Structural abnormality of the corticospinal tract in major depressive disorder

Participants were 14 women diagnosed with MDD and 18 healthy, age-matched, female controls (CTLs) 18–55 years of age. The Structured Clinical Interview for DSM-IV-TR Axis I (SCID-I) [17] was used to establish a psychiatric diagnosis of MDD based on DSM-IV-TR criteria. To qualify for study entry, individuals in the CTL group could not have met criteria for any past or current DSM-IV-TR Axis I disorder. Exclusion criteria for both MDD and CTL participants included current alcohol or substance abuse or dependence and head trauma resulting in loss of consciousness greater than 5 min. During the SCID-I, to assess age of onset of depression, depressed participants were asked at what age they first experienced a depressive episode. A trained interviewer also completed the Global Assessment of Functioning (GAF) scale [18]. This scale indexes, from 1 to 100 (sickest to healthiest), the level of the participants’ social, occupational, and psychological functioning. Severity of depression was assessed with the Beck Depression Inventory-II (BDI-II [19]). The Stanford University Institutional Review Board approved the study and informed consent was collected from each participant.

Whole-brain diffusion-weighted and high-resolution T1-weighted images were collected using a Discovery MR750 3.0 T MR system (GE Medical Systems, Milwaukee, WI, USA), housed at the Stanford Center for Neurobiological Imaging. The T1-weighted images were used for anatomical registration (spoiled gradient echo (SPGR) pulse sequence; repetition time (TR) = 6,240 ms; echo time (TE) = 2.34 ms; flip angle = 12°; resolution = 0.9 mm isotropic; 186 slices; scan duration = 5 min 15 s). The diffusion-weighted scan was a single-shot, dual-spin-echo, echo-planar imaging sequence (96 unique directions; b = 2,000 s/mm²; TR = 8,500 ms; TE = 93.6 ms; resolution = 2 mm isotropic; 64 slices; scan duration = 15 min 1 s). Nine non-diffusion-weighted (b = 0 s/mm²) volumes were additionally collected for anatomical localization and registration purposes.

AFQ systematically uses whole-brain tractography methods to characterize major white matter fiber tracts. Here we briefly describe the AFQ procedure (see Additional file 1 for more detail). First, diffusion data were preprocessed, including motion correction, data alignment, resampling, and trilinear interpolation [20]. Tensors were then fit at each voxel using a robust tensor fitting method [21], and FA was computed as the normalized standard deviation of the tensor’s eigenvalues. FA ranges from 0 (perfectly isotropic) to 1 (perfectly anisotropic diffusion). Following this, tractography was estimated using a deterministic streamline tracing algorithm [22, 23]. Then, waypoint ROIs labeled on the MNI template were warped into participant-specific diffusion space, and fibers intersecting these ROIs were identified. After a series of fiber cleaning and tract refinement steps, the central portion of each fiber tract was located and diffusion metrics were computed along this core, resulting in a “tract profile”. These tract profiles permit the systematic and unbiased assessment of group differences in diffusion metrics, FA in this study. After identifying tract profiles, we computed the mean FA along each white matter tract.

The MDP approach allows for the automated assessment of compact and localized white matter paths on an individual-participant basis [15] (see Additional file 1 for more detail). Because MDPs are smaller, more numerous, and associated with major white matter tracts (i.e., they are located in major tracts or in areas to which these tracts project), they offer additional information to that obtained using AFQ. We identified MDPs in 50 white matter regions described in the Johns Hopkins University white matter atlas, resulting in a total of 67 MDPs (several regions have more than one MDP). To implement this procedure, we first corrected the diffusion data for eddy current and motion effects; next, we estimated whole-brain tractography using an optimized global probabilistic tractography method [24]. Then, from the whole-brain tractography computed using the global method, we created fiber density images for each white matter ROI by identifying fibers that intersect with the ROI (AFQ-identified fibers intersected pairs of ROIs). The next step employed graph theoretical analysis. Specifically, fiber density graphs were created with nodes as voxel locations and edges as density information. Seed points identified in the white matter atlas were then warped into each fiber density graph image. Using an optimized grid search method to find the path of highest density [25], MDPs were identified between each pair of seed points. The resulting paths were compact representations of the given tract’s scale/size, location, and geometry/shape. Finally, paths were registered spatially across individuals using a geodesic curve registration procedure [26, 27], allowing us to conduct between-group comparisons of FA in a point-wise manner.

