Prenatal social disadvantage is associated with alterations in functional networks at birth

Abstract

Childhood exposure to social disadvantage is a major risk factor for psychiatric disorders and poor developmental, educational, and occupational outcomes, presumably because adverse exposures alter the neurodevelopmental processes that contribute to risk trajectories. Yet, given the limited social mobility in the United States and other countries, childhood social disadvantage is frequently preceded by maternal social disadvantage during pregnancy, potentially altering fetal brain development during a period of high neuroplasticity through hormonal, microbiome, epigenetic, and immune factors that cross the placenta and fetal blood-brain barrier. The current study examines prenatal social disadvantage to determine whether these exposures in utero are associated with alterations in functional brain networks as early as birth. As part of the Early Life Adversity and Biological Embedding study, mothers were recruited during pregnancy, prenatal social disadvantage was assessed across trimesters, and their healthy, full-term offspring were imaged using resting-state functional magnetic resonance imaging during the first weeks of life. Multivariate machine learning methods revealed that neonatal functional connectivity (FC) varied as a function of prenatal exposure to social disadvantage (n = 261, R = 0.43, R² = 0.18), with validation in an independent sample. Alterations in FC associated with prenatal social disadvantage occurred brain-wide and were most pronounced in association networks (fronto-parietal, ventral attention, dorsal attention) and the somatomotor network. Amygdala FC was altered at birth, with a pattern shared across subcortical structures. These findings provide critical insights into how early in development functional networks begin to diverge in the context of social disadvantage and elucidate the functional networks that are most impacted.

Keywords: functional connectivity; infant; prenatal exposure; social disadvantage; socieconomic status.

Fig. 1.

Study overview. Prenatal social disadvantage was measured as a latent composite of maternal education, health insurance status, income-to-needs ratio, area deprivation index, and a healthy eating index. Infant FC was measured by imaging neonates within the first weeks of life, extracting fMRI timeseries from 333 parcels covering 12 functional networks and 14 subcortical regions, and calculating the temporal correlation between each pair of regions and/or parcels. In a 10-fold cross validation framework, SVR was used to train a model to detect differences in prenatal social disadvantage from patterns in infant FC. Trained models were evaluated and interrogated during the testing phase. (1) As part of the 10-fold cross validation framework, the trained model was evaluated using the 10% of the eLABE sample left out of training and infant FC was used to estimate prenatal social disadvantage for the left-out sample. Variance explained (R²) was quantified by comparing the estimated social disadvantage from FC and true social disadvantage across all individuals in the eLABE sample. (2) An independent validation sample was used to evaluate the generalizability of the trained model and infant FC was used to estimate economic disadvantage for the CUDDEL sample. Ten estimates of social disadvantage were generated for each individual in the CUDDEL sample corresponding to the 10 folds of cross-validation. Variance explained (R²) was quantified by comparing the average estimate of social disadvantage based on FC with the true Income-to-Needs Ratio across all individuals in the CUDDEL sample. (3) To determine whether each individual functional network contains FC that varies with social disadvantage, SVR was used to train a model to detect differences in prenatal social disadvantage using only the FC from a single functional network (within and between network connections). Variance explained (R²) was quantified by comparing the estimated social disadvantage based on FC from a single functional network and the true social disadvantage across all individuals in the eLABE sample. (4) To interrogate how each individual functional network contributes to the estimation of prenatal social disadvantage from brain-wide patterns of FC, FC from a single functional network (within network and between network connections) or a single parcel/subcortical ROI (similar to a seed map) was randomly permuted across the left-out 10% of subjects during the testing phase of 10-fold cross validation, removing any variation related to prenatal social disadvantage. Reduction in variance explained (ΔR²) was quantified by comparing the variance explained when estimating prenatal social disadvantage using with and without permuted functional connections.

Fig. 2.

Prenatal exposure to social disadvantage can be estimated from infant FC. (A) Estimates of social disadvantage based on infant FC were generated through 10-fold cross-validation in the eLABE sample. In turn, 10% of the sample was kept aside and the remaining 90% of the sample was used to identify patterns of infant FC that could detect differences in prenatal social disadvantage. The 10% of the sample that was left out of training was then used for testing, such that each subject in the eLABE sample has an estimate of social disadvantage based on infant FC. The actual social disadvantage depicted is the residuals after controlling for head motion, amount of data, sex, maternal medical risk, tobacco, and cannabis. (B) Estimates of social disadvantage generated in the independent validation sample (CUDDEL, red) using the models derived from the eLABE sample. Estimates of social disadvantage based on infant FC were generated for each of the 10 models derived during 10-fold cross-validation in the eLABE sample and then were averaged. Because not all measures in the composite social disadvantage variable were available in the CUDDEL study, the model was evaluated by comparing the average estimate of social disadvantage based on FC with the true economic disadvantage (i.e., inverse of Income-to-Needs Ratio) across all individuals in the CUDDEL sample. The relationship between estimates of social disadvantage based on infant FC in the eLABE sample (black) and economic disadvantage is provided for reference.

Fig. 3.

Interrogating which functional networks contribute to the detection of prenatal social disadvantage from FC. (A) For each functional network (FP: frontoparietal, DAN: dorsal attention, CO: cingulo-opercular, SAL: salience, VAN: ventral attention, DMN: default mode, SMH: somatomotor-hand, SMM: somatomotor-mouth, AUD: auditory, RST: retrosplenial, VIS: visual, CP: cinguloparietal), patterns of FC involving only parcels from a single functional network that were related to social disadvantage were identified with SVR in the eLABE sample. Variance explained (R²) by each network (circle) is depicted in relation to the variance explained by chance (dashed line) such that P < 0.05. (B) The relation between the detection of prenatal social disadvantage and functional network identity as a function of the number of parcels sampled from a single functional network. Patterns of FC involving a set number (e.g., 20) of randomly selected parcels from a single network that were related to social disadvantage were identified with SVR in the eLABE sample. This process was repeated 50 times for each possible subnetwork size for each functional network (e.g., 1 to 41 for the default-mode network, 1 to 4 for the SAL network). Curves represent the average variance explained across the 50 random repetitions. (C) Reduction in variance explained when the FC from a single functional network (circle) was randomly permuted before estimating prenatal social disadvantage. Black dots represent the null, where the FC from a matched number of randomly selected connections was permuted before estimating prenatal social disadvantage. For each network, a null was generated from 50 random repetitions. Asterisks represent an adjusted P-value < 0.05 when using false discovery rate. (D) Functional networks whose FC is uniquely important for detection of prenatal social disadvantage.

Fig. 4.

Interrogating which subcortical structures contribute to the detection of prenatal social disadvantage from FC. (A) Reduction in variance explained when the FC from a single subcortical structure (average of Left and Right) was randomly permuted before estimating prenatal social disadvantage (circle). Black dots represent the null, where the FC from a matched number of randomly selected connections was permuted before estimating prenatal social disadvantage. A null was generated from 1,000 random repetitions. Asterisks represent an adjusted P-value < 0.05 when using false discovery rate. (B) The association of prenatal social disadvantage and the FC between each subcortical structure and the whole brain with is displayed for each of seven subcortical structures. Blue arrows indicate the lower FC between each subcortical structure and the medial prefrontal cortex associated with greater social disadvantage. Yellow arrows indicate the higher FC between each subcortical structure and the sensorimotor cortices associated with greater social disadvantage. (C) The spatial similarity of the relationship between infant FC and prenatal social disadvantage for each pair of subcortical structures.