Evidence suggests that treatment-resistant depression (TRD) is associated with more prominent and widespread brain alterations in areas related to mood and cognition compared with non-treatment-resistant depression (nTRD). However, direct comparisons of brain functioning between TRD and nTRD are scarce.
MethodsWe applied graph theory-based resting-state functional magnetic resonance imaging analysis to compare the degree centrality (DC) of brain functional connectivity among 31 patients with TRD, 28 patients with nTRD, and 30 healthy controls. Cognitive function was assessed using working memory and go/no-go tasks.
ResultsCompared with controls, patients with TRD exhibited reduced DC in the left cuneus, right frontal operculum cortex, cerebellum vermis I, II, and IX, and left cerebellum lobule X. The DC in the right cerebellum lobule III was lower in patients with TRD compared with those with nTRD. Among patients with TRD, we discovered positive associations between mean reaction time on the go/no-go task and DC in the left cuneus (r = 0.44, p = 0.015) and the right frontal operculum cortex (r = 0.41, p = 0.025). Conversely, the mean reaction time on the working memory task was inversely correlated with DC in the left cerebellar lobule X (r = −0.43, p = 0.019).
ConclusionOur findings highlight the important roles of the cerebellum (specifically, lobule X, and the right lobule III), frontal operculum, and cuneus in TRD. Dysfunction in these brain regions, which are integral to the salience and default mode networks, is likely associated with TRD-related cognitive dysfunction.
The sequenced treatment alternatives to relieve depression (STAR*D) study indicated that approximately 40 % of patients with major depressive disorder did not experience symptomatic remission after undergoing at least two antidepressant trials. Researchers often categorize these patients as having treatment-resistant depression (TRD).1,2 Subsequent analysis revealed that, after four trials with various antidepressant treatments, including combination therapy and augmentation therapy, 33 % of these patients continued to exhibit significant symptoms of depression.3 TRD is known to be associated with several negative outcomes, including functional impairment, adverse quality of life, suicide ideation and attempts, self-injurious behavior, and a high rate of relapse.1,2 Despite these findings, the underlying neuro-pathomechanisms differentiating TRD from non-TRD (nTRD) remain unclear.
Limited studies have directly compared brain functioning between individuals with TRD and those with nTRD. However, research focusing on each condition separately suggests that patients with TRD exhibit more prominent and widespread brain dysfunction in areas related to mood and cognition.4-7 For example, a study examined 28 patients with TRD, 32 patients with nTRD, and 48 healthy controls by using resting-state functional connectivity magnetic resonance imaging (MRI). The study discovered that those with TRD primarily exhibited widespread disrupted functional connectivity in bilateral prefrontal and thalamus areas.4 Conversely, those with nTRD demonstrated reductions in functional connectivity, particularly in the anterior cingulate cortex (ACC), bilateral amygdala, hippocampus, and insula.4 Sun et al. demonstrated that compared with the nTRD group, the TRD group exhibited more pronounced functional dysconnectivity in various brain networks, including affective (i.e., between subgenual ACC and middle frontal cortex), cingulo-opercular salience (i.e., between anterior insula and cerebellum crus I), and cognitive control (i.e., between dorsolateral prefrontal cortex and middle occipital cortex) networks.7
Given the involvement of large-scale brain network alterations in TRD and nTRD,4-7 graph theory is a powerful method for analyzing brain connectivity. This approach conceptualizes the brain as a graph composed of nodes and edges, representing functional connectivity among the nodes.8-10 Yang et al. applied graph theory to examine the antidepressant effect of serotonin-norepinephrine reuptake inhibitors on the whole-brain functional connectome in major depressive disorder, identifying the node of the thalamus as a key mediator in the treatment response.10 Similarly, Jia et al. applied graph theory-based functional connectivity analysis to discover that patients with major depressive disorder characterized by prominent rumination symptoms, a potential clinical marker of TRD, exhibited reduced degree centrality (DC) in the superior frontal cortex.11 DC measures a node's connectivity within the network, offering valuable insights into its influence and integration.12 Research indicates that a decrease in DC is associated with a decrease in network integration.12
In the present study, we employed graph theory-based resting-state functional connectivity MRI analysis to compare brain functioning between patients with TRD and those with nTRD. We hypothesized that patients with TRD exhibit lower DC in depression-related regions, such as the prefrontal cortex and insula, compared with their nTRD counterparts and healthy individuals.
