Neural correlates of reduced depressive symptoms following cognitive training for chronic traumatic brain injury

Abstract Depression is the most frequent comorbid psychiatric condition among individuals with traumatic brain injury (TBI). Yet, little is known about changes in the brain associated with reduced depressive symptoms following rehabilitation for TBI. We identified whether cognitive training alleviates comorbid depressive symptoms in chronic TBI (>6 months post‐injury) as a secondary effect. Further, we elucidated neural correlates of alleviated depressive symptoms following cognitive training. A total of seventy‐nine individuals with chronic TBI (53 depressed and 26 non‐depressed individuals, measured using the Beck Depressive Inventory [BDI]), underwent either strategy‐ or information‐based cognitive training in a small group for 8 weeks. We measured psychological functioning scores, cortical thickness, and resting‐state functional connectivity (rsFC) for these individuals before training, immediately post‐training, and 3 months post‐training. After confirming that changes in BDI scores were independent of training group affiliation, we identified that the depressive‐symptoms group showed reductions in BDI scores over time relative to the non‐depressed TBI controls (p < .01). Within the depressive‐symptoms group, reduced BDI scores was associated with improvements in scores for post‐traumatic stress disorder, TBI symptom awareness, and functional status (p < .00625), increases in cortical thickness in four regions within the right prefrontal cortex (p vertex < .01, p cluster<.05), and decreases in rsFC with each of these four prefrontal regions (p vertex < .01, p cluster < .0125). Overall, these findings suggest that cognitive training can reduce depressive symptoms in TBI even when the training does not directly target psychiatric symptoms. Importantly, cortical thickness and brain connectivity may offer promising neuroimaging markers of training‐induced improvement in mental health status in TBI.

Psychological intervention (Clark & Beck, 2010) is one of the promising options to treat depression in individuals with TBI, though more research is needed to provide stronger evidence for its efficacy (Fann, Hart, & Schomer, 2009;Gertler, Tate, & Cameron, 2015).
Further, little is known about underlying brain-based changes associated with reduced depression severity following psychological interventions for TBI. As depression alters brain structure and function (Mayberg, 1997;Price & Drevets, 2012), elucidating the neural mechanisms by which interventions reduce depressive symptoms could allow us to better understand the effects of cognitive interventions for depression in TBI.
Brain morphometry and resting-state functional connectivity (rsFC) are of particular interest to elucidate the underlying neural plasticity associated with reduced depressive symptoms following interventions.
Previously, we reported the efficacy of a strategy-based reasoning intervention for chronic TBI, compared to an information-based intervention (Vas et al., 2016), and the underlying neuroplasticity following intervention (Han, Davis, Chapman, & Krawczyk, 2017). Both intervention programs in our previous study included social interactions in small group settings with regular meetings comprised of multiple individuals with TBI and social contact with instructors who fostered an empathetic environment for the participants along with cognitive activities.
In light of previous studies about the impacts of social interaction on depression (Ashman et al., 2014;Cruwys et al., 2013;Cruwys et al., 2014) and neural effects of cognitive intervention on depression (Beevers et al., 2015;DeRubeis, Siegle, & Hollon, 2008;Li et al., 2016), our intervention study design provided an opportunity to characterize the impact of reduced depressive symptom severity on other psychological functioning scores and the neural correlates of reduced depressive symptoms following cognitive intervention for TBI.
In the current study, we assessed whether small-group-based cognitive interventions reduce depressive symptom severity in individuals with TBI as a secondary effect. Further, we sought to identify the effects of reduced depressive symptoms severity on other psychological functions. Lastly, we utilized cortical thickness and rsFC to examine the neural plasticity associated with reduced depressive symptom severity following cognitive interventions. We hypothesized that smallgroup-based TBI cognitive interventions would reduce depressive symptoms severity in individuals who reported depressive symptoms at baseline relative to individuals with minimal-to-no depressive symptoms. We further hypothesized that such reductions in depressive symptoms following cognitive training would induce changes in cortical thickness and rsFC in individuals after TBI.

| M A TER I A LS A N D M ETH OD S
We conducted a two arm, double-blind (i.e., types of intervention were blinded to participants and data acquisition team) randomized control study with a 3-month follow-up phase (see (Krawczyk et al., 2013) for detailed methods). The presence of depressive symptoms was blinded to intervention instructors. We acquired participants' depressive symptom severity, psychological functioning scores, and MRI data prior to training (TP 1 ), after training (TP 2 ), and 3 months later (TP 3 ).

