White matter architecture in major depression with anxious distress symptoms

BACKGROUND
Comorbid anxious distress is common in Major Depressive Disorder (MDD), and associated with significantly worse clinical course and treatment response. While DSM-5 recently introduced the Anxious Distress (AD) specifier as a potentially useful symptom-based subtyping scheme for MDD, its neurobiological underpinnings remain unclear. The current study hence uniquely probed whether MDD with co-occurring AD (MDD/AD+) relates to distinct perturbations in frontolimbic white matter (WM) pathways tentatively theorized in MDD/AD+ pathophysiology.


METHODS
Tract-based spatial statistics (TBSS) was therefore used to analyze diffusion tensor imaging data on WM microstructure, in MDD/AD+ patients (N = 20) relative to MDD patients without AD (MDD/AD-; N = 29) and healthy controls (HC; N = 39). Using TBSS, we probed fractional anisotropy and axial/radial/mean diffusivity as proxies for WM integrity. Categorical (between-groups) and dimensional (within-patients) analyses subsequently assessed how Anxious Distress in MDD impacts frontolimbic WM connectivity. Receiver-Operating Characteristics additionally assessed classification capabilities of between-groups WM effects.


RESULTS
Compared to MDD/AD- and HC participants, MDD/AD+ patients exhibited diminished integrity within the anterior thalamic radiation (ATR). Higher AD specifier scores within MDD patients additionally related to diminished integrity of the uncinate fasciculus and cingulum pathways. These effects were not confounded by key clinical (e.g., comorbid anxiety disorder) and sociodemographic (e.g., age/sex) factors, with altered ATR integrity moreover successfully classifying MDD/AD+ patients from MDD/AD- and HC participants (90% sensitivity | 73% specificity | 77% accuracy).


CONCLUSIONS
These findings collectively link MDD/AD+ to distinct WM anomalies in frontolimbic tracts important to adaptive emotional functioning, and may as such provide relevant, yet preliminary, clues on MDD/AD+ pathophysiology.