Materials and methodsParticipantsIn the present study, 31 adult patients with TRD, 28 adult patients with nTRD, and 30 age-/sex-matched healthy controls were included for the neuroimaging analysis. All participants in the TRD and nTRD groups met the criteria for major depressive episodes during the baseline neuroimaging assessment. The baseline neuroimaging assessment was performed prior to the start of the treatment in both the TRD and nTRD groups. Baseline neuroimaging data were analyzed and compared among groups. Patients with TRD were selected from our previous clinical trial of a single low-dose ketamine infusion treating TRD.13,14 In our clinical trial, TRD was defined as major depressive disorder with ≥ 3 failures of antidepressant treatments with adequate doses and treatment durations.13,14 Patients with TRD were randomly assigned to three infusion groups: 0.05 mg/kg vs. 0.02 mg/kg ketamine vs. normal saline.13,14 Age-and sex-matched patients with nTRD and healthy controls were randomly selected from our neuroimaging cohort of 160 patients with major depressive disorder and 160 healthy controls.15 Patients with nTRD were those who responded to an antidepressant treatment with adequate doses and treatment durations.15 Depressive symptoms were measured by the Montgomery-Asberg Depression Rating Scale (MADRS).16 Healthy individuals had no Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition, diagnoses. Exclusion criteria included major medical or neurological diseases or a history of alcohol or substance use disorders in current study. This study was conducted in accordance with the Declaration of Helsinki and was approved by the Institutional Review Board of Taipei Veterans General Hospital. All participants gave their written informed consent.
Measurement of neurocognitive functionsThe present study examined working memory and inhibitory control function through 2-back working memory and go/no-go tasks. In the 2-back working memory task, participants were asked to respond as quickly as possible when they saw a number that appeared again only separated by one other number (i.e., 13–35–13; participants responded to the second 13 as quickly as possible).17,18 In the go/no-go task, participants were asked to respond as quickly as possible after the × symbol appeared. They were not to press the key when the + symbol appeared. After they completed the pretest with all correct responses, the formal test was then administered to record their correct responses, errors, and reaction times (mean). The 2-back working memory and go/no-go tasks were commonly used in our previous studies.19-21
Image acquisitionAll MRI images were acquired using a 3-Tesla scanner (GE Healthcare Life Sciences, Little Chalfont, UK) with a quadrature head coil at the Taipei Veterans General Hospital. Anatomical whole-brain T1-weighted images were acquired using magnetization-prepared rapid acquisition gradient-echo three-dimensional T1-weighted sequence with the following parameters: TR = 12.2 ms, TE = 5.2 ms, 168 axial slices, flip angle = 12°, FOV = 256×256 mm, matrix size = 256×256, and slice thickness = 1 mm. Resting-state functional MR images were obtained using a T2*-weighted gradient-echo approach, echo-planar sequence with the following parameters: TR = 2500 ms, TE = 30 ms, 43 axial slices, flip angle = 90°, voxel size = 3.5 × 3.5 × 3.5 mm. Each subject underwent the acquisition of two hundred MRI volumes while maintaining closed eyes, a state of mental neutrality, and refraining from any movement or falling asleep. The total sequence duration for T1-weighted MRI was 359 s, while for resting-state fMRI it was 500 s.
