Community Socioeconomic Disadvantage in Midlife Relates to Cortical Morphology via Neuroendocrine and Cardiometabolic Pathways

Abstract

Residing in communities of socioeconomic disadvantage confers risk for chronic diseases and cognitive aging, as well as risk for biological factors that negatively affect brain morphology. The present study tested whether community disadvantage negatively associates with brain morphology via 2 biological factors encompassing cardiometabolic disease risk and neuroendocrine function. Participants were 448 midlife adults aged 30–54 years (236 women) who underwent structural neuroimaging to assess cortical and subcortical brain tissue morphology. Community disadvantage was indexed by US Census data geocoded to participants' residential addresses. Cardiometabolic risk was indexed by measurements of adiposity, blood pressure, glucose, insulin, and lipids. Neuroendocrine function was indexed from salivary cortisol measurements taken over 3 days, from which we computed the cortisol awakening response, area-under-the-curve, and diurnal cortisol decline. Community disadvantage was associated with reduced cortical tissue volume, cortical surface area, and cortical thickness, but not subcortical morphology. Moreover, increased cardiometabolic risk and a flatter (dysregulated) diurnal cortisol decline mediated the associations of community disadvantage and cortical gray matter volume. These effects were independent of age, sex, and individual-level socioeconomic position. The adverse risks of residing in a disadvantaged community may extend to the cerebral cortex via cardiometabolic and neuroendocrine pathways.

Introduction

Health is not randomly distributed in space: it is patterned by place of residence. As evidence, residing in communities of socioeconomic disadvantage confers risk not only for early mortality (Waitzman and Smith 1998; Borrell et al. 2004; van Lenthe et al. 2005; Ross et al. 2013), but also dysphonic mood (Mair et al. 2008), cognitive dysfunction and decline (Seeman and Crimmins 2001; Lang et al. 2008; Shih et al. 2011; Zeki Al Hazzouri et al. 2011; Clarke et al. 2012; Sisco and Marsiske 2012), and diverse chronic medical diseases and pathologic conditions. The latter include cardiovascular disease (Diez Roux et al. 2001; Sundquist et al. 2004), cerebrovascular disease (Lisabeth et al. 2007), chronic kidney disease (Merkin et al. 2005), respiratory disease (Schreier and Chen 2013), elevated blood pressure (McGrath et al. 2006), and obesity (Grafova et al. 2008; Berry et al. 2010; Moore et al. 2013). Somewhat surprisingly, however, the risks for adverse health outcomes conferred by community socioeconomic disadvantage are not uniformly attributable to disadvantage measured at the level of the individual (Robert 1999; Pickett and Pearl 2001; Diez Roux and Mair 2010). Hence, even after accounting for educational, financial, and occupational measures of socioeconomic disadvantage among individuals, indicators of disadvantage measured at the level of their residential communities still predict health and disease outcomes. Thus, individual and community socioeconomic disadvantage are not redundant with each other, which agrees with the theoretical perspective that community disadvantage defines a unique source of disease risk rooted in complex social and environmental exposures (Diez-Roux et al. 2001). Notwithstanding cumulative evidence for the unique and diverse risks of community disadvantage, an open question is whether these risks extend to the brain. At present, 2 lines of evidence support this possibility.

The first is neuroimaging evidence showing that socioeconomic disadvantage measured at the level of the individual associates with presumptively negative changes in brain morphology, particularly reductions in tissue volume (e.g., Gianaros et al. 2007; Butterworth et al. 2011; Cavanagh et al. 2013; Brito and Noble 2014; Holz et al. 2015). Functionally, these associations with brain morphology may partly account for the diverse and negative cognitive, behavioral, and other brain-emergent endpoints linked to aspects of socioeconomic disadvantage experienced across the lifespan (Gianaros and Hackman 2013). Presently unknown from this neuroimaging evidence, however, is whether community socioeconomic disadvantage predicts similar morphological outcomes and whether it does so independently of individual socioeconomic disadvantage, which would 1) mirror cumulative epidemiological findings on community disadvantage and health and 2) agree with the conceptual possibility that social and environmental exposures may exert unique influences on the brain.

The second line of supporting evidence encompasses epidemiological findings showing that community disadvantage associates with cardiovascular, metabolic, and neuroendocrine risk factors for chronic diseases, which themselves may plausibly exert diffuse and negative effects on brain morphology. Hence, residing in disadvantaged communities confers risk for insulin resistance (Diez Roux et al. 2002; Auchincloss et al. 2007; Do et al. 2011) and the related clustering of multiple cardiovascular and metabolic (cardiometabolic) disease determinants (i.e., dyslipidemia, elevated blood pressure, adiposity, and glucose dysregulation), a clustering referred to as the metabolic syndrome (Chichlowska et al. 2008; Ngo et al. 2013). Importantly, cardiometabolic disease determinants associate with negative changes in brain morphology, including reductions in brain tissue volume (Yates et al. 2012; Alosco and Gunstad 2014; Onyewuenyi et al. 2014; Song et al. 2014). The latter associations have been interpreted from conceptual models positing a central role for the cumulative burden of systemic pathophysiology on the brain—particularly in the context of socioeconomic disadvantage (McEwen and Gianaros 2010). Another component of this second line of supporting evidence is that residing in disadvantaged communities associates with functional alterations in glucocorticoid (cortisol) release by the hypothalamic-pituitary-adrenocortical (HPA) axis (Dowd et al. 2009; Do et al. 2011; Hackman et al. 2012; Karb et al. 2012; Rudolph et al. 2014), which may result from greater exposure to hardships or unfavorable (e.g., stressful, uncertain, uncontrollable) life circumstances presumptively engendered by community-level disadvantage (Miller et al. 2009; Gianaros and Manuck 2010; Hackman et al. 2010). Like cardiometabolic risk factors, different indicators of cortisol that may indicate HPA dysregulation exhibit similar—though not uniformly—negative associations with aspects of brain morphology, presumably via the diffuse and negative effects of glucocorticoids on neural tissue (Lupien et al. 1998; Wolf et al. 2002; MacLullich et al. 2005; Bruehl et al. 2009; Kremen et al. 2010; Frodl and O'Keane 2013). Unclear from these 2 lines of evidence, however, is whether cardiometabolic and neuroendocrine (HPA) factors might lie along the intermediate or etiological pathways linking community disadvantage to putatively negative changes in brain morphology.