Introduction
Major Depressive Disorder (MDD) is one of the most prevalent and debilitating psychiatric disorders, affecting hundreds of millions of people worldwide (WHO, 2017). Despite these concerns, a significant portion of MDD patients do not fully benefit from currently available treatment options and are severely impaired by MDD symptomatology (Lener and Iosifescu, 2015). Many scholars and clinicians point to the unitary diagnosis of MDD, based on heterogeneous symptom profiles, as one of the main reasons for our inability to fully understand MDD pathophysiology and formulate targeted, personalized treatments. To address this issue and encourage subtype-specific research and treatment initiatives, the newly revised DSM-5 included several specifiers for the diagnosis of MDD, aimed at meaningful subgrouping of MDD patients based on etiology, symptomatology, and treatment responsivity (APA, 2013;Gaspersz et al., 2018;Ionescu et al., 2013).
The UNC is a long-range WM tract that connects key limbic regions such as the amygdalar and hippocampal complexes in the anterior temporal lobe to ventral frontal lobe territories, including ventromedial/orbitofrontal sections of the PFC and ventral ACC regions (Olson et al., 2015). The location and connectivity profile of the UNC pinpoint it as an integral component of adaptive frontolimbic interactions, and abnormalities in its structural integrity are often linked to abnormal emotional processing and susceptibility to affective psychopathology (Bracht et al., 2015;Cullen et al., 2010;LeWinn et al., 2014;Olson et al., 2015;Sexton et al., 2009;Von Der Heide et al., 2013). The CIN bundle, on the other hand, allows for more widely-distributed interconnections by linking prefrontal, cingulate, parietal and medial temporal sites, whilst also interlinking key subcortical nuclei (e.g., thalamic nuclei) to various cingulate regions (Bubb et al., 2018;Wu et al., 2016). It is involved in a wide range of processes, such as emotion processing, motivational drive, executive functioning, pain and memory (Bubb et al., 2018). It is noteworthy that studies indicative of UNC microstructural alterations in MDD also implicate involvement of the CIN in MDD pathophysiology (Bhatia et al., 2018;Bracht et al., 2015;Cullen et al., 2010), further corroborating the crucial role of both tracts in mood dysregulation (Bhatia et al., 2018;Keedwell et al., 2012;Keedwell and Linden, 2013;Sexton et al., 2009). However, whether specific alterations in UNC and CIN microstructure may also relate to MDD/AD+ diagnosis, symptomatology, and severity has yet to be investigated.
It is worth mentioning that altered WM microstructure is also found in pathways other than UNC and CIN among MDD patients, with some alterations even occurring early in the course of MDD (Aghajani et al., 2014;Zhu et al., 2011). Commonly affected pathways include the corpus callosum, superior/inferior longitudinal fasciculus, forceps, internal capsule, and anterior thalamic radiation, with generally lower integrity of WM bundles being reported (Bracht et al., 2015;Dillon et al., 2018;Jiang et al., 2017;Liao et al., 2013;Shen et al., 2017;Wen et al., 2014). It should be noted though that WM findings in MDD are not always unequivocal, with several important studies reporting opposing effects or even null findings (e.g., see meta-analysis Choi et al., 2014). These mixed findings might, among other things, be ascribed to the heterogeneous nature of MDD, with various subgroups of patients presenting different etiology, symptomatology, and treatment responsivity profiles (Belmaker and Agam, 2008). A subtype-specific approach towards MDD is therefore anticipated to tackle these conflicting findings, improve our understanding of underlying neurobiology, and aid personalized treatment. That said, little is yet known of whether DSM-5 informed MDD subtypes, and in particular MDD/AD+, may involve specific WM anomalies. A handful of studies have tried to address this issue, though fairly indirectly by either using a AD specifier not conforming to DSM-5 criteria or comparing MDD patients with vs. without comorbid anxiety on WM integrity (e.g., Delaparte et al., 2017;Xia et al., 2018). However, MDD with comorbid anxiety disorder tends to capture a partially different group of patients than MDD/AD+ (Gaspersz et al., 2017a;Rosellini et al., 2018), as evidenced by the poor overlap between the presence of the AD specifier and comorbid anxiety disorders among MDD patients (Gaspersz et al., 2017a). The DSM-5 informed AD specifier is additionally deemed more useful than comorbid anxiety diagnoses in formulating meaningful MDD subtypes, partly due to its higher discriminative power on key clinical outcomes (Gaspersz et al., 2017a;Gaspersz et al., 2018;Rosellini et al., 2018). Examining WM microstructure in MDD/AD+ patients thus seems rather pertinent, as it may provide deeper insights into the underlying neuropathology and aid biomarker development in the future.
We therefore employed tract-based spatial statistics (TBSS: Smith et al., 2006) to analyze diffusion tensor imaging (DTI) data on WM microstructure, in MDD/AD+ patients relative to MDD/AD-and healthy control (HC) participants. Our main parameter of interest was fractional anisotropy (FA), which reflects the tendency of water molecules in WM to diffuse in one direction as opposed to all other directions, thereby providing a general proxy for WM microstructural integrity (e.g., myelin thickness, membrane integrity) (Kochunov et al., 2007). FA, however, is less informative about the specific type of alterations that may present in WM fiber bundles, so additional parameters are commonly calculated and examined to gain more insight in underlying mechanisms (Alexander et al., 2007). We therefore additionally probed axial (A.Dif), radial (R.Dif), and mean (M.Dif) diffusivity, which provide complementary information on axonal integrity, fiber bundle coherence, and myelination, thereby aiding interpretation of FA changes (Alexander et al., 2007;Budde et al., 2009;Horsfield and Jones, 2002;Song et al., 2002) (see Fig. 1 for schematic representation). Specifically, A.Dif indexes water diffusion along the principal direction of the fiber, and is deemed sensitive to axonal integrity and fiber bundle coherence. R.Dif tends to index diffusion perpendicular to the principal direction of the fiber, and is more indicative of myelination level. Finally, M.Dif reflects average water diffusion in all directions within the fiber bundle, providing a proxy marker for cellularity.
Given the proposed frontolimbic circuit dysfunction in MDD (Disner  et al., 2011;Mayberg, 2003;Phillips et al., 2015;Price and Drevets, 2010) and the relevance of UNC and CIN bundles to frontolimbic connectivity (Cohen et al., 2008;Concha et al., 2005;Petrides and Pandya, 2002;Swartz and Monk, 2014), we focused primarily on these two WM tracts. For completeness though, we also performed wholebrain analyses to capture potential WM changes in other tracts, including those considered pertinent to adaptive frontolimbic communication and affective psychopathology (e.g., anterior thalamic radiation, forceps minor, and internal capsule) (Bracht et al., 2015;Dillon et al., 2018;Jiang et al., 2017;Liao et al., 2013;Shen et al., 2017;Wen et al., 2014). We employed categorical between-groups (based on the presence of the AD specifier) and dimensional within-patient analyses (based on the total AD specifier score) to thoroughly examine the impact of the AD specifier on WM architecture in MDD patients. We hypothesized diminished WM integrity as indexed by lower FA values among MDD/AD+ patients versus MDD/AD-and HC participants, with these group differences following a linear trend (MDD/AD+ < MDD/ AD-< HC), echoing the more severe symptomatology and neuropathology of MDD/AD+ (Gaspersz et al., 2017b;Gaspersz et al., 2017;Gaspersz et al., 2018;Kessler et al., 2015). We additionally speculated that higher AD specifier total scores in MDD patients would relate to lower FA values. Post-hoc analyses moreover probed the impact of key clinical and sociodemographic factors on potential WM changes, and additionally assessed whether the presence of the AD specifier vs. comorbid anxiety disorders relates to distinct/overlapping WM alterations. Receiver-Operating Characteristics were finally computed to assess diagnostic sensitivity of between-groups (categorical) WM effects in classifying MDD/AD+ patients from MDD/AD-and HC participants.