Image data processingThe preprocessing of the functional and anatomical data was conducted through a comprehensive pipeline using CONN toolbox.22 The initial step involved realignment using the SPM12 (Statistical Parametric Mapping, https://www.fil.ion.ucl.ac.uk/spm/) realign & unwarp procedure. All scans were coregistered to a reference image (the first scan of the first session) using a least squares approach and a 6 parameter (rigid body) transformation. Subsequently, b-spline interpolation was employed for resampling to correct for motion. Temporal misalignment between different slices of the functional data, which were acquired in an interleaved bottom-up order, was rectified using the SPM slice-timing correction procedure. This involved sinc temporal interpolation to resample each slice BOLD timeseries to a common mid-acquisition time. Outlier scans were identified using Artifact detection tools (ART)23 as acquisitions with framewise displacement above 0.9 mm or global BOLD signal changes above 5 standard deviations.24 Participants were excluded from analysis if the movement exceeded framewise displacement thresholds, or has >10 invalid volumes. A reference BOLD image was computed for each subject by averaging all scans excluding outlier images. The functional and anatomical data were then normalized into standard MNI space, segmented into grey matter, white matter, and CSF tissue classes, and resampled to 2 mm isotropic voxels. This was achieved using a direct normalization procedure25 that employed the SPM unified segmentation and normalization algorithm with the default IXI-549 tissue probability map template.26,27 Finally, the functional data were smoothed using spatial convolution with a Gaussian kernel of 6 mm full width half maximum (FWHM). This comprehensive preprocessing pipeline ensured that the data were optimally prepared for subsequent analysis. Moreover, a standard denoising pipeline28 was applied to the functional data. This involved regressing out potential confounding effects represented by motion parameters and their first-order derivatives,29 outlier scans,24 session effects and their first-order derivatives, and linear trends within each functional run. Subsequently, bandpass frequency filtering of the BOLD timeseries was conducted, limiting the frequency range between 0.009 Hz and 0.08 Hz.
Measurement of resting network characteristicsROI-to-ROI connectivity (RRC) matrices were computed to characterize the functional connectivity between each pair of regions among the 132 Harvard-Oxford atlas ROIs.30 The functional connectivity strength was expressed through Fisher-transformed bivariate correlation coefficients obtained from a general linear model. This model was separately estimated for each pair of ROIs, delineating the association between their BOLD signal timeseries. To address potential transient magnetization effects at the start of each run, individual scans were weighted using a step function convolved with an SPM canonical hemodynamic response function and rectified. The RRC matrices of each subjects then further processed through GRETNA toolbox31 for the topological properties of resting network. The degree of an individual node (indicating the importance of the node in the network) represents the number of edges connected to other nodes in this network.12 Density was defined as the mean degree of the entire network. The resting state connectivity were evaluated by computing the DC. Degree centrality of node i, DCi, was defined as the summation of all neighboring edge weights, as follows: Degree centrality (dCi) = Σj∈N,j≠iaij, where aij is the connection status between node i and node j, and N is the set of all nodes in the network.
Statistical analysisStatistical analyses were conducted using SPSS 25 (IBM Corp., Armonk, NY; Version 25) and MATLAB (Mathworks, Natick, MA; version 9). In assessing demographic and clinical characteristics, we employed one-way ANOVA to compare continuous variables among the three groups (healthy controls, nTRD, and TRD). For comparisons specifically between nTRD and TRD groups, two-sample t-tests were used for continuous variables, while chi-square tests were applied to nominal variables. The distinctions in network metrics, adjusted for age and sex among the three groups, were assessed using one-way ANOVA (p < 0.005, uncorrected), followed by post hoc two-sample t-tests (p < 0.005, uncorrected). Results were visualized using BrainNet Viewer.32 The Pearson correlation coefficient assessed the association between psychological measurements and DC, with the significance level set at α = 0.05.
ResultsTable 1 presents the demographic and clinical characteristics of our analysis sample, revealing no difference in age at disease onset (p = 0.852) between patients with TRD and those with nTRD. The TRD group exhibited higher total MADRS scores (p = 0.001) than did the non-TRD group (Table 1). Patients in the TRD and nTRD groups performed worse (correct response and mean response time, all p < 0.05) on the working memory task than the control subjects, without a difference between the two depression groups (Table 1). Patients with TRD performed worst on the mean reaction time (p = 0.002) of the go/no-go task compared with those with nTRD and the controls (Table 1).
Demographic and clinical characteristics between groups.
TRD: treatment-resistant depression; SD: standard deviation; MADRS: Montgomery-Asberg Depression Rating Scale.
Patients with TRD showed lower DC in the right frontal operculum cortex (p = 0.003), left cuneus (p = 0.002), left cerebellum lobule X (p = 0.001), cerebellum vermis I and II (p = 0.005), and cerebellum vermis IX (p = 0.004) compared with the controls (Table 2, Fig. 1). Furthermore, the DC in the right cerebellum lobule III was significantly lower (p = 0.003) in patients with TRD than in those with nTRD (Table 2, Fig. 1). Finally, we found positive associations of mean reaction time on the go/no-go task with the DC in the left cuneus (r = 0.44, p = 0.015) and the right frontal operculum cortex (r = 0.41, p = 0.025) among patients with TRD (Fig. 2A and B). However, the reaction time on the working memory task was negatively related to the DC in the left cerebellum lobule X (r = −0.43, p = 0.019) among patients with TRD (Fig. 2C).
