Nonlinear associations between body mass index and brain microstructure across adolescence in the ABCD Study

Body mass index (BMI) is widely used to screen for weight-related health risks during adolescence. Prior neuroimaging studies have assumed a linear relationship between BMI and brain microstructure, potentially obscuring how this association varies across the BMI distribution. Using restriction spectrum imaging (RSI) in the Adolescent Brain Cognitive Development (ABCD) Study, previous work has identified positive linear associations between BMI and weight-related metrics and the restricted normalized isotropic (RNI) signal fraction in subcortical structures, but it remains unclear whether these associations are uniform across the full BMI spectrum or driven by particular portions of the distribution.

We examined the relationship between BMI percentile and voxelwise RNI in subcortical gray matter and white matter structures using data from the ABCD Study 6.1 release, which includes four imaging timepoints spanning ages 9-18 years (22,011 observations from 10,465 unique participants). Sex-stratified generalized additive mixed-effects models with smooth terms for BMI percentile, age, and pubertal development were used to model the shape of the BMI-microstructure association across the full percentile range, controlling for genetic principal components, household income, parental education, and MRI scanner/software version.

The association between BMI percentile and RNI was nonlinear in the bilateral nucleus accumbens, caudate, pallidum, putamen, thalamus, and forceps minor. A modest, positive association was present across most of the BMI range, but the rate of change accelerated markedly above the 80^th percentile. This pattern was consistent across structures and sexes, though the overall magnitude of the partial effect was higher for males across most structures, while females showed steeper rates of change in most structures above the 80^th percentile. Voxelwise analyses revealed spatial heterogeneity within structures, with stronger effects concentrated in specific subregions including the posterior forceps minor, dorsal pallidum, anterior putamen, and posterior thalamus.

The relationship between BMI and subcortical brain microstructure during adolescence is not uniform but instead accelerates at the upper end of the BMI distribution, suggesting that prior linear estimates may reflect a blended average of a modest slope across most of the range and a steep slope above the 80^th percentile. These findings extend the existing literature by capturing a wider developmental window, employing voxelwise rather than ROI-averaged analyses, identifying the forceps minor as a novel region of interest, and highlighting the advantages of nonlinear modeling in revealing dynamic associations.