Participants
The current sample comprised a subset from the most recent wave (9-year follow-up) of the Netherlands Study of Depression and Anxiety (NESDA) (Penninx et al., 2008), and only included MDD and HC participants who had complete DTI datasets and sufficient data to formulate the AD specifier (N = 91). Visual inspection and quality control of DTI data resulted in exclusion of 3 participants due to: missing DTI scan in anterior-to-posterior direction (1), differing gradients in the anterior-to-posterior and posterior-to-anterior DTI scan (1), and missing volumes in the posterior-to-anterior DTI scan (1). The dataset of 88 subjects consisted of 49 patients with current MDD diagnosis (mean age = 48.98, SD = 8.57; 49% female) and 39 HC participants (mean age = 52.10, SD = 8.44; 39% female). The groups were matched on age, sex, and education. MDD patients were recruited through general practitioners, primary care, and specialized mental care institutions, while HC participants were approached via local advertisements. DSM diagnoses of current (6-month recency) MDD were established using the Composite International Diagnostic Interview lifetime version 2.1 (Andrews and Peters, 1998).
Patients were excluded if they presented Axis I disorders other than MDD or anxiety disorder (i.e., panic disorder, social anxiety disorder and/or generalized anxiety disorder), and used psychotropic medications other than selective serotonin reuptake inhibitors (SSRIs) on a stable basis or benzodiazepine on an infrequent basis (oxazepam/diazepam, maximum of three times a week, max 10 mg, and not within 48 h before scanning). Exclusion criteria for HC participants were lifetime diagnosis of MDD or other psychopathologies, and use of any psychotropic medication, with the exception of infrequent benzodiazepine use (< 1 days per week). All participants were asked not to use benzodiazepines or recreational drugs in the 48 h preceding the scan.
Exclusion criteria for the entire group were: the presence or history of major internal or neurological disorder, dependency or recent abuse (past year) of alcohol or drugs, hypertension, severe head trauma, insulin dependence, presence of MRI-contraindications, and not being fluent in Dutch language. Ethical Review Boards of the three participating centers (VU Medical Center = VUMC; Leiden University Medical Center = LUMC; University Medical Center Groningen = UMCG) approved this study, and written informed consent was obtained from all participants. Detailed information on sample characteristics is provided in Table 1.

Anxious distress specifier
As per prior work (Gaspersz et al., 2017a;Gaspersz et al., 2017b;Gaspersz et al., 2017;Gaspersz et al., 2018), the AD specifier for MDD patients was constructed based on 5 self-reported items from the Inventory of Depressive Symptomatology (IDS (Rush et al., 1996)) and the Beck Anxiety Inventory (BAI (Beck et al., 1988)). These items directly matched with the DSM-5 criteria for the AD specifier: (1) feeling keyed up or tense (IDS item 7); (2) feeling unusually restless (IDS item 24); (3) difficulty concentrating because of worry (IDS item 15); (4) fear that something awful might happen (BAI item 5); (5) feeling that the individual might lose control of self (BAI item 14). Both questionnaires assess symptoms in the past week on a 0-3 (not at all-to-severe) scale, and symptoms are considered present when a patient indicates a score ≥2 (i.e., moderate or severe). The presence of the specifier was assessed according to the criterion of the DSM-5, requiring that the MDD patient presents/fulfills at least 2 of 5 the symptoms. In addition to this dichotomous approach (i.e., specifier was present or absent), we also examined the total specifier severity score (sum score of 5 AD items: 0-15) for within-patients examination of WM. Earlier studies (Gaspersz et al., 2017a;Gaspersz et al., 2017b) have shown that the AD specifier has an adequate internal consistency (Crohnbach's α = 0.71) and predictive validity for subsequent course and treatment response in depressed patients (Gaspersz et al., 2018). Based on the criteria mentioned above, 20 of the 49 MDD patients met the criteria for the AD specifier (see Table 1 for characteristics).

