiences of chronic stress present in discrepancy and stimulation exposures ( Figure S3) (McLaughlin et al., 2021;Nelson III & Gabard-Durnam, 2020;Takesian & Hensch, 2013). Further study will be needed to modify the conceptual framework.
In addition, while environmental domains related to neural regions that were originally hypothesized by the SDD theory (i.e., cingulate and insular regions for discrepancy, prefrontal regions for deprivation), there were also a substantial portion of unpredicted findings (see figure 1 in the study by   Perhaps subcortical regions such as the hippocampus and amygdala relate to chronic stress more broadly (Bauer, 2008;Juster et al., 2010;McEwen, 2017;, rather than specifically by one type of exposure over and above the others. With regards to the expected co-occurrence of multiple environmental exposures within individuals, co-occurrence of exposures was not a major concern, as strong associations did not emerge between environmental dimensions in most cases, except for analyses on the high population density subsample. Given the high association between neighborhood crime and population density in that set of analyses, results ought to be taken as preliminary and interpreted with caution-future investigations will be needed to determine whether observed relations are generalizable to geographic locations beyond the current sample. More broadly, it will be crucial for future investi- As mentioned, one of the strengths of the study lies in investigating environmental factors during a dynamic period of neurodevelopment: pre-adolescence (Jeon et al., 2015;Lyall et al., 2015;Tamnes et al., 2017;Wierenga et al., 2014). Foundational processes readying the organism for puberty (including adrenarche and gonadarche), along with widespread pruning and specialization, make this stage an impactful period to understand (Mills, Lalonde, Clasen, Giedd, & Blakemore, 2014). With the developmental timing of thickness and surface area varying by cortical region, and cortical thinning and pruning processes being highly active during this time, there is marked neural reorganization occurring. As such, this period is prime for assessing for plasticity, or sensitivity to environmental influences (Nelson & Gabard-Durnam, 2020;Pechtel & Pizzagalli, 2011). Future studies will be crucial in determining effects of neurodevelopmental stage on observed associations, as these may not generalize to other developmental stages. By the same token, it is necessary to contextualize results in the developmental stage that they were observed: preadolescence.
Future studies would also benefit from incorporating multiple time points to better infer possible mechanisms of influence and incorporate multiple developmental sensitive periods. Given that exploring relations to biological sex was outside the scope of the current study, sex was controlled for in all analyses. However, future studies would benefit from investigating sex-specific relationships, given the key developmental of prepuberty, which includes several biological and neurological sequelae that differ based on biological sex (Mills et al., 2014). Future investigations could also build on this early work by exploring questions related to hemisphere asymmetry and laterality. Expanding beyond gray matter morphometry to functional imaging and white matter would also enrich understanding of possible underlying mechanisms of influence. Further teasing out functional and symptom outcomes relating to the exposures is also a crucial future direction. In all, the present study offers a first step toward unmasking neural correlates of systemic environmental exposures, which ultimately could inform public health policy, prevention and intervention efforts for vulnerable populations.

ACKNOWLEDGMENTS
Data used in the preparation of this article were obtained from the Adolescent Brain Cognitive Development (ABCD) Study (https:// abcdstudy.org), held in the NIMH Data Archive (NDA). This is a multisite, longitudinal study designed to recruit more than 10,000 children age 9-10 and follow them over 10 years into early adulthood.