Association of Prenatal Exposure to Early-Life Adversity With Neonatal Brain Volumes at Birth

Key Points Question Is prenatal exposure to maternal social disadvantage and psychosocial stress associated with global and relative infant brain volumes at birth? Findings In this longitudinal, observational cohort study of 280 mother-infant dyads, prenatal exposure to greater maternal social disadvantage, but not psychosocial stress, was associated with statistically significant reductions in white matter, cortical gray matter, and subcortical gray matter volumes and cortical folding at birth after accounting for maternal health and diet. Meaning These findings suggest that prenatal exposure to social disadvantage is associated with global reductions in brain volumes and folding in the first weeks of life.


Maternal Measures in the Social Disadvantage Construct.
Health insurance status (private, public/no insurance) and highest educational level were obtained at study entry (during the first trimester). Household income and composition were obtained in each of the three trimesters and generated the Income to Needs Ratio (I/R). 1 An I/R of 1.0 is equivalent to the federal poverty line. Home addresses were obtained at birth and used to calculate the national Area Deprivation Index (ADI) percentile. The ADI scores neighborhood disadvantage using US Census data regarding poverty, education, housing, and employment, with higher values indicating greater disadvantage. 2 The Diet History Questionnaire II (DHQ-II), 3 was obtained at the time of neonatal scan. The DHQ-II is a yearly food frequency measure used to characterize nutrition via the Healthy Eating Index (HEI). 4

Maternal Measures in the Psychosocial Stress Construct. Psychological measures of the
Perceived Stress Scale (PSS) 5 and the Edinburgh Postnatal Depression Scale (EPDS) 6 were collected in each trimester. The Stress and Adversity Inventory (STRAIN) 7 is a composite measure of stressful and traumatic life experiences that was obtained at time of neonatal scan (n=186) or at follow-up at one or two years (n=77). On post-hoc analyses, we did not find differences in the STRAIN stressful/traumatic life event count (t-statistic=.85, two-tailed p=0. 4) or severity (t-statistic=1.01, two-tailed p=0.3) between mothers who had this collected at birth or at subsequent follow up. The Everyday Discrimination Scale (EDS) was obtained at time of neonatal scan and was scored for the "day-to-day" experience of racial discrimination, with participant response choices that ranged from "never" or "less than once a year" to "every day". 8 Latent Constructs. Confirmatory factor analysis, distinct from exploratory factor analysis, confirms that variables identified a priori load on each factor. MPlus software was used to validate our a priori grouping of early life adversity variables into a Social Disadvantage latent factor (variables listed above) and a Psychosocial Stress factor (variables listed above).
Maximum likelihood estimation was used to derive latent factor scores for these two composite measures for all participants, despite occasional missing datapoints in observed variables. 9 Selfreported race was highly correlated with Social Disadvantage, offering no additional improvement to the model after other variables were accounted for and, thus, it was not included in either the latent Social Disadvantage or Psychosocial Stress composites.
Additionally, maternal substance use, health, and BMI all have complex relationships with both SES and psychosocial factors. Therefore, we analyzed these measures independently of our defined constructs of Social Disadvantage and Psychosocial Stress. The T2-weighted images were then preprocessed using the following standard steps: gradient and readout distortion correction using the Human Connectome Project preprocessing pipeline, 10 FSL axis reorientation to the MNI152 standard-space template, 11 image denoising using Advanced Normalization Tools for Brain and Image Analysis (ANTS) Registration Suite, 12 and co-registration using the Washington University School of Medicine Neuroimaging Laboratory (NIL)'s 4-dimensional floating point (4dfp)-based image analysis. 13 The resulting T2 images were then used as input for Melbourne Children's Regional Infant Brain atlas Surface (M-CRIB-S) segmentation and surface extraction toolkit, which automatically generated anatomical volume segmentations and reconstructed cortical surfaces. 14,15 The M-CRIB-S toolkit included N4 bias field correction and brain extraction, as well as automatic segmentation into white and gray matter, cerebellum, brainstem, and subcortical gray matter subdivisions corresponding to FreeSurfer-like labeling. Curvature-based spherical registration and mapping, alignment, and averaging were performed, allowing for spatial normalization within the cohort and to the M-CRIB atlas.

MRI Data Collection, Preprocessing, and Brain Volumetric
The segmentation volumes and the cortical surfaces were then projected on the T2 images  Step 1 Step 2 .46 a Standardized coefficient values. b Birthweight was not included as an independent variable for relative region of interest volumes adjusted for total brain volume to avoid overfitting. Standardized region of interest volumes were computed as the volume of the region divided by total brain volume. Step 1