DTI data acquisition and preprocessing
Participants were scanned at one of the three participating centers: VUMC (N = 41), LUMC (N = 29), and UMCG (N = 18). DTI data were collected using a Philips 3 T MRI scanner (Philips Medical Systems, The Netherlands) with a 32-channel phased array head coil. A single-shot echo-planar imaging (EPI) sequence was used with the following parameters: TR = 7351 ms, TE = 71 ms, flip angle = 90°, b factor = 1000 s/mm 2 , FOV = 240 × 240 × 150 mm, voxel dimensions = 1.88 × 2.35 × 2.00 mm, number of transverse slices = 75, and no slice gap. DTI data were acquired along 32 directions, together with a baseline image having no diffusion weighting (b = 0). Data was collected with reversed phase-encode directions, resulting in pairs of images with distortions going in opposite directions. The total scanning time was approximately 8 min: 4 min acquisition in anterior to posterior direction and 4 min posterior to anterior direction. Collected DTI data were preprocessed and analyzed using FSL version 5.0.10. (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FSL) (Smith et al., 2004).
After visual inspection of the data and quality control, FSL's Topup Tool (Andersson et al., 2003;Smith et al., 2004;Van Essen et al., 2012) was used to estimate susceptibility-induced off-resonance field maps based on the b0-images and acquisition parameters. These off-resonance field maps -together with brain-extracted mask images -were then fed into FSL's Eddy Tool  to correct for distortion and motion artefacts induced by eddy currents or by simple head motions. To correct for movement that coincided with the diffusion encoding part of the sequence (causing partial or complete signal dropout), FSL's Outlier Replacement Tool  was additionally used, which detects dropout-slices and replaces them with Gaussian Process prediction. Finally, to generate individual FA, A.Dif, R.Dif and M.Dif maps for each participant, the standard diffusion tensor model was fitted to each voxel using FSL's Diffusion Toolbox (Behrens et al., 2003). A.Dif was defined as the largest eigenvalue vector (λ1), R.Dif was calculated as the average of the two small eigenvalues ((λ2+λ3)/2), and M.Dif was calculated as the average of the three eigenvalues ((λ1+λ2+λ3)/3) (https://fsl.fmrib.ox.ac.uk/ fsl/fslwiki/TBSS). Whereas A.Dif is deemed sensitive to axonal integrity and fiber bundle coherence, R.Dif is more indicative of myelination level, with M.Dif providing a proxy marker for cellularity (Alexander et al., 2007)

TBSS analyses
Tract-based spatial statistics (TBSS) (Smith et al., 2006) version 1.2 was used for voxelwise analyses of the preprocessed DTI data. First, all individual FA images were aligned to the FMRIB58_FA standard-space image using FMRIB's non-linear registration tool (FNIRT). Next, a mean FA image was created and skeletonized, representing the centers of all major tracts common to the entire group. This skeleton was then thresholded at an FA value of 0.35 to exclude peripheral tracts and minimize partial voluming (Aghajani et al., 2014;Westlye et al., 2009).
Finally, each participant's aligned FA images were projected onto the mean FA skeleton. In a similar manner, A.Dif, R.Dif and M.Dif data were projected onto the skeleton using the FA registration and skeleton projection parameters. In line with prior work, FA was examined throughout the WM skeleton, while A.Dif, R.Dif and M.Dif were only examined in regions showing significant FA changes to aid interpretation of these changes.
To test for regional specific FA alterations in UNC and CIN, bilateral masks for these tracts were first created as per prior work (Aghajani et al., 2014;Westlye et al., 2011;Westlye et al., 2009), using the probabilistic Johns Hopkins University and Juelich Histological WM atlases (Eickhoff et al., 2007;Mori et al., 2005) provided by FSL. The UNC and CIN masks were then skeletonized against the thresholded mean FA skeleton to include only voxels comprised in the individual masks and the mean skeleton. This confines the statistical analyses to voxels from the center of the tract, thereby minimizing anatomic intersubject variability, registration errors and partial voluming (Aghajani et al., 2014;Westlye et al., 2009). The resulting UNC and CIN masks were subsequently used for voxel-wise permutation-based statistical analyses. For completeness, we also performed whole-brain voxel-wise analyses to capture potential WM changes in other tracts, including those considered pertinent to adaptive frontolimbic communication and affective psychopathology (e.g., anterior thalamic radiation, forceps minor, and internal capsule) (Bracht et al., 2015;Dillon et al., 2018;Jiang et al., 2017;Liao et al., 2013;Shen et al., 2017;Wen et al., 2014).
The statistical inferences included both a categorical -between-groups  Values are given as mean ± standard deviation or number. Categorical variables (sex, scan location, medication, and presence of comorbid anxiety disorder) were tested using Chi-square test, whereas continuous variables were tested with nonparametric Kruskal-Wallis ANOVA (IDS, BAI, and AD Specifier Total) or parametric ANOVA (age, education). a Not significant (P > 0.05). b Significant at P < 0.001. c Only tested within-patients, not applicable to HC participants.
-approach (based on the presence of the AD specifier) and a dimensional -within-patients -approach (based on the total AD specifier score) to thoroughly examine the impact of the AD specifier on WM architecture in MDD patients. The categorical between-groups analysis comprised MDD/AD+, MDD/AD-, HC participants, and probed whether MDD/AD + involves diminished FA values compared to MDD/AD-and HC, according to a linear trend (MDD/AD+ < MDD/AD-< HC). The dimensional within-patients analysis included only the MDD patients and probed whether higher AD specifier total scores (mean-centered) among individual patients would relate to lower FA values. Statistical analyses were performed using FSL's permutation-based Randomise tool, which included 5000 random permutations to build up the null distribution of the cluster size statistic, while testing our contrasts of interest in the categorical and dimensional analyses. Statistical maps were thresholded using threshold-free cluster enhancement (TFCE (Smith and Nichols, 2009)) with cluster normalization (Salimi-Khorshidi et al., 2011), and family-wise error (FWE) corrected for multiple comparisons at P < 0.05. Post-hoc analyses probed whether adjusting for antidepressant use, current comorbid anxiety disorders, presence of melancholic and atypical MDD symptoms (bases on IDS items: Gaspersz et al., 2017;Khan et al., 2006;Novick et al., 2005), sociodemographics (age, sex, education), and scan location might impact any of the emerging findings (all covariates mean-centered). Post-hoc analyses additionally assessed whether the presence of the AD specifier vs. current comorbid anxiety disorders may relate to distinct or overlapping WM correlates. Finally, to explore the sensitivity and specificity of between-groups (i.e., categorical) WM effects in classifying MDD/AD+ patients versus MDD/ADand HC participants, Receiver-Operating Characteristics (ROC) analyses were conducted.