| Participants
We analyzed 79 individuals (age 20-65) at the chronic stage of TBI (>6 months post-injury) ranging from lower moderate disability to lower good recovery on the Extended Glasgow Outcome Scale (GOS-E) (Wilson, Pettigrew, & Teasdale, 1998). Fifty-seven out of the 79 participants completed MRI scans and passed the quality assurance (QA) procedures described below. In our previous studies, we demonstrated that a subset of the current sample sustained TBI-related abnormalities in neuropsychological performance, rsFC, and white matter integrity, relative to healthy individuals (Han et al., 2015;Han, Chapman, & Krawczyk, 2016). We recruited our sample from the Dallas-Ft. Worth community and conducted a phone-screen interview before inclusion.
The inclusion criteria for the participants specified (1) a GOS-E score between four and seven, consistent with the likely ability to successfully engage in training programs, (2) the presence of TBI identified by the TBI screening form (Corrigan & Bogner, 2007), (3) >6 months postinjury time at the beginning of training, (4) age between 19 and 65 years old for the duration of the study, (5) the ability to understand, read, and speak English, (6) no history of significant clinically-diagnosed neurological or psychiatric co-morbidities other than self-reported depressive symptoms, (7) no current use of any prescription medications that may have affected depressive symptoms or cognitive performance, and (8) not currently pregnant. The primary TBI mechanisms of the participants included blasts, blunt force trauma, falls, athletic impacts, vehicular accidents, or combinations thereof. Initial injury severity was retrospectively estimated utilizing the Ohio State University TBI identification method (OSU TBI-ID; Corrigan & Bogner, 2007). The rationale for utilizing the OSU TBI-ID is described in detail in our previous study (Han et al., 2017). Both civilians and military veterans were included (See Table 1 for demographics). We confirmed that no participants showed visible focal lesions, contusions, mass shifting, or extreme cortical thinning on structural MRI, therefore minimizing the potential effects of macro structural injuries on preprocessing for cortical surface reconstruction and rsFC analyses. All participants provided written informed consent, and this study was conducted in compliance with the Declaration of Helsinki. The study was approved by the Institutional Review Boards of the University of Texas at Dallas and University of Texas Southwestern Medical Center.

| Experimental design
All participants were randomly assigned into one of the two training groups: (1) a strategy-based reasoning training called Strategic Memory Advanced Reasoning Training (SMART; n 5 40) or (2) the informationbased training called Brain Health Workshop (BHW; n 5 39). Both training programs comprised of 12 sessions (1.5 hr per session) for 8 weeks with quizzes, homework assignments, and projects conducted in small group settings (4-5 participants) that involved social interactions. The SMART group focused on selective attention, abstract reasoning and other thinking strategies (Vas, Chapman, Cook, Elliott, & Keebler, 2011), and the BHW group focused on education regarding brain anatomy, function, the effects of TBI, neural plasticity, and the effects of diet, sleep, stress, social activity, and exercise on the brain performance (Binder, Turner, O'connor, & Levine, 2008). More specifically, the SMART group was trained to (1) manage information by blocking distractions and irrelevant information by avoiding multitasking, (2) increase the ability to understand general ideas and extract take-home messages from information, and (3) interpret information from divergent perspectives. This set of strategies was aimed at

| Assessment of depressive symptoms
Depressive symptom severity was quantified using the Beck Depression Inventory-II (BDI-II; Beck, Steer, & Brown, 1996). According to BDI scoring guidelines for identifying the presence of depressive symptoms (Beck et al., 1996), we subdivided TBI participants into two groups using baseline total BDI scores: those with TBI plus mild-to-severe depressive symptoms (n 5 53; BDI 5 14-63) and those with TBI only/minimal depressive symptoms (n 5 26; BDI 5 0-13). The cutoff score for the presence of depressive symptoms was based on the suggested total BDI score for a diagnosis of depression from the BDI-II manual (Beck et al., 1996).
Since the two BDI-based groups participated in one of the two training programs, the participants were divided into four groups: (1) TBI-plus-depressive symptoms who participated in the SMART, (2) TBI-plus-depressive symptoms who participated in the BHW, (3) TBIonly who participated in the SMART, and (4) TBI-only who participated in the BHW (see Table 2). We first set up a full model including interaction terms between training types and the presence of depressive symptoms, and performed linear mixed effects model (LME; Bernal-Rusiel, Greve, Reuter, Fischl, & Sabuncu, 2013) analysis of total BDI scores using a piece-wise linear model with a break-point at TP 2 and a randomly varying intercept. We utilized the LME model because it is more appropriate for longitudinal data with increased sensitivity, specificity and reliability than other alternatives such as repeated-measures ANOVA (Bernal-Rusiel et al., 2013).
Using the full model, we confirmed that there were no statistically significant interactions between training types and the presence of depressive symptoms for temporal changes in total BDI scores. This was presumably due to the reduced sample size for each of the cells (see Table 2). To obtain a parsimonious statistical model, we thus performed the LME analysis of a simplified model with group membership based on either training types only or the presence of depressive symptoms only. This procedure was aimed to determine whether traininginduced changes in total BDI scores were specific to exclusively training types or the presence of depressive symptoms.