Comparison of degree centrality between groups.
Values are presented as mean (SD, standard deviation). TRD, treatment-resistant depression; L, left; R, right. (Tukey's post-hoc two-sample t-test; * p < 0.005 uncorrected, TRD < healthy control; # p < 0.005, TRD < non-TRD; two-sample t-test).
Correlation between cognitive function and degree of centrality among patients with TRD. (A) degree centrality of the left cuneus with mean time on go/no-go task; (B) degree centrality of the right frontal operculum cortex with mean time on go/go-no task; (C) degree centrality of the left cerebellum X with mean time on working memory task.
In our study, the application of graph theory-based resting-state functional connectivity MRI analysis revealed a significant reduction in DC in the right cerebellum lobule III in patients with TRD compared with those with nTRD. Additionally, patients with TRD exhibited lower DC in several brain regions, including the right frontal operculum cortex, left cuneus, left cerebellum lobule X, and cerebellum vermis I, II, and IX, compared with controls. This study also found associations between cognitive and behavioral functions and specific brain regions in patients with TRD: worse working memory was correlated with lower DC in the left cerebellum lobule X, whereas worse impulse control was related to higher DC in the left cuneus and the right frontal operculum cortex.
Our findings unexpectedly highlighted widespread cerebellum dysfunction in patients with TRD, specifically reduced DC in the right cerebellum lobules III and X and in the cerebellum vermis I, II, and IX. The cerebellum is composed of 10 lobules, each with an unpaired medial (i.e., vermis) portion and a bilateral hemispheric portion, and these lobules play a critical role in the corresponding brain networks.33,34 For example, cerebellum lobules VII and IX are important to the default mode networks; cerebellum lobule VI is associated with the salience and affective networks; and cerebellum lobule VII plays an important role in the cognitive control network.33,34 Rabellino et al. indicated the important role of the cerebellum lobule X, also known as the vestibulocerebellum, in body-self integration and bodily self-consciousness.35 Studies utilizing the amplitude of low-frequency fluctuations (ALFF) approach have indicated differences in brain circuits between patients with TRD and those with nTRD, predominately in the cerebellum and networks related to visual recognition and default mode.36 Guo et al. observed decreased regional homogeneity, particularly in the left insula, inferior frontal cortex, and anterior lobe of the cerebellum (including cerebellum lobules I to V) among patients with TRD.37 Sun et al. identified decreased ALFF in the bilateral posterior lobes of the cerebellum (including cerebellum lobules VI to IX) among patients with TRD compared with those with nTRD.38 These findings suggest that patients with major depressive disorder, especially those with TRD, may experience difficulty in disassociating their emotional processing from cognitive and sensorimotor functions, potentially due to impaired functioning of the vestibulocerebellum.39,40
As mentioned, the dysfunction of the salience (i.e., insula) and visual recognition (i.e., cuneus) networks may be associated with the pathomechanisms of TRD. This hypothesis aligns with our study findings, which demonstrated that patients with TRD exhibited the lowest DC in the right frontal operculum cortex and left cuneus compared with those with nTRD and healthy controls.7,36 Goldin et al. assessed the neural base of cognitive reappraisal and expressive suppression in emotional processing; the results demonstrated that reappraisal produced significant responses in the medial prefrontal cortex, insula, and cuneus during the early stages of emotional processing, whereas suppression generated significant responses in the dorsolateral prefrontal cortex, dorsal ACC, and middle and superior occipital cortex during later stages.41 Yang et al. highlighted the important role of the cuneus in self-regulating aversive emptions, including perception, inhibition, and modulation.42 Guo et al. also reported lower resting state activity in the cuneus in patients with TRD compared with those with nTRD.36 Increasing evidence indicates the critical role of the frontal operculum cortex in emotional regulation.43,44 Wager et al. revealed that the frontal operculum cortex with the nucleus accumbens mediates emotional processing, which contributes to successful cognitive reappraisal.43 He et al. identified an association between cingulo-opercular network dysfunction and suicide severity among patients with major depressive disorder.45 A clinical trial investigating an 8-week treatment with selective serotonin reuptake inhibitors in patients with major depressive disorder demonstrated that reduced functional connectivity between the amygdala and central opercular cortex was associated with a poor treatment response.46
Finally, we found that among patients with TRD, those who exhibited poorer performance on the working memory task, as indicated by longer mean reaction times, had lower DC in the left cerebellum lobule X. Conversely, poorer performance on the go/no-go task, again marked by longer mean reaction times, was associated with greater DC in the left cuneus and right frontal operculum cortex. These inconsistent correlations may reflect the neuropsychological results comparing patients with TRD to those with nTRD. Both groups exhibited poor working memory, suggesting a potential association between reduced DC in the left cerebellum lobule X and impaired working memory in patients with major depressive disorder, especially those with TRD. However, only patients with TRD exhibited the poorest performance on the go/no-go task compared with patients with nTRD and the control group, implying a compensatory relationship between impulse control and DC in the left cuneus and right frontal operculum cortex.