Sample characteristics
As shown in Table 1, MDD/AD+ patients did not differ in age, sex, education and scan location from MDD/AD-and HC participants. Kruskal-Wallis ANOVA revealed a linear trend in which MDD/AD+ patients scored the highest on the IDS, BAI and AD specifier total, followed by MDD/AD-and HC, respectively. Of note, the three groups still differed on the AD specifier total scores when correcting for melancholic and atypical symptoms (univariate model: F(2,83) = 59.43, P < 0.001), further showcasing the robustness and specificity of the AD construct in the current study. Finally, MDD/AD+ and MDD/ADpatients did not differ in terms of antidepressant use and presence of current comorbid anxiety disorders (P's > 0.05).

Categorical approach
The categorical analyses focusing on UNC and CIN microstructure revealed no significant between-groups differences in FA within these tracts. Complementary whole-brain TBSS analysis, however, revealed a significant FA effect within the left anterior thalamic radiation (ATR), wherein MDD/AD+ patients exhibited the lowest FA values in comparison to MDD/AD-and HC participants, respectively (P = 0.01, TFCE and FWE corrected) (see Fig. 2). Closer examination revealed that the effect site was in the vicinity of the brainstem/substantia nigra region, in which the change in FA was mainly driven by higher A.Dif, R.Dif, and M.Dif levels among MDD/AD+ patients relative to MDD/AD-and HC participants, respectively (P's < 0.01, TFCE & FWE corrected). No other effects emerged from these between-group analyses.

Dimensional approach
The dimensional within-patient analysis, however, did reveal an association between MDD patients' AD specifier total scores and UNC and CIN microstructure. Specifically, higher AD scores related to lower FA values in the OFC section of the left UNC and ventral ACC segment of the right CIN (P's < 0.02, TFCE & FWE corrected) (see Fig. 3 &  Fig. 4). Examination of diffusivity measures revealed that these changes in FA were both mainly driven by decreased A.Dif, increased R.Dif, and preserved M.Dif (P's < 0.01, TFCE & FWE corrected). Patients' AD specifier scores were not associated with elevated FA levels, nor were any associations found between AD scores and FA in other WM tracts (i.e., complementary whole-brain analysis).

