Major depressive disorder (MDD) is the most common psychiatric disorder in the United States [1], accounting for approximately half of disability-adjusted life years worldwide [2], with major economic and personal costs [3]. MDD involves a wide range of symptoms, including most prominently negative affect and anhedonia (loss of pleasure), as well as difficulties in psychomotor functioning, sleep, and weight changes. If we could better understand the neural basis of MDD, we may be able to better prevent and treat this debilitating disorder.

Across different areas of neuroscience, there has been growing interest in delineating brain networks, in contrast to examining specific brain regions in isolation. Networks of brain regions have been increasingly implicated in depressive pathology, underscoring the need to understand depression-related anomalies in the connections among these regions [4–6]. In this context, diffusion-weighted imaging can assess diffusion properties of white matter and can be used to infer brain connectivity. Using diffusion tensor imaging (DTI), water diffusion can be quantified using fractional anisotropy (FA), which measures the degree of directional preference in water diffusion. FA, the most commonly used diffusion metric, is influenced by intra-voxel orientation dispersion, axonal myelination and packing density, membrane permeability, the number of axons, and partial volume effects [7]. Moreover, tractography algorithms can use diffusion tensor information to estimate the location and direction of fiber tracts. DTI has been used to characterize abnormal white matter diffusion properties in a range of diseases, including psychiatric disorders involving psychosis and disturbances in mood and attention [4, 5, 8, 9].

To our knowledge, there have been three reviews documenting diffusion abnormalities in MDD [4–6]. Across these three reviews, there has been considerable discrepancy in the direction and location of effects of white matter abnormalities in MDD. This may be because of significant heterogeneity in participant samples (e.g., half of the studies included in one review assessed elderly individuals [5]), meta-analytic methods (e.g., qualitative [5], signed differential mapping (SDM) [6], and activation likelihood estimation (ALE) [4]), individual study analysis techniques (e.g., tractography, voxel-based analysis (VBA), or tract-based spatial statistics (TBSS)), and/or study inclusion criteria (e.g., only analyzing decreases in FA [6]). Thus, our current understanding of white matter pathology in MDD is based on relatively few studies that, themselves, incorporate heterogeneous methodological approaches. Most studies of diffusion in MDD have not assessed tractography, but instead examined FA or other diffusion measures in specific regions of interest (ROIs), or, globally, using VBA or TBSS [4–6]. Tractography uses directional information from the diffusion data to extract diffusion properties from specific fiber tracts, and it may offer greater power to detect disease-related abnormalities than do VBA and TBSS [10].

Few studies have used tractography-based methods to characterize white matter connectivity in MDD. Zhang et al. first used tractography to identify the cingulum bundle and uncinate fasciculi and then estimated diffusion properties in these fiber tracts. These investigators found that FA was lower and mean diffusivity was higher in the right uncinate fasciculus in depressed individuals relative to nondepressed controls [10]. In a second study, Zhang and colleagues found MDD-related reductions in FA in tractography-identified anterior limb of the internal capsule, an important component of the cortico-striatal-pallidal-thalamic (CSPT) circuit [11]. Finally, in a connectomics framework, tractography and graph theory have been used to explicate large-scale network abnormalities in depression [12, 13].

Whole-brain tractography commonly includes tens of thousands of fibers; consequently, findings using this technique in isolation can be difficult to interpret. To better understand such massive amounts of data, whole-brain tractography is often summarized. One data reduction method identifies key fiber tracts by requiring manual tracing of an ROI which is followed by algorithmic assessment of the fibers that pass through it (as in [10, 11]). This manual identification of ROIs is time consuming, however, and limits the number of fascicles that can be assessed. In addition, manual tracing methods can introduce investigator bias during selection and tracing of ROIs. In contrast, clustering methods permit the automated, unbiased summarization of fiber tract information, by using anatomical and DTI information to locate important fiber tracts. Automated fiber quantification (AFQ) [14] and the maximum density path (MDP) [15] approach are two such clustering methods. Briefly, AFQ identifies important white matter tracts by assessing sets of fibers that intersect pairs of waypoint ROIs. Similarly, the MDP procedure uses a graph search method in a set of white matter ROIs to identify abnormalities in fiber paths. MDPs are smaller and more numerous than the AFQ-identified tracts and provide complementary anatomical information.