Accordingly, the present study of midlife adults tested for the first time to our awareness whether community socioeconomic disadvantage relates—independently of individual disadvantage—to in vivo indicators of cortical and subcortical brain morphology (total cortical and subcortical tissue volumes, cortical surface area, and cortical thickness) and whether this association is mediated by 2 plausible paths: one corresponding to cardiometabolic disease risk and the other to dysregulated HPA activity. An indicator of cardiometabolic risk was derived by combining canonical measures and correlates of the metabolic syndrome that have been linked to socioeconomic disadvantage, as well as adverse changes in brain functionality and morphology: waist circumference, body mass, blood pressure, and fasting levels of glucose, insulin, triglycerides, and high-density lipoprotein cholesterol. Indicators of HPA activity were derived from salivary measurements of cortisol obtained over multiple days to compute aggregate measures of the cortisol awakening response (CAR), the slope of the diurnal decline in cortisol, and cumulative cortisol release (area-under-the-curve). Indicators of socioeconomic position (SEP) were measured at the individual level by self-reported education, occupational status, and family income. Socioeconomic disadvantage was measured at the community level by aggregating several measures from the year 2010 US Census, which were geocoded to participants' residential addresses. Based on existing evidence, we hypothesized that greater community disadvantage would associate with reductions in brain tissue volume. We further explored whether these predicted negative associations would be apparent for cortical surface area and cortical thickness. Finally, we tested whether any observed associations 1) were independent of individual SEP; 2) could be accounted for by targeted psychosocial and demographic factors that may plausibly relate to socioeconomic disadvantage (i.e., low social integration, depressive symptoms, poor sleep quality, low positive emotionality, and high negative emotionality); and 3) were statistically mediated by greater cardiometabolic risk and signs of HPA dysregulation. These hypotheses were tested in 448 healthy and employed men and women.

Materials and Methods

Participants

Data for the present study were derived from 448 of 490 community-dwelling adult participants (aged 30–54 years; 236 women) of the Adult Health and Behavior project, phase-II (AHAB-II). AHAB-II is an epidemiological registry of biological, behavioral, sociodemographic, and neural correlates of cardiometabolic risk. Details regarding recruitment, eligibility, reasons for missing data from 42 participants, and other aspects of the protocol can be found in Supplementary Material. Importantly, ancillary analyses demonstrated that the 448 participants comprising the present sample did not differ significantly from the 42 AHAB participants with missing data in age, sex, race, ethnicity, or other predictor, mediator, covariate, or outcome variables described below (see Supplementary Material). Participants' average IQ, as estimated by the 4 subtests of the Wechsler Abbreviated Scale of Intelligence (Wechsler 1997), was 113.10 (SD = 12.59; range 72–140). All participants provided informed consent. The University of Pittsburgh Institutional Review Board granted study approval. Table 1 provides sample characteristics.

Table 1.

Descriptive statistics for demographic, psychological, and socioeconomic variables, and their univariate correlations with individual-level SEP and community-level socioeconomic disadvantage in 448 midlife adults (212 men, 236 women)

Characteristic			Univariate correlations
	Descriptive statistics		Individual	Community disadvantage
	M	SD	SEP r	r
Demographic variables
Age (years)	42.618	7.341	0.022	−0.006
Sex (female, N)	236		−0.115*	0.015
Cardiometabolic risk variables a
Waist circumference (inches)	35.553	5.558	−0.129**	0.128**
BMI	26.839	5.139	−0.206***	0.188***
Triglycerides (mg/dl)	109.221	68.546	−0.064	0.026
HDL (mg/dl)	55.960	15.010	−0.075	0.073
Glucose (mg/dl)	98.194	11.105	−0.059	0.101*
Insulin (uU/ml)	12.424	6.368	−0.123**	0.106*
Systolic BP for metabolic syndrome (mmHg)	115.037	11.078	−0.141***	0.174***
Diastolic BP for metabolic syndrome (mmHg)	72.262	8.168	−0.081	0.111*
Cardiovascular metabolic risk factor	0.000	0.668	−0.164***	0.170***
Community socioeconomic variables b
% Households on public assistance	2.815	3.234	−0.226***	0.714***
% Households below poverty level	12.660	10.274	−0.168***	0.813***
% Without a high school diploma for population 25 years and over	7.859	5.296	−0.284***	0.769***
% Total population in work force unemployed for population 16 years and over	6.994	4.860	−0.220***	0.701***
Median household income	53847.888	23465.893	−0.279***	0.903***
Disadvantageous community-level SEP	−0.140	0.728	−0.302***
Individual socioeconomic variables
Years attended school	16.971	2.853	0.689***	−0.188***
Subject job code	6.232	1.963	0.781***	−0.217***
Family income adjusted for number of people living in the household	35.537	6.647	0.789***	−0.277***
Personal-level SEP	0.004	2.262		0.302***
Brain morphometric variables
Cortical gray matter volume (mm3)	469578.416	52930.459	0.166***	−0.176***
Cortical white matter volume (mm3)	488858.433	59592.511	0.139***	−0.167***
Cortical thickness (mm)	2.509	0.098	0.093*	−0.128**
Cortical surface area (mm2)	169955.662	17013.044	0.135***	−0.138***
Subcortical gray matter volume (mm3)	187127.185	20844.735	0.143***	−0.091
Hippocampus volume (mm3)	8355.319	925.535	0.156***	−0.094*
Amygdala volume (mm3)	3219.632	404.874	0.109*	−0.062
Anterior cingulate volume (mm3)	9170.382	1508.835	0.088	−0.109*
Cortisol
Awakening response c	0.630	0.693	−0.083	0.066
Diurnal slope	−0.041	0.013	−0.197***	0.185***
Area under the curve d	7.414	2.182	0.080	−0.077
Psycho-social behavioral factors
Sleep quality e	5.029	2.636	−0.080	0.036
Social integration
Network diversity f	6.234	1.993	0.120*	−0.112*
Network size g	21.497	8.486	0.117*	−0.056
Depressive symptom h	8.520	7.860	−0.112*	0.023
PNSX: Positive affect i	33.984	5.912	0.044	0.008
PNSX: Negative affect j	15.457	5.085	−0.017	−0.032

^a,bSome variables were transformed prior to analysis (see text).

^c Sample size = 440.

^d Sample size = 443.

^e Pittsburgh sleep quality index: Global PQSI score.

^f Social network index: Number of high-contact roles. Sample size = 445.

^g Social network index: Number of people in social network. Sample size = 445.

^h Center for epidemiologic studies depression scale total: CESDTOT. Sample size = 442.

ⁱ General dimension scale: Positive affect. Sample size = 446.