AD specifier vs. comorbid anxiety disorder
Post-hoc analyses additionally assessed whether the presence of the AD specifier vs. current comorbid anxiety disorders may relate to distinct or overlapping WM effects. To this end, the categorical TBSS analyses were rerun, but instead of using the AD specifier for MDD groupings, the presence of comorbid anxiety disorders was now used to form MDD groups. MDD patients with comorbidity (N = 21), without comorbidity (N = 28), and HC participants (N = 39) were subsequently compared on WM integrity, using exactly the same analytical approach as the primary categorical TBSS analyses. The analyses, however, revealed no significant effects within the UNC, CIN, ATR, or any other WM tracts (P's > 0.20, TFCE & FWE corrected). This suggests a weak overlap between WM correlates of the AD specifier and comorbid anxiety disorders among our MDD patients, cautiously hinting that the findings are specific to MDD/AD+.

Receiver-operating characteristics (ROC)
To explore the sensitivity and specificity of the between-groups ATR effect in classifying MDD/AD+ patients versus MDD/AD-and HC participants, ROC analyses (SPSS, Version 25, IBM Corp., Armonk, N.Y., USA) were conducted. Similar to prior work (Ellard et al., 2018), mean FA values from ATR regions of significant effect were entered as the test variable, with diagnostic group entered as the state variable, assuming a nonparametric distribution. Resulting cutoff scores representing the greatest balance between highest sensitivity and specificity (i.e., Youden Index) were used to calculate sample-based sensitivity/specificity proportions. Microstructural integrity within the ATR effects site was moderately successful in classifying MDD/AD+ patients versus MDD/AD-and HC participants (AUC = 0.83, SE = 0.05, P < 0.001, CI = 0.74-0.92), with 90% sensitivity, 73% specificity, and 77% accuracy (see Fig. 5). When looking at the two patient groups alone, ATR microstructural integrity was still fairly successful in discriminating MDD/AD+ from MDD/ AD-patients (AUC = 0.78, SE = 0.07, P = 0.001, CI = 0.65-0.91), with 90% sensitivity, 66% specificity, and 75% accuracy (see Fig. 5).

Discussion
The current study uniquely probed whether MDD/AD+, as informed by DSM-5, relates to distinct perturbations in WM architecture. Whereas categorical analyses revealed diminished ATR integrity in MDD/AD+ patients relative to MDD/AD-and HC participants, dimensional analyses within MDD patients linked elevated AD specifier scores to diminished integrity of the UNC and CIN tracts. These effects were not confounded by key clinical and sociodemographic factors, with altered ATR integrity moreover successfully classifying MDD/AD + patients from MDD/AD-and HC participants. These novel findings collectively link MDD/AD+ to distinct WM anomalies in frontolimbic tracts important to adaptive emotional functioning. The findings may thus provide relevant, yet preliminary, clues on MDD/AD+ pathophysiology by highlighting WM microstructural changes as a putative pathomechanism.
The ATR is a frontolimbic tract that links various thalamic subregions to lateral and medial PFC territories (Bracht et al., 2015;Cho et al., 2015), and is deemed relevant to the expression of emotions (Coenen et al., 2012;Deng et al., 2018;Spalletta et al., 2013). In fact, abnormalities in the ATR have been tentatively theorized to upset the dynamic equilibrium between positive and negative affective states, potentially rendering an individual vulnerable to mood dysregulation (Coenen et al., 2012). The ATR is furthermore argued to play an important role in physiological distress and arousal (Coenen et al., 2012;Kirouac, 2015), as it connects PFC and thalamic regions with central autonomic structures such as the brainstem/substantia nigra territory (Cho et al., 2015). Interestingly, this is also where we observed alterations in ATR microstructure among MDD/AD+ patients (Fig. 2). The brainstem/substantia nigra territory was also implicated in a prior DTI study on MDD with anxiety features (Blood et al., 2010), though this concerned more the grey matter segments of this territory rather than its ATR section, wherein the FA was intriguingly elevated. While differences in methodology (TBSS vs. Voxelwise DTI) and effect site (white vs. grey matter) may explain the opposing FA effects in this region, they collectively converge on the relevance of the brainstem/ substantia nigra territory to MDD with anxiety features. Taken as a whole, the ATR effect documented here might thus provide preliminary insights into a putative substrate for MDD/AD+, which ostensibly carries relevance for emotional homeostasis on the cognitive, behavioral, and physiological level. Our classification analyses seem to partly support this tentative notion by additionally showing the sensitivity and specificity of these ATR effects. The findings, however, warrant cautious interpretation, further investigation, and future replication, given their exploratory nature and our incomplete understanding of ATR contributions to (mal)adaptive emotional function.