Given the likely importance of anomalies in white matter connectivity in MDD, the inconsistency in the literature concerning diffusion-related findings in this disorder, and the recent development of sensitive, automated, tractography clustering methods, the present study was designed to use the AFQ and MDP tractography clustering methods to automatically characterize properties of white matter diffusion in MDD. First, we used AFQ to identify depression-related anomalies in FA in 18 major white matter paths. MDPs allow for additional and complementary information relative to tract properties derived by AFQ, given their small size, large number, and association with major white matter tracts. After identifying abnormal white matter pathways using AFQ, we conducted secondary analyses in a subset of MDPs that were associated with these specific paths. In addition, given evidence that age of onset of depression and severity of disorder are related to abnormalities in white matter properties [6, 16], we assessed the relations between these two variables as well as the level of global functioning and diffusion properties of abnormal white matter paths.

Thus, we used information from tractography and took advantage of the lower bias and higher efficiency of two automated clustering methods to study major white matter paths in MDD. We hypothesized that FA would be lower in depressed individuals in the uncinate fasciculus, which links regions associated with emotion processing (e.g., hippocampus, amygdala) with regions implicated in cognitive control (e.g., prefrontal cortex).

The literature examining white matter abnormalities in MDD is methodologically varied and sparse and has yielded inconsistent findings. In this context, the current study was designed to capitalize on the improved detection power of tractography, the reduced bias of automated clustering methods, and more systematic and data-driven analysis, to assess abnormalities in white matter in MDD and begin to yield a more systematic and reliable connectomics of depression. Indeed, this is the first study to use automated tractography clustering to characterize white matter in depression. Our analyses included fiber tracts that have been previously studied in this disorder, in addition to several tracts that have not before been examined.

Using AFQ [14], we found that MDD was characterized by abnormalities in FA in the bilateral corticospinal tracts. We then used the MDP procedure [15] to further probe localized abnormalities that were associated with these group differences. This analysis revealed, for the first time, increased FA in the bilateral posterior limbs of the internal capsule, right superior corona radiata, and the left external capsule in MDD.

Previous studies have primarily documented reduced FA associated with MDD. In contrast, the current results include almost exclusively increased FA in participants diagnosed with this disorder. This discrepancy may be a result of the small number of studies considered in the previous reviews (12 studies [5], 11 studies [4], 7 studies [6]). In addition, Liao et al. only analyzed data indicating increases in FA in MDD [4]; the two quantitative reviews [4, 6] excluded tractography studies; and the third review [5] included only one tractography study in the discussion of MDD. Because tractography incorporates directionality information that is used to identify important white matter pathways, it may allow for greater detection power than do whole-brain voxel-wise techniques (e.g., VBA or TBSS) and, thus, may explain why previous research did not report increases in FA in CST in MDD. It is also possible that the automated clustering methods that we implemented yield greater spatial specificity than do earlier methods, and that decreases in FA in MDD that were previously identified using VBA and TBSS were, in fact, inaccurately reported to be localized to major fiber tracts.

It is important to also note that other investigators have reported depression-related increases in FA. For example, Blood et al. found that the ventral tegmental area was associated with greater FA in MDD than in control participants [29]. Moreover, several studies have identified regions of increased FA in bipolar disorder (BD) [5] in areas of the corpus callosum [30] and the frontal lobe [31], including the uncinate fasciculus, optic radiation, and anterior thalamic radiation [32]. Given recent interest in examining transdiagnostic factors in the proposed NIMH RDoC framework, it will be important in future research to investigate how tract-specific FA may correspond to signs and symptoms of disorders of emotion and mood with respect to specific RDoC domains and constructs.

Previous research has suggested that increased FA of the CSTs is related to decreases in FA of the superior longitudinal fasciculi (SLF). Specifically, Douaud et al. reported increases in CST FA in individuals with mild cognitive impairment and Alzheimer’s disease compared to healthy controls; moreover, using a method of quantitative crossing fiber tractography, Douaud et al. found that the increases in CST FA were associated with reduced FA of SLF association fibers in a crossing fiber region at the level of the centrum semiovale [33]. Although these findings raise the intriguing possibility that MDD-related increases in FA of the CST are related to selective sparing of this tract with concurrent abnormality in the SLF, we did not find SLF abnormalities in the current study. Future research using imaging and tractography methods that allow for greater resolution of crossing fibers may permit a better assessment of whether increases in CST FA in MDD are related to abnormalities in regions of crossing fibers.