^j General dimension scale: Negative affect. Sample size = 446.

*P < 0.05; **P < 0.01; ***P < 0.005.

Assessment of Socioeconomic Indicators

Individual-Level Indicators

Individual-level socioeconomic indicators were comprised of: 1) educational attainment, 2) occupational status, and 3) occupant-adjusted household income. Participants' educational attainment was defined by the number of years of schooling completed prior to study participation. Occupational status was determined by the Hollingshead Index of Social Position (Hollingshead 1975). Pretax household income in US dollars was coded as follows: <$10K; $10–14 999K; $15–24 999K; $25–34 999K; $35–49 999K; 50–64 999K; 65–79 999K; 80–94 999K; 95–109 999K; 110–124 999K; 25–139 999K; 140–154 999K; 155–169 999K; 170–185K; and >185K. Following our prior work (Gianaros et al. 2013), adjusted household income was computed as one-third of the median household income, weighted by the square root of the number of occupants in each household. A composite score representing individual-level SEP was computed as the sum of the standardized aforementioned variables (M = 0.047, SD = 2.24, interquartile range = −1.47 to 1.72). Accordingly, a “higher” SEP score denotes comparatively “greater” socioeconomic advantage at the individual level (Table 1).

Community-Level Indicators

To derive a composite indicator of community-level socioeconomic disadvantage, participants' addresses were submitted to a geographical coding procedure to derive US Census Bureau variables measured at the level of census tracts. Based on prior work (Gianaros et al. 2013) and for all tracts, the following were extracted from the Year 2010 Census Report (http://www.census.gov/2010census/): 1) percentage of households on public assistance (e.g., from food stamps or a supplemental nutrition assistance program); 2) percentage of households below the federal poverty line; 3) percentage of residents without a high school diploma for the population aged 25 years and over; 4) percentage of the working aged population who are unemployed; and 5) median household income (reverse-coded). All variables were logarithm (base-10) transformed to correct for skewed distributions. For each tract, a composite community-level disadvantage score was computed by (z-score) standardizing and then averaging the 5 transformed census variables (M = −0.14, SD = 0.73, interquartile range = −0.70 to 0.24). Accordingly, a relatively “higher” community-level score denotes “greater” socioeconomic disadvantage at the tract level of residence, which subsumes blocks and groups of blocks comprising residential areas of ∼3505 (median) individuals and ∼1553 (median) households per tract in this sample (N = 213 unique tracts) (Table 1). The correlations between individual-level SEP and community-level indicators and components are in Table 1.

Assessment of Mediators and Alternative Explanatory Factors

HPA Activity

Salivary cortisol was collected by participants on 4 monitoring days (3 workdays, 1 nonworkday). Each day, participants were prompted by an electronic diary to collect 5 saliva samples, the first immediately after awakening, at 30 min after awakening, 4 and 9 h after awakening, and at bedtime. Cortisol, expressed as nmol/L, was assayed in duplicate by a time-resolved fluorescence immunoassay and a cortisol–biotin conjugate as a tracer in the laboratory of Clemens Kirschbaum (Dresden, Germany). The sensitivity was 0.43 nmol/L. The intra-assay coefficient of variability was <10%, and interassay coefficient of variability was <12%. Cortisol values were log transformed prior to computing the diurnal slope using linear regression, whereas raw cortisol values were used to compute the percent increase in cortisol from waking to 30 min and total cortisol output (area-under-the-curve, AUC) by the methods described next. Descriptive statistics for HPA variables are in Table 1, and additional methodological details are in Supplementary Material.

Cortisol Awakening Response

CAR was calculated as percent change from the first awakening sample to 30 min after awakening. Cortisol samples not taken within 10 min of the 30-min postawakening instruction were excluded.

Diurnal Slope

Diurnal slope was calculated by fitting a regression line to log cortisol values for each participant, where successive cortisol measurements were predicted from hours since awakening (Matthews et al. 2006) to yield the linear equation: $\hat{Y}=a+b{x}_{}.$ Here, a denotes the intercept and b the slope of the diurnal rhythm. The second cortisol sample was excluded from the calculation of diurnal slope (and the AUC metric described below) to minimize the influence of the awakening response.

Area-Under-the-Curve (AUC)

AUC with respect to ground (omitting the 30-mi cortisol sample) was calculated by previously described methods (Pruessner et al. 2003).

Averaging Across Days

Preliminary analyses (repeated-measures ANOVAs) demonstrated that cortisol indices and the levels from which they were derived differed between nonworkday and workdays, with values obtained on the single nonworkday being lower than that obtained on each workday and the latter values not differing from each other (results available on request). Accordingly, the 3 workday indices were averaged and used for primary analyses and nonworkday indices were treated separately in ancillary analyses. Notably, all results reported here were comparable when examining nonworkday cortisol values separately and when averaging them with workday values (results available on request). As described later, of the 3 cortisol indices derived, CAR, AUC, and diurnal slope, only a shallower (flatter) diurnal slope was significantly associated with greater community-level disadvantage (r = 0.14, P = 0.003) after controlling for age, sex, and individual-level SEP. Hence, the workday mean of the diurnal slope of cortisol level was the primary HPA mediator tested here.

Cardiometabolic Risk

Components of a cardiometabolic risk metric used here—henceforth referred to as CMR—were assessed in the morning after a 12 h, overnight fast, as described previously (Jennings et al. 2013). First, systolic (SBP) and diastolic (DBP) blood pressures were measured by sphygmomanometry, as the mean of 2 consecutive readings obtained in a seated position. Then, a nurse determined body mass index (BMI in kg/m²) and waist circumference (measured at end-expiration to the nearest one-half inches with a tape measure centered at the umbilicus) and drew a 40-mL blood sample. Fasting serum lipids, glucose, and insulin were determined in the Heinz Nutrition Laboratory, University of Pittsburgh Graduate School of Public Health. A composite index of CMR was computed from the following variables, as in our prior work (Marsland et al. 2010): BMI, waist circumference, high-density lipoproteins (reverse-coded), triglycerides, glucose, insulin, SBP, and DBP (Table 1). Prior to analysis, glucose and insulin were winsorized at the 1st and 99th percentiles, and logarithm (base 10) transformations were applied to triglyceride and insulin values to correct for skewed distributions. A composite CMR index was then derived by averaging the standardized (z-score) values of the above variables.

