Neighborhood Disadvantage and Neural Correlates of Threat and Reward Processing in Survivors of Recent Trauma

This cross-sectional study investigates associations between neighborhood deprivation and neural reactivity to treat and reward among survivors of recent trauma.

This supplemental material has been provided by the authors to give readers additional information about their work.time of scanning), and participants were asked about their symptoms over the past 2 weeks (i.e., since the trauma).In addition, at the 2-week visit, participants completed the Lifetime Events Checklist for DSM-5 (LEC-5) which evaluated whether the participant had experienced, witnessed, or learned about 16 different stressful or traumatic events (e.g., "Serious accident at work, home, or during recreational activity").A total score was created by summing all responses.

Functional Tasks
In the threat task, participants viewed blocks of faces depicting fearful or neutral expressions (from the Ekman library).Each of the blocks (15 neutral; 15 fearful) were presented for 8s.Block order was counterbalanced across participants.Within a block, 8 different faces were presented for 500ms with an intertrial interval of 500ms.After every 10 blocks, participants were given a rest period of 10s and were instructed to relax with their eyes open.The reward task was a modified high/low card guessing game 4 .Participants viewed cards with a question mark and had 2s to guess whether the card's value was higher or lower than $5.Following a short, jittered delay (2-4s), the card's value and monetary outcome was displayed.Prior to the task, participants were informed they would win $1 for each correct guess and lose $0.50 for each incorrect guess.A total of 40 cards were presented, with a predetermined 20 gains and 20 losses (participants always won $10).
Frame-wise displacement was calculated for each functional run using the implementation of Nipype.ICA-based Automatic Removal Of Motion Artifacts (AROMA) was used to generate aggressive noise regressors as well as to create a variant of data that is non-aggressively denoised 11 .For more details of the pipeline see https://fmriprep.readthedocs.io/en/stable/workflows.html.An overall motion threshold was also implemented such that any participant's task data with >15% volumes and ≥1-mm framewise displacement were excluded.
In first-level analyses, as the threat task was a block design, blocks were modeled using separate boxcar functions whereas the event-related reward task gain and loss trials were modeled as separate events convolved with a canonical hemodynamic response function.Contrasts for region of interest extraction (ROI) included fearful > neutral blocks for the threat task and gains > losses trials for the reward task.ROIs were selected based on previous work 12 and defined anatomically using the Automated Anatomical Atlas 12 .ROIs for the threat task included the amygdala, insula, Brodmann's area (BA) 25 (corresponds to the subgenual anterior cingulate cortex) and BA32 (corresponds to the ACC).Reward ROIs included the nucleus accumbens (NAcc), OFC, BA32, insula, and amygdala.A-priori selection of these ROIs was not based on the participant's showing significant task-related activation.The mean across all voxels in each ROI was extracted from first-level contrasts.For ROIs with hemisphere differentiation, the activity was averaged across left and right seeds.
As described in prior work 13 , diffusion weighted images (DWI) were preprocessed using guidelines from the ENIGMA consortium to derive measures of structural connectivity (http://enigma.ini.usc.edu/protocols/dti-protocols/).Briefly, DWI data were first corrected for susceptibility using nonlinear warping to the participant's T1w image.Motion and eddy effects were reduced using FSL's eddy function.Finally, DWI data were fit with a tensor model prior to Tract-Based Spatial Statistics processing.

MRI Quality Control and Exclusions
For T1w images, we assessed sitewise differences in MRI quality control metric in MRI-QC 14 : coefficient of joint variation (CJV) of gray and white matter, median intensity nonuniformity (INU), and the signal-to-noise ratio (SNR).CJV is thought to reflect the presence of heavy head motion and artifacts where lower scores reflect better quality data.INU reflects non-anatomically variation in signal intensity across the volume.SNR is the signal to noise ratio of the entire T1-volume across tissue types.
For DWI data, we assessed sitewise differences in MRI quality control metrics derived from initial processing: temporal signal-to-noise ratio (TSNR), maximum outlier voxel intensity (OUTMAX), mean absolute motion (MEANABS), and maximum absolute motion (MAXABS).TSNR is the temporally averaged signal to noise ratio for each dataset.
For functional data, we assessed sitewise differences in MRI quality control metrics in MRI-QC: the AFNI Quality Index (AQI), FD, DVARs, and temporal-signal-to-noise ratio (TSNR).AQI is a general and crude screening tool for motion or scanner artifacts in 4D datasets.AQI is calculated as an average of 1 minus the Spearman rank correlation coefficient for every volume to the median volume in the dataset.
Framewise displacement is an estimation of head-movement across the dataset.DVARs is calculated as the derivative of the root-mean-square variance over dataset voxels 15 .
Table S1 describes the harmonized MRI acquisition parameters across each site.

