Association of Childhood Violence Exposure With Adolescent Neural Network Density

Key Points Question Are violence exposure and social deprivation associated with person-specific patterns (heterogeneity) of adolescent resting-state functional connectivity? Findings In this cohort study of 175 adolescents, childhood violence exposure, but not social deprivation, was associated with reduced adolescent resting-state density of the salience and default mode networks. A data-driven algorithm, blinded to childhood adversity, identified youth with heightened violence exposure based on resting-state connectivity patterns. Meaning Childhood violence exposure appears to be associated with adolescent functional connectivity heterogeneity, which may reflect person-specific neural plasticity and should be considered in neuroscience-based interventions.


Supplemental Participant Information
Five hundred and six families from Detroit, Toledo, and Chicago area who had participated in the FFCWS were contacted. Of the 506 FFCWS families contacted, 237 families participated in SAND data collection, and 183 of those families had teens who were eligible and completed the neuroimaging tasks. 54 teens either were not eligible (i.e., braces, etc.) to complete the neuroimaging portion of the study (N=28) or did not complete the full protocol (N=26).
Families that agreed to participate in SAND data collection did not differ from families that refused or were unreachable on annual household income at the six waves of the FFCWS (ps = .11 -.84). However, nonparticipation was associated with mother reports of community violence exposure when adolescents were ages 3, 5, and 9 (χ2 = 6.72, df = 1, p < .05, V = .12).
Sixty-three percent of all mothers from families that participated in SAND endorsed at least one form of community violence when adolescents were ages 3, 5, or 9 compared to 51.10% of families that did not participate. Within the Detroit, Toledo, and Chicago subsamples, mothers' self-report of race/ethnicity at the time of the child's birth did not differ between mothers who did and did not participate in SAND (ps = .13-.49). However, mothers' self-report of race/ethnicity at the time of the child's birth differed between the SAND sample and the larger FFCWS (χ2 = 86.32, df = 3, p < .01, V = .13). The majority of mothers who participated in SAND data collection identified as Black/African American (75%), and also included mothers identifying as White/Caucasian (16%), Hispanic (6%), and other (3%). In the FFCWS, 46% of mothers identified as Black/African, 21% identified as White/Caucasian, 29% identified as Hispanic, and 4% as other.
Of the 183 SAND teens with resting-state MRI data, three participants were excluded due to artifacts in the functional or structural MRI data, four participants had excessive motion (as defined by average relative framewise displacement greater than 0.5mm), and one person had signal dropout in the areas of the brain included in the present analysis. The sample included for the neuroimaging analyses did not differ from the full SAND sample (ps 0.670 -0.997) (Supplemental Table 1). This information has also been reported in other work from our group 1,2 .
Resting State MRI Acquisition. Slices were prescribed parallel to the AC-PC line (same locations as structural scans). Images were reconstructed into a 64x64 matrix (TR=2000 ms, TE=30 ms, flip angle=90°, FOV=22 cm, voxel size=3.44mm x 3.44mm x 3mm, bottom-up interleaved). Slices were acquired contiguously, which optimized the effectiveness of the movement post-processing algorithms. Images were reconstructed off-line using processing steps to remove distortions caused by magnetic field inhomogeneity and other sources of misalignment to the structural data, which yields excellent coverage of subcortical areas of interest. These methods are identical to those described in previous task-based fMRI research using this sample 3 .