| Psychological functioning
We selected a subset of psychological functioning scores that were affected by depressive symptom severity in individuals with TBI in our previous study (Han et al., 2015) and additional survey-based functional outcome measures that are known to be related to depressive symptoms (Goverover & Chiaravalloti, 2014 & Keane, 1993), and the Satisfaction With Life Scale (Diener, Emmons, Larsen, & Griffin, 1985). The additional survey-based functional outcome measures included TBI Awareness Questionnaire (Sherer, 2004), and Functional Status Examination (FSE; Dikmen, Machamer, Miller, Doctor, & Temkin, 2001). We administered the survey-based measures using the corresponding forms on the same day that we assessed the participants' psychological functioning. We also acquired full scale intelligent quotient-2 (FSIQ-2) from the Wechsler Abbreviated Scale of Intelligence for estimated current IQ (Wechsler, 1999), and FSIQ from the Wechsler Test of Adult Reading for estimated premorbid IQ (Wechsler, 2001). 20 (11,9) 14 (4, 10) 20.6 6 1.5 a Only MRI scans that passed the quality assurance procedures were reported.
Abbreviations: TP 1 5 within one month prior to training; TP 2 5 immediately after training completed; TP 3 5 3 months after training completed.

| MRI data acquisition
Participants underwent structural MRI scans in a Philips Achieva 3T scanner. In each imaging session, T1-weighted sagittal magnetization prepared rapid acquisition gradient echo (MPRAGE) images were acquired using a standard 32-channel head coil (repetition time (TR)/ echo time (TE) 5 8.1/3.7 ms; flip angle (FA) 5 128; field of view (FOV) 5 25.6 3 25.6 cm; matrix 5 256 3 256; 160 slices, 1.0 mm thick). In this imaging session, one or two 416-s runs of rsfMRI scans were also acquired with the same head coil using a T Ã 2 -weighted image sequence (TR/TE 5 2,000/30 ms; FA 5 808; FOV 5 22.0 3 22.0 cm; matrix 5 64 3 64; 37 slices, 4.0 mm thick). During rsfMRI acquisition, the participants were asked to remain still with their eyes closed. Early stage observations revealed that the QA procedures with only one rsfMRI run yielded high rates of participant exclusion. Thus, we acquired two rsfMRI runs for the remainder of data collection. Refer to the rsfMRI data analysis section for our strategy to account for differences in total number of rsfMRI scans across participants.

| Cortical thickness analysis 2.6.1 | Cortical surface reconstruction and cortical thickness measurement
We reconstructed the cortical surface from each MRI scan with FreeSurfer v.5.3.0 (http://surfer.nmr.mgh.harvard.edu/). The cortical surface reconstruction procedures from T1-weighted MPRAGE images included nonuniform intensity correction, volumetric transformation to Talairach space, intensity normalization, skull stripping, segmentation of white and gray matter structures, tessellation of the gray/white matter boundary, smoothing of the gray/white matter boundary, surface inflation, topology correction, spherical mapping, and spherical registration. The gray/white matter boundary and pial surfaces were placed at the locations where the greatest intensity gradient occurred. More technical details of cortical surface reconstruction procedures have been described elsewhere Fischl et al., 1999;Fischl, 2012). We obtained cortical thickness as the closest distance between the estimated gray/white matter boundary and pial surfaces at each vertex across the cerebral cortex (Fischl & Dale, 2000).

