*2.5. MRI Data Processing*

The T1 MP-RAGE structural image was put into the automated segmentation software, FreeSurfer version 6.0 hippocampal module (http://surfer.nmr.mgh.harvard.edu/, RRID:SCR\_001847) to examine total hippocampal grey matter volume and grey matter volume in the hippocampal subfields. The procedure uses Bayesian inference and a probabilistic atlas of the hippocampal formation based on manual delineations of subfields in ultra-high-resolution MRI scans [38]. Manual quality check of automated hippocampal segmentation was performed for each participant following an existing protocol [39]. The segmentation of the hippocampus was visually assessed by an individual trained in hippocampal neuroanatomy and then given a rating of "pass", "pass on condition", and "fail". Images that failed to have defined landmarks due to motion artifacts or segmentation error were excluded. Although twelve subfield volumes are generated by FreeSurfer 6.0, we only included subfields that have been shown to be preferentially affected by high sugar diets, including the CA1, CA2/3, CA4, DG (granule cell layer) and subiculum [40,41]. Previous studies in children have used FreeSurfer to segment the hippocampus and hippocampal subfields [42,43]. The raw volume data were included in the supplemental materials (Table S1).

Tractography models were created from the diffusion-weighted MRI (dMRI) data using FSL [44] and the Quantitative Imaging Toolkit (QIT) [45]. The dMRIs were first skull stripped using FSL BET and then corrected for motion and eddy current artifact using FSL FLIRT. For this, each diffusion scan was affinely registered to the baseline scan using the mutual information metric, and the associated gradient orientations were rotated to account for the registration. Diffusion tensor models were then estimated from the dMRI using QIT, and the following tensor parameters were extracted: fractional anisotropy (FA), mean diffusivity (MD), axial diffusivity (AD), and radial diffusivity (RD). Tensor images were upsampled to 1 mm<sup>3</sup> using model-based interpolation in QIT [46], and a deformation field was computed using DTI-TK [47] to register the data to the IIT brain template [48]. Tractography models of the bundles-of-interest were created using a framework for deterministic streamline integration [49]. For each bundle, seed, inclusion and exclusion masks were manually drawn in the IIT template [50] in reference to a white matter atlas [51]. The template masks were then resampled in each subject's native space image to constrain tractography. Other tractography parameters included a step size of 1.0 mm, a maximum angle of 45 degrees, a minimum FA of 0.15, and 25000 seeds per bundle. Bundle-specific metrics of fractional anisotropy (FA), mean diffusivity (MD), axial diffusivity (AD), and radial diffusivity (RD) were then computed. FA and MD were used as primary metrics of microstructure, with many studies showing that FA increases with age while MD decreases with age [52–55]. AD and RD were used as post-hoc measures in regions that showed a significant effect on MD. The MD measure is a weighted average of AD and RD, which themselves are more biologically specific than MD alone. The bundles of interest included major connections between the hippocampus and the rest of the brain, specifically: the uncinate fasciculus, fornix, and cingulum bundle (separated into prefrontal and temporal lobe sections) (See Figure 1A).