Alternative Explanatory Factors

The following measures were examined, as they may plausibly relate to community socioeconomic disadvantage or account for study findings. These included social network diversity and size (as proxies of social integration), sleep quality, depressive symptoms, and trait positive and negative affect (Table 1). Social network composition was measured with the social network index (Cohen et al. 1997); namely, social network diversity was assessed as self-reported participation in 12 types of social relationships and social network size was assessed as the total number of people with whom the respondent has regular contact (i.e., at least once every 2 weeks). Quality of sleep was measured by the Pittsburgh Sleep Quality Index (PSQI), which ranks total sleep quality on a scale between 0 and 21 (Buysse et al. 1989). Depressive symptoms were assessed using the 20-item version of the Center for Epidemiologic Studies Depression Scale, CES-D (Weissman et al. 1977). Trait positive and negative affect were assessed using the Positive Affect (PA) and Negative Affect (NA) Schedule (Watson and Clark 1994).

MR Image Acquisition

MRI scans were collected on a 3T Trio TIM whole-body scanner (Siemens) using a 12-channel head coil. High-resolution T₁-weighted 3D magnetization-prepared rapid gradient echo neuroanatomical images were acquired over 7 min 17 s by these parameters: FOV = 256 × 208 mm, matrix = 256 × 208, TR = 2100 ms, TI = 1100 ms, TE = 3.31 ms, and FA = 8° (192 slices, 1 mm thick, no gap).

Image Processing

FreeSurfer 5.3.0 (http://surfer.nmr.mgh.harvard.edu) was used to compute cortical and subcortical volumetric data, as well as total cortical surface area and mean cortical thickness (Fischl and Dale 2000). Image preprocessing steps included affine-registration to the Talairach atlas, intensity bias correction, removal of nonbrain tissue, and normalization. Normalized images were used to estimate subcortical volumes by an automatic segmentation method that labels each voxel based on an atlas of 40 structures containing probabilistic information on structure location, intensity, and spatial relationships with nearby subcortical structures (Fischl et al. 2002). Cortical measurements, including tissue volumes, surface area, and thickness, were estimated using a surface-based approach wherein the cortical surfaces were extracted and reconstructed (Dale et al. 1999; Fischl, Sereno and Dale 1999). To this end, the 2 hemispheres were separated and boundaries between white and gray matter (i.e., white surface) and gray and pia mater (i.e., pial surface) were determined and extracted. The cortical surfaces were reconstructed by triangular tessellation. Cortical thickness was measured as the distance between the white and pial surface at each vertex on the tessellated surface. The surfaces were then inflated and registered to a spherical atlas that aligns cortical folding patterns to match cortical geometry on an average map for computing surface area (Fischl, Sereno, Tootell et al. 1999). Total cortical volume was calculated as the product of surface area and cortical thickness (Fischl et al. 2004). ICV estimates for each subject were determined by a scaling factor from the Talairach transformation step (Buckner et al. 2004).

Brain Morphology Measures

Total cortical gray matter volume, total cortical white matter volume, total subcortical gray matter volume, total cortical surface area, and mean cortical thickness were treated as primary and a priori outcome variables for each of the multiple mediator models described later. Hippocampal and amygdala volumes were included as subcortical outcome variables for secondary analyses, as reduced volume of these regions has been associated with individual-level socioeconomic hardship in adulthood (Butterworth et al. 2011) and implicated in HPA and cardiometabolic dysregulation in the context of socioeconomic disadvantage (McEwen and Gianaros 2010).

Control Variables

The following were selected as a priori covariates because they may plausibly confound study findings: age, gender, individual-level SEP, and intracranial volume (ICV).

Mediation Testing

Path analyses tested whether community disadvantage (X) was associated with brain morphology measures (Y) and whether any observed associations were statistically mediated by CMR and diurnal cortisol slope (multiple M's). Associations of community disadvantage with CMR (M₁) and diurnal cortisol slope (M₂) were tested as the effects of X on M_i (X→M₁, X→M₂), corresponding to the a_i path (where i = 1–2). Associations of mediators with brain morphology estimates, controlling for community disadvantage, were tested in separate models for each morphology outcome variable as the effects of M_i on Y (M_i→Y), corresponding to the b_i path. Associations reflecting the total effect of community disadvantage on brain morphology, without controlling for the mediators, was tested as the unadjusted effect of X on Y (X→Y) or c path. Associations reflecting the direct effect of community disadvantage on brain morphology while controlling for mediators were tested as the adjusted effects of X and Y (X→Y) or c′ path. The indirect effects reflecting the association of community disadvantage and brain morphology measures—as mediated by CMR and diurnal cortisol slope (X→M_i→Y)—were tested as the product of Paths a_i and b_i (a_i×b_i). The total indirect effects of the multiple mediator models corresponded to the sum of the specific indirect effects (a₁×b₁ + a₂×b₂). All effects were estimated using ordinary least square regressions. We did not employ multilevel models due to insufficient nesting of individuals within tracts (see Supplemental Material). The reason for insufficient nesting was because of practical recruitment and sampling constraints, wherein we did not oversample within tracts to achieve more representative sampling across tracts with this sample size. Statistical testing was done using nonparametric bootstrapping (5000 iterations), with 95% confidence intervals (CIs) for specific indirect (a_ib_i) and total indirect (Σa_ib_i) effects, generated by the percentile method. Mediation modeling was executed in the SPSS macro, PROCESS v2.13 (Preacher and Hayes 2008; Hayes 2013).

Results

Community Disadvantage and Brain Morphology

Prior to mediation modeling, we tested for associations between community disadvantage and cortical morphology. Total cortical gray and white matter volumes, total cortical surface area, and mean cortical thickness were all inversely and significantly associated with community disadvantage in univariate analyses (Table 1). Moreover, after statistically controlling for age, sex, ICV, and individual-level SEP in partial correlation analyses, the associations of community disadvantage with all indicators of cortical morphology persisted (P = −0.10 to −0.18, P < 0.05). These negative associations were unchanged after controlling for age, sex, ICV, and individual-level SEP, as well as other potential correlates of community disadvantage and cortical morphology; namely, social network size and diversity, depressive symptoms, sleep quality, and positive and negative affect (all P < 0.05). These findings indicate that the associations of community disadvantage with cortical morphology are independent of individual-level SEP and the above covariates and additional factors.

As shown in Table 1, analyses of subcortical morphology showed that only hippocampal volume exhibited an inverse and significant univariate association with community disadvantage. However, in contrast to findings for cortical morphology, after statistically controlling for age, sex, ICV, and individual-level SEP, the associations of community disadvantage with total subcortical volume (P = −0.04, P = 0.35), amygdala volume (P = −0.02, P = 0.66), and hippocampal volume (P = −0.03, P = 0.49) did not reach statistical significance. Notably, the elimination of the only significant univariate association (i.e., between hippocampal volume and community disadvantage) in these partial correlation analyses appeared to be equally driven by ICV and individual-level SEP, as inclusion of either factor reduced the association to below statistical significance (results available on request). These findings indicate that the associations of community disadvantage with subcortical morphology are not independent of individual-level SEP and the above covariates.