| Longitudinal analysis of cortical thickness associated with BDI
To obtain a reliable longitudinal analysis of cortical thickness, we utilized the longitudinal processing stream (Reuter, Schmansky, Rosas, & Fischl, 2012) in FreeSurfer. This included skull stripping, Talairach transforms, atlas registration, and spherical surface maps and parcellations initialized from an unbiased within-subject template (Reuter, Rosas, & Fischl, 2010). For the group analysis, we resampled cortical thickness for each of the scans on a standard template, followed by surface smoothing with a 10 mm full-width-at-half-maximum (FWHM) Gaussian kernel.
We performed the LME analysis on the preprocessed longitudinal data using a piece-wise linear model with a break-point at TP 2 , randomly varying intercept, age covariate, and between-and withinsubject BDI covariates for each BDI group. Specifically, the cortical thickness of subject i at time point j, y ij , can be written as: where t ij is the time of measurement for subject i at time point j,t is an average time of measurement at TP 2 , S i is an indicator function for the SMART group for subject i, A i is the age for subject i, B i is the average BDI score over time for the subject i, B ij is the BDI score for subject i at time point j, D i is an indicator function for the TBI-plus-depressive symptoms group for subject i, N i is an indicator function for the TBI-only group for subject i, b i is a subject-specific intercept (cortical thickness of subject i at TP 1 ), e ij is the measurement error for subject i at time point j, and H(Á) represents the Heaviside step function. In this LME model, b 1 ; . . . ; b 6 account for training-specific temporal changes in cortical thickness, which are covariates of no interest in this study. We included BDI covariates according to the BDI-based group membership as temporal changes in total BDI scores were only depend on the presence of depressive symptoms at TP 1 (see result section and Figure 1).
We performed subsequent statistical inferences for the withinsubject BDI covariates for the BDI groups (i.e., b 9 and b 11 ). Each of the statistic maps were then thresholded at p vertex < .01 and p cluster < .05 to identify cortical regions that showed statistically significant associations between temporal changes in cortical thickness and temporal changes in total BDI scores in each of the BDI groups.
2.7 | rsfMRI data analysis 2.7.1 | Volumetric rsfMRI preprocessing Volumetric rsfMRI data were preprocessed in a subject-native space in AFNI (Version AFNI_17.3.02; Cox, 1996). For each rsfMRI run, the initial four frames were discarded, followed by despiking, slice timing correction, motion correction, spatial resampling (4 mm isotropic), normalization to whole-brain mode of 1,000, band-pass filtering (0.009 < f < 0.08 Hz), and linear regression. In the linear regression, we detrended the rsfMRI time series (third order) and included nuisance variables for the six rigid body motion profiles. Additional nuisance variables included cerebrospinal fluid (CSF) and white matter signals that were obtained from averaging over the lateral ventricles and deep cerebral white matter, respectively. We also included the first temporal derivatives of the aforementioned parameters (i.e., the motion profiles, and the average CSF and white matter signals Note that we performed spatial registration to a template and spatial smoothing after volume-to-surface mapping for the fMRI data. Refer to the next section for details.

| Surface mapping
Using FreeSurfer, we projected the preprocessed volumetric functional time series and subject masks onto mesh surfaces of each subject by averaging the time series across the voxels along the line between two matching nodes of the white and pial surfaces, followed by resampling onto the same standard template that was used for the cortical thickness analysis and surface smoothing at 10 mm FWHM. We also obtained an intersection mask for the mesh nodes in which rsfMRI signals existed across all subjects and time points.

| Longitudinal analysis of seed-based connectivity associated with BDI
Our previous study demonstrated that brain regions where traininginduced cortical thickness changes occurred also impacted corresponding resting-state functional connectivity (Han et al., 2017). Thus, we selected four seed regions where statistically significant associations between temporal changes in cortical thickness and total BDI scores occurred, and we assessed their seed-based connectivity. These four seed regions were the right ventrolateral prefrontal cortex, anterior prefrontal cortex, and dorsal prefrontal cortex within Brodmann Area 9 and Brodmann Area 8B (see the results section). We obtained the seed regions from the standard template that was used for the cortical thickness analysis. With these seeds we obtained seed-based connectivity for each of the scans, followed by a Fisher's Z-transform and scaling to Z-scores.
We performed the LME analysis with the same model used for the cortical thickness analysis, excluding the age covariate (after confirming no significant effects of age on rsFC). In this LME analysis, we assessed associations of training-related changes in rsFC to changes in total BDI-II scores. We also assessed relationships between rsFC and each of the Buckley three-factor scores (Buckley, Parker, & Heggie, 2001) because of reported sensitivity of rsFC to the subtypes of depressive symptoms (Han et al., 2015). The Buckley three-factor model decomposes the total BDI-II scores into cognitive, affective, and somatic symptoms (Buckley et al., 2001). The cognitive factor includes items regarding sadness, pessimism, past failure, guilty feelings, punishment feelings, self-dislike, selfcriticalness, suicidal ideation, and worthlessness. The affective factor includes items probing loss of pleasure, crying, loss of interest, and indecisiveness. The somatic factor includes the agitation, loss of energy, sleep disturbance, irritability, appetite disturbance, concentration difficulty, fatigue, and loss of sexual interest test items. The Buckley factor model provides a better characterization of depressive symptom severity of psychiatric patients over alternative models (Vanheule, Desmet, Groenvynck, Rosseel, & Fontaine, 2008) and is the best model for veterans with polytrauma (Palmer et al., 2014). Statistically significant associations were identified at p vertex < .01 and p cluster < .0125(5.05/4) to correct for multiple comparisons across vertices and the number of seeds.