Assessment of Missing Data and Lost to Follow-up Participants
Figure S1 provides an overview of reasons participants were excluded from the reward and threat samples.A total of 234 participants had both reward and threat task neuroimaging data.An additional 10 completed only reward data and 46 completed only threat data.Independent t-tests revealed there were no significant differences between the threat sample and the reward sample in regard to ADI (t(522) = 0.42, p = .68),income (t(522) = 0.05, p = .95),PCL-5 scores (t(522) = 0.81, p = .42),age (t(522) = 0.25, p = .80),or lifetime trauma (t(522) = 0.83, p = .41).

Threat vs Reward β Comparison
To further establish whether the effect of ADI was valence specific, we compared the standardized regression coefficients between the significant ROIs of the threat task to the reward task.
Comparisons of the β-values using z-tests confirmed that the effect of ADI on threat-related ACC and insula reactivity was significantly different compared to the association between ADI and reward-related ACC and insula reactivity, respectively (ACC: z-value = 2.34, p = .02;insula: z-value = 2.26, p = .02).

DWI
Following the significant effects of ADI on ACC and insula activity to fearful > neutral faces, we selected (post-hoc) white matter tracts which are related to activity in these regions: the cingulumcingulate gyrus (CGC) and the cingulate-hippocampal gyrus (CGH).The CGC and CGH were selected because of their relevance to the ACC activity.The CGC is the upper segment of the cingulum running adjacent to ACC whereas the CGH is the lower segment and connect the cingulum to the hippocampus (Mori et al., 2008).
Our primary aim was to test whether ADI was associated with fractional anisotropy (FA) values in these tracts.However, to evaluate robustness of the effect of ADI on a different feature of white matter microstructure, we also tested whether ADI was associated with mean diffusivity.Results from the GLMs conducted to evaluate the effect of ADI on the tracts are presented in Table S2.Higher ADI rankings were associated with greater CGC FA values and lesser CGC MD values.These findings suggest the effect of ADI is tract-specific and impacts multiple features of microstructure.

Post-Hoc Voxelwise ACC ROI analysis
Following the significant effects of ADI on ACC and insula activity to fearful > neutral faces, we examined the surface area and volume of these two regions (see main text).As the BA32 seed used for functional task traversed multiple Freesurfer parcellations corresponding to the ACC, we first conducted a supplementary ROI analysis to determine whether the effect of ADI was located in the rostral or caudal ACC.Results of the 3dttest in AFNI revealed a significant effect of ADI after adjusting for age, sex, and income in the caudal section of the ROI (k = 12, X: 4.5, Y: -33.5, Z = 29.5; Figure S4).

Neighborhood disadvantage is associated with ACC and insula macrostructure
Given our findings of ADI on threat-related function and structure, we completed exploratory post-hoc GLMs to examine the effects of ADI on cortical thickness and surface area of regions which showed significant task-related effects.We selected only the caudal ACC and insula for these analyses given the aforementioned findings.The caudal ACC was chosen based on voxelwise analyses of the BA32 seed region (described above) to determine if the original association was most pronounced in either rostral or caudal subregions given the seed traversed both subregions in the Desikan-Killainy atlas.S3.There was a significant relationship between ADI and caudal ACC cortical thickness (t(273) = -2.29,β = -0.13,p corrected = .02)(Figure S5).