Imaging Data Analysis
MRI Preprocessing. Anatomical images were homogeneity-corrected using SPM8, then skull-stripped (f=0.25) using the Brain Extraction Tool (BET) in FSL (version 5.0.7) 4,5 . The functional imaging data then had the following preprocessing steps applied: removal of large temporal spikes in k-space data (> 2 std dev), field map correction and image reconstruction using custom code in MATLAB; noise from cardiac and respiratory motion were removed using RETROICOR, and slice-timing correction using SPM8 (Wellcome Department of Cognitive Neurology, London, UK; http://www.fil.ion.ucl.ac.uk). Additionally, the first 10 volumes of functional data were removed to ensure the stability of signal intensity. Lastly, the structural images were segmented into gray matter, white matter, and CSF using FSL's FAST 6 .
Following these initial preprocessing steps, the resting state functional data underwent further preprocessing using FEAT (FMRI Expert Analysis Tool) Version 6.00, part of FSL (FMRIB's Software Library, www.fmrib.ox.ac.uk/fsl). Registration to high resolution structural and/or standard space images was carried out using FLIRT 7,8 . The following pre-statistics processing was applied: motion correction using MCFLIRT 8 ; non-brain removal for the functional images using BET 4 ; spatial smoothing using a Gaussian kernel of FWHM 6.0mm; grand-mean intensity normalization of the entire 4D dataset by a single multiplicative factor. ICA-AROMA 9 was used to remove motion-related artifacts in the data. Nuisance signal derived from the white matter and cerebrospinal fluid (CSF) were regressed out of the data and then the data was high-pass filtered to remove signal below 0.01Hz.

Violence Exposure and Social Deprivation Composite Scores
The procedure creating these composite scores were first utilized in previous work from our lab 2 and has been used previously in research on cortisol reactivity in this sample 10 . Data for these composite scores come from primary caregiver report on survey measures at 3, 5, 9 years.
Childhood Exposure to Violence. Included in this composite was the primary caregiver's report of child physical and emotional abuse based on items from the Parent-Child Conflict Tactics Scale 11 that have been used in previous research 12,13 . Five items were used to assess physical abuse including, "hit him/her on the bottom with a hard object" and "shook him/her" and five items were used to assess emotional abuse including whether the parent/caregiver has "sworn or cursed at," or "called him/her dumb or lazy or some other name like that." Each item was rated on a 7-point Likert scale ranging from "never happened" to "more than 20 times." The primary caregiver's report of the child's exposure to or victimization of violence in the neighborhood 14 was also included in the composite. This was measured using the primary caregiver's report of the child witnessing or being the victim of beating, attacks with a weapon, shootings, and killings (witness only) on a 5-point Likert scale ranging from "never" to "more than 10 times." At age 9, the primary caregiver was not asked about whether the child had witnessed killings or it they had been the victim of a shooting, so these items were only included for ages 3 and 5 years. Lastly, the child's mother reported on intimate partner violence (IPV) (physical-2 items, emotional-3 items, or sexual-1 item) in the home at each wave 13 . Each item was rated on a 3-point Likert scale ranging from "never" to "often." Physical IPV items included "he slapped or kicked you" and "he hit you with his fist or a dangerous object." Emotional IPV items included "he tried to isolate you from family and friends," and "he tried to prevent you from going to work and/or school." The sexual IPV was "he tried to make you have sex or do sexual things you didn't want to do." The child's exposure to IPV against the mother was coded as missing for a given wave if the child did not live with their mother at least 50% of the time. We considered violence exposure to exist on a continuum in which high scores represent violence exposure and low scores represent safety.
Childhood Exposure to Social Deprivation. Included in this composite was the primary caregiver's report of child physical and emotional neglect based on items from the CTS-PC 11 that have been used in previous research 12,13 . Four items from the CTS-PC were used to assed physical neglect including whether the parent was ever "so drunk or high that you had a problem taking care of your child." One item, whether the parent was "ever so caught up in your own problems that you were not able to show or tell your child that you loved him/her," was used to asses emotional neglect. These items from the CTS-PC were reported on the same 7-point Likert scale as the items in the violence exposure composite. The primary caregiver's report of social cohesion in the neighborhood was also included in this composite (reverse coded such that higher scores corresponded to lower cohesion). The items were selected based on previous research in the FFCWS linking neighborhood cohesion and adolescent mental health 15 and were adapted from previous neighborhood research 16 . This included 4 items, such as "this is a closeknit neighborhood," rated on a 5-point Likert scale ranging from "strongly agree" to "strongly disagree." Lastly, the child's mother reported on the level of intimate partner support for each wave using six items, such as "how frequently (the current romantic partner) expresses love and affection (for the mother)," that were rated on a 3-point Likert scale ranging from "never" to "often" 17 . This was also reverse coded such that high scores represent low support. Child exposure to the mother's intimate partner support was coded as missing for a given wave if the child did not live with their mother at least 50% of the time. We considered social deprivation to exist on a continuum in which high scores (e.g., the child experienced either high neglect or witnessed low social support for their mother or low neighborhood social cohesion) represent deprivation and low scores (e.g., the child experienced little neglect or witnessed high social support for their mother or high neighborhood social cohesion) represent social support.
Composite Score Calculation. To calculate composite scores, the Z scores for each of the childhood experiences (i.e., child abuse, exposure to intimate partner violence, community violence, child neglect, lack of romantic partner support, lack of neighborhood social cohesion) were summed for each of the childhood experiences within a dimension (i.e., violence exposure and social deprivation) 18 and then divided by the number of childhood experiences within a dimension for each participant, thus maximizing the number of participants and the diversity of the sample by minimizing drop out due to missing data at any given wave. This procedure has been previously described 2 .
In an exploratory attempt to characterize the sample, we examined the abuse and neglect subscales of the Conflict Tactics Scale (CTS). We found that averaged across ages 3, 5, and 9 years, the focal children experienced greater than 3 of the 10 abuse categories (M=3.67, SD=1.61, range: 0-10) and less than 1 of the 5 neglect categories (M=0.19, SD=0.37, range: 0-4) in the CTS scale.