Cardiometabolic and HPA Mediators

We next examined the relationships between community disadvantage and targeted mediator variables in partial correlation analyses that controlled for age, sex, and individual-level SEP (univariate results are in Table 1). As expected, greater community disadvantage was related to a greater expression of cardiometabolic disease determinants, as reflected by a greater CMR score (P = 0.13, P = 0.006). Greater community disadvantage was also associated with a flatter diurnal cortisol slope (P = 0.14, P = 0.004). As noted above, however, greater community disadvantage was not associated with the CAR (P = 0.05, P = 0.26) or area-under-the-curve (P = −0.05, P = 0.28). Finally, accounting for social network size and diversity, depressive symptoms, sleep quality, positive affect, and negative affect did not change the direction or statistical significance of any of these associations. Given that the CAR and area-under-the-curve metrics did not relate to community disadvantage, these variables were excluded from further analyses (see Discussion).

We next tested associations of each potential mediator variable (CMR and diurnal cortisol slope) with cortical morphology indicators that also associated with community disadvantage (total cortical gray and white matter volumes, cortical surface area, and mean cortical thickness) while accounting for age, sex, ICV, and individual-level SEP. Greater CMR was associated with reduced total cortical gray matter volume (P = −0.14, P = 0.003), reduced total cortical white matter volume (P = −0.13, P = 0.005), and reduced total cortical surface area (P = −0.15, P = 0.002). CMR was not associated with mean cortical thickness (P = −0.01, P = 0.85). A flatter diurnal cortisol slope was associated with reduced total cortical gray matter volume (P = −0.11, P = 0.02), but not with other indicators of cortical morphology (P > 0.10). Thus, of the cortical morphology indicators, only total cortical gray matter volume showed simultaneous associations with community disadvantage, CMR, and the diurnal cortisol slope. Additionally, total cortical white matter volume and surface area also showed simultaneous associations with community disadvantage and CMR. Finally, CMR and the diurnal slope were themselves not significantly correlated (r = −0.02, P = 0.62), suggesting that they may correspond to independent (parallel) biological mediating factors—particularly linking community disadvantage and cortical morphology.

Mediation Tests of Community Disadvantage and Brain Morphology

We tested the indirect paths potentially mediating observed community disadvantage and morphology associations, while accounting for age, sex, ICV, and individual-level SEP. For completeness of reporting, the results of models for all morphology outcomes investigated in this study are in Table 2. Greater CMR and a flatter diurnal cortisol slope indirectly and simultaneously mediated the relationship of community socioeconomic disadvantage and cortical gray matter volume. Furthermore, we observed evidence for a direct effect of community disadvantage on cortical gray matter volume (c′ = −0.12, P < 0.01), suggesting partial, but not full, mediation. To convey the magnitude of effect sizes for the different paths of the mediation model, Figure 1 illustrates associations of community disadvantage, cortical gray matter volume, and the 2 mediator variables.

Table 2.

Summary of multiple mediator models

Dependent variables	Mediator variables	Path a X→M	Path b M→Y	Indirect effect a×b X→M→Y			Total effect c X→Y	Direct effect c′ X→Y (M adj.)
		Point estimation (SE)	Point estimation (SE)	Point estimation (SE)	95% CI Lower	95% CI Upper	Point estimation (SE)	Point estimation (SE)
Cortical gray matter volume	Total			−0.0301* (0.0106)	−0.0523	−0.0110	−0.1452*** (0.0382)	−0.1151** (0.0383)
	CMR	0.1139** (0.0414)	−0.1479*** (0.0433)	−0.0168* (0.0078)	−0.0340	−0.0035
	Cortisol	0.0012** (0.0004)	−11.3568* (4.4862)	−0.0133* (0.0063)	−0.0269	−0.0024
Cortical white matter volume	Total			−0.0201 (0.0119)	−0.0451	0.0010	−0.1415*** (0.0448)	−0.1214** (0.0453)
	CMR	0.1139** (0.0414)	−0.1538*** (0.0511)	−0.0175* (0.0090)	−0.0382	−0.0031
	Cortisol	0.0012*** (0.0004)	−2.2021 (5.2991)	−0.0026 (0.0065)	−0.0157	0.0108
Cortical thickness	Total			−0.0130 (0.0126)	−0.0400	0.0102	−0.1347* (0.0635)	−0.1216 (0.0647)
	CMR	0.1139** (0.0414)	0.0029 (0.0730)	0.0003 (0.0084)	−0.0178	0.0177
	Cortisol	0.0012*** (0.0004)	−11.4336 (7.5719)	−0.0134 (0.0094)	−0.0336	0.0031
Cortical surface area	Total			−0.0274* (0.0120)	−0.0523	−0.0062	−0.1055* (0.0442)	−0.0781 (0.0455)
	CMR	0.1139** (0.0414)	−0.1755*** (0.0502)	−0.0200* (0.0096)	−0.0417	−0.0039
	Cortisol	0.0012*** (0.0004)	−6.3427 (5.2055)	−0.0074 (0.0064)	−0.0210	0.0045
Subcortical gray matter volume	Total			−0.0143 (0.0097)	−0.0359	0.0023	−0.0385 (0.0415)	−0.0242 (0.0423)
	CMR	0.1139** (0.0414)	−0.0496 (0.0477)	−0.0056 (0.0065)	−0.0202	0.0052
	Cortisol	0.0012*** (0.0004)	−7.4036 (4.9499)	−0.0086 (0.0067)	−0.0235	0.0024
Hippocampus volume	Total			−0.0164 (0.0132)	−0.0463	0.0064	−0.0371 (0.0540)	−0.0207 (0.0550)
	CMR	0.1139** (0.0414)	−0.0937 (0.0621)	−0.0107 (0.0093)	−0.0323	0.0036
	Cortisol	0.0012*** (0.0004)	−4.9110 (6.4383)	−0.0057 (0.0087)	−0.0245	0.0098
Amygdala volume	Total			−0.0018 (0.0111)	−0.0247	0.0195	−0.0244 (0.0554)	−0.0226 (0.0566)
	CMR	0.1139** (0.0414)	−0.0183 (0.0639)	−0.0021 (0.082)	−0.0199	0.0141
	Cortisol	0.0012*** (0.0004)	0.2115 (6.6297)	0.0002 (0.0078)	−0.0154	0.0164

Notes: Covariates: age, sex, individual-level SEP, ICV. N = 448.