| Quality assurance
Only participants who passed the following quality assurance procedures were included in analysis: less than 5 missed training sessions, no significant brain atrophy on structural MRI scans, a minimum of 4 min of rsfMRI volumes after motion scrubbing (van Dijk et al., 2010), and acquisition timing of MRI scans or psychological functioning scores within the two-standard-deviation band from the mean at each time point. Further, we excluded MRI scans from the LME analysis if corresponding BDI scores were not available. See Table 2 for the number of MRI scans that survived QA procedures.

| Statistical analyses
All statistical analyses were conducted in MATLAB R2013a. First, we performed the Shapiro-Wilk normality test on demographic data (age, years of education, current IQ, premorbid IQ, and post-injury time) within each group at a 5 0.05. Age, years of education, premorbid IQ, and post-injury time did not pass the Shapiro-Wilk normality test. Thus, the Mann-Whitney U test was used to compare these demographics between the groups. We performed t tests for group comparisons of current IQ. The Fisher's exact test was used to compare the groups according to gender distributions, proportion of civilians and veterans, and proportion of SMART and BHW participants. We used the likelihood ratio chi-square test to compare the distribution of estimated initial injury severity and primary cause of injury between the groups.
Similar to the analyses of cortical thickness and rsFC, we performed the LME analysis on psychological functioning scores. In these analyses, we additionally included a years-of-education covariate for age-adjusted scores from the color-word and verbal fluency tests. Both age and education covariates were included for immediate and delay recall scores.
For scores from the PCL-S, TBI awareness questionnaire, FSE, and satisfaction with life scale, we did not include age and education covariates, and found no statistically significant effects for these measures.
We also obtained effect sizes for demographics, and correlations between reduced depressive symptoms and psychological functioning, cortical thickness and resting-state functional connectivity, respectively.

| Control analyses 2.10.1 | Motion analysis
To identify whether there were systematic differences in participants' head motion during rsfMRI scans, we performed LME analyses on (1) FD after motion censoring and trimming, and (2) on the percentage of motion censored volumes across scans.

| The assessment of the effects of injury characteristics
To assess the effects of initial injury severity on our findings, we repeated the LME analyses with an additional covariate: worst-injury from the OSU TBI-ID (2$5). To identify potential effects of diverse post-injury times on our findings, we also performed the LME analyses with the covariate of post-injury time. Lastly, we divided the participants into mild and moderate/severe TBI groups and assessed whether there were group differences in reductions in BDI-II scores over time.

| Assessments of civilian versus veteran participants
We identified whether there were differences in initial TBI severity between civilians and veterans by performing a Fisher's exact test comparing the proportion of mild versus moderate/severe TBI among civilians and veterans. We also assessed the temporal changes in total BDI scores of veterans versus civilians. There was not an adequate sample size to allow for comparisons of the BDI covariates for civilians compared to veterans on psychological functioning scores, cortical thickness, or functional connectivity (e.g., N 5 4 with rsfMRI scans for veterans within the TBI-only group at TP 2 [refer to Table S1]).

| The assessment of cortical thickness versus resting-state functional connectivity
We obtained correlations between age-adjusted, average cortical thickness of the four seed regions and corresponding seed-based restingstate functional connectivity at each time point. We then identified whether the spatial patterns of these correlations differed across time points and groups.