Simple Mediation Model
Prior to testing the moderated mediation model presented in the main text, we examined whether CGC FA values mediated the relationship between ADI and ACC reactivity to fearful > neutral faces (Figure S6; N = 280).A mediation model (PROCESS model 4) showed there was a significant indirect effect of CCG FA values between neighborhood disadvantage and ACC reactivity (β = -0.035,SE = 0.017, 95% CI [-0.0722, -0.059].There was a significant relationship between ADI and CCG FA values (a path: β = 0.21; SE < 0.01, t = 3.48, p < .001)after adjusting for sex, age, income, and lifetime trauma.
There was also a significant relationship between CGC FA values and ACC activity (b path: β = -0.17;SE = 0.24, t = -2.69,p = .008),after adjusting for ADI, sex, age, income, and lifetime trauma.The direct effect was also significant (c' path: β = 0.20, SE < 0.01 t = 3.20, p = .002).This pattern of results revealed an inconsistent mediation.More specifically, neighborhood disadvantage was associated with greater ACC reactivity and higher CGC FA value (positive a and c paths), however, CGC values were associated with decreased ACC threat reactivity (negative b path).This pattern suggested another variable (e.g., PTSD symptoms) may be affecting the relationship between CCG FA values and ACC threat reactivity.

Moderated Mediation Analysis with Depression Symptoms
We repeated the moderated mediation analysis (PROCESS macro; model 14; 10,000 bootstrapping iterations) with depression symptoms as the moderator to determine if the observed relationship was specific to PTSD symptoms or related to general distress (Figure S7).Depression symptoms did not significantly moderate the effect of CGC integrity on ACC reactivity to threat (interaction β = -.07,SE = 0.06, t = -1.13,p = .26).The overall moderated mediation was not significant (index of moderated mediation = -0.02;standard error: 0.02; bootstrapped 95% CI[-0.05,0.01].

1 . 1 .eFigure 2 . 3 . 4 .eFigure 5 . 6 .
Harmonized Magnetic Resonance Imaging Sequences Across Study Sites eTable 2. General Linear Models for White Matter Tracts Among 280 Participants eTable 3. General Linear Models for Macrostructure Among 280 Participants eFigure Flowchart of Study Participants Who Met Inclusion Criteria for Threat or Reward Analyses Neighborhood Disadvantage and Income by Race and Ethnicity eFigure Neighborhood Disadvantage and Income by Study Site eFigure Anterior Cingulate Cortex and Neighborhood Disadvantage Region of Interest Analysis Neighborhood Disadvantage and Macrostructure eFigure Associations Between Neighborhood Disadvantage, Microstructure, and Threat Reactivity eFigure 7. Associations Between Neighborhood Disadvantage, Microstructure, Threat Reactivity, and Depression Symptoms eReferences.

Figure S1 .
Figure S1.Flowchart of AURORA study participants who met inclusion criteria for the threat or reward analyses.

Figure
Figure S2.[A] ADI and [B] Income significantly differed by ethnoracial group (ps < .001;N = 280; threat sample is depicted).ADI ranges from 1-100 (a 1 is indicative of the most advantaged neighborhood relative to all other neighborhoods in the country and a 100 is indicative of the most disadvantaged neighborhood.Income ranged from 1-6 (a 1 is indicative of an income <$19,000 and a 6 is indicative of greater than $100,000).Asterisks depict significant Tukey post-hoc tests (p < .05 with Hom-Bonferroni correction applied).

Figure S4 .
Figure S4.Results of the ROI analysis (N = 280) revealed the significant main effect of neighborhood disadvantage (blue cluster; k = 12, X: 4.5, Y: -33.5, Z = 29.5)was in the caudal section of the ACC seed (yellow).

Figure S5 .
Figure S5.Higher ADI rankings were associated with lower caudal anterior cingulate cortex [A] cortical thickness [B] and higher [C] surface area after adjusting for income, lifetime trauma, sex, age, and total intracranial volume (N = 280).Greater neighborhood disadvantage was also associated with lower insula [D] cortical thickness [E] and higher [F] surface area after adjusting for covariates.These are marginal effects plots depicting predicted values (orange regression line) for cortical thickness and surface area at each ADI ranking (shaded line: 95% confidence interval for the marginal effects; datapoints: observed data).
S7. [A]A moderated mediation model revealed depression symptoms did not moderate the association between white matter tract integrity and threat activity.[B] Conditional indirect effects of ADI and ACC activity to fearful versus neutral faces via CGC FA values did not significantly differ at higher or lower scores of depression symptoms (ps > .05).Coefficients are standardized.

Table S2 .
General Linear Models for White Matter Tracts (N = 280)