Statistical Analysis

In the resting-state fMRI data, we first performed a one-sample t-test to determine in each group the significant clusters following the backward regression to individual ICA components. This t-test identified significant ICA components in each group, and provided confirmation that the technique was successful. Between-group comparisons were examined by submitting functional connectivity z-scores, FW, and FA_T values to voxel-wise, independent-samples t-tests (i.e., Dyt1 KI versus WT). Within- and between-group differences within a single voxel were thresholded at a voxel-level of P = 0.005. Functional and diffusion MRI data were corrected for multiple comparisons using a Monte Carlo simulation. The signi cance level of the contrasts of interest were set at minimum cluster size of 0.200 mm³, the equivalent of P < 0.05 corrected using the family-wise error rate (FWER). Notably, one Dyt1 KI mouse was excluded from fMRI analyses due to incomplete volume acquisition during scanning and five mice (one Dyt1 KI and four WT) were removed from dMRI analyses due to motion-related signal distortions and issues with partial alignment during data pre-processing. For subjects included in both analyses (18 Dyt1 KI, 16 WT), we computed the relation between FW values and functional connectivity z-scores using Pearson’s correlation coefficients. Correlations were performed by combining FW and functional connectivity values across Dyt1 KI and WT mice. P-values were adjusted for multiple comparisons at a false discovery rate (FDR) of 0.05 using the Benjamini-Hochberg-Yekutieli method in MATLAB.

In a subsequent analysis, we sought to determine if resting-state functional connectivity differences alone could distinguish between mouse genotypes (i.e., Dyt1 KI and WT). To accomplish that objective, in a training cohort of 24 (12 Dyt1 KI, 12 WT) randomly selected mice we extracted the average z-score values from component clusters where significant between-group effects were detected – and submitted data to a linear kernel support vector machine classification algorithm with 10-fold cross-validation using the library for Support Vector Machines (LIBSVM) (Chang and Lin, 2011). Reliability of the algorithm was evaluated in a testing cohort of 15 mice (seven Dyt1 KI, eight WT).