| Demographics
All participants were in the chronic phase of TBI ($9 years post-injury on average). All demographics were matched at a 5 0.05 (Table 1).  (Figure 1a). These baseline BDI-II scores were comparable to previous studies in psychological intervention for depression in TBI (Ashman et al., 2014;B edard et al., 2014;D'Antonio et al., 2013). Baseline BDI-II scores for the TBI-only group (7.0) was higher than those of healthy individuals (3.6) from a previous study (Han et al., 2015). The TBI-plus-depressive symptoms group showed a monotonic decrease in total BDI scores over time, relative to the TBI only group (p < .01). No statistically significant change in BDI scores was identified in the TBIonly group (Figure 1a). Average BDI-II scores for the TBI-plusdepressive symptoms group at the follow-up stage was approximately 16, slightly greater than the threshold for depressive symptom (14).
Reductions in total BDI scores were independent of training types (Figure 1b). Within the TBI-plus-depressive symptoms group, 16/43 participants (37%) right after training and 24/42 participants (57%) at the follow-up reported minimal depressive symptoms (BDI total <14; Figure   1c,d). Subsequent Fisher's exact test on the proportion of SMART and BHW participants within the TBI-plus-depressive symptoms group revealed no statistically significant training-specific effects at any time point at a 5 0.05.

| Improved psychological functioning versus reduced depressive symptoms severity
The average times between the assessment of psychological functioning were 9 (TP 2 ) and 18 weeks (TP 3 ) from baseline (Table 2). Within the TBI-plus-depressive symptoms group, statistically significant (p < .00625(5.05/8)) associations of reduced BDI total scores over time co-occurred with changes in scores from the PCL-S, TBI awareness, and FSE (Table 3, Figure 2). Within the TBI-only group, there were no statistically significant associations with any of the assessed psychological functioning scores at a 5 0.00625 (Table 3, Figure 2).

| Changes in cortical thickness versus reduced depressive symptoms severity
The QA procedures resulted in cortical thickness analysis for 143 structural MRI scans from 57 participants (TBI-plus-depressive symptoms group n 5 36; Tables 2). The average times between MRI scans that passed QA procedures were 9 (TP 2 ) and 21 weeks (TP 3 ) from baseline ( Table 2).
The whole-brain LME analysis exhibited spatial patterns in regions where temporal changes in cortical thickness were associated with reduced BDI total scores ( Figure 3). Overall, only the TBI-plus-depressive symptoms group showed statistically significant (p vertex < .01, See Table 4 for coordinates and surface areas of these regions. The directionality of the associations over all four regions were negative, indicating that reduced total BDI scores following training were linearly correlated with increases in cortical thickness.
BDI total scores and seed-based connectivity occurred over multiple regions only within the TBI-plus-depressive symptoms group (Figure 4). Positive associations indicated that reductions in BDI total scores following trainings were linearly correlated with decreases in functional connectivity.
Among the three Buckley factors, only changes in the cognitive factor scores in the TBI-plus-depressive symptoms group were associated with seed-based connectivity for the R VLPFC, R APFC, R DPFC1, and R DPFC2 in multiple regions (Figure 4). These regions were inclusive of regions associated with BDI total scores. Further, the patterns of associations with the cognitive factor were more prominent than those with the BDI total scores. This suggests that reductions in symptoms related to the Buckley cognitive factor may be the primary driver of decreases in functional connectivity following cognitive training in chronic TBI.

| The effect sizes for neural correlates with reduced depressive symptoms severity
When BDI-II total scores for the TBI-plus-depressive symptoms were reduced by 10, cortical thickness was increased by 0.04-0.09 mm (Figure 5), and z-scores for functional connectivity were decreased by 0.5-1.5 (Figure 6 left). When Buckley cognitive factor scores for the TBIplus-depressive symptoms were reduced by 4, Z-scores for functional connectivity were decreased by 0.5-1.5 (Figure 6 right).

| Motion analysis results
There were no statistically significant group differences in trajectories of the percentage of motion censored volumes or FD after censoring and trimming at a 5 0.05 (Supporting Information Table S2).

| The effects of injury characteristics
The inclusion of covariates for initial injury severity or post-injury time did not alter the observed patterns of (1) BDI changes over time (Supporting Information Figure S1) or (2) associations between changes in BDI scores and improvement in psychological functioning scores (Table   S3). The uncorrected (p vertex < .01) maps for correlates of changes in cortical thickness and BDI scores were not affected by inclusion of covariates for initial injury severity or post-injury time (Supporting Information Figure S2). Compared to uncorrected (p vertex < .01) maps for the LME analysis on rsFC (Supporting Information Figure S3), the inclusion of covariates for initial injury severity (Supporting Information Figure S4) or post-injury (Supporting Information Figure S5) did not alter the main findings. When we divided participants into mild and moderate/severe TBI groups, the mild TBI group showed higher baseline BDI-II scores (p 5 .02), which resulted in marginal (p 5 .06) group differences in reductions in BDI-II scores over time (Supporting Information Figure S6a). This is consistent with other investigations that have found higher levels of depression in mild TBI compared to moderate and severe TBI (Homaifar et al., 2009;Rowland, Lam, & Leahy, 2005).