In the first stage of analysis, we used two-sample t-tests to compare the 18 AFQ-identified fiber tract core mean FA values for the MDD and CTL groups. To correct for false positive inflation as a result of multiple comparisons, we implemented a false discovery rate (FDR) procedure (q = 0.05) [28]. This analysis identified major fiber bundles in which there were depression-related abnormalities. Using Pearson linear partial correlation (controlling for age), we assessed the correlations between identified abnormal fiber tracts and age of onset of depression, severity of depression (BDI-II scores), and level of global functioning (GAF scores) in the MDD group.

Means and standard deviations for demographic and clinical variables for the 14 depressed and 18 control female participants are presented in Table 1. The two groups did not differ in age (t(30) = -1.53, p > 0.10), handedness (χ²(1) = 0.14, p > 0.10), or level of education achieved (t(30) = -1.34, p > 0.10). As expected, the depressed participants had significantly higher BDI-II scores than did the never-depressed controls. Half of the participants in the depressed group met criteria for at least one anxiety disorder, and three depressed participants were currently taking psychotropic medications (see Table 2).

Across the entire sample, AFQ was unable to characterize 5 of the 576 fiber tracts (i.e., N × [number of fiber tracks] = 32 × 18): the callosum forceps major for three participants and the callosum forceps minor for two participants. These participants were excluded from analyses involving these particular fiber tracts. The failure of AFQ to identify these fibers may be a result of crossing fibers, noise in the data, abnormal anatomy that caused problems for automated segmentation, or small fiber tracks for which it is difficult to compute statistics. Of the 18 analyzed fiber groups (Table 3), two distinguished MDD from CTL participants after correcting for multiple comparisons: the left corticospinal tract (CST) (t(30) = 3.45, p < 0.002) and the right CST (t(30) = 3.79, p < 0.001) (Table 4, Figure 1). Both CSTs were characterized by greater FA in the MDD than in the CTL group (Table 4). As an exploratory analysis, we divided the group of MDD participants into two subgroups based on the presence or absence of comorbid anxiety; these two subgroups did not differ in FA for either the left or the right CST (p’s > 0.10). In addition, the group differences in CST were unchanged after removing the three MDD participants who were taking psychotropic medications (left CST: t(27) = 3.45, p < 0.002; right CST: t(27) = 3.11, p < 0.005). Finally, in the MDD group, we correlated age of onset and severity of depression, and level of global functioning, with mean FA individually for both left and right CST. No significant correlations were obtained (p > 0.10).

For the second stage of the analysis plan, we identified MDPs associated with the AFQ-identified CST. This resulted in the identification of 24 unique MDPs from seven white matter ROIs (from a total of 50). The 24 MDPs (12 in each hemisphere) comprised 35.8% of the total set of 67 MDPs (Table 5). These MDPs overlap, are along the same white matter bundle, or are spatially proximal to the CST projections identified in the AFQ analysis.

We conducted two-sample t-tests on a point-by-point basis along each of the identified MDPs. After correcting for multiple comparisons by using FDR independently for each MDP, the MDD and CTL groups exhibited point-wise differences in four MDPs: the left posterior limb of the internal capsule, the right posterior limb of the internal capsule, the right superior corona radiata, and the left external capsule (Table 6, Figure 2). Because each of the four MDPs was located in a unique region, and because each of these regions included two analyzed MDPs, only one of the two MDPs identified for each implicated white matter region yielded point-wise group differences after FDR correction. Of the identified points that differed between groups (27 total across the four MDPs), all except three of the six points of the left external capsule were characterized by greater FA in the MDD than in the CTL group.