Analyses examined whether alcohol consumption (aAUDIT-C) predicted GMV only within regions of interest (ROIs) where associations were observed in the discovery DNS sample (Figure 1, Table S1 in Supplement 1). ROIs were defined by the Automated Anatomic Labeling atlas (45). A voxelwise generalized linear model regression was conducted using multilevel block permutation–based nonparametric testing (FSL PALM v.alpha103; tail approximation p < .10 with 5000 permutations), which accounts for the family structure of the HCP data while correcting for multiple comparisons (46–48). Covariates included sex, age, self-reported race and/or ethnicity, intracranial volume, twin and/or sibling status (dizygotic or not, monozygotic or not, half-sibling or not), presence of a diagnosis other than alcohol or substance abuse or dependence, perceived stress, education level, and SES. Analyses were thresholded at p < .05 familywise error corrected with a cluster extent threshold of 10 contiguous voxels (ke = 10).