| Civilians versus veterans
The civilian participant group was comprised of 33 mild TBI and 20 moderate/severe TBI cases. The veteran participant group was comprised of 20 mild TBI and 6 moderate/severe TBI cases. There were no statistically significant differences in the proportion of mild and moderate/severe TBI between civilians and veterans at a 5 .05. The veteran participants had higher baseline depressive symptoms severity than the civilian participants (p < .01). However, both civilians and veterans showed reductions in scores for depressive symptoms severity, yielding no statistically significant group-by-time interactions at a 5 .05 (Supporting Information Figure S6b).

| Correlations between cortical thickness and resting-state functional connectivity
There were correlations between cortical thickness and resting-state connectivity (p vertex < .01, p cluster < .05). The spatial patterns of these   [Evans et al., 1993].

| D I SCUSSION
We demonstrated reduced depressive symptom severity following cognitive training to address chronic TBI symptoms, its impact on improvement in psychological functioning, and its correlation to changes in cortical thickness and resting-state functional connectivity. To our knowledge, this is the first study to report neural plasticity associated with reduced depressive symptom severity following cognitive intervention for TBI. Strengths of our study include our study design (instructors were blinded to the presence of depressive symptoms in participants, and participants were unaware of the type of intervention they were in) and the methodology that allowed us to identify the effects of time-varying covariates (i.e., BDI scores) on other timevarying measures (e.g., rsFC).

| Reduction in depressive symptom severity
The TBI-plus-depressive symptoms group showed reduced BDI-II scores over time relative to the controls (Figure 1a), consistent with previous studies in the remediation of depression following training in individuals after TBI (Ashman et al., 2014;D'Antonio et al., 2013; (Fann et al., 2009;Gertler et al., 2015). We utilized an active control group (i.e., the TBI-only group) who underwent the same intervention programs and assessed group-by-time interaction effects-the gold standard for training-induced changes. This feature of the design helps to address the test-retest reliability of BDI-II scores. Note that the TBI-only group unlikely experienced 'floor effects' because of (1) higher baseline BDI-II scores than healthy non-TBI individuals in a previous study (7.0 vs. 3.6) (Han et al., 2015) and (2) slight monotonic reductions in BDI-II scores over time (7.0 at TP 1 vs. 5.9 at TP 3 ).
Reductions in depressive symptoms severity following cognitive intervention were independent of the employed intervention program (Figure 1b,d). This phenomenon is consistent with other studies that administered multiple psychological intervention programs for comorbid depression in TBI (Ashman et al., 2014;D'Antonio et al., 2013;Ponsford et al., 2016). The strong inclusion of social interactions in our training programs may explain the reductions in depressive symptoms in TBI, as previous studies reported that the effects of supportive psychotherapy promoting participants' social interactions with instructors and cognitive behavioral therapy were comparable (Ashman et al., 2014;D'Antonio et al., 2013).

| Improvement in psychological functioning
Within the TBI-plus-depressive symptoms group, reductions in depressive symptom severity following cognitive intervention were associated with improvements in PTSD-related symptoms severity, the level of self-perceived disability relative to the pre-injury state (i.e., TBI awareness score), and functional status (Table 3). This may not be surprising, as several studies in TBI reported that BDI scores are associated with PTSD scores from the PCL (Esterman et al., 2013;Han et al., 2016), TBI awareness (

| Changes in cortical thickness
Reduced depressive symptom severity after cognitive training for TBI was related to increased cortical thickness over time only within the TBI-plus-depressive symptoms group (Figure 3) (Peterson et al., 2009;Schmaal et al., 2017) and depression plus blunt TBI (Mollica et al., 2009). These previous studies support the observed increases in cortical thickness following intervention in current study.
Increases in cortical thickness occurred in the prefrontal cortex, which is associated with controlling emotional responses reflected in limbic activity (Ochsner & Gross, 2005