*P < 0.05; **P < 0.01; ***P < 0.005.

Figure 1.

A multiple mediator path analysis demonstrated that greater community socioeconomic disadvantage was associated with reduced cortical gray matter volume, and this association was mediated by greater cardiometabolic risk (CMR) and a flatter diurnal decline in cortisol (see Table 2). (A–F) Illustrations of the relationships between individual components of the mediator model, with all variables being (z-score) standardized and adjusted for covariates (age, sex, individual-level SEP, and ICV). (A) Cortical gray matter volume as a function of community disadvantage, controlling for age, sex, individual-level SEP, and ICV (total effects, or c path); (B) CMR as a function of community disadvantage, controlling for age, sex, individual-level SEP, and ICV (a₁ path); (C) cortical gray matter volume as a function of CMR, controlling for age, sex, individual-level ICV, ICV, diurnal cortisol slope, and community disadvantage (b₁ path); (D) cortical gray matter volume as a function of community disadvantage, controlling for age, sex, individual-level SEP, ICV, CMR, and diurnal cortisol slope (direct effects, or c′ path); (E) diurnal cortisol slope as a function of community disadvantage, controlling for age, sex, individual-level SEP, and ICV (a₂ path); (F) cortical gray matter volume as a function of diurnal cortisol slope, controlling for age, sex, individual-level SEP, ICV, CMR, and community disadvantage (b₂ path).

In addition to total cortical volume, the inverse associations of community disadvantage with total cortical white matter volume and cortical surface area were significantly mediated by CMR, but not the diurnal slope in cortisol, net the influences of age, sex, ICV, and individual-level SEP (Table 2). Hence, in addition to cortical gray matter volume, CMR appears to also link community disadvantage to other features of cortical morphology (total white matter and surface area). We note that the inverse association between community disadvantage and cortical thickness was not statistically accounted for by study covariates (age, sex, ICV, and individual-level SEP) or significantly mediated by CMR or the diurnal slope in cortisol, suggesting other as yet undetermined mediators of this association (Table 2).

In post hoc and ancillary analyses, we did not find support for the interpretation that indicators of brain morphology themselves statistically mediated the associations of community disadvantage and CMR or any indicator of cortisol functioning. Specifically, we executed a series of multiple meditator models using community disadvantage as the predictor variable (X), brain morphology indicators as multiple mediators (Ms), and then cardiometabolic risk or our cortisol metrics as the outcome variables (Ys), effectively reversing our mediators and outcomes. For consistency and comparability, we controlled for age, sex, ICV, and individual-level SEP. Critically, across all models, the 95% confidence intervals for the M_i mediation (a×b) paths included 0, meaning that none of the morphology indicators reached significance as indirect mediators (results available on request). These findings seem to suggest that indicators of brain morphology are more accurately modeled as the outcome variables than as mediators in this context.

Finally, we executed another series of post hoc and exploratory hierarchical regression analyses to test the possibility that community disadvantage and individual-level SEP “interacted” to predict brain morphology indicator variables. These analyses addressed the question of whether community and individual-level disadvantages synergistically relate to brain morphology. Results from these analyses showed that none of the brain morphology indicators was significantly associated with the interaction between community disadvantage and individual-level SEP, after accounting for the covariates above and the main effects of these variables, all ΔR² values for the interaction terms were <0.1% and all P > 0.2 across the regression models.

Community versus Individual-Level Effects

We considered the possibility that individual-level SEP might exhibit associations with the above indicators of cortical morphology that might be likewise independent of community socioeconomic disadvantage. In partial correlation analyses, however, individual-level SEP was not significantly associated with “any” indicator of cortical morphology after accounting for age, sex, ICV, and community disadvantage (all P < 0.08, all P > 0.10). The latter findings appear to suggest unique associations between community disadvantage and cortical morphology that are not similarly observed for individual-level SEP.

Exploratory Tests of Regional Specificity

The primary observations above concerned whole-brain associations of community disadvantage with cortical tissue volume, thickness, and surface area. It can also be asked whether community disadvantage associates with these aspects of morphology on a regional basis. Accordingly, we examined in post hoc analyses the associations of community disadvantage and morphology for the 4 cortical lobes separately, as well as for cortical regions of interest (ROIs) after employing false-discovery rate (FDR) correction for multiple exploratory testing and after controlling for age, sex, individual SEP, and ICV in all analyses (see Supplementary Material). Greater community disadvantage was associated with a greater reduction in cortical gray matter volume in all cortical lobes, as well as a greater reduction in white matter volume in the frontal, parietal, and temporal lobes (all P < 0.05_{FDR-corrected}). Greater community disadvantage was also associated with a greater reduction in the surface area of the frontal and temporal lobes, as well as a greater reduction in the cortical thickness of the parietal lobe (all P < 0.05_{FDR-corrected} [see Supplementary Table 1]). Moreover, exploratory ROI analyses showed that greater community disadvantage was associated with a greater reduction in gray matter volume throughout the cerebral cortex (Fig. 1, Supplemental Table 1), suggesting more diffuse rather than highly localized effects. In partial contrast to cortical gray matter volume effects, community disadvantage associations with white matter volume, surface area, and thickness tended to be less diffuse on average (Supplemental Figs 1–3 and Table 1).

Discussion

The novel findings of the present study are that 1) residing in communities of socioeconomic disadvantage relates to reduced cortical gray matter volume in midlife and 2) this relationship is accounted for by greater cardiometabolic risk and a flatter diurnal decline in the glucocorticoid, cortisol. Moreover, socioeconomic characteristics measured at the individual level (education, income, occupation) and self-reports of social integration, depressive symptoms, sleep quality, and trait emotionality did not explain the associations of community disadvantage with cortical morphology. Greater community disadvantage was further associated with reduced cortical white matter volume and reduced cortical surface area, net the effects of age, sex, and individual-level socioeconomic characteristics. And, the latter associations were statistically mediated by greater cardiometabolic risk. Together, these findings agree with epidemiological evidence regarding the adverse health effects of socioeconomic inequalities at the level of residential communities. They also extend a growing literature on the neurobiology of socioeconomic disadvantage and ill health, which has largely focused on individual-level socioeconomic indicators and not residential communities.