Dimensional approach
Our within-patients dimensional analysis additionally revealed that higher AD scores related to lower FA values in the OFC section of the UNC and ventral ACC segment of the CIN. Examination of diffusivity measures suggested these FA changes to be largely driven by decreased A.Dif, increased R.Dif, and preserved M.Dif, thus putatively linking AD To explore the sensitivity and specificity of the between-groups ATR effect in classifying MDD/AD + patients versus MDD/AD-and HC participants, ROC analyses were conducted. Microstructural integrity (i.e., mean FA) within the ATR effects site (A & B) was moderately successful in classifying MDD/AD+ patients versus MDD/ AD-and HC participants (AUC = 0.83, SE = 0.05, P < 0.001, CI = 0.74-0.92), with 90% sensitivity, 73% specificity, and 77% accuracy (C, green ROC curve). When looking at the two patient groups alone, ATR microstructural integrity was still fairly successful in discriminating MDD/AD+ from MDD/AD-patients (AUC = 0.78, SE = 0.07, P = 0.001, CI = 0.65-0.91), with 90% sensitivity, 66% specificity, and 75% accuracy (C, blue ROC curve). Red squares represent the point of greatest balance between maximum sensitivity and specificity (i.e., Youden Index). Brain images in A & B provide a representative view of the WM skeleton (beige/grey) and the ATR tract (red), along with the ATR subregion (green) that exhibited significantly reduced FA in MDD/AD+ patients relative to MDD/AD-and HC participants, respectively. symptoms in MDD to demyelination (i.e., increased R.Dif) and axonal disintegrity (i.e., decreased A.Dif) within the UNC and CIN (Alexander et al., 2007). Various studies implicate abnormal UNC and CIN microstructure in the pathophysiology of depression and anxiety, with generally lower FA values being found within these tracts in depressed and/or anxious individuals (Bhatia et al., 2018;Bracht et al., 2015;Cullen et al., 2010), thus echoing the findings presented here. The UNC mainly connects the amygdalar/hippocampal complexes to ventral frontal territories (e.g., OFC) (Von Der Heide et al., 2013), while CIN bundles allow for more widely-distributed interconnections by linking prefrontal, cingulate, parietal and medial temporal territories (Bubb et al., 2018;Wu et al., 2016). Their combined connectivity profiles are therefore deemed crucial to adaptive frontolimbic communications that serve to support various aspects of human emotional behavior (Bhatia et al., 2018;Bracht et al., 2015;Cullen et al., 2010;Keedwell et al., 2012;Keedwell and Linden, 2013;Sexton et al., 2009).
Alterations in UNC and CIN microstructure documented here thus seem to reverberate the centrality of these WM tracts to the frontolimbic system dysfunction that is commonly hypothesized in affective psychopathologies, and in specific MDD/AD+ (Gaspersz et al., 2017b;Gaspersz et al., 2017;Gaspersz et al., 2018;Phillips et al., 2015). The location of WM alterations we find for the UNC and CIN (OFC and ventral ACC region, respectively), however, seems more suggestive of perturbations in the ventral subsegment of this frontolimbic system (Phillips et al., 2015;Phillips et al., 2008). This ventral frontolimbic subsystem comprises the amygdala, subgenual/ventral ACC, ventrolateral PFC, and OFC, and is among other things, heavily involved in encoding the emotional significance of a stimulus and producing an affective response to it (Phillips et al., 2015;Phillips et al., 2008;Stein et al., 2007). As the OFC and ventral ACC crucially link subcorticallygenerated information with higher-order cortical computations, and vice versa, (Etkin et al., 2011;Fatahi et al., 2018;Phillips et al., 2015;Zhu et al., 2011), our findings may point to aberrant frontolimbic transmission of emotional/affective information, which speculatively could impede how emotions are perceived, processed, and controlled (Gaspersz et al., 2018;Phillips et al., 2015;Phillips et al., 2008). The fact that the ventral frontal segments of the UNC and CIN are prime target sites for deep-brain stimulation in treatment-resistant depression seems to further highlight their importance to adaptive emotional function Howell et al., 2019;Lujan et al., 2012;Noecker et al., 2018;Riva-Posse et al., 2018;Riva-Posse et al., 2014). The ventral/subgenual ACC section of the CIN, however, seems of particular relevance in this context, for its stimulation tends to produce far more superior and enduring antidepressant effects than stimulating the "confluence point" of ventrofrontal UNC and CIN regions (Howell et al., 2019). While the data presented here highlight the importance of the UNC and CIN to (mal)adaptive emotional function, future studies are warranted for further exploration and validation, given the intricate modulation of human emotions by frontolimbic circuits.