| Changes in functional connectivity
Within the TBI-plus-depressive symptoms group, training-induced reductions in depressive symptoms severity was associated with reduced rsFC over time ( Figure 4). As with previous reports describing changes in rsFC following pharmacological treatments (Li et al., 2013), electroconvulsive therapy , and cognitive behavioral therapy (Chattopadhyay et al., 2017), our findings demonstrated the potential utility of rsFC for identifying neuroimaging markers associated with reduced depressive symptoms following interventions. Previously, we reported elevated amgydala connectivity in TBI individuals with depressive symptoms, relative to healthy individuals (Han et al., 2015). Thus, the positive association between rsFC and depressive symptom severity indicates that cognitive training restored elevated rsFC by TBI and depression to healthier status.
Among the three Buckley BDI factors, only changes in the cognitive factor for the TBI-plus-depressive symptoms group were associated with training-induced neural plasticity of rsFC ( Figure 4). This finding exhibited the specificity of rsFC with regard to subtypes of depressive symptoms in TBI. Further, there were more prominent patterns of neural correlates with the cognitive aspect of depressive symptoms (i.e., the Buckley cognitive factor) than with total scores (Figure 4). This suggests that reductions in the Buckley cognitive factor score led to neural correlates of reduced overall depressive symptoms severity after cognitive training.
The Buckley cognitive factor includes items related to sadness, pessimism, past failure, guilt, punishment feelings, self-dislike, self-criticalness,

| Explaining reduced depressive symptom severity
First, social interactions during small-group-based interventions might have reduced depressive symptom severity. Previous studies reported positive effects of increased social activity and group-based cognitive behavioral therapy on depressive symptoms (Ashman et al., 2014;Cruwys et al., 2013;Cruwys et al., 2014). Further, older participants enrolled in a social service program aimed at improving memory and executive function through social engagement demonstrated increases in brain activity in the prefrontal cortex relative to controls (Carlson et al., 2009). The observed neural correlates of reduced depressive symptoms in prefrontal regions (Figures 3 and 4) may be relevant in this regard. The frontopolar region is involved in social judgment and behavior (Cicerone, Levin, Malec, Stuss, & Whyte, 2006), and thinner cortex in this region leads to problematic social behavior of children with TBI (Levan et al., 2016). The VLPFC is associated with social exclusion (Eisenberger, Lieberman, & Williams, 2003) and has been linked to social problem-solving capability (Barbey et al., 2014).
Second, improved emotion regulation after cognitive training might lead to reduced depressive symptom severity. Cognitive behavioral therapy is thought to stabilize brain activity between the prefrontal Third, improved cognitive functioning following training may have contributed to reductions in related depressive symptoms. TBI-induced cognitive impairment is associated with depressive symptoms (Rapoport et al., 2005), and cognitive training for depression improves cognitive functions (Motter et al., 2016;Tiersky et al., 2005). Our TBI participants with depressive symptoms showed improvement in cognitive and dailylife functioning, regardless of training group, and this improvement was associated with reduced depressive symptoms ( Table 3).
Note that all these factors remain speculative without direct evidence. Thus, other factors might contribute to reduced depressive symptom severity in this report.

| Limitations and future directions
None of the participants sought medical attention for depression prior to the current study, although the TBI-plus-depressive symptoms group reported mild-to-severe depressive symptoms according to the BDI manual (Beck et al., 1996). Whereas previous studies demonstrated that the BDI-II is a reliable and valid measure and group-by-time analyses in current study support the test-retest reliability of it, the BDI-II is a self-report measure of depression. This could lead to bias in the severity level reported in this population (Malec et al., 2007).
Similar to several studies investigating the treatment for depression after TBI (Fann et al., 2009), addressing depression was not the primary purpose of our cognitive intervention programs. We prospectively included depressive symptoms severity as one of the outcome measures in our clinical trial, but the primary purpose of the two interventions was to improve cognitive functioning by learning strategy-based reasoning skills or new information about the brain. Thus, we interpreted reduced depressive symptom severity after intervention as secondary effects.
Unlike our previous report (Vas et al., 2016), we did not find training-specific effects on depressive symptoms. Apparent discrepancy may be attributable to different sample inclusion criteria, heterogeneity, and sample sizes. Additionally, we did not assess neural correlates of reduced depressive symptoms in subcortical regions. Based on reported alterations in subcortical regions in depression (Mayberg, 1997), future directions include the assessment of subcortical regions in our participants.

| CON CL U S I ON
We demonstrated reduced overall depressive symptom severity following cognitive interventions for chronic TBI and its correlation to psychological functioning score dynamics, cortical thickness, and rsFC.
Intervention-induced reductions in rsFC were associated with the Buckley cognitive factor (related to self-perception) of the BDI only among the three Buckley factors. Our findings suggest that cortical thickness and rsFC may be promising biomarkers sensitive to evaluating reductions in depressive symptom severity following cognitive intervention for chronic TBI.

ACKNOWLEDGMENTS
First of all, we would like to thank all study participants and their families that supported this study. Our thanks also go out to Mr. Jameson