To elaborate, epidemiological evidence has linked community disadvantage with a greater risk for a broad spectrum of adverse health outcomes (Diez Roux 2001, 2007; Diez Roux and Mair 2010). Critically, this evidence suggests that the risks for ill health conferred by disadvantaged communities are not explained by individual-level factors, including personal attributes of education, income, and occupation. Our novel results agree with the latter findings and extend them to the level of the brain. In this regard, our results also agree broadly with existing studies of the neural correlates of socioeconomic disadvantage. For example, regional reductions in cortical volume, cortical surface area, and cortical thickness have been reported among 21 men residing in the most disadvantaged areas of Glasgow, Scotland, compared with 21 men residing in more advantaged areas (Krishnadas et al. 2013). Notably, these regional reductions linked to area disadvantage agree with findings from exploratory analyses of cortical lobes and subregions in the present study (Fig. 2, Supplementary Table 2). The former findings among a relatively small and all male sample, however, employed an extreme groups approach, unlike the present study, which modeled a continuous range of community-level socioeconomic disadvantage. Furthermore, with respect to cortical gray matter volume, our findings suggest more widespread and diffuse effects of community disadvantage rather than highly localized effects within particular regions of the cortex (see Fig. 1 and Supplementary Fig. 1). Also unaddressed in prior work is the issue of whether the associations of residential disadvantage and brain morphology are independent of (not confounded with) individual-level socioeconomic factors, as was shown here. Also shown here was that community and individual-level indicators did not appear to interact in the prediction of brain morphology. Other studies focusing on morphological correlates of socioeconomic disadvantage in adulthood have observed reductions in hippocampal and amygdala volumes among individuals reporting financial hardship (Butterworth et al. 2011). It is possible that the potentially adverse effects of socioeconomic disadvantage on brain morphology may originate early in development, with possible long-term consequences extending into adulthood and midlife. For example, parental socioeconomic disadvantage predicts reduced hippocampal volume among children (Hanson et al. 2011), as well as reduced whole-brain cortical surface area among typically developing individuals aged 3–20 years (Noble et al. 2015). In aggregate, our findings are thus consistent with existing evidence regarding the negative associations between indicators of socioeconomic disadvantage and brain morphology. We add, however, that our findings indicate that some of the previously reported associations between brain morphology and individual-level indicators of socioeconomic disadvantage may be partly explained by community-level influences and not the converse. Furthermore, in contrast to our findings for cortical morphology, we observed that community-level disadvantage did not correlate significantly with subcortical, amygdala, or hippocampal volumes, especially after accounting for ICV and individual-level SEP. These findings could suggest some dissociation in the morphological (cortical vs. subcortical) neural correlates of socioeconomic disadvantage across levels of analysis. At present, the functional mechanisms and implications of these potentially dissociable correlates of disadvantage are not yet clear.

Figure 2.

Labeled regions shown in color correspond to cortical areas exhibiting significant reductions in gray matter volume as a function of increasing community-level socioeconomic disadvantage, P < 0.05_{FDR-corrected}, after controlling for age, sex, individual-level SEP, and ICV. Areas coded in gray did not survive FDR correction for multiple testing. (A,B) Regions determined by the Desikan–Killiany atlas in Freesurfer (see Supplementary Material). In (A) are regions shown from a lateral view of the left hemisphere. In (B) are regions shown from a medial view of the left hemisphere. All regions are shown on the inflated surface template in Freesurfer. All analyses of ROIs were bilateral (see Supplemental Material). CN, cuneus; IPG, inferior parietal gyrus; ITG, inferior temporal gyrus; LOC, lateral occipital cortex; LOFC; lateral orbital frontal cortex; MTG, middle temporal gyrus; PCL; paracentral lobule; PCN, precuneus; PO, pars opercularis; POG, postcentral gyrus; PRG, precentral gyrus; RMFG, rostral middle frontal gyrus; SFG, superior frontal gyrus; SMG, supramarginal gyrus; SPG, superior parietal gyrus; STG, superior temporal gyrus; TTC, transverse temporal cortex.

At a conceptual level, findings linking disadvantage to brain morphology have often been interpreted from theoretical models emphasizing the adverse role(s) of stress and negative emotions, which may be occasioned by socioeconomic inequality (Shonkoff et al. 2009; Gianaros and Manuck 2010; Hackman et al. 2010; Miller et al. 2011). Hence, communities of disadvantage are more likely to be characterized as areas of social fragmentation, ambient incivility, incommodious housing, lesser access to resources to promote physical activity in safe environments, and greater exposure to putatively health damaging media, nonnutritious food outlets, and greater exposure to psychosocial stress (Diez Roux 2001, 2007; Diez Roux and Mair 2010). Indeed, socioeconomic disadvantage measured across multiple levels of analysis relates to increased exposure to life stressors that are thought to impact health across the lifespan (Mair et al. 2008; Chen and Miller 2013). To date, however, there is little evidence to support the notion that indicators of psychological stress and related emotional factors, particularly when measured by self-report, fully moderate or mediate the associations of socioeconomic disadvantage to particular behavioral, biological, or neural endpoints (Matthews et al. 2010; Matthews and Gallo 2011). Moreover, in prior neuroimaging studies, measures of psychological stress and negative emotionality have not been widely tested as candidate mediators or confounders of the associations between socioeconomic indicators and measures of brain morphology. In view of existing work and based on our null findings regarding social integration, sleep quality, depressed mood, and affect, other candidate mediators linked to the residential environment may represent alternative pathways by which socioeconomic disadvantage relates to biobehavioral health outcomes and neurobiological endpoints. The latter interpretation is consistent with our earlier findings demonstrating that negative and self-reported psychological and social factors do not explain the associations of 1) lower individual-level subjective socioeconomic status and reduced volume of the anterior cingulate cortex (Gianaros et al. 2007) and 2) lower individual and community-level SEP and global reductions in white matter integrity throughout the brain (Gianaros et al. 2013). On balance, however, we appreciate that a larger and more diverse sample exposed to a broader range or more severe sources of life adversities than that represented here might yield different results. Moreover, the use of self-report measures of stress and related factors in this area may not adequately assess stressor or other relevant exposures across individuals, when compared with interview, observation, or biologically based approaches (see below).