This study has several limitations that merit consideration. First, we did not discontinue the medications used by patients during MRI and cognitive examinations. Given the ethical considerations, particularly for patients with TRD, continuing medication was deemed more appropriate to prevent exacerbation of the disease and relapse. However, future studies may require a drug-free study design to confirm our findings. Second, the study's scope was intentionally limited to working memory and go/no-go tasks to take into consideration the severe depressive symptoms and profound cognitive impairment in patients with TRD. More research may be required to comprehensively assess cognitive function among patients with TRD and those with nTRD. Third, while we have chosen uncorrected thresholds to maintain sensitivity in detecting potentially significant correlations, we recognized that this approach, along with the lack of multiple comparison corrections, may inflate Type I error rates. Therefore, careful interpretation of the results was necessary, and our exploratory findings would warrant further investigation to validate the identified patterns.
In conclusion, our study found that patients with TRD were more likely to exhibit lower DC in the right frontal operculum cortex, left cuneus, left cerebellum lobule X, and cerebellum vermis I, II, and IX compared with the control group. In addition, we observed that patients with TRD exhibited significantly decreased DC in the cerebellum lobule III compared with patients with nTRD. The DC in the cerebellum lobule X, cuneus, and right frontal operculum cortex was associated with impairments in working memory and impulse control among patients with TRD. Our findings underscore the important role of the cerebellum, salience (i.e., frontal operculum), and visual recognition (i.e., cuneus) network-related brain regions in the pathomechanism of TRD. Further studies may be required to elucidate whether direct modulation (i.e., repetitive transcranial magnetic stimulation) to the cerebellum, salience, and visual recognition network-related brain regions may be beneficial to TRD.
Ethical considerationsThis study was conducted in accordance with the Declaration of Helsinki and was approved by the Institutional Review Board of Taipei Veterans General Hospital. All participants gave their written informed consent.
Funding sourceThe study was supported by grant from Taipei Veterans General Hospital (V113C-039, V113C-011, V113C-010, V114C-089, V114C-064, V114C-217), Yen Tjing Ling Medical Foundation (CI-113-32, CI-113-30, CI-114-35), Ministry of Science and Technology, Taiwan (MOST110-2314-B-075-026, MOST110-2314-B-075-024 -MY3, MOST 109-2314-B-010-050-MY3, MOST111-2314-B-075 -014 -MY2, MOST 111-2314-B-075-013, NSTC113-2314-B-075-042), Taipei, Taichung, Kaohsiung Veterans General Hospital, Tri-Service General Hospital, Academia Sinica Joint Research Program (VTA112-V1-6-1, VTA114-V1-4-1) and Veterans General Hospitals and University System of Taiwan Joint Research Program (VGHUST112-G1-8-1, VGHUST114-G1-9-1), Cheng Hsin General Hospital (CY11402-1, CY11402-2). The funding source had no role in any process of our study.
ContributionsDr MHC and Prof LFC designed the study; Drs LKC and PCT analyzed the neuroimaging data; Drs MHC and LKC drafted the manuscript; Drs TPS, CTL, WCL, YMB, and SJT enrolled the candidate patients and performed the literature reviews; all authors reviewed the final manuscript and agreed for the publication.
None of the authors in this study had any conflict of interest to declare.
We thank Mr I-Fan Hu for his support and friendship.