Covariates
With the exception of mean framewise displacement, all covariates have been previously Race. Race was operationalized as a set of two dummy-coded variables based on three race categories reported by the teen: African American (Hispanic and Non-Hispanic), White/Caucasian, and Other. Other included Hispanic, Asian, multi-racial participants, Native American, and Unknown/Not Reported (Table 1). If teen report was not available (N = 12), then parent report was used.
Pubertal Development. Self-report of pubertal status was assessed using child report of the Pubertal Development Scale 22 that measures growth spurt in height, public hair, and skin change in boys and girls; facial hair growth and voice change in boys only; and breast development and menarche in girls only (Table 1). When adolescent report was not available (N=8), parent report was used. Pubertal development scores reported by parents were not significantly different from those reported by the adolescent (t(9.04)=-0.74, p=0.48).
Maternal Covariates at Birth. Maternal self-report of marital status at birth (yes/no) and education at birth (1 -less than high school, 2 -high school or equivalent, 3 -some college/technical school, 4 -college or graduate school) was assessed.
Framewise Displacement. Framewise displacement (FD) is a metric that evaluates motion in the scanner and is the average of rotation and translation parameter differences 23 . FD was quantified using fsl_motion_outliers (FSL v.5.0.7).

Sensitivity Analyses
Extreme Outliers. Predictor variables (i.e., violence exposure and social deprivation) were checked for extreme outliers, which were considered to be individuals with a value less than Q1 -2.2*Interquartile Range (IQR) or greater than Q3 + 2.2*IQR 24 . Four individuals were considered to be extreme outliers. Analyses were run with and without these participants, and inferences did not change. Thus, the participants were not excluded in order to retain as much data as possible.
Protecting Against Model Overfitting. To protect against model overfitting, the psi matrices for each individual were examined -these matrices summarize the variance that was not explained for each network node for each participant. Participants were flagged (N=26 or 14.8%) if they had a psi value greater than 1. To ensure that these participants were not driving effects, all analyses were run with and without participants with high psi values, and inferences did not change. Thus, the participants were not excluded in order to retain as much data as possible.  Note. A significant b-weight indicates the beta-weight and semi-partial correlation are also significant. b represents unstandardized regression weights. beta indicates the standardized regression weights. r represents the zero-order correlation. LL and UL indicate the lower and upper limits of a confidence interval, respectively. * indicates significant predictor of network density 1 Motion is measured using mean relative framewise displacement 2 Dummy coded variables represented 3 category race variable (African American, Caucasian, Other) 3 Current life stress is measured using the Adolescent Life Events Scale 4 Maternal variable at the child's birth eTable 4 Node degree for ROIs that were not significantly associated with violence exposure or social deprivation using the Bonferroni-corrected significance threshold.