Our findings also agree with prior evidence linking socioeconomic disadvantage to greater cardiometabolic risk, including studies linking community disadvantage to insulin resistance (Diez Roux et al. 2002; Auchincloss et al. 2007; Do et al. 2011), the metabolic syndrome (Chichlowska et al. 2008; Ngo et al. 2013), and measures of systemic pathophysiology (Dowd et al. 2009; Bird et al. 2010). They also agree with animal evidence suggesting that low social status leads to adverse cardiometabolic changes (Tamashiro 2011). We extend prior findings to suggest that the community-level patterning of cardiometabolic risk statistically accounts for associated reductions in cortical gray matter volume, cortical white matter volume, and cortical surface area (Table 2). The latter findings themselves build on neurobiological evidence indicating that greater cardiometabolic risk associates with negative changes in brain morphology, with implications for premature cognitive declines in later life (Yates et al. 2012). To speculate, greater cardiometabolic risk observed among those from disadvantaged communities could be attributable to aspects of the built environment that foster the development of energy imbalance (e.g., limited safe or green spaces for physical activity, higher density of nonnutritious foods), with potentially reversible consequences on cortical morphology (Diez Roux and Mair 2010). For example, physical activity interventions that reduce cardiometabolic risk also positively affect cortical morphology (Erickson, Leckie et al. 2014). In this regard, our findings provide the first evidence to our knowledge that cardiometabolic risk patterned by community disadvantage associates with negative changes in brain morphology, net individual-level influences. In turn, these findings might provide an empirical basis for community interventions designed to improve aspects of the built environment that promote energy balance.

Evidence regarding associations between socioeconomic disadvantage and HPA functioning is mixed (Dowd et al. 2009). Hence, a recent review of the literature suggests that indicators of socioeconomic disadvantage—largely measured at the individual level—are not uniformly related to different cortisol metrics (e.g., CAR, AUC, slope, etc.), with many prior studies finding no associations (Dowd et al. 2009). Across prior studies, however, greater socioeconomic disadvantage was most consistently related to a blunted pattern of diurnal cortisol secretion, which is taken to indicate HPA-axis dysfunction and steeper decline taken to indicate a normal glucocorticoid release. Our findings indicated that greater community socioeconomic disadvantage was associated with a flatter diurnal decline in cortisol, which partly agrees with cumulative individual-level findings and more recent findings specifically focusing on neighborhood disadvantage. In the Multi-Ethnic Study of Atherosclerosis (Do et al. 2011), for example, a flatter diurnal cortisol decline among midlife adults was associated with neighborhood poverty, net the influences of individual-level SEP and other confounders. It was also reported that greater levels of neighborhood violence were associated with a flatter diurnal decline in cortisol. It is possible that our cortisol findings may reflect a psychosocial pathway (e.g., corresponding to exposure to chronic stressors, threats, or risks to safety) linking residential community disadvantage to brain morphology. However, we note that compared with diurnal cortisol, our self-report measures of affect and depressive symptoms may not have adequately assessed this putative psychosocial or “stress-related” pathway. Furthermore, the functional importance of associations between community socioeconomic disadvantage, a diurnal flattening of cortisol release, and reduced cortical gray matter volume are not yet clear. In animal models, it has been shown that circadian glucocorticoid oscillations promote learning-dependent synapse formation and maintenance in the cerebral cortex, particularly by affecting the formation and stability of dendritic spines (Liston et al. 2013). In epidemiological work, a flatter decline in cortisol has also been linked to increased atherosclerotic vascular disease risk (Matthews et al. 2006), which could possibly affect cerebral hemodynamic support and impact cortical morphology and neurocognitive functioning (Vuorinen et al. 2014). Accordingly, it will be important for future research to examine the potential functional implications of these effects, possibly by examining the extent to which neuroendocrine-related changes in cortical morphology account for negative cognitive outcomes linked to community-level disadvantage (Seeman and Crimmins 2001).

We recognize that our cross-sectional findings leave several questions unanswered. The first is whether (or how) community socioeconomic disadvantage and cortical morphology are causally or temporally related to each other. According to an indirect selection perspective, it is possible that some “third” variable (e.g., a genetic factor or intellectual ability) partially accounted for the present associations (Black 1988). We note, however, that adding IQ as a covariate in post hoc partial correlation analyses did not affect the direction or statistical significance of the associations between community socioeconomic disadvantage and our mediator variables (CMR, diurnal cortisol) and cortical morphology indicators (gray matter volume, white matter volume, surface area, and thickness), all P < 0.05 (results available on request). We also note that alternative mediation models did not provide evidence suggesting that indicators of brain morphology themselves mediated the associations of community disadvantage with CMR and diurnal cortisol. Rather, brain morphology indicators appeared to be more appropriately modeled at a conceptual level as outcome variables or so-called targets of dysregulated aspects of peripheral physiology (cf., Erickson, Creswell et al. 2014). With respect to temporal ordering, interventions designed to reduce the adverse effects of cardiometabolic risk on brain morphology could help clarify the nature of at least some of the present associations by measuring socioeconomic indicators and testing them as effect modifiers. In this context, promoting health behaviors, such as physical activity, that reduce cardiometabolic risk and alter neuroendocrine function may be especially important for individuals living in disadvantaged communities. And finally, it will be important to examine other candidate mediating pathways linking community disadvantage to brain morphology, which were not studied here. For example, we note that there were direct and unexplained effects of community disadvantage on brain morphology (see Table 2). Candidate mediators that could partly explain these effects and linked functional outcomes (e.g., related to cognition) could include systemic inflammation (Marsland et al. 2015), dietary factors (Nicklett et al. 2011), and developmental and environmental correlates of residential areas (e.g., physical pollutant and toxin levels) (Evans and Kantrowitz 2002) that could plausibly impact brain morphology.

According to the World Health Organization, residing in areas of socioeconomic disadvantage confers early mortality risk and for living with a chronic disease for nearly 17 years of a person's life. Our findings suggest that the adverse health effects of community socioeconomic disadvantage appear to extend as well to the brain—particularly the cerebral cortex—via cardiometaolic risk and neuroendocrine pathways, and that these effects are not explained by individual-level socioeconomic factors. Our observations may have implications for understanding the neural bases of how community-level disadvantage relates to neurocognitive and other health outcomes observed across the lifespan.

Authors’ Contributions

P.J.G., A.L.M., and S.B.M. designed and performed the research; P.J.G., L.K.S., K.G.M., and D. C-H. K. analyzed the data; P.J.G., D. C-H. K., K.G.M., A.L.M., D.A.H., and S.B.M wrote the paper.

Supplementary Material

Supplementary material can be found here.

Funding

This work was supported by National Institutes of Health grants PO1 HL040962 (S.B.M.), R01 HL089850 (P.J.G.), and the Robert Wood Johnson Foundation Health & Society Scholars Program (D.A.H.).