Given the role of the CSTs in motor processes, it is possible that our findings of anomalous FA in these structures are related to psychomotor symptoms that often characterize MDD [34]. More specifically, motor retardation and agitation, criterion symptoms of MDD, may result from aberrations in white matter microstructure connecting the brainstem to motoric regions of cortical gray matter [35]. The current findings offer a foundation from which future research might explore this hypothesis. Importantly, although the white matter of primate CSTs is understood to arise primarily from the primary motor cortex, projections from the somatosensory, cingulate, and insular cortices are also represented [35]. Thus, the CST is likely to be involved in a variety of functions and thus may be related to a range of depression-related functioning (e.g., somatosensory, affective, and cognitive). Future research, therefore, may profitably assess the relations of these important symptom domains with CST diffusion properties in MDD. Moreover, given the varied projection profile of the CST, future research should assess relations between abnormalities in CST FA and gray matter properties (e.g., volume) in this disorder.

Although controversial, findings of abnormal fronto-striatal networks in MDD have led to the formulation that this disorder is a “disconnection syndrome” characterized by reduced connectivity between cortical and subcortical brain regions [4, 5, 36]. Evidence for this formulation includes observations that frontal white matter FA is reduced in MDD [37] and is correlated with remission from depression [38]. The current findings provide evidence that MDD may be characterized by abnormalities in connectivity between subcortical and brainstem structures and cortical gray matter regions. In future studies, investigators might use the AFQ and MDP procedures to examine the viability of the disconnection syndrome formulation more systematically, given that these procedures yield increased spatial specificity in neuroanatomical abnormalities associated with MDD.

The uncinate fasciculus and thalamic radiation are the most commonly studied fiber tracts in mood disorders [5]; indeed, we had hypothesized that we would find abnormalities of the uncinate fasciculus in MDD. The interest in these white matter tracts is due primarily to their potential involvement in abnormal cognitive control over emotion processing. Specifically, the uncinate fasciculus includes connections between medial temporal lobe regions associated with emotion processing (e.g., the hippocampi and amygdala) and the frontal cortex (involved in cognitive control); similarly, the white matter of the thalamic radiation links the frontal cortex with the thalamus (potentially a key connection in the disconnection syndrome formulation). Notably, we did not find abnormalities in these two tracts. This may be due in part to the location of the default AFQ waypoint ROIs, which are placed to identify the fiber tract cores and, thus, limit the assessment of variability more proximal to cortex.

The current results indicate that AFQ and MDPs are complementary techniques for the quantification and characterization of white matter paths in psychiatric populations and represent an important step towards the automated and efficient characterization of psychopathology, as demonstrated here in MDD. Given the sensitivity and automated nature of these methods, they may prove useful in identifying and characterizing biomarkers that can facilitate efforts to prevent and treat psychiatric disorders.

Despite the strengths of these procedures, we should note three limitations of the present study. First, the sample size in the current study was relatively small; thus, it is possible that the analyses are underpowered to find reductions in FA that have been previously reported, or significant relations between CST FA and age of onset or severity of depression, or level of global functioning. Second, our sample of depressed participants was heterogeneous with respect to the presence of anxiety comorbidities and medication use. We did not find differences in CST FA between the comorbid and non-comorbid participants in the MDD sample, nor did the effects we report appear to be determined by psychotropic medication use in a small subset of our depressed sample. Thus, it seems that these factors did not confound our results. Third, as with all FA-related results, the biological basis of the observed abnormality is unclear, as many factors can influence this metric: the level of orientation dispersion, myelination, numbers of axons, membrane permeability, axonal packing density, geometric properties of the tract, partial volume effects, and influences from branching, merging, or crossing fibers [7].

Using tractography clustering methods, we identified abnormalities in major white matter paths in MDD, specifically in the CSTs and several related pathways, including bilateral posterior limb of the internal capsules, right superior corona radiata, and left external capsule. These are the first results to implicate abnormality of the CST and related pathways in MDD. These findings highlight important future research directions, including increasing our understanding of CST abnormalities in the context of depressive symptoms and in relation to RDoC domains and constructs. Finally, the current study demonstrates that tractography clustering techniques can be used to increase our understanding of white matter abnormalities in MDD.