Categorical versus dimensional approach
Due to the prevalence of comorbidity and multi-sourced etiologies, psychopathological research benefits from an integrative approach combining categorical and dimensional measures of psychiatric entities (Fuchs, 2010;Fuchs and Pallagrosi, 2018;Hudziak et al., 2007). Though the categorical approach at first sight may seem more appealing and clinically relevant, especially in the context of DSM, it may not fully pick up the inter-subject variability that is unmistakably present among MDD patients (Fuchs, 2010;Hudziak et al., 2007). Hence, many scholars suggest the complementary application of categorical and dimensional analyses in psychiatry (Fuchs and Pallagrosi, 2018), even when examining a specific subgroup/subtype based on a predefined specifier (Dillon et al., 2018;Hudziak et al., 2007). Consonant with this notion, the current study innovatively probed the impact of the DSM-5 Anxious Distress specifier for MDD on WM architecture both categorically and dimensionally. While both approaches interestingly revealed the same trend/effect as a function of Anxious Distress symptoms (i.e., reduced WM integrity), the location of WM changes differed between approaches (categorical = ATR tract; dimensional = UNC & CIN tracts). This could speculatively suggest that ATR microstructure might be more relevant for differentiating MDD/AD+ from other MDD subtypes and healthy controls, a notion partly supported by our classification analyses, whereas alterations in UNC and CIN seem more indicative of MDD/AD+ severity. Future research on MDD subtyping should further explore and validate these tentative interpretations, ideally employing both categorical and dimensional approaches and comparing different subtypes simultaneously, in search for commonalities and unique subtype characteristics.

Study limitations and strengths
The cross-sectional nature of this study does not allow for firm conclusions regarding causality. Hence, we cannot ascertain whether microstructural alterations reported in this study preceded or followed the onset of MDD/AD+. The sample size is moreover relatively small, though fairly similar to prior structural MRI work on DSM-informed MDD subtypes (Bracht et al., 2014;Dillon et al., 2018;Ota et al., 2015;Zhao et al., 2017). Longitudinal research in larger samples with a wide age range, and DTI studies of individuals at risk for affective disorders, could tackle these limitations. Although DTI is a useful tool for characterizing WM microstructure (Lim and Helpern, 2002), it only provides a proxy for actual WM integrity (Beaulieu, 2002;Jones et al., 2013;Soares et al., 2013). Caution is thus warranted in interpreting FA changes as actual changes or differences in WM "integrity". While our post-hoc analyses revealed no significant impact of key clinical and sociodemographic confounds on documented WM effects, cautious interpretation of the findings is nevertheless warranted, given the difficulties to fully tease out/eliminate confounders (especially in smaller samples). The same cautiousness also applies to our classification analysis/finding, which necessitates replication and further exploration in preferably larger studies to reaffirm its accuracy.
Notwithstanding these limitations, our findings do merit attention as the first evidence for fairly specific and discriminant WM anomalies in MDD patients that meet the DSM-5 informed Anxious Distress specifier, which is novel within the field of MDD subtyping. Previous DTI studies have tried to address this issue, though fairly indirectly by either using a AD specifier not conforming to DSM-5 criteria or comparing MDD patients with vs. without comorbid anxiety on WM integrity (e.g., Delaparte et al., 2017;Xia et al., 2018). However, MDD with a comorbid anxiety disorder captures a partially different group of patients than MDD/AD+ (Gaspersz et al., 2017a;Rosellini et al., 2018), as evidenced by the poor overlap between the presence of the AD specifier and comorbid anxiety disorders among MDD patients (Gaspersz et al., 2017a). The current study corroborates this premise, as MDD/AD+ and MDD/AD-patients did not differ in terms of comorbid anxiety diagnoses (see Table 1), while our post-hoc analyses found no overlap between WM correlates of the AD specifier and comorbid anxiety disorders in MDD patients. This may cautiously hint that the findings documented here are specific to MDD/AD+, and plausibly allow deeper insights into MDD/AD+ pathophysiology. The fact that melancholic and atypical MDD symptoms did not affect any of the WM effects documented here seems to further corroborate their specificity to AD symptoms. The conjoint categorical and dimensional examination of AD-specific effects on WM, matching of groups on sociodemographics, and correction for key confounds to the best of our abilities, tend to moreover increase the reliability of the findings.

Conclusions
The data presented here collectively link MDD/AD+ to distinct WM anomalies in frontolimbic tracts important to adaptive emotional functioning. The findings seem to provide relevant, yet preliminary, clues on MDD/AD+ pathophysiology by highlighting WM microstructural changes as a putative pathomechanism. For a deeper understanding of WM architecture in MDD/AD+, it would be important to examine whether WM alterations predict susceptibility, chronicity and treatment response in this specific class of MDD patients. Ideally, these future studies should be longitudinal and employ multimodal imaging, in which structural and functional connectivity of relevant frontolimbic tracts are simultaneously investigated.