Exploratory Probabilistic Tractography: Clusters (resulting from between group TBSS analysis) were transformed back to each subject’s native space to create the individualized seed masks. As discussed in the main text, probabilistic tracking was then carried out separately for both conditions (i.e. free and constrained) seeding from the clusters. Using bedpostX, estimates of fiber orientation and their uncertainty were calculated at each voxel. This model also accounts for the possibility of crossing fibers within each voxel41. We used the default parameters with 5000 sample pathways per each seed voxel with a curvature threshold of 0.2 (corresponding to ±80°). Pathways were also terminated after 2000 steps, using a step length of 0.5 mm. Finally, to compute percentage of SWM-restricted tracts for each cluster, the resulting waytotals for both rounds of tractography were then fed into the Equation 1 for each subject. Further, population probability maps for the tractography outputs seeded from each cluster were created as follows: first, binary maps were generated by thresholding each subject's individual tractography result at 0.01 of waytotal. Next, the resulting binary maps were nonlinearly transformed to the MNI space by applying warp fields generated during TBSS registration stage. Finally, population probability maps were created by averaging individual binary maps in the MNI space. For each healthy subject, this process was repeated twice for tractography outputs of both conditions: i) Free (without any exclusion mask); ii) Constrained (with exclusion mask). For illustration purposes, the resulting probability maps were thresholded at 10%. Supplementary Figure 1 depicts population-averaged tractography outcomes for each cluster in both conditions (free and constrained). The figure shows that fiber tracking constrained with the SWM mask selectively excluded long range connections by eliminating tracts extending to the deep white matter. Figure S1. Thresholded population probability maps of tractography outcomes seeded from different clusters with and without exclusion mask. Table S1. Cognitive scores of healthy subjects and patients with schizophrenia. Cognitive Task Healthy Subjects (SD) Schizophrenia Patients (SD) PBonferroni-corrected Letter Number Sequence 16.5 (3.4) 12.1 (4.2) <0.0001 Trails B -2.4 (1.1) -4.7 (3.0) 0.0002 Digit Span 12.1 (2.3) 10.0 (2.4) 0.0006 Digit Symbol Coding Task 55.5 (10.2) 43.7 (9.3) <0.0001 Story Recall 10.6 (1.3) 7.8 (2.6) <0.0001 Table S2. Percentage of superficial white matter restricted streamlines originating from each cluster in probabilistic tractography. %SWM-restricted (SD) Seed Masks P-value Schizophrenia Patients Healthy Subjects Cluster-1 60.6 (5.6) 63.9 (8.0) 0.031 Cluster-2 74.1 (3.0) 79.8 (6.8) <0.0001* Cluster-3 86.9 (4.0) 90.7(5.6) 0.0007* Cluster-4 69.9 (6.8) 68.0 (12.7) 0.37 Cluster-5 39.2 (11.5) 42.9 (28.1) 0.44 Whole-brain SWM 83.4 (2.3) 83.4 (1.9) 0.97 SWM: superficial white matter Due to between-group differences in percentage of SWM-restricted streamlines seeding from Cluster 2 and Cluster 3, we re-analyzed between-group differences in SWM-FA including SWM-restricted streamline percentage as a covariate for all 5 clusters. SWM-FA in all 5 clusters remained significantly different between groups, at the Bonferroni-